More testing of the good old AI MCP skills and tools for CVNA.

For what it’s worth this quarter shouldn’t be too notable results wise. Expecting ~199k units which is right about in line of estimates. July so far is on track for ~40% YoY growth which is a slight uptick from some slightly slower Q2 growth in May/June.

The company continues to grow rapidly at great economics. Kudos to the team and happy to continue owning the shares.

CVNA — Q2 FY2026 Earnings Prep

Reporting Date: Wednesday, July 29, 2026 (After Market Close)
Conference Call: 5:30 p.m. ET

What to Expect

1. Record Retail Unit Sales (~200,000 Units): Driven by peak Q2 spring/summer buying seasonality, expanded Inspection & Reconditioning Center (IRC) throughput (1.5M unit annual built-out capacity), and accelerated market share gains (1.6% full-year 2025 baseline). Management guided to a sequential increase over Q1 FY2026’s record of 187,393 units (+40.0% YoY). The model projects 200,000 retail units (+39.6% YoY).

2. All-Time High Adjusted EBITDA & Operating Profit Expansion: CFO Mark Jenkins guided to a sequential increase in Adjusted EBITDA beyond Q1 FY2026’s record of $672M. With high structural conversion of Adjusted EBITDA to GAAP Operating Income (~90%), GAAP Operating Income is projected to reach $670M (+31.1% YoY from $511M in Q2 FY2025), supported by fixed cost leverage and sustained monetization in finance/ancillary services.

3. Retail GPU Sequential Recovery vs. Year-over-Year Headwinds: Management explicitly cautioned that Retail GPU will expand sequentially from Q1 FY2026 ($3,165 GAAP) but drop YoY compared to Q2 FY2025. This YoY margin compression reflects a $100–$200 per-unit headwind from narrower industry wholesale-to-retail price spreads, non-vehicle inflation, and the lapsing of prior-year tariff benefits (~$100/unit). Total Non-GAAP GPU is modeled at $7,025/unit.

4. SG&A Cost Discipline Following Q1 Step-Up: After SG&A overhead stepped up to $690M in Q1 FY2026, management characterized Q1 as establishing a stable “new baseline level” rather than the start of accelerated expense growth. The model projects Q2 FY2026 operating expenses at $710M (+28.4% YoY), driving operating leverage as total revenue scales +44.2% YoY to $6.98B.

5. Accelerated Balance Sheet Deleveraging: Net debt to trailing 12-month Adjusted EBITDA declined to 1.1x in Q1 FY2026 (down from 1.5x in Q3 FY2025). Peak seasonal cash generation and earnings expansion in Q2 are expected to reduce leverage to 1.0x, bringing Carvana closer to investment-grade credit metrics.

Guidance & Consensus

Note on Stock Split: Carvana executed a 5-for-1 forward stock split on May 7, 2026. Post-split consensus EPS estimates reflect this change (~$0.38–$0.43 per share), while pre-split historical figures are noted where appropriate.

Historical Trends

(Values in millions of USD, except per-share and per-unit data)

*Note: Q4 FY2025 Net Income included a $685M non-cash tax benefit.

Trajectory Highlights & Margin Inflections

• Volume Acceleration: Retail units sold expanded from 133,898 units in Q1 FY2025 to 187,393 units in Q1 FY2026 (+40% YoY), with market share expanding to 1.6%.

• Operating Leverage Trend: After SG&A peaked as a percentage of revenue during early restructuring, SG&A overhead declined to -0.93% YoY in FY2025. Despite a sequential bump in Q1 FY2026 to $690M (driven by stock-based compensation vesting, winter weather, and AI tech investments), operating margin is projected to re-expand to 9.6% in Q2 FY2026.

• Deleveraging Path: Net Debt / Trailing 12M Adjusted EBITDA improved from 1.5x in Q3 FY2025 to 1.1x in Q1 FY2026, driven by cumulative Adjusted EBITDA growth ($2.237B in FY2025).

Segment & KPI Outlook

1. Retail Vehicle Sales

• Recent Trajectory: Revenue grew from $2,980M in Q1 FY2025 to $4,828M in Q1 FY2026 (+62.0% YoY). Retail units sold hit consecutive record highs, reaching 187,393 in Q1 FY2026.

• Management Commentary: CEO Ernie Garcia highlighted that fastest-turnaround vehicle processing dropped below 5 days from purchase to retail listing. Annual built-out reconditioning capacity stands at 1.5M units across 34 capable locations.

• Q2 FY2026 Expectation: Retail unit volume is expected to reach 200,000 units (+39.6% YoY), generating ~$5.25B in retail vehicle revenue.

2. Wholesale Vehicle Sales

• Recent Trajectory: Revenue expanded from $863M in Q1 FY2025 to $1,078M in Q1 FY2026 (+24.9% YoY). Wholesale unit volume reached 83,574 units in Q1 FY2026 (+31.7% YoY), generating $1,328 in Wholesale GPU per unit.

• Management Commentary: Transition to the ADESA Clear digital auction platform and integration of ADESA physical sites (10 mega sites integrated in 2025; targeting 6–8 in 2026 at $30M–$35M per site) continue to boost wholesale marketplace efficiency.

• Q2 FY2026 Expectation: Wholesale volume will track higher trade-in volumes alongside record retail sales, with revenue projected at ~$1.15B.

3. Other Sales & Revenues (Finance & Ancillary)

• Recent Trajectory: Revenue increased from $389M in Q1 FY2025 to $526M in Q1 FY2026 (+35.2% YoY), yielding $2,807 in Other GPU per retail unit.

• Management Commentary: Carvana secured $12B in 2-year institutional loan purchase agreements ($4B new partner agreement through Dec 2027, plus $6B with Ally through Oct 2026). End-to-end automation tool “Sebastian” now enables >30% of retail buyers and >60% of vehicle sellers to complete transactions with zero human contact.

• Q2 FY2026 Expectation: Other GPU is expected to hold near ~$2,800/unit, driving total segment revenue to ~$560M on higher retail volume.

4. Operations & Reconditioning Cost Efficiency

• Recent Trajectory: Operations SG&A per retail unit dropped to $1,549 in Q2 FY2025 before experiencing a temporary spike in Q4 FY2025 due to rapid facility scaling.

• Management Commentary: In Q1 FY2026, management deployed new software algorithms, manager decision tools, and labor productivity tracking across IRC facilities, bringing April labor efficiency close to all-time company bests.

• Q2 FY2026 Expectation: Reconditioning efficiency gains will flow through to inventory, stabilizing total Non-GAAP GPU at $7,025/unit.

Recent Developments

1. Corporate Actions & Stock Split

• 5-for-1 Forward Stock Split: Carvana completed a 5-for-1 stock split on May 7, 2026. Post-split trading settled in the $67–$80 range (equivalent to $335–$400 pre-split).

2. Competitor Read-Throughs

• CarMax (KMX) Q1 FY2027 Earnings (June 17, 2026): CarMax reported a $230 YoY decline in retail GPU ($2,177/unit) due to price cuts aimed at preserving unit volume. CVNA stock dipped ~8%–10% in sympathy. However, analysts noted Carvana’s volume growth (+40% YoY vs flat/negative comp sales at CarMax) and vertical GPU integration provide structural insulation.

3. Infrastructure & Operational Expansion

• ADESA Hybrid Expansion: Added full Inspection & Reconditioning capabilities at ADESA Sarasota (June 2026) and ADESA Chicago (May 2026).

• Delivery & Digital Upgrades: Expanded Same-Day Delivery to Milwaukee (July 2026) and launched ADESA Timed, expanding digital B2B auction capabilities.

4. Insider Activity Context

• 10b5-1 Clustered Sales: Significant insider sales occurred on May 1 and June 1, 2026. CFO Mark Jenkins and COO Benjamin Huston exercised deeply in-the-money options ($2.01, $8.41, $10.39 strike prices post-split) and monetized shares under pre-arranged 10b5-1 plans (CFO sold ~$4.48M in June; COO sold ~$3.52M in June).

• Derivative Transaction by Major Shareholder: On May 1, 2026, 10% owner Ernest Garcia II (via ECG II SPE, LLC) sold 4,000,000 covered call options (2M at $450 strike, 2M at $500 strike pre-split, expiring May 21, 2027), generating $272.23M in cash premiums.

• Open Market Buys: Zero open-market purchases were recorded.

Risks & Catalysts

Upside Catalysts

• Reconditioning Labor Productivity Gains: Rapid absorption of new software tools across IRCs could lower reconditioning costs faster than anticipated.

  • Estimated Impact: +$50 to +$100 Retail GPU uplift (+$10M to +$20M Gross Profit).

• Finance Monetization & Loan Spreads: Sustained high institutional demand for auto loan originations across the $12B financing pipeline could drive higher loan monetization premiums.

  • Estimated Impact: +$50 Other GPU uplift (+$10M Gross Profit).

• Market Share Outperformance: Retail sales exceeding 205,000 units (+43% YoY).

  • Estimated Impact: +$175M Revenue / +$30M Adj. EBITDA.

Downside Risks

• Wholesale-to-Retail Spread Squeeze: Broader vehicle depreciation could cause the wholesale-to-retail spread compression to exceed the guided $100–$200 range.

  • Estimated Impact: -$150 Retail GPU (-$30M Gross Profit).

• Industry Price Competition (CarMax Read-Through): Promotional price cutting by traditional dealers could pressure retail average selling prices (ASPs).

  • Estimated Impact: -$100 to -$200 Retail GPU (-$20M to -$40M Gross Profit).

• SG&A Overhead Creep: AI and technology investments, combined with peak seasonal marketing spend, pushing operating expenses above $730M.

  • Estimated Impact: -$20M Operating Income.

Questions for the Call

1. Wholesale-to-Retail Spread Squeeze: Management guided to a $100–$200 YoY Retail GPU headwind in Q2 from narrower spreads. Has this spread compression stabilized entering Q3, and how is it trending relative to initial expectations?

2. Reconditioning Labor Efficiency: You noted that labor efficiency in April approached all-time bests following new software tool rollouts. Given that costs lag unit sales, how much of this efficiency gain is recognized in Q2 gross margin versus flowing into Q3 inventory?

3. Competitive Pricing Environment: Competitors like CarMax reported dropping retail GPU by over $200 per unit to defend volume. How is Carvana balancing volume growth versus retail GPU unit economics in the current pricing environment?

4. Overhead & Technology Run-Rate: Following the SG&A step-up to $690M in Q1, how should we think about the run-rate for overhead expenses in H2 2026 as AI investments (such as “Sebastian”) continue?

5. ADESA Buildout Cadence: What is the current construction timeline and cap-ex cadence for the 6 to 8 ADESA full site build-outs planned for FY2026 at $30M–$35M per site?

6. Balance Sheet & Credit Goals: With Net Debt to Trailing 12-Month Adjusted EBITDA reaching ~1.0x, what are the primary milestones or leverage targets required to achieve investment-grade credit ratings or optimize long-term corporate debt costs?

The primary goal is to make it dramatically easier and cheaper to utilize AI for investment research. To keep a long story short I am currently trying to do that by chunking up the work as much as possible. It saves tokens, increases accuracy, improves quality, etc.

I’ll provide a more thorough coverage of how everything works here soon. For now, I wanted to share an earnings preview skill I’ve been testing for the upcoming and earnings reports expected later today. Our MCP is currently not yet live as we finish up some boring backend optimizations, but the idea is we’d basically provide pre-built skills and data templates to power your AI assisted research anywhere.

It’s also very likely that automated reporting via Substack for a variety of tickers becomes something that is consistently published. These sample reports take ~3-5 minutes to generate and cost about $1-$2 by utilizing MCP tools to efficiently parse numerous quarters of historical information, automatically retrieving filings/earnings transcripts/etc.

Feel free to let me know what you think and if this kind of coverage is interesting. I’ll definetly write more about the process at the very least. Cheers.

NOW — Q2 FY2026 Earnings Prep

What to Expect

ServiceNow (NOW) reports Q2 FY2026 earnings with strong operational momentum, though investors are scrutinizing top-line organic performance versus M&A contribution, as well as near-term margin pressure from acquisitions. Key focus areas for the upcoming release:

1. Armis Acquisition Contribution vs. Margin Drag: Q2 FY2026 represents the first full quarter including the Armis acquisition (closed early April 2026). Armis is expected to contribute 125 bps to Q2 subscription revenue growth but creates a 125 bps headwind to Non-GAAP operating margin, driving guided operating margin down to 26.5% (vs. 32.0% in Q1). Management committed to reducing this acquisition margin drag to 0 bps by year-end FY2026.

2. Generative AI & Agentic AI Run-Rate Acceleration: Following a raise in the FY2026 AI commitment run-rate from $1.0B to $1.5B in Q1, markets will evaluate whether Now Assist Pro Plus adoption, Autonomous Workforce AI specialist roles (launched May 2026), and RaptorDB Pro continue to drive net new ACV (NNACV). In Q1 FY2026, $1M+ ACV Now Assist customer counts expanded +130% YoY.

3. cRPO Growth & On-Prem Deal Recovery: Q2 constant currency (CC) cRPO growth is guided to 19.5% YoY, decelerating from 21.0% CC in Q1 FY2026. A primary focal point is whether the ~75 bps headwind from delayed Middle East on-premise deals in Q1 closed as expected during Q2.

4. Expansion of Non-Seat Pricing Models: Management disclosed that 50% of net new business in Q1 FY2026 was structured on non-seat pricing models (consumption, tokens, capacity, connectors). Investors will track whether this transition mitigates market concerns over software seat compression.

Guidance & Consensus

ServiceNow issued target guidance for Q2 FY2026 during its Q1 FY2026 earnings release on April 22, 2026.

Q2 FY2026 Guided Metrics

FY2026 Full-Year Guidance Summary

Historical Trends

Key Financial Trajectory (Q1 FY2025 – Q1 FY2026)

Key Inflection Points & Trends:

• Top-Line Resiliency: Subscription revenue growth maintained a steady 19.0%–21.5% CC range over the last 5 quarters.

• Margin Dynamics: Non-GAAP operating margin expanded to 32.0% in Q1 FY2026 (+100 bps YoY), but Q2 FY2026 guidance introduces a planned seasonal and M&A-driven reset to 26.5%.

• Public Cloud & Data Center Drag: Non-GAAP subscription gross margin is projected to contract 50 bps in FY2026 to 81.5% due to public cloud hyperscaler migration and AI inference infrastructure investments.

Segment & KPI Outlook

                  NOW Growth Engines & KPI Drivers
┌──────────────────────────────────────────────────────────────────┐
│                    Generative AI / Now Assist                    │
│   • Run-rate target raised to $1.5B (up from $1.0B in Q4 FY25)    │
│   • $1M+ ACV AI Customers: +130% YoY in Q1 FY2026                │
└──────────────────────────────────────────────────────────────────┘
                                │
                                ▼
┌──────────────────────────────────────────────────────────────────┐
│                   Customer & Platform Traction                   │
│   • Deals >$1M Net New ACV: +30% YoY in Q1 FY2026                │
│   • Renewal Rate: Maintained at 97.0% - 98.0%                    │
│   • Non-Seat Pricing: 50% of NNACV from consumption/tokens       │
└──────────────────────────────────────────────────────────────────┘
                                │
                                ▼
┌──────────────────────────────────────────────────────────────────┐
│                   Acquisitions & Expansion                       │
│   • Armis: Adds 125 bps to Q2 Sub Rev / -125 bps Op Margin drag  │
│   • Moveworks / Employee Works: Rebranded unit grew 5x YoY       │
└──────────────────────────────────────────────────────────────────┘

1. Large Customer Acceleration & Deal Volume

• Recent Trajectory: Large deals with >$1M Net New ACV grew +30% YoY in Q1 FY2026, up from 244 total mega-deals closed in Q4 FY2025. Customers generating >$5M ACV increased from 508 in Q1 FY2025 to 603 in Q4 FY2025.

• Management Commentary: CEO Bill McDermott highlighted that core enterprise IT systems of record are experiencing a “resurgence” as customers standardize on ServiceNow to build AI workflows.

• Q2 Outlook: Management expects sustained large-deal activity, backed by enterprise multi-suite expansion and government sector renewals.

2. Generative AI (Now Assist) Suite

• Recent Trajectory: Now Assist ACV crossed $600M in Q4 FY2025 and pushed the total AI commitment run-rate toward $1.5B in Q1 FY2026. Now Assist Pro Plus tiers continue to command a >30% price uplift over standard Pro SKUs.

• Management Commentary: CPO Amit Zavery highlighted strong adoption of RaptorDB Pro (+80% YoY deal volume in Q1 FY2026) and sales CRM NNACV (>5x YoY).

• Q2 Outlook: Focus centers on the rollout of 20 pre-configured Autonomous AI Specialist roles (launched May 2026) and token-based usage billing.

3. Moveworks & Security Integration (Armis / Veza)

• Recent Trajectory: Moveworks was integrated into Employee Experience to form Employee Works, which grew 5x YoY in Q1 FY2026. Armis closed early in April 2026.

• Management Commentary: Armis CEO Yevgeny Dibrov was appointed to lead the combined security business. CEO Bill McDermott promised that Armis dilution would be brought down to 0 bps by Q4 FY2026.

• Q2 Outlook: Armis will contribute ~$45M–$50M (125 bps) to Q2 subscription revenues, with integration expenses fully reflected in the guided 26.5% operating margin.

Recent Developments

External news retrieval tools were not enabled; analysis is based on corporate SEC disclosures, financial statements, and insider transaction filings.

Insider Activity Highlights

                Recent Open-Market Insider Activity
┌──────────────────────────────────────────────────────────────────┐
│  Feb 27, 2026 | CEO Purchase (William McDermott)                 │
│  • Shares: 28,682 @ ~$104.60 per share ($3.0M total value)       │
│  • Executed via Non-Rule 10b5-1 Plan                             │
└──────────────────────────────────────────────────────────────────┘
                                │
                                ▼
┌──────────────────────────────────────────────────────────────────┐
│  May 2026 | Planned Exec & Director Sales (Post-RSU Vesting)      │
│  • Paul Fipps (Pres. Customer Ops): 14,536 shares ($1.51M)       │
│  • Anita Sands (Director): 16,445 shares ($1.48M)                │
│  • Executed via Rule 10b5-1 Plans & Scheduled Tax Withholdings   │
└──────────────────────────────────────────────────────────────────┘

1. Significant Open-Market CEO Purchase: Chairman & CEO William R. McDermott purchased 28,682 shares on February 27, 2026, at an average price of $104.60 per share ($3,000,058 total value) under a non-Rule 10b5-1 trading arrangement.

2. Routine Executive Dispositions: Following the May 2026 RSU vesting period, President of Global Customer Operations Paul Fipps sold 14,536 cumulative shares ($1.51M) under a Rule 10b5-1 plan adopted November 19, 2025. Director Anita Sands sold 16,445 shares ($1.48M) on May 14, 2026.

3. Share Repurchase Execution: Following the January 2026 authorization of an incremental $5.0B share repurchase program, ServiceNow executed an immediate $2.0B Accelerated Share Repurchase (ASR), reducing the FY2026 diluted share count target to 1.04B shares.

Risks & Catalysts

Upside Catalysts

• Middle East Deal Slip Catch-Up (+50 to +75 bps Subscription Revenue Impact): Closing the on-premise government and enterprise deals delayed in Q1 could drive revenue toward the high end of the guided range ($3.820B).

• Accelerated Armis Cross-Selling (+50 bps cRPO Impact): Rapid go-to-market integration of Armis into ServiceNow’s security and asset workflow workflows could expand net new ACV.

• AI Consumption Monetization Acceleration (+100 bps ACV Impact): Faster adoption of non-seat models (tokens, capacity) driven by Autonomous AI Specialists could expand top-line expansion rates.

Downside Risks

• Public Cloud & Inference Cost Inflation (-50 to -100 bps Gross Margin Risk): Shift toward hyperscaler cloud deployments and higher LLM inference workloads could pressure subscription gross margins below the 81.5% FY2026 target.

• Armis Integration Execution Drag (-50 bps Operating Margin Impact): Delayed operational synergies could prolong acquisition margin dilution beyond management’s Q4 FY2026 target.

• Seat Rationalization Delays (-100 bps cRPO Impact): Enterprise budget tightening or slower core seat expansion could limit net new ACV growth.

Questions for the Call

1. Armis Integration & Synergies: Management committed to eliminating Armis operating margin dilution (125 bps in Q2) by Q4 FY2026. What specific cost synergies and operating efficiencies drive this margin recovery over the next two quarters?

2. On-Premise Middle East Deal Closures: Did the delayed Middle East on-premise deals that created a ~75 bps revenue headwind in Q1 fully close during Q2, and are regional deal execution timelines back to normal baselines?

3. Non-Seat Pricing Traction: With 50% of Q1 net new business structured on non-seat models, how are customer consumption patterns for tokens and capacity trending in Q2? Does this shift alter ACV visibility or contract duration?

4. Autonomous AI Specialist Monetization: Following the launch of ~20 pre-configured AI specialist roles in May 2026, how are enterprise customers pricing and packaging these autonomous agents relative to traditional seat-based Pro Plus upgrades?

5. Subscription Gross Margin Pressures: Given the full-year 50 bps reduction in subscription gross margin to 81.5%, what proportion of this decline stems from public cloud migration versus higher AI inference costs, and when do hyperscaler economies of scale begin to offset these investments?

6. Core IT Workflows Resurgence: Management noted a major resurgence in core IT systems of record. To what extent is this demand driven by mandatory platform upgrades required to support Now Assist and RaptorDB Pro?

TSLA — Q2 FY2026 Earnings Prep

Executive Overview & Strategic Focus

This pre-earnings report prepares portfolio management for Tesla, Inc.’s (TSLA) upcoming Q2 FY2026 earnings release and conference call. The period represents an operational turning point for Tesla as it transitions away from legacy vehicle lines (Model S/X final assembly ended in early May 2026) to scale dedicated autonomous hardware (Cybercab production began in April 2026 at Giga Texas) and humanoid robotics (retooling Fremont for Optimus starter production).

Concurrently, capital expenditures are accelerating to >$25.00B for FY2026, which management expects will generate negative Free Cash Flow (FCF) for the remainder of the fiscal year.

What to Expect

Lead items to monitor for the Q2 FY2026 print and conference call:

CapEx Expansion Run-Rate & Free Cash Flow Drain

• Why it matters: Management raised FY2026 CapEx guidance from >$20.00B to >$25.00B on the Q1 FY2026 call, explicitly warning of negative FCF for the remainder of 2026. Investors need to evaluate the quarterly cash burn rate against Tesla’s liquid cash and short-term investments balance ($36.56B as of Q4 2024).

• What the data suggests: Heavy front-loaded spending on AI compute capacity, the ~$3.00B Giga Texas Research Fab (partnering with Intel on the 14A node), and the Fremont factory retooling will push quarterly CapEx past $6.00B–$6.50B in Q2 FY2026.

Automotive Gross Margin (ex-Reg Credits) & Product Mix Shift

• Why it matters: Automotive gross margin excluding regulatory credits expanded sequentially over three quarters: 15.4% (Q3 2025) → 17.9% (Q4 2025) → 19.2% (Q1 2026). Sustaining margins near or above 19.0% validates core cost-reduction efforts amid hardware transitions.

• What the data suggests: The shutdown of Model S/X in early May eliminates high-ASP but lower-volume vehicles. Early low-rate Cybercab production (started April 2026) and global vehicle inventory normalizing from 27 days of supply in Q1 2026 will test unit cost economics.

Energy Storage Rebound & Margin Compression Headwinds

• Why it matters: Energy storage deployments dropped 38.0% QoQ to 8.80 GWh in Q1 2026 (down from a record 14.20 GWh in Q4 2025). Rebounding toward 10.0–12.0 GWh is necessary to hit full-year guidance of exceeding FY2025’s 46.70 GWh total.

• What the data suggests: While Energy segment gross margins were strong throughout FY2025 (28.6%–31.4%), CFO Vaibhav Taneja warned of incoming gross margin compression due to low-cost battery competition, policy uncertainties, and tariff friction.

FSD Software Monetization & Hardware 3 Upgrade Liabilities

• Why it matters: Active FSD paid subscriptions grew to 1.28M in Q1 2026 (+51.0% YoY). Management has shifted to an exclusive subscription model for FSD, treating vehicles primarily as hardware delivery mechanisms.

• What the data suggests: Key software catalysts due in Q2 include EU-wide FSD regulatory approval (following Netherlands clearance in April 2026) and the end-of-June release of FSD V14 for Hardware 3 (HW3). However, offering trade-in discounts and camera/computer retrofits for HW3 owners upgrading to AI4 will create near-term operating expense drag.

Guidance & Consensus

Note: Quantitative statistical model predictions were not generated/available for this reporting period. The table below compares official company guidance against historical track records and recent management updates.

Q2 FY2026 & FY2026 Official Guidance Tracker

Historical Trends

Financial Inflections & Trajectory Heading into Q2 FY2026

• Revenue Trough & Recovery: Total revenue hit a cyclical low of $19.34B in Q1 2025 before peaking at $28.10B in Q3 2025. Automotive sales contracted YoY in 3 of the 4 reported quarters of FY2025, making Energy (+25.4% YoY in Q4 2025) and Services (+18.4% YoY in Q4 2025) essential growth buffers.

• Margin Expansion Peak: Total Gross Margin expanded sharply to 23.91% in Q4 2025, fueled by peak Energy GM (28.62%) and $542M in regulatory credits. Auto GM ex-reg credits steadily climbed to 19.20% in Q1 2026.

• Inventory Accumulation: Days of supply surged from 15 days in Q4 2025 to 27 days in Q1 2026, signaling fleet staging ahead of the Model S/X retirement and transition to Cybercab production lines.

Segment & KPI Outlook

1. Automotive Segment & Cybercab

• Recent Trajectory: Total automotive revenue stood at $17.69B in Q4 2025 (-10.6% YoY). Automotive GM ex-credits reached 19.2% in Q1 2026, benefiting from unit COGS remaining below $35,000.

• Management Commentary: Model S/X assembly lines were shut down in early May 2026, with dismantling taking ~4 months to convert the space into an Optimus factory. Cybercab production commenced at Giga Texas in April 2026. Management reiterated that vehicles are now viewed as delivery mechanisms for FSD software subscriptions.

• Q2 FY2026 Expectation: Expect Automotive Revenue to reflect the loss of ~12,000–15,000 Model S/X deliveries per quarter, partially offset by initial low-volume Cybercab deliveries. Automotive GM ex-credits will face retooling cost absorption headwinds.

2. Energy Generation & Storage

• Recent Trajectory: Energy segment revenue grew to $3.84B in Q4 2025 (+25.4% YoY) with record 14.20 GWh deployed. Q1 2026 saw a sharp sequential pullback to 8.80 GWh (-38.0% QoQ). Segment gross margin remained robust at 28.62% in Q4 2025.

• Management Commentary: CFO Vaibhav Taneja reconfirmed full-year 2026 storage deployments will exceed FY2025 (46.70 GWh), backed by the Megapack 3 and Mega Block launches out of Houston. However, management warned of gross margin compression due to low-cost Asian competition and tariffs.

• Q2 FY2026 Expectation: Deployments should rebound to 10.50–12.00 GWh, driving Energy revenue back above $3.20B. Watch whether Energy gross margin holds above 28.0% or compresses toward 25.0% under tariff pressure.

3. Services, FSD & AI Compute Infrastructure

• Recent Trajectory: Services & Other revenue reached $3.37B in Q4 2025 (+18.4% YoY). Active FSD paid subscribers expanded to 1.28M in Q1 2026 (+51.0% YoY).

• Management Commentary: Transitioned entirely to subscription FSD. FSD received approval in the Netherlands in April 2026, setting up EU-wide clearance in late Q2 2026. China approval is targeted for Q3 2026. CapEx guidance was raised to >$25.00B to fund Texas AI compute expansion, the $3.00B Research Fab partnership with Intel (14A node), and metropolitan upgrade micro-factories for HW3-to-HW4 vehicle retrofits.

• Q2 FY2026 Expectation: Services revenue should surpass $3.50B, supported by 18.7% YoY growth in Supercharger connectors (79,918 in Q1 2026) and expanding FSD subscription revenue.

Recent Developments

Note: External media search was not conducted; information is sourced from SEC filings, company disclosures, and official insider reporting.

1. Key Insider Transactions (Late 2025 – May 2026)

• Massive Open-Market CEO Purchase: On September 12, 2025, CEO Elon Musk purchased 1,207,757 shares in the open market across 13 tranches at an average price of ~$388.62 per share, totaling $469.36M. This non-10b5-1 purchase represents direct open-market buying by the CEO.

• Restricted Equity Forfeiture: On April 21, 2026, Elon Musk forfeited 96,000,000 restricted shares ($0 transaction value) due to legal trigger events related to the Tornetta litigation proceedings.

• Systematic 10b5-1 Director Sales: Director Kathleen Wilson-Thompson executed regular option exercises ($14.99 strike) and 10b5-1 sales:

  • Feb 25, 2026: Sold 25,681 shares for $10.67M ($415.35 avg).

  • Mar 30, 2026: Sold 28,309 shares for $10.19M ($359.80 avg).

  • Apr 30, 2026: Sold 26,413 shares for $9.99M ($378.10 avg).

• Director & Officer Activity: Director James Murdoch sold 50,000 shares for $22.29M ($445.80 avg) on Jan 2, 2026 under a 10b5-1 plan. CFO Vaibhav Taneja exercised options and sold 3,000 shares for $1.35M ($450.00 price) on May 13, 2026 under a 10b5-1 plan.

2. Operational & Strategic Events Disclosures

• Model S/X Line Shutdown: Final assembly ended early May 2026 at Fremont. The space is being cleared to build the Optimus starter factory, targeted for initial production in late July/August 2026.

• Cybercab Production Launch: Start of production achieved in April 2026 at Gigafactory Texas.

• Semiconductor Fab Initiative: Finalized plans for a ~$3.00B Texas Research Fab partnering with Intel on its 14A process node to secure logic, memory, and packaging capabilities.

Risks & Catalysts

Upside Catalysts

Catalyst EventTarget PeriodEstimated Financial / Operational ImpactPrimary Metrics AffectedEU-Wide FSD ApprovalLate Q2 2026+$150M–$250M annualized high-margin subscription revenueServices Revenue, Total Gross MarginAccelerated Cybercab Ramp2H 2026+5,000–10,000 units added to 2H26 production run-rateAutomotive Revenue, Total DeliveriesEnergy Storage Margin ResilienceQ2–Q4 2026Storage gross margins staying >30.0% (vs feared compression)Energy Segment Gross Profit, Total GM

Downside Risks

Risk FactorTarget PeriodEstimated Financial / Operational ImpactPrimary Metrics AffectedCapEx Acceleration & Cash BurnFY2026FY26 CapEx >$25.00B driving -$1.5B to -$2.5B quarterly negative FCFFree Cash Flow, Cash & ST InvestmentsFremont Retooling & Retaining CostsQ2–Q3 2026Dismantling Model S/X and HW3 micro-factory setups drag Auto GM by -50 to -100 bpsAutomotive GM ex-credits, OpExTariff Friction on Megapack 3Q2–Q4 2026Asian battery cell tariffs compressing Energy GM from ~28.6% toward ~24.0%Energy Gross Margin, Net Income

Questions for the Earnings Call

1. CapEx Cadence & Liquidity Minimums: With FY2026 CapEx guided above $25.00B and negative FCF expected for the rest of the year, what is the expected quarterly spending profile across the $3.00B Texas Research Fab, Optimus retooling, and AI compute? What minimum cash reserve balance does management intend to hold?

2. Cybercab Unit Economics & Production Rate: Following the April 2026 start of production at Giga Texas, what total delivery volume is anticipated for Cybercab in Q2 and Q3? At what annual production volume does Cybercab reach gross margin parity with Model 3/Y?

3. HW3 Retrofit Liability & Micro-Factory Costs: For the 1.28M active FSD subscriber base, what percentage of vehicles currently run on Hardware 3? What is the estimated total capital and operating cost per vehicle for the HW3 trade-in discounts and camera/computer retrofits?

4. Fremont Retooling & Optimus Timeline: With Model S/X production ended in early May, are line dismantling and Optimus equipment installation progressing on schedule for the late July/August starter production window? When is the second Optimus factory at Giga Texas scheduled to break ground?

5. FSD International Monetization Rate: Following the April Netherlands approval and expected late-Q2 EU-wide clearance, what initial subscription conversion rate is expected across European fleet owners? What is the current status of regulatory discussions in China ahead of the Q3 target?

6. Energy Storage Margin Mitigation: How did Energy segment gross margins perform in Q2 following Q1’s 8.80 GWh deployment trough? How are tariff headwinds and cell input costs impacting pricing structures for Megapack 3 and Mega Block orders?

7. Semiconductor Fab Governance & Partnership Structure: Could management clarify the capital split, intellectual property ownership, and operational division between Tesla, SpaceX, and Intel regarding the $3.00B Texas Research Fab and the long-term TerraFab strategy?

Alphabet Inc. (GOOGL) — Q2 FY2026 Earnings Prep

What to Expect

Google Cloud Operating Margin Trajectory & Wiz Integration

• Why it matters: Google Cloud operating income surged to $6.60B in Q1 FY2026 (up 203.10% YoY from $2.18B in Q1 FY2025). However, management guided to a low single-digit percentage point headwind to Cloud’s operating margin for the remainder of FY2026 following the March 2026 closing of the Wiz acquisition.

• What the data suggests: Cloud backlog expanded to $462.00B in Q1 FY2026 (nearly doubling QoQ), with >50% expected to convert into revenue over the next 24 months. GenAI product revenue growth reached ~800% YoY in Q1 FY2026. Key to watch in Q2 is whether top-line expansion can absorb Wiz integration costs without causing a sharper-than-expected margin contraction.

CapEx Ramp & Long-Term Infrastructure Commitments

• Why it matters: On the Q1 FY2026 call, CFO Anat Ashkenazi raised the FY2026 full-year CapEx target range to $180.00B – $190.00B (up from $175.00B – $185.00B) to incorporate the acquisition of Intersect (closed March 2026) and infrastructure buildouts. Furthermore, management explicitly signaled that 2027 CapEx will significantly increase compared to 2026.

• What the data suggests: Ramping CapEx directly elevates depreciation and data center operating costs (depreciation grew +43.60% YoY to $6.04B in Q4 FY2025). Analysts will monitor whether higher capital intensity pressures consolidated operating margins (36.12% in Q1 FY2026) in Q2.

Moderating FX Tailwinds & Google Services Growth

• Why it matters: Google Services generated $40.59B in operating income in Q1 FY2026. Management guided for FX tailwinds to moderate from 3 percentage points in Q1 FY2026 to ~1 percentage point in Q2 FY2026.

• What the data suggests: Paid subscriptions across YouTube and Google One reached 350M in Q1 FY2026. Monetization will be tested in Q2 as YouTube Premium Lite expands to more than a dozen new countries (up from 23 countries in Q1).

Direct TPU Hardware Revenue Recognition Dynamics

• Why it matters: Alphabet introduced direct on-premises TPU hardware sales (e.g., TPU 8t and TPU 8i) to select enterprise and AI lab clients.

• What the data suggests: Management noted that a small percentage of TPU hardware revenues will be recognized in FY2026, with the vast majority realized in FY2027. Quarterly hardware shipments are expected to create revenue lumpiness starting in Q2 FY2026.

Guidance & Consensus

FY2026 & Q2 FY2026 Management Guidance

Historical Financial Trends

(Dollar values in millions, except per share data)

Trajectory & Inflection Analysis

• Revenue Acceleration: YoY consolidated revenue growth re-accelerated to +21.79% in Q1 2026 ($109.90B) compared to +12.04% in Q1 2025 ($90.23B), driven by enterprise AI adoption and expanding cloud workloads.

• Margin Expansion: Consolidated operating margin expanded 220 bps YoY to 36.12% in Q1 2026, powered by record Google Cloud operating profit ($6.60B, margin ~33%+ prior to Wiz dilution).

Segment & KPI Outlook

1. Google Cloud

• Recent Trajectory: Operating income increased from $2.18B in Q1 2025 to $6.60B in Q1 2026 (+203.08% YoY). Backlog surged from $106.00B in Q2 2025 to $240.00B in Q4 2025 and $462.00B in Q1 2026.

• Management Commentary: Demand remains extremely high, though management noted operating under compute capacity constraints. Wiz acquisition closed in March 2026 and will be included within Cloud segment reporting.

• Q2 FY2026 Expectations: Continued top-line momentum supported by backlog burn (>50% over 24 months) and GenAI growth (~800% YoY in Q1 2026). Operating margin will reflect a low single-digit percentage point headwind from Wiz integration costs.

2. Google Services (Search, YouTube, Subscriptions)

• Recent Trajectory: Operating income grew from $32.68B in Q1 2025 to $40.59B in Q1 2026 (+24.20% YoY). Total paid subscriptions reached 350M in Q1 2026 (up from 270M in Q1 2025).

• Management Commentary: Daily AI mode queries per user doubled since launch in the U.S., with queries running 3x longer than traditional searches. Universal Commerce Protocol (UCP) added Amazon, Meta, Microsoft, Salesforce, and Stripe to its technical council.

• Q2 FY2026 Expectations: Moderate sequential growth with ~1 percentage point FX tailwind (down from 3 percentage points in Q1). Subscription expansion will be supported by YouTube Premium Lite launches across 12+ new countries.

3. Key Operational KPIs

Recent Developments

Insider Activity

• Insider Trends: Zero open-market purchases occurred during the period. On April 8, 2026, the executive team received a synchronized grant totaling 361,325 GSUs. CEO Sundar Pichai liquidated 32,500 shares for ~$10.01M on March 18, 2026.

Note on Market Context & News: External web search tools were not enabled for this briefing; analysis is derived exclusively from SEC filings, company financial disclosures, and quarterly earnings transcripts.

Risks & Catalysts

Upside Catalysts

1. Accelerated Backlog Revenue Recognition ($462.00B Backlog)

   • Metric Impact: Google Cloud Revenue & Operating Income.

   • Detail: Cloud backlog grew to $462.00B in Q1 FY2026. Faster-than-anticipated customer onboarding or earlier delivery of TPU 8t/8i hardware instances could exceed top-line expectations and offset Wiz integration margin drag.

2. Subscription Scaling via YouTube Premium Lite

   • Metric Impact: Google Services Subscriptions, Platforms & Devices Revenue.

   • Detail: Expanding YouTube Premium Lite into more than 12 new countries in Q2 FY2026 builds on the 350M total paid subscriptions reported in Q1 FY2026, lifting recurring high-margin revenue.

Downside Risks

1. Depreciation Acceleration from CapEx Ramping ($180.00B – $190.00B Range)

   • Metric Impact: Consolidated Operating Income & Operating Margin.

   • Detail: Full-year CapEx was raised to $180.00B – $190.00B due to the Intersect transaction and AI compute buildout. Depreciation expense growth (which accelerated to +43.60% YoY in Q4 FY2025) will place persistent upward pressure on operating costs.

2. Wiz Operating Margin Dilution

   • Metric Impact: Google Cloud Operating Margin.

   • Detail: Management guided to a low single-digit percentage point headwind to Cloud operating margins for the rest of FY2026 from the Wiz integration. Dilution higher than expected could depress segment operating profit.

3. FX Tailwind Moderation

   • Metric Impact: Total Consolidated Revenue.

   • Detail: Consolidated FX tailwinds decrease from 3 percentage points in Q1 FY2026 to ~1 percentage point in Q2 FY2026, creating a sequential growth headwind of several hundred million dollars.

Questions for the Call

1. Cloud Backlog Burn & Conversion: “With Cloud backlog reaching $462.00B in Q1 FY2026 and guidance calling for >50% conversion over 24 months, what sequential conversion rate should we expect in Q2 FY2026, and how much backlog is driven by software consumption vs. direct TPU silicon sales?”

2. Wiz Operating Margin Headwind: “Could you quantify the specific operating expense drag from Wiz in Q2 FY2026 within the low single-digit margin headwind guidance, and what is the expected integration timeline to return Cloud operating margins to pre-acquisition trajectories?”

3. CapEx Cadence & Infrastructure Spend: “With full-year FY2026 CapEx updated to $180.00B–$190.00B and 2027 CapEx expected to increase significantly, how will CapEx be distributed across Q2, Q3, and Q4, and what proportion is allocated to data center power/energy infrastructure vs. server silicon?”

4. Direct TPU Hardware Sales Contribution: “As direct on-premises TPU shipments (TPU 8t/8i) begin in FY2026, how will direct hardware sales impact Google Cloud’s gross margin profile relative to traditional GCP cloud instance rentals?”

5. AI Search Monetization & Token Volume: “First-party token processing reached 16B/min in Q1 FY2026, and daily AI mode queries per user doubled. What specific ad load, click-through rate (CTR), and cost-per-click (CPC) trends are you seeing in AI mode queries compared to traditional search layouts?”

6. Waymo Scaling & Other Bets Loss Cadence: “Waymo surpassed 500,000 weekly autonomous rides across 11 cities in Q1 FY2026. How should we model Other Bets operating losses (which reached -$2.10B in Q1 FY2026) as Waymo expands into Nashville, the UK, and Japan?”

Today I’m happy to announce the first step in what is hopefully a very long journey, the beta launch of Clarity AI, which you can try for free today.

Link here: https://www.clarity-markets.com/

As I’ve written before and my ~30% Google position suggests, I’m a staunch AI bull. The capabilities are staggering and continuously improving.

I’ve spent the past 6 months working with models to code, to analyze filings, to analyze financials, to generate charts, to write reports, browse the news, create financial models, and so much more.

Like many of you I’ve had mixed results, especially when it comes to the required 100% accuracy for many financial use cases.

Clarity is my attempt to solve that problem. Here’s a quick rundown of how that works.

Step 1: The Data Backbone

The primary reason your Claude co-work or ChatGPT feed you bad results is a lack of proper context.

Creating a generic model that can fulfill any task is a herculean effort that requires not only razor sharp model capabilities, but also hundreds or thousands of tools at the models disposal. Unfortunately we aren’t quite there yet.

With Clarity, our focus is specifically on analyzing the market, which lets us point the models focus where it matters, the source data to answer your questions.

That means we went ahead and grabbed SEC filings, earnings transcripts, price data, and 10 years of historical financials to power what we do. A nice benefit of that is they’re human readable too. Let’s explore.

Financials:

Reliable financials are key to any equity research platform, and they aren’t easy.

With Clarity, we use custom software to export the Income Statement, Balance Sheet, and Cash Flow Statement for each ticker directly from the SEC. We then utilize AI to standardize reporting. For ease of mind, we provide click-through citations for every number and explanations of any calculations.

Here’s some visual examples:

Need the source for peace of mind? Let’s click on Total Revenue for FY2025

We also offer “As Reported” financials with more detail specific to each company.

For calculated values, we define the calculation and show you the components

Paid users also have access to export functionality, as well as AI charting capabilities

There’s always more to do and AI makes more possible every day. We’d love to add custom KPI extraction, add more metrics and ratios, and more. Whether you browse for free or pay for additional features, I hope you love it.

All of this information also has the dual purpose of feeding our models, so humans and machines both win.

Filings:

Did you know that by default models can’t access the SEC site directly? It’s a bit of a headache when trying to get results grounded in reality.

So we do that for the model, while also creating a document browse and search section for us humans that remain in the loop.

We currently grab all 10-K/10-Q/8-K/S-1/DEF 14A and earnings transcripts for tickers we cover, with plans to expand to all filing types very soon.

You can browse and read filings directly within our site, as well as search and chat directly with the documents

We also have document search capabilities, combined with AI summaries. For example, say you were curious about recent discussion of “AI strategies” in filings? Just send out a search and click summarize.

How has inflation impacted CVNA over the years? Let the AI figure it out.

We also support direct filing imports to our AI Chat and AI Reports so you can be sure you’re chatting with the source.

There’s again so much to do here. Some of our plans include Form 4 trackers that utilize AI to summarize the transaction type, no more guessing what each transaction code means. We also are working on dramatically improving the context compression and search capabilities so you can chat with dozens of filings at once and easily compare competitors or get up to speed on different industries.

More Data:

Beyond the financials, filings, and transcripts, we are also in the early stages of expanding our general data infrastructure.

Our early targets include FRED data releases so you can track all the important macro data you wish

As well as the emerging prediction markets information ecosystem

While very early, longer term if there’s data that impacts an investment decision, we want to get it in our platform.

Step 2: AI Integrations

Now that you know what kind of information we’re working with, let’s talk about how it’s used.

The general hypothesis with Clarity is that AI is at it’s best when it’s proactive. Chat bots are nice and all, but since I’ve worked on analyzing stocks for years now, why not attempt to automatically generate some of the research outputs I’d be making anyways?

Our AI integrations have 3 primary touch points which we’ll explore below.

The Dashboard:

After creating an account and logging in, whenever you go to the Clarity Markets site you will be greeted by your home dashboard.

This dashboard acts as the glue between all of the data and AI integrations within Clarity. You can hop directly to any portion of the site, as well as browse our curated AI reports.

Each day we have a Morning Briefing and Afternoon Breakdown that cover general market news for the day. We additionally have public AI Market Reports which hosts AI workflows we’ve created. Here’s a sample of what’s included in the CVNA daily stock brief for example

Our goal is to create thousands of curated research reports browsable any time. Have 3 portfolio companies reporting earnings the same day? Catch up with an AI report. Stock on your watchlist dropped 10%? Find out why with an AI report. Slept through your alarm and wake up to everything red? Find out what new special military operation happened overnight with an AI report.

We’re currently working on more report templates, as well as more ways to custom tailor your hub to you.

Reports:

One of the AI integrations powering the hub is of course our AI reports. These reports can either be viewed within the site in the reports/hub sections, or via your email.

In the reports section you can browse, view, and subscribe to AI reports. These reports utilize our data and frontier models to create scheduled multi-prompt outputs with no input required. Pretty neat.

Our paid users (with some tinkering) can create custom reports either explicitly with our report creator or with the help of our AI builder.

We plan on continuing to flesh out the reports section with more pre-built templates, more robust scheduling options, more data, and an even better experience.

Chat:

Every AI app has chat, and Clarity is no exception.

We do try to make it a little better though

Our chat again utilizes explicit context management so you have full control over what’s included. Want to make sure the numbers are coming from a filing? Either use our smart chat helper or set up your own chat with explicitly defined context.

The context management tool lets you upload your own files or chat with specific SEC filings and earnings transcripts. You can also enable web search for real time information access, and paid users have access to Deep Research and Image Generation capabilities.

Our chats also include citations of the website or file that provided specific information, so if you’re in doubt, you can double check.

We’re currently working on expanding the amount of context available in a given chat window (if you add too many files it will error currently), as well as attempting to have click-through citations that get you straight to the source sentence. We also are planning on adding charting and better ability to compare tickers.

Currently chat has usage limits based on your tier, but free users do get to try it!

Earnings:

Last but not least, we’re also working on our earnings portal. Here’s a sample for mid April.

For each ticker in our coverage we track the earnings date, basic consensus EPS/revenue, historical results, recent price changes, and have an AI preview available for a quick breakdown of what to look out for.

Step 3: You

I’m really excited about what we’ve accomplished and plan to accomplish with Clarity. The only key ingredient left is passionate users, you.

Much of the site is currently available entirely for free, with paid options available at $20 or $200 per month to unlock additional AI features and data access. I think that’s decently competitive even without the AI additions.

For full disclosure, pricing is subject to change based on the AI usage habits. While AI is dramatically cheaper than a full time human, it’s not free, and we’ll adjust if needed. Our goal is to have a very robust baseline data and documents platform for humans that’s quite cheap, which also happens to support premium AI analysis comparable to many teams of humans.

It’d be great to get a bunch of passionate users, invest more in engineering and data to continuously improve the platform, and fully re-imagine the process of financial research.

We’d greatly appreciate your support, and thanks for reading.

PS:

We’re a small team and only in beta, so there may be the occasional bug with functionality you run into. If so, feel free to reach out and I’ll be happy to fix ASAP. Same goes for any feature requests, ticker requests, data requests, and more.

I will also be using our tools and AI reports to create more investing content in the near future. Stay tuned!

Subscribe now

My counter-claim is that the majority of hallucinations are basically just user error and the models are, and have been, unequivocally genius.

Now instead of just saying this like I have for awhile, I actually got around to testing it with the release of Gemini 3.

Subscribe now

The Experiment.

Today we are going to follow along with an experiment I ran which I will describe now.

HYPOTHESIS: LLM incorrect responses are due to the model not being given proper context for user queries, NOT a fundamental failing of LLM’s.

EXPERIMENT: Ask the three most recent Gemini models (2.5 Pro/Flash and 3 Pro Preview) to provide financial metrics for given tickers with varying context provided.

PARAMETERS: Each model is tested with the same prompt asking it to return the following list of values/calculations:

We do so for the tickers NFLX/TSLA/GOOGL/AMZN/META to get some variety in reporting.

The correct answer is determined using a data provider that provides these values using human labor/deterministic software. These values are directly reported in the most recent 10-Q or calculated using values within said 10-Q.

There are four different context types tested:

1. No Context - Simply provide the prompt and make sure the model has no search tool or coding capabilities.

2. Google Search Enabled - Enable the model to browse the web for information

3. SEC filing provided - Enable the model to browse the SEC filing directly

4. Correct answers provided - Provide the model with the actual values, and then ask it to return them.

Values provided by the LLM within 1% of actuals are counted as correct to avoid non-material errors from rounding.

Simple enough really, we just ask the LLM for some line item metrics that would be included in the typical financial model. So let’s explore.

The Results:

We’ll break these down both by model type as well as context type and include a bit of discussion on where the model struggled and why.

First off, no context (search disabled, no filing provided)

No Context:

Gemini 2.5 Flash - No Context

Gemini 2.5 Flash with no context scored 0/17 on all 5 stocks for a total of 0/85 and 0% accuracy!

Gemini 2.5 Pro - No Context

Gemini 2.5 Pro with no context scored 0/17 on all 5 stocks for a total of 0/85 and 0% accuracy!

Gemini 3 Pro Preview - No Context

Gemini 3 Pro Preview with no context scored 0/17 on all 5 stocks for a total of 0/85 and 0% accuracy!

Summary:

As it turns out, when the model has no capability to index the information you are asking about, it cannot return to you correct information!

This is a very serious consideration whenever asking a model a question where you care about the accuracy of the output! The model *NEEDS* context for anything outside of training data. Training data is going to be information that was in the model providers corpus of training documents 12-24 months ago.

What’s a bit awkward is the majority of users interact with models via the standard web UI’s/apps. Those products attempt to contextually enable search for users, but they cannot spin proprietary documents out of thin air, search is not always enabled when it should be, and users then complain about “hallucinations” because the model cannot arbitrarily pull up to date estimates from Bloomberg.

Our prompt is essentially just “grab these line items from the most recent quarterly financial statements”. The models were returning information from 2023/2024 instead of Q3 2025, so naturally all incorrect.

SEC Filings Provided

Gemini 2.5 Flash - SEC Filings

Gemini 2.5 Flash had a total accuracy of 82% when provided SEC context!

So where did the model struggle?

depreciationAndAmortization: All models and contexts struggled with calculating D&A similarly to the data provider. I imagine this is due to the somewhat arbitrary nature of the metric as reporting isn’t always standardized. It was hard to trace exactly why this was consistently a problem.

ebit: EBIT of course is not typically reported directly and instead the model must understand was EBIT means, and then find the proper lines to calculate. A common mistake was the model mistaking Operating Income, Net Income, and Income before taxes. When those line items were asked for directly the model could figure it out, but the added abstraction of being asked for EBIT led to failure in properly determining where to start. The model did not get a single one of these correct.

ebitda: Naturally if having issues with D&A and EBIT a model will struggle with EBITDA. The Flash model was unable to get a single one of these correct besides NFLX. This is because both EBIT and EBITDA were ~$300m off, but for EBITDA that error was within 1%. For the most part the model was consistent with combining its D&A and EBIT values, it just made errors on one of those two prior to the calculation.

interestExpense: For TSLA the model returned the interest expense with the wrong sign.

operatingExpense: Operating expenses posed some difficulty for the model as it struggles due to definitions. Operating Expense is typically defined as

OpEx = Total Revenue - Operating Income - Cost of Goods Sold (COGS)

The model was instead calculating total operating costs, aka

Total Operating Cost = Cost of Goods Sold (COGS) + Operating Expenses (OpEx)

It was only doing this occassionally, with about a 50/50 failure rate on 2.5 flash and pro

Every single other metric was correct, specifically there were no errors in the more deterministic line items like revenue/gross profit/operating income/net income/etc.

Gemini 2.5 Pro - SEC Filings

Pro did slightly better at 87% accuracy vs the 82% for flash. Areas of struggle were similar:

ebit and d&a and ebitda: the source of almost all errors, again mostly due to starting with the wrong value for E or D&A being off.

NetIncome for TSLA was marginally off for all models using SEC filings, reported as $1.39b instead of $1.37b. The model actually isn’t wrong here, it’s just that the model is using Net Income as reported, and most providers use Net Income Attributable to Common Stockholders!

operatingExpense suffered the same issue in pro as flash with it mixing up Operating Expense and Operating Costs

Every other metric was correct.

Gemini 3 Pro - SEC Filings

Gemini 3 Pro was a bit better here at 89% accuracy, not a dramatic bump from the 2.5 models.

Areas of struggle:

For META cost of revenue was reported with an extra 0, oops! For this experiment the model did have to convert the values from the filing into my strict formatting. Occasionally you see some errors there.

D&A was again a miss here. Gemini 3 however refused to answer when it was unsure while the other models just guessed.

EBIT was actually spot on with 3. EBITDA it refused to answer for a couple tickers as it did not have D&A.

OpEx it improved and only missed NFLX whereas Flash and Pro missed META/NFLX/AMZN

So only 89% headline accuracy but in reality it just missed D&A and NFLX OpEx with a formatting oopsie on META Cost of Revenue. The mistakes were pretty much entirely definitional/formatting and it refused to answer unknown D&A.

Summary:

SEC filings as a source are an infinite % improvement over no context! Surprise! Even 2.5 Flash is very close to perfect outside of D&A with a little bit of help on definitions. Keep in mind this is without any data cleaning, just telling the model to go find random metrics in a 100 page plus document. The average prompt is also completed in 15-30 seconds for Flash and up to a minute or so for 3, which I wager is quite a bit faster than a human.

When testing re-tries (run the prompt 5 times, take the mode) the formatting errors cease. When testing better definitions I was able to get all models to answer everything correctly besides D&A (and thus EBITDA). D&A is just a bit tricky if its only reported on a YTD basis and I’m just feeding the model Q3 reports, so I don’t blame the model.

So with some tuning we can get ~100% accuracy, not bad!

Search Enabled

Search actually sees a pretty interesting distribution of results and is part of why I believe Gemini 3 is a huge unlock!

Gemini 2.5 Flash - SEC Filings

2.5 Flash Accuracy: 33%

For whatever reason the model was able to figure out a large portion of META and GOOGL, then failed everything else. It was able to get revenue for every stock though! For META and GOOGL it was able to get most of the higher level metrics but fell apart on similar areas as with SEC context.

I would not trust 2.5 Flash with search for pretty much anything, which makes sense as it’s not intended to be a reasoning model with sophisticated tool use.

Gemini 2.5 Pro - SEC Filings

2.5 Pro Accuracy: 59%

Pro search was pretty evenly distributed with slight underperformance for AMZN and overperformance for TSLA. META/NFLX/GOOGL were all the same.

Similar error distribution as with the SEC filings but also the occasional random error on things like gross profit or income before tax that even Flash 100%’s with SEC filing context.

Revenue and SG&A were 100% accurate but other lower level metrics just have random fails. Taking the mode of 5 requests will help. Part of the issue with 2.5 Pro is that sources can be total junk and the browse tool that actually pulls context from Google searches also fails occasionally.

On the bright side, 2.5 Pro starts to not return an answer with search if unsure, which it doesn’t do with SEC filings.

Gemini 3 Pro - SEC Filings

3 Pro Accuracy: 86%

Gemini 3 Pro makes it much much less punishing to not provide context, which is super super important!

Most users are bad at providing context. Gemini 3 pro seems to be much better at tool use with dramatically lower errors! D&A and EBITDA are about 50/50 with 3 search. EBIT was perfect.

Interest Expense and Interest Income for META did not return a value, which is promising for recognizing it doesn’t know what’s going on.

Other than that, it had a slight mistake with OpEx by instead calculating total cost of operations for AMZN and was perfect on the other metrics.

If we ignore D&A/EBITDA that’s a 96% accuracy rating with one issue being a definition oopsie! Near perfect

Asking the model later specifically for interest income it was able to one shot it, so likely just a bit of a hiccup and again taking the mode of 3-5 answers is a great way to basically eliminate the small mistakes.

Summary:

Search matters a lot for context with the web UI, which is how the majority of people will interact with AI. Gemini 3 search is now almost at parity with providing raw context (specifically for SEC filings, many other use cases it will fail).

Now the model can still fail to properly enable search, but if we naively assume accuracy goes from 59% to 86% that’s a ~65% reduction in “hallucinations”. Pretty big change for the average user!

Source Context Provided:

Gemini 2.5 Flash - Source Context

Gemini 2.5 Flash with source context scored 17/17 on all 5 stocks for a total of 85/85 and 100% accuracy!

Gemini 2.5 Pro - Source Context

Gemini 2.5 Pro with source context scored 17/17 on all 5 stocks for a total of 85/85 and 100% accuracy!

Gemini 3 Pro - Source Context

Gemini 3 Pro with source context scored 17/17 on all 5 stocks for a total of 85/85 and 100% accuracy!

Summary:

The basic lesson here is that context is a super power and the closer you can get to perfect source context the closer you get to perfect recall! No context, no recall!

It’s imperative to provide full explanations for what you want, full data for what you want to be explored, and retry if something looks off.

Also, never argue with a model! Once it outputs something wrong, just try to identify “why” and adjust your prompt and ask again. Arguing, especially when something is impossible, will just produce hallucinations.

Fin:

If exploring model intricacies is not your idea of a good time and you’d prefer to just research stocks, you can use the tool I’m building with Clarity AI which manages context, prompting, and model settings for you, fine tuning to achieve better results!

It now also comes with fun graphics

Beyond self promoting, I can’t emphasize enough how much AI usage will shift over the next 6-18 months. It’s clear from these tests that tools like Search provide infinitely better results for some use cases, and search is just the beginning. Model intrinsic capabilities and tool calling capabilities are rapidly scaling, dramatically increasing usability in repeatable enterprise workflows.

The world will look very different by 2027.

Cheers.

Gemini 3 is the new State of the Art (SOTA) model, but what does that mean for finance and the stock? I’ll explore Gemini 3 more directly likely in a later article as I’m also exploring a basic benchmark for financial use cases like modeling/research/etc since for whatever reason the official Gemini release has vending machine management benchmarks but not any financial ones! So stay tuned for the full deep dive as I continue to play with it. I will say that my incessant “You don’t own enough $GOOGL” has gone pretty well thus far and I imagine will continue to do so!

Before doing the full Gemini article, I’m going to shill for Clarity and the AI tools I’m building.

Clarity Shilling:

Clarity AI now has the capability to utilize Gemini 3 for pre-defined financial research, so you can try it out directly on some use cases relevant to you before it hits the web client (Gemini 3 is currently only available in AI studio or via API)

If you’d like to try it out and follow along feel free to test one of our 3 public workflows by clicking the hyperlinks below. All reports are currently free!

Daily Stock Brief

Stock Primer

Mag-7 Capex Rundown

As a note, this is the first time this is working publicly and still very early feature wise, so if you encounter any bugs, have any suggestions on features or data to include, or just want to chat, feel free to reach out via Twitter or email.

Let’s go through the Daily Stock Brief together. As of now it should support all NASDAQ 100 stocks plus a select few others like CVNA/CDLX/HIMS and such. In the future the UI will do a better job at telling you if the ticker you enter is invalid, for now if you enter something not supported you may get some junk in your inbox.

So let’s try it out with GOOGL.

You can now create an account, press start this report workflow, and within 5-10 minutes receive high quality reporting from the new Gemini 3 SOTA model. For this workflow, it also has automatic scheduling.

So if you click start and enter a supported ticker, then every day at 9:00am EST it sends you an email report covering relevant news/filings/conferences/sell-side updates/etc from the past 24 hours.

I will say that now with Gemini 3 these reports are probably better than just about any junior you can find, and if you don’t believe me you can try it yourself, or read the Google brief I’ve attached at the bottom of this article.

Currently the site supports the three pre-made workflows I’ve created, which I’ll be expanding on in the near future. My goal is to create a vast library of public workflows covering numerous aspects of the analyst job, letting users automate away a bunch of the standard analysis so they can focus on their proprietary research stack.

You can also create your own workflows and either keep them private or share them publicly. At some point I’d like to enable users to bill subscriptions to proprietary workflows, so if you’re a budding analyst feel free to tinker with the creation tool which we’ll explore more below.

Clarity Workflow Creation:

Creating workflows is a little janky currently as it’s actually a quite complicated engineering problem, so I’ll provide a bit of a tutorial here as well as discussing planned future specs.

First off, you can find the create page here if you’d like to follow along

This section is rather self explanatory. I would add that Name/Description/Tags can be used to filter and search workflows as the library grows, so if you want to be a power user, be descriptive!

User inputs can be used as variables so for example instead of having 500 different flows to analyze SPY constituents, we can have one that allows users to insert specific tickers.

Scheduling currently supports on-demand runs or scheduled runs. On demand runs will send you a report whenever you click start within 5-10 minutes. Recurring runs will run immediately then again once per day/week/month at the scheduled time. This section is currently a high priority to allow more customization in the future. I’d like to add more granularity so you can do things like every Monday, each quarter end, or event driven things like government data releases, earnings dates, price changes, etc.

If you have any suggestions for scheduling please reach out!

Next up is the query creation itself. Currently each workflow can host up to 10 queries. If for whatever reason you need more, let me know.

The workflow itself once generated will be sent as one email. Each query is seperated in that email based on the label. The Daily Stock Brief for example has query 1 as The “Core Company” Monitor so it creates a nice little separator in the email.

Typically model accuracy improves when breaking requests up into numerous queries with more isolated context. If you ask about daily stock price and volume in the same query as asking for in depth research, the model is more likely to hallucinate than doing those separately.

You can also choose the AI model to run the query with and it currently supports Gemini 2.5 Flash/Pro as well as Gemini 3.0 Pro Preview. I’d love to add more models here as time goes on, especially if I get around to making a finance benchmark it’d be interesting to compare.

Query specific scheduling is also a WIP, but currently only workflow level scheduling is supported.

The advanced settings are where the magic is made. Temperature is basically “creativity” of the model. If you ask the model “1+1=” then you definitely don’t want it getting creative and citing Terrence Howard to say the answer is 3. On the flip side, if you want the model to go find LinkedIn contact info for Lithography experts to have a call with, you might want some more creativity.

Enabling Google Search is also basically a requirement if you need any info you are not directly providing the model. This isn’t foolproof as many sites disallow robots. The SEC for example does not allow models to access the Edgar database via the web, which is why hallucinations for SEC filing related questions are so common! It is however excellent for news aggregation, academic paper finding, some web scraping, and more.

You can try adding specific URL’s as well. They may work, they may not, it’s up to the website host.

File uploads are also supported. Currently these are only really for context, like providing an example of your diligence formatting for the model, or providing proprietary data and asking for analysis. In the future I’d like to enable flows like model updates and the like, but currently xlsx and some other file types require additional tools to allow model edits.

Custom File and Data selectors are where I get to do a lot of work so you don’t have to!

Currently those options support importing recent 10-Q’s/10-K’s/8-K’s/Earnings Transcripts/Stock Data/etc. You can either use user input variables or directly enter a ticker.

It finds the most recent filings either from now or within the date range and adds them to the model context. So if you’d like to analyze say the past year of earnings, you could select 10-K and 10-Q, enter 4, then add another file selector and add 4 earnings transcripts.

The UI here will get some updates soon as docs besides 10-K/10-Q/Earnings transcripts will need some better selection criteria, but feel free to tinker and inevitably encounter bugs.

The purpose of this section is to try and manage as much context as possible for users. As I said models by default can’t access Edgar right? They also can’t access alt data, sell-side reports, Factset/CapIQ/Bloomberg, and many more info sources necessary for the job. My goal is to get as much of that information into these context windows as possible, which dramatically reduces hallucinations and enables workflows not possible via the web client!

If there’s a data source you’d like to see added, or have any thoughts on the UI/UX of this section, shoot me a DM. Tentatively I’d love to add the CVNA/KMX alt data I produce as well as prediction market data on earnings/fed events/macro next, and work my way up to things like credit card data and the like.

Fin:

Anyways, hope you enjoyed the brief rundown of Clarity and give it a shot with the new Gemini 3.0 upgrades!

If you burn through a bunch of my money and I have to paywall it, I’ll consider it a success and explore the best ways to continue granting as much access as possible to as much data as possible.

Feel free to subscribe for my more targeted Gemini breakdown in the future and other such articles. Cheers!

Subscribe now

Google Brief 11/19/2025:

Analysis of GOOGL (Alphabet Inc.) News & Filings

Date: November 19, 2025

1. SEC Filings

• Company Filings: There were no direct filings (8-K, 10-Q, or Form 4) by Alphabet Inc. or its executives in the last 24 hours. The most recent company-specific filings occurred earlier in November (e.g., Form 144s and Form 4s filed around Nov. 12–17).

• Institutional Filings (13F): Several institutional investors (e.g., Soltis Investment Advisors, Golden State Wealth Management, Coleford Investment Management) filed Form 13Fs in the last 24 hours disclosing changes to their GOOGL holdings for the quarter ended September 30, 2025. These are standard quarterly disclosures by funds, not corporate actions by Alphabet.

2. Product News

• Launch of Gemini 3: Google officially launched Gemini 3, its newest and most advanced AI model, on November 18–19.

  • Integration: The model is immediately being embedded into Google Search via a new “AI Mode” for premium subscribers (AI Pro and Ultra plans).

  • Capabilities: Executives describe it as having “state-of-the-art reasoning” and “deep multimodal understanding.”

• “Antigravity” Platform: Alongside Gemini 3, Google launched Antigravity, a new AI-first coding platform. It utilizes the Gemini 3 Pro model to allow AI agents to act as independent software developers, capable of debugging and verifying code autonomously.

• Play Store “Best of 2025”: Google announced its “Best of 2025” lists for apps and games on the Play Store, highlighting a surge in AI-powered applications.

3. Corporate Action

• DOJ Antitrust / Chrome Divestiture (Major Event): Reports confirmed that the U.S. Department of Justice (DOJ) is formally asking a federal judge to force Google to sell off its Chrome browser.

  • Context: This follows a roughly year-long trial where a judge ruled Google maintained an illegal monopoly in search. The DOJ is also seeking remedies related to data decoupling and restrictions on Android, though it reportedly stopped short of demanding a sale of Android itself.

• Stock Performance: Despite the regulatory threat, GOOGL shares hit an all-time high, crossing the $300 mark for the first time. The surge is attributed to the positive reception of the Gemini 3 launch, which investors seemingly prioritized over the antitrust news.

4. Sentiment

Tone: Polarized (Bullish Market / Bearish Regulatory)

• Bullish (Market/Product): Investors and analysts are reacting euphorically to the Gemini 3 product cycle, viewing it as a decisive answer to competitors like OpenAI.

  • Headline: “Alphabet (GOOGL) stock jumps over 6% to hit $300 for the first time in history as Gemini 3 powers Alphabet’s biggest breakout of the year” — The Economic Times

  • Headline: “Google soars on positive reception for Gemini 3” — Robinhood/Financial Feeds

• Bearish (Regulatory): The regulatory landscape has darkened significantly with the DOJ’s “nuclear option” proposal to break up the company.

  • Headline: “DOJ Pushes for Google Chrome Sell-Off to Break Search Monopoly” — Tech Research Online

  • Headline: “DOJ Sways Judge To Force Sale Of Google’s Chrome” — MediaPost

Date: 2025-11-19 Ticker: GOOGL (Alphabet Inc.) Focus: Analyst ratings, price targets, and institutional flow

Analyst Activity & Significant Market Commentary

Summary: Alphabet (GOOGL) shares are trading at all-time highs today (touching ~$294-$300 levels) driven by a “perfect storm” of positive catalysts: a high-profile institutional endorsement from Warren Buffett, the launch of its new state-of-the-art AI model (Gemini 3), and sweeping relief across the tech sector regarding antitrust regulatory risks following a major court victory for Meta Platforms yesterday.

Upgrades/Downgrades

• Loop Capital: Upgraded to Buy from Hold.

  • Price Target: Raised to $320 from $260.

  • Reasoning: The firm cited increasing confidence in Google’s AI product cycle, specifically noting that the new Gemini 3 model effectively closes the gap with competitors, and that search revenue remains resilient despite “generative answer” fears.

• Roth Capital: Raised Price Target to $265 (Reiterated Buy).

• Consensus Sentiment: Most analysts are reiterating “Strong Buy” ratings today, pointing to the removal of the “breakup” tail risk and the validation of valuation by Berkshire Hathaway’s entry.

Commentary

• DA Davidson (Gil Luria): In a note released today following the Gemini 3 launch, Luria described the model as “deserving of the top overall spot” in the AI landscape. He highlighted its “outstanding benchmark scores” and superior reasoning capabilities compared to GPT-5 competitors, effectively debunking the narrative that Google is lagging in the AI arms race.

• Stratechery (Ben Thompson): Published analysis on “Gemini 3, Winners and Losers,” suggesting that Google’s integration of this new model into its ecosystem (Search, Workspace) creates a defensive moat that is proving harder for startups to disrupt than previously feared.

Market Chatter

• “The Buffett Seal of Approval”: The dominant discussion on trading desks is the Berkshire Hathaway Q3 13F filing (disclosed earlier this week but dominating today’s flow), which revealed a massive ~$4.3–$4.9 billion stake in Alphabet. Traders view this as a historic “value” signal, marking Buffett’s first major pure-play entry into the company and validating the stock’s valuation relative to other “Mag 7” peers.

• Antitrust Relief Rally: Market chatter is heavily focused on the Department of Justice (DOJ) vs. Big Tech narrative shifting in favor of the companies. Following a federal judge’s ruling yesterday (Nov 18) dismissing the FTC’s antitrust case against Meta (Instagram/WhatsApp), investors are pricing in a significantly lower regulatory risk premium for Google. This follows Google’s own victory in September 2025, where Judge Amit Mehta ruled against the DOJ’s request to force a divestiture of the Chrome browser, a decision that is being re-circulated and celebrated today in light of the Meta news.

• Rumors: There is speculation that the DOJ may scale back its remaining remedies phase for Google Search given the recent string of legal defeats for regulators, further fueling the rally.

Institutional Flow

• Berkshire Hathaway: The headline flow is the accumulation of ~17.8 million shares by Berkshire, making it a top-10 holding for the conglomerate. This has triggered “copycat” buying from other value-oriented funds today.

• Options Activity: Bullish. There is “unusual options activity” reported today with heavy volume in November and December Calls (strikes $300 and $310), suggesting traders expect the momentum to carry through year-end. Sentiment indicators show a “significant flip” to bullishness in the last 48 hours.

Based on the current date of Wednesday, November 19, 2025, and the latest search results, here is the analysis for Alphabet Inc. (GOOGL).

Top Direct Competitors

1. Microsoft (MSFT): Direct rival in cloud (Azure vs. Google Cloud), search (Bing vs. Google), and AI (OpenAI partnership vs. Gemini).

2. Meta Platforms (META): Primary competitor in digital advertising and consumer AI.

3. Amazon (AMZN): Leading competitor in cloud computing (AWS) and a growing rival in digital advertising.

4. Nvidia (NVDA): While primarily a hardware supplier, Nvidia competes for AI capital and sovereignty; its performance is the bellwether for the AI industry that drives Google’s valuation.

1. Competitor Earnings/Guidance

• Nvidia (NVDA) – Reporting Today (After Market Close):

  • Event: Nvidia is set to report Q3 earnings after the bell today, Nov 19, 2025. This is described as “Wall Street’s Super Bowl,” with analysts expecting EPS of ~$1.25 on ~$55B revenue.

  • Read-Through for GOOGL:

    • Strength: A “beat and raise” from Nvidia would confirm that demand for AI infrastructure remains insatiable. This validates Google’s massive CapEx spending (which investors have fretted over) and suggests continued strong growth for Google Cloud, as enterprises rent GPUs/TPUs to run models.

    • Weakness: Any sign of slowing demand or “air pocket” in Nvidia’s orders would severely punish GOOGL stock. It would signal an “AI Bubble” pop, validating CEO Sundar Pichai’s own recent warning about market “irrationality.”

• Target (TGT) – Reported Today:

  • Result: Target reported a profit beat but revenue miss and warned on the holiday outlook.

  • Read-Through: Weakness in consumer retail spending is a negative signal for Google’s advertising revenue, particularly for retail/shopping ads leading into Q4.

2. Competitor Products

• OpenAI (Microsoft backed) – ChatGPT-5.1:

  • News: OpenAI recently released ChatGPT-5.1 (Nov 12, 2025).

  • Threat: This places immediate pressure on Google’s retention of search traffic.

  • GOOGL Response: Google launched Gemini 3 yesterday (Nov 18, 2025). Sundar Pichai claims it outperforms ChatGPT-5.1 and Claude Sonnet 4.5 on reasoning benchmarks.

  • Analysis: The rapid counter-punch with Gemini 3 significantly de-risks the “Google is falling behind” narrative. The “product war” is now a stalemate rather than a rout, which is a relative positive for GOOGL shares.

• Amazon – “Project Prometheus” & Job Cuts:

  • News: Reports indicate Jeff Bezos has launched a new AI startup, “Project Prometheus,” while Amazon cuts 14,000 jobs to pivot toward AI.

  • Threat: While Bezos’ startup is separate, Amazon’s aggressive restructuring suggests a leaner, more AI-focused AWS is coming to battle Google Cloud.

3. Industry News & Read-Throughs

• Antitrust Victory for Meta (Positive Second-Order Effect):

  • News: On Nov 18, 2025, a federal judge ruled that Meta is not a monopoly in social networking and dismissed the FTC’s attempt to break up Instagram/WhatsApp.

  • Mechanism: This is a massive positive read-through for GOOGL. Google is currently entangled in similar DOJ antitrust battles regarding Search and AdTech. The judiciary’s high bar for proving monopoly power (as seen in the Meta ruling) suggests Google may escape a breakup or severe structural remedies, lowering the “regulatory discount” on the stock.

• “AI Bubble” Commentary:

  • News: Alphabet CEO Sundar Pichai spoke to the BBC (Nov 18/19), admitting that “irrationality” has crept into the AI market and “no company is immune” to a shakeout.

  • Mechanism: While honest, this creates short-term jitter. If the CEO of the company leading the AI charge is cautious, investors may rotate out of high-beta tech. However, it also positions Google as the “adult in the room” with a sustainable balance sheet compared to unprofitable startups.

Summary Sentiment for GOOGL

• Short-Term: Neutral/Cautious. The market is paralyzed waiting for Nvidia’s earnings tonight.

• Relative Value: Bullish. Google’s successful launch of Gemini 3 (defending its moat) combined with Meta’s antitrust win (lowering regulatory risk) makes GOOGL look attractive relative to peers, especially given it is trading at a lower multiple (approx. 26x-28x) than the broader AI basket.

The following analysis covers materially relevant economic, regulatory, and commodity updates for Alphabet Inc. (GOOGL) from the last 24 hours as of November 19, 2025.

Economic Prints

• Data Point: Atlanta Fed GDPNow Estimate (Q3 2025)

  • Release: Updated today, November 19, 2025.

  • Print: 4.2% (up from 4.1%).

  • Impact on GOOGL: This robust growth forecast signals a strong macroeconomic environment, which directly correlates with corporate willingness to spend on digital advertising—Google’s core revenue driver. A growing economy supports sustained consumer demand for hardware (Pixel/Nest) and services (YouTube Premium).

• Data Point: U.S. International Trade Balance (Aug 2025)

  • Release: November 19, 2025.

  • Print: Deficit narrowed to $59.6 billion.

  • Impact on GOOGL: The narrowing deficit, driven by increased exports, suggests healthy cross-border commerce. For a global platform like Google, strong trade activity often parallels increased demand for international cloud and enterprise services.

Regulation

• Major Ruling: Meta Platforms Antitrust Victory (Relevant Precedent)

  • Update: A federal judge dismissed the FTC’s antitrust lawsuit against Meta, ruling the company does not hold a monopoly in personal social networking.

  • Relevance to GOOGL: This is a highly positive read-through for Alphabet. The court accepted the argument that “new entrants” (like TikTok) constitute real competition. Google can leverage this same logic in its own antitrust battles (specifically regarding Search and Ad Tech) by pointing to the rapid rise of Generative AI (e.g., OpenAI, Perplexity) as evidence that its market dominance is not unassailable.

• Legislative Shift: EU AI Act & GDPR Adjustments

  • Update: Reports today indicate the European Commission may “water down” strict provisions of the AI Act and GDPR under pressure from the U.S. and Big Tech lobbying.

  • Relevance to GOOGL: A relaxation of these rules would be a material tailwind for Google DeepMind and Gemini development, lowering compliance costs and data usage barriers in the critical European market.

Inputs/Commodities

• Commodity: Crude Oil (WTI & Brent)

  • Movement: Prices fell approximately 3% today (WTI to ~$59/bbl).

  • Reason: Rising U.S. crude inventories and oversupply concerns outweighed geopolitical risks in Ukraine.

  • Impact on GOOGL: Bullish. Lower energy prices reduce operating expenses (OpEx) for Google’s massive data center footprint, which consumes vast amounts of power for AI training and inference.

• Rate: 10-Year U.S. Treasury Yield

  • Movement: Held steady/rose slightly to ~4.11% - 4.15%.

  • Impact on GOOGL: Neutral to Slight Headwind. Yields remain elevated but stable. While high rates keep the cost of capital (WACC) higher for growth initiatives, the lack of a sudden spike today provides a stable valuation floor for long-duration tech assets like GOOGL.

The Lead GOOGL is staging a structural breakout today, decoupling from the broader tech sell-off driven by the dual catalysts of the Gemini 3.0 launch and a massive $4.3B Berkshire Hathaway stake revealed in 13Fs.

Price Action Context

• Relative Strength: GOOGL is trading up ~3-4% intraday (currently ~$294), significantly outperforming the Nasdaq and “Magnificent 7” peers, which are largely red on sector-wide weakness.

• Technical: The stock hit an all-time high of $304 earlier in the session before consolidating. The ability to hold the $294 level on heavy volume (23M+ shares) despite the broader market drag confirms strong institutional accumulation.

Key Fundamental Developments

• Product Cycle (Gemini 3.0): Google officially launched Gemini 3.0, with early benchmarks and reviews characterizing it as a “step change” in reasoning capabilities, effectively reclaiming the AI performance crown from competitors (OpenAI/Anthropic).

• Institutional Validation (The “Buffett Put”): Berkshire Hathaway’s 13F revealed a new 17.8M share position (~$4.3B). This is a critical sentiment shifter, validating GOOGL’s valuation (~22x forward P/E) as a “value” play within the AI hyper-growth theme.

• Cloud Momentum: Sentiment remains buoyed by recent data showing Google Cloud accelerating to a 34% growth rate ($50B annualized run rate), quieting fears that AI capex spending isn’t yielding immediate ROI.

The “So What?” This is not a noise event; it is a regime shift for the stock. For the past year, the bear case on GOOGL was “search decay” and “falling behind in AI.” Today’s news dismantles both: Gemini 3.0 proves technical leadership, and Buffett’s entry signals deep value support. We are seeing a flight to quality where GOOGL is viewed as the “safe” AI compounder (Value + Growth). We recommend using the intraday consolidation around $294 to increase exposure; the “Buffett Floor” dramatically improves the risk/reward profile here.

The goal here really is to make it much easier to increase both breadth and depth of analysis. Like for a typical analyst I can’t imagine more than 10-20% of time spent is actually unique value additive work instead of something that’s already been done before. Hypothetically a lot of that can be compressed down via AI. The grunt work to do so really is hooking up the AI to the right context, defining the right prompts, scheduling out that work, etc.

Thus far I’ve gotten some helpful feedback and totally redesigned the experience. I’m hoping to be able to publicly test it at scale in the next couple weeks as the backend gets more stable (SWE is still not entirely solved by AI!)

What I’d also like to enable once the core filings/transcripts/etc basic data is integrated, is alternative data sources, utilizing users FactSet/CapIQ/BBG data, etc. There’s a lot that AI unlocks in terms of data collection and data aggregation that wasn’t always possible. This post for example could easily include GPU spot price scrape data, GPU per watt performance data, new build contracts popping up in government databases, etc. The world is y(our) oyster.

Anyways, here’s a sample of the CapEx report which users could run on demand at any time, or schedule for automatic generation. Cheers.

Subscribe now

GOOGL:

Alphabet Inc. (GOOGL) Capital Expenditure Analysis

As of October 31, 2025

1. Quarterly Capex Trend (Last 12 Months)

Based on the “Purchases of property and equipment” line item from the Consolidated Statements of Cash Flows and management commentary, the capital expenditure for each of the last four quarters is as follows:

• Q4 2024 (Three months ended Dec 31, 2024): $14.0 billion

• Q1 2025 (Three months ended Mar 31, 2025): $17.2 billion

• Q2 2025 (Three months ended Jun 30, 2025): $22.4 billion

• Q3 2025 (Three months ended Sep 30, 2025): $24.0 billion

The quarterly trend shows a significant and accelerating rate of investment throughout the past year.

Total Trailing Twelve Months (TTM) Capex: $77.6 billion

2. Management Guidance vs. Actuals

Management’s guidance for full-year 2025 capital expenditures has increased substantially over the course of the year.

• Initial Guidance (as of Q4 2024 earnings): The initial quantitative guidance was for approximately $75 billion for the full year 2025. CFO Anat Ashkenazi stated in the Q4 2024 earnings call, “We expect to invest approximately $75 billion in CapEx in 2025.” This guidance was reiterated in the Q1 2025 earnings call.

• Guidance Changes:

  • Q2 2025 Update: Guidance was raised by $10 billion. In the Q2 2025 earnings call, the CFO stated, “we now expect to invest approximately $85 billion in CapEx in 2025, up from a previous estimate of $75 billion.” This increase was attributed to stronger demand for Cloud products and an acceleration in data center construction.

  • Q3 2025 Update: Guidance was raised again. The CFO noted in the Q3 2025 earnings call, “We now expect CapEx to be in the range of $91 billion to $93 billion in 2025, up from our previous estimate of $85 billion.”

• Current Full-Year Guidance: As of the latest report (Q3 2025), the current guidance for full-year 2025 capex is a range of $91 billion to $93 billion.

3. Primary Drivers of Capex (The “Why”)

The documents consistently identify investment in technical infrastructure as the primary driver of capital expenditures. This infrastructure is essential for supporting the company’s company-wide artificial intelligence (AI) initiatives.

Key Investment Areas: Management commentary specifies that the bulk of this investment is directed towards:

• Servers: This is consistently cited as the largest component of technical infrastructure spending.

• Data Centers: Includes construction and outfitting of new and existing facilities.

• Networking Equipment: Necessary to connect the servers and data centers.

Strategic Rationale and Key Commentary: The strategic rationale is to build the necessary computational capacity to power AI advancements across all of Alphabet’s businesses, with a notable acceleration driven by enterprise demand.

• Supporting AI Across All Segments:

“Our reported CapEx in the fourth quarter was $14 billion, primarily reflecting investments in our technical infrastructure, with the largest component being investment in servers, followed by data centers, to support the growth of our business across Google Services, Google Cloud, and Google DeepMind.” - Anat Ashkenazi, Q4 2024 Earnings Call Transcript

• Meeting Surging Cloud Customer Demand: The narrative evolved throughout 2025 to emphasize that the acceleration in spending was a direct response to stronger-than-anticipated demand from Google Cloud customers for AI services.

“Our updated outlook reflects additional investment in servers, the timing of delivery of servers and an acceleration in the pace of data center construction, primarily to meet cloud customer demand.” - Anat Ashkenazi, Q2 2025 Earnings Call Transcript “we’re continuing to invest aggressively due to the demand we’re experiencing from cloud customers, as well as the growth opportunities we see across the company.” - Anat Ashkenazi, Q3 2025 Earnings Call Transcript

• Powering Next-Generation Models and Products: This infrastructure is the foundation for developing and deploying more advanced AI models like Gemini.

“We are set up well for continued growth... In Search, AI overviews are now available in more than 100 countries... we have seven Products and Platforms with over 2 billion users and all are using Gemini.” - Sundar Pichai, Q4 2024 Earnings Call Transcript

The narrative has remained consistent in its focus on AI infrastructure, but the primary justification for accelerating that spend has shifted from a broad-based need to a specific and urgent need to meet intense Google Cloud customer demand.

4. Segmental Capex Breakdown (Services vs. Cloud vs. Bets)

The provided documents do not offer a quantitative breakdown of capital expenditures by segment. However, management commentary allows for a qualitative synthesis of how these investments support each area. The vast majority of technical infrastructure capex is a shared resource that benefits both Google Services and Google Cloud.

• Google Services (Search, YouTube, Ads): The infrastructure investments directly power the computationally intensive generative AI features being integrated into core consumer products. This includes running AI Overviews in Search, which now serve over 1.5 billion users monthly, and powering AI tools for YouTube creators. The CFO stated capex supports “Google Services” in multiple calls.

• Google Cloud: This segment is a primary beneficiary and the key driver for the acceleration in capex during 2025. The investments build capacity for enterprise clients to train and run their own AI models on Google’s infrastructure (TPUs and GPUs). As stated by the CFO in the Q2 and Q3 transcripts, the increases in the 2025 capex forecast were “primarily to meet cloud customer demand.” Strong backlog growth in the Cloud segment ($49 billion sequential increase in Q3) corroborates this intense demand.

• Other Bets (e.g., Waymo): The massive technical infrastructure spend is not directed at Other Bets. Waymo’s capital needs are distinct and relate to its own hardware. The Q4 2024 transcript notes Waymo is “developing the sixth-generation Waymo Driver, which will significantly lower hardware costs,” indicating a focus on reducing its own capex, separate from the server and data center build-out.

In summary, the capital expenditure is a foundational investment in AI compute that is leveraged by both the consumer-facing Services segment and the enterprise-focused Cloud segment, with Cloud’s surging demand being the primary catalyst for the increased spending in 2025.

5. Future Outlook

Management has provided clear qualitative guidance indicating that the current high level of investment will not only continue but increase into next year.

• Expected Trend for 2026: Management expects a significant increase in capital expenditures for 2026.

“Looking out to 2026, we expect a significant increase in CapEx, and we’ll provide more detail on our fourth-quarter earnings call.” - Anat Ashkenazi, Q3 2025 Earnings Call Transcript “Looking out to 2026, we expect a further increase in CapEx due to the demand we’re seeing from customers as well as growth opportunities across the company.” - Anat Ashkenazi, Q2 2025 Earnings Call Transcript

• Stated Reasons: The rationale for the continued increase in 2026 is the same as the drivers for 2025: sustained high demand from cloud customers and the pursuit of growth opportunities across the company, all fueled by the advancement of AI.

6. Summary of Analysis

Alphabet is in the midst of a massive and accelerating capital investment cycle, fundamentally driven by the strategic imperative to build leadership in the era of artificial intelligence. Over the past twelve months, capex has surged from $14.0 billion in Q4 2024 to $24.0 billion in Q3 2025, totaling $77.6 billion. This pace has caused management to repeatedly raise its full-year 2025 guidance from an initial estimate of $75 billion to a current range of $91-$93 billion, signaling that the demand for computational power is outstripping original forecasts.

The “story” of this spending is twofold. These investments in technical infrastructure—primarily servers and data centers—are the engine for Alphabet’s next phase of growth. They support the company’s consumer-facing Google Services segment by powering new, computationally expensive generative AI features in Search and YouTube. Concurrently, and more critically to the accelerated spending, this build-out is essential to meet the overwhelming demand from its Google Cloud customers, who are increasingly turning to Google for the infrastructure needed for their own AI training and inference workloads. This enterprise demand has become the primary catalyst for the upward revisions in spending plans throughout 2025. This intensive investment phase is set to continue, with management explicitly guiding for another “significant increase” in capex in 2026, underscoring a long-term commitment to securing the foundational compute capacity required to win across both consumer and enterprise AI markets.

META:

Based on my analysis of the provided documents, here is a synthesis of Meta’s capital expenditures.

1. Quarterly Capex Trend (Last 12 Months)

The total capital expenditure (capex), including purchases of property and equipment and principal payments on finance leases, for each of the last four quarters is as follows:

• Q3 2025 (quarter ended September 30, 2025): $19.4 billion (Source: META_Transcript_Q3_2025)

• Q2 2025 (quarter ended June 30, 2025): $17.0 billion (Source: META_Transcript_Q2_2025)

• Q1 2025 (quarter ended March 31, 2025): $13.7 billion (Source: META_Transcript_Q1_2025)

• Q4 2024 (quarter ended December 31, 2024): $14.8 billion (Source: META_Transcript_Q4_2024)

The total capex for the trailing twelve months (TTM) is the sum of these four quarters:

• Trailing Twelve Months (TTM) Capex: $64.9 billion

2. Management Guidance vs. Actuals

Management’s guidance for full-year 2025 capital expenditures has increased throughout the year.

• Initial Annual Capex Guidance: The initial guidance for full-year 2025 was $60 billion to $65 billion, as provided in the Q4 2024 earnings call. (Source: META_Transcript_Q4_2024)

• Guidance Changes:

  • Q1 2025: Guidance was increased to $64 billion to $72 billion. (Source: META_Transcript_Q1_2025)

  • Q2 2025: Guidance was narrowed and raised to $66 billion to $72 billion. (Source: META_Transcript_Q2_2025)

• Current Full-Year Capex Guidance: As of the latest report (Q3 2025), the current guidance has been raised again to $70 billion to $72 billion. (Source: META_Transcript_Q3_2025)

3. Primary Drivers of Capex (The “Why”)

Based on management commentary, the primary driver for this substantial level of capital expenditure is an aggressive, long-term investment in Artificial Intelligence (AI) infrastructure, including servers, data centers, and network capacity. This investment is aimed at establishing Meta as a leader in AI and building what the company refers to as “superintelligence.”

Key Commentary on Strategic Investments:

• The company’s strategy is to “aggressively front-load building capacity” to be prepared for the rapid advancement of AI. As CEO Mark Zuckerberg stated, “if superintelligence arrives sooner, we will be ideally positioned for a generational paradigm shift and many large opportunities. If it takes longer, then we’ll use the extra compute to accelerate our core business -- which continues to be able to profitably use much more compute than we’ve been able to throw at it.” (Source: META_Transcript_Q3_2025)

• These investments are massive in scale. Zuckerberg noted, “we’re building out several multi-gigawatt clusters. Our Prometheus cluster is coming online next year, and we think it’s going to be the world’s first gigawatt-plus cluster. We’re also building out Hyperion, which will be able to scale up to 5 gigawatts over several years...” (Source: META_Transcript_Q2_2025)

• The investments support both long-term generative AI efforts and the immediate needs of the core business. CFO Susan Li confirmed, “We expect CapEx growth in 2025 will be driven by increased investment to support both our generative AI efforts and our core business. The majority of our CapEx in 2025 will continue to be directed toward our core business.” (Source: META_Transcript_Q4_2024)

Evolution of the Narrative:

Over the past four quarters, the narrative has evolved from a general focus on AI infrastructure to a more ambitious and urgent push toward building foundational “superintelligence.”

• Initially (Q4 2024), the focus was on supporting genAI and the core business.

• By Q1 2025, the company announced it was “accelerating some of our efforts to bring capacity online more quickly.”

• In Q2 2025, the ambition was clarified with specific multi-gigawatt cluster projects and the explicit goal of building “personal superintelligence for everyone.”

• By Q3 2025, the rationale crystallized into a deliberate strategy of “front-loading” investments to prepare for the earliest possible arrival of superintelligence, with management noting that compute needs have “continued to expand meaningfully, including versus our own expectations last quarter.”

4. Segmental Capex Breakdown (FoA vs. RL)

The provided documents do not offer a direct quantitative breakdown of capital expenditures between the Family of Apps (FoA) and Reality Labs (RL) segments. However, management’s qualitative commentary consistently indicates that the vast majority of this infrastructure spending supports the core FoA business.

• Majority Supports FoA: CFO Susan Li repeatedly stated that “The majority of our CapEx... will continue to be directed toward our core business.” (Sources: META_Transcript_Q4_2024, META_Transcript_Q1_2025)

• Shared Infrastructure: The capex is largely for shared infrastructure (servers, data centers) that powers all of Meta’s initiatives. However, the most immediate and profitable returns are seen in FoA. These investments directly enhance FoA’s advertising systems and content recommendation engines, which drive engagement and revenue. For example, improvements in AI recommendation systems have led to a “5% increase in time spent on Facebook and 6% on Instagram just this quarter.” (Source: META_Transcript_Q2_2025)

• Supporting Long-Term R&D: While FoA is the primary beneficiary, this same infrastructure is foundational for the long-term, more speculative work being done in Reality Labs (e.g., Metaverse, AI Glasses) and at the “Meta Superintelligence Labs” (MSL). The CFO noted that for 2026, capex growth will come from “MSL, core AI, as well as non-AI spend... but the MSL AI needs are growing the most.” (Source: META_Transcript_Q3_2025). This indicates that while the base of the capex supports FoA, an increasing portion of the growth in capex is dedicated to long-term, foundational AI research that will benefit both segments in the future.

5. Future Outlook

Management has provided a clear, albeit qualitative, outlook indicating that capital expenditures are expected to increase substantially again next year.

• 2026 Capex Trend: For 2026, management’s “current expectation is that capex dollar growth will be notably larger in 2026 than 2025.” (Source: META_Transcript_Q3_2025)

• Stated Reasons for Future Outlook: This continued acceleration is driven by compute needs that are expanding faster than previously anticipated. The CFO stated, “As we have begun to plan for next year, it’s become clear that our compute needs have continued to expand meaningfully... We expect to invest aggressively to meet these needs both by building our own infrastructure and contracting with third party cloud providers.” (Source: META_Transcript_Q3_2025) These investments are required to support initiatives across Meta Superintelligence Labs, core AI, and general infrastructure.

6. Summary of Analysis

Over the past twelve months, Meta has embarked on an aggressive and accelerating capital expenditure program, spending $64.9 billion and signaling even larger investments ahead. The quarterly spending trend has steadily increased from $14.8 billion in Q4 2024 to $19.4 billion in Q3 2025. This escalation is reflected in management’s full-year 2025 guidance, which was raised three times during the year from an initial range of $60-65 billion to its current level of $70-72 billion.

The strategic driver behind this massive investment is a company-wide pivot to establish a commanding lead in Artificial Intelligence. The narrative has sharpened over the past year from general AI investments to a clear and ambitious goal: front-loading the construction of unparalleled compute capacity (including multi-gigawatt data centers) to be prepared for the arrival of “superintelligence.” While this infrastructure is foundational for long-term projects in Reality Labs and the newly formed Meta Superintelligence Labs, management consistently justifies the spending with its immediate, profitable application in the core Family of Apps segment. The majority of this capex directly powers the AI recommendation and advertising systems that are measurably increasing user engagement and driving revenue, providing a strong financial rationale and backstop for the ambitious long-term vision.

Looking forward, Meta has signaled that this intense investment cycle is not slowing down. Management expects capital expenditures in 2026 to grow by a dollar amount “notably larger” than the substantial increase seen in 2025. This outlook underscores that Meta’s compute requirements are expanding even faster than its own aggressive plans had anticipated, cementing its strategy to leverage its financial strength to build a foundational infrastructure advantage for the next era of AI and computing.

AMZN:

Here is a detailed analysis of Amazon’s capital expenditures based on the provided documents.

1. Quarterly Capex Trend (Last 12 Months)

Based on the Consolidated Statements of Cash Flows in the provided 10-K and 10-Q filings, the total capital expenditure for each of the last four quarters, defined as “Purchases of property and equipment” less “Proceeds from property and equipment sales and incentives,” is as follows (all figures in millions):

• Q4 2024 (ending Dec 31, 2024):

  • Purchases: $14,925

  • Proceeds: $990

  • Net Capex: $13,935 (Calculated from FY 2024 10-K data less 9M 2024 data provided in the Q3 2025 10-Q)

• Q1 2025 (ending Mar 31, 2025):

  • Purchases: $25,019

  • Proceeds: $764

  • Net Capex: $24,255 (Source: AMZN 10-Q Q1 2025)

• Q2 2025 (ending Jun 30, 2025):

  • Purchases: $32,183

  • Proceeds: $815

  • Net Capex: $31,368 (Source: AMZN 10-Q Q2 2025)

• Q3 2025 (ending Sep 30, 2025):

  • Purchases: $35,095

  • Proceeds: $867

  • Net Capex: $34,228 (Source: AMZN 10-Q Q3 2025)

The total capital expenditure for the trailing twelve months (TTM) is the sum of these four quarters:

• TTM Capex: $13,935 + $24,255 + $31,368 + $34,228 = $103,786 million (or $103.8 billion)

2. Management Guidance vs. Actuals

• Initial Annual Capex Guidance: In the Q4 2024 Earnings Call Transcript, CFO Brian Olsavsky provided initial commentary for 2025, stating: “Capital investments were $26.3 billion in the fourth quarter, and we think that run rate will be reasonably representative of our 2025 capital investment rate.” This implied a full-year 2025 capex of approximately $105 billion ($26.3B x 4).

• Changes in Guidance:

  • In the Q1 2025 transcript, management noted plans “to bring on an increasing amount of capacity in the back half of the year,” signaling that spending would likely ramp up from the Q1 level of $24.3 billion.

  • In the Q2 2025 transcript, the CFO updated this view, stating, “We expect Q2 CapEx to be reasonably representative of our quarterly capital investment rate for the back half of this year.” With Q2 capex at $31.4 billion, this implied a new full-year total of approximately $118.5 billion ($24.3B for Q1 + 3 x $31.4B for Q2-Q4).

• Current Full-Year Capex Guidance: As of the latest report (Q3 2025 Earnings Call), the guidance was updated and made more explicit. The CFO stated: “Looking ahead, we expect our full-year cash CapEx to be approximately $125 billion in 2025...” This represents a significant increase from both the initial and mid-year guidance.

3. Primary Drivers of Capex (The “Why”)

Based on management commentary, capital expenditures are overwhelmingly driven by investments in technology infrastructure for AWS to meet soaring demand for Artificial Intelligence (AI) and core cloud services, with secondary investments focused on optimizing the e-commerce fulfillment network.

Key Commentary and Quotes:

• AWS Infrastructure, AI & Data Centers:

  • The primary driver is consistently identified as AWS. The Q4 2024 transcript notes, “the majority of the spend will be to support the growing need for technology infrastructure. This primarily relates to AWS, including support demand for our AI services.”

  • CEO Andy Jassy has framed this as a critical strategic investment. In the Q3 2025 transcript, he stated: “To enable customers to [run AI workloads in AWS], we need to have the requisite capacity, and we’ve been focused on accelerating capacity the last several months, adding more than 3.8 gigawatts of power in the past 12 months, more than any other cloud provider.”

  • He further elaborated on the urgency and rationale: “You’re going to see us continue to be very aggressive investing in capacity because we see the demand. As fast as we’re adding capacity right now, we’re monetizing it. It’s still quite early and represents an unusual opportunity for customers in AWS.” (AMZN Transcript Q3 2025)

• Custom Silicon (Trainium): A significant portion of the investment is for Amazon’s custom-designed AI chips.

  • CEO Andy Jassy (Q3 2025 Transcript): “This capacity consists of power, data center, and chips, primarily our custom silicon Trainium and NVIDIA... Trainium 2 continues to see strong adoption, is fully subscribed, and is now a multi-billion-dollar business that grew 150% quarter over quarter.”

• Fulfillment & Transportation Network: While a smaller portion of the spend, investment in the stores business is focused on efficiency and speed.

  • CFO Brian Olsavsky (Q2 2025 Transcript): “Additionally, we continue to invest in our fulfillment and transportation network to support growth of the business, improve delivery speeds, and lower our cost to serve by investing in same-day delivery facilities as well as robotics and automation.”

Evolution of the Narrative: Over the past four quarters, the narrative has shifted from a more balanced discussion of investing in both AWS and fulfillment to an intense focus on the massive, urgent opportunity in Generative AI. While fulfillment optimization was a key topic at the beginning of the period, the commentary in the most recent quarters is dominated by the need to aggressively build out AWS capacity (power, data centers, chips) to meet demand that management describes as outstripping current supply.

4. Segmental Capex Breakdown (AWS vs. E-commerce)

The provided documents do not offer a direct quantitative breakdown of capex by business segment. However, management’s qualitative commentary consistently indicates that the vast majority of capital expenditure is for the AWS segment, with the remainder supporting the North America and International e-commerce segments.

• AWS Investment (Growth and Capacity Driven): The commentary positions AWS capex as an investment in growth to meet immediate and future demand for AI and core cloud services. The spending is for foundational infrastructure:

  • Generative AI Infrastructure: Building out massive compute capacity, including data centers, power, and procuring/developing both third-party (NVIDIA) and custom (Trainium) AI chips. Management emphasizes that they have “more demand than we have supplied for at the moment” (Q2 2025 Transcript), justifying the accelerated pace of investment.

  • Core Cloud Services: Expanding the global data center footprint to support the ongoing migration of enterprise IT workloads to the cloud.

• E-commerce Investment (Efficiency and Optimization Driven): Commentary on capex for the North America and International segments is focused on improving the existing network to lower costs and increase delivery speeds.

  • Fulfillment & Logistics: Investments are directed toward “refining our newly redesigned inbound network, building out our same-day delivery sites and adding additional robotics and automation” (Q1 2025 Transcript). This is about making the network more efficient and faster, rather than a large-scale expansion.

In summary, the pace of investment for AWS is aggressive and accelerating to capture a massive new market opportunity, while the investment for e-commerce is more measured and focused on optimizing an existing, mature network.

5. Future Outlook

Management has provided clear qualitative guidance that capital expenditures will continue to rise next year.

• Expected Capex Trend for 2026: In the Q3 2025 Earnings Call Transcript, CFO Brian Olsavsky explicitly stated: “...we expect our full-year cash CapEx to be approximately $125 billion in 2025, and we expect that amount will increase in 2026.”

• Stated Reasons for Future Outlook: The rationale is a direct continuation of the current strategy—investing heavily to capture the long-term opportunity in AI.

  • The CFO explained in the Q3 2025 transcript: “We’ll continue to make significant investments, especially in AI, as we believe it to be a massive opportunity with the potential for strong returns on invested capital over the long term.”

  • This is further supported by the CEO’s long-term view in the same transcript, where he noted that AWS is “on track to double [power capacity] again by 2027,” indicating a multi-year investment cycle.

6. Summary of Analysis

Amazon’s capital expenditure story over the past 12 months is one of significant and accelerating investment, overwhelmingly driven by the strategic imperative to build out infrastructure for Amazon Web Services (AWS) to meet explosive demand for Generative AI. The company’s TTM capex reached $103.8 billion, with a sharp quarterly ramp-up from $13.9 billion in Q4 2024 to $34.2 billion in Q3 2025. This acceleration led management to repeatedly raise its full-year 2025 capex guidance from an initial run rate of approximately $105 billion to a final forecast of $125 billion.

The primary driver for this capital intensity is the build-out of AWS data centers, securing massive amounts of power, and investing heavily in both third-party NVIDIA chips and Amazon’s own custom AI silicon, Trainium. Management commentary has become increasingly bullish, stating that demand for AI services currently outstrips supply and that they are monetizing new capacity as quickly as it comes online. In contrast, capex supporting the e-commerce segments is focused on optimization and efficiency. These investments in regionalization, robotics, and same-day delivery facilities are aimed at lowering cost-to-serve and increasing delivery speeds within the existing network, rather than a large-scale expansion. Looking forward, Amazon has clearly signaled that this period of intense investment will continue, with guidance indicating that capital expenditures are expected to increase again in 2026 to further capitalize on what management views as a massive, long-term opportunity in AI.

As an expert financial analyst, based exclusively on the provided documents, here is a detailed analysis and synthesis of Microsoft’s capital expenditures.

1. Quarterly Capex Trend (Last 12 Months)

Based on the cash flow statements and management commentary in the earnings call transcripts, Microsoft’s total capital expenditures, which include purchases of property and equipment and finance leases, for the last four quarters are as follows:

• Q2 FY2025 (Quarter ended December 31, 2024): $22.6 billion

  • Source: MSFT Transcript Q2 2025

• Q3 FY2025 (Quarter ended March 31, 2025): $21.4 billion

  • Source: MSFT Transcript Q3 2025

• Q4 FY2025 (Quarter ended June 30, 2025): $24.2 billion

  • Source: MSFT Transcript Q4 2025

• Q1 FY2026 (Quarter ended September 30, 2025): $34.9 billion

  • Source: MSFT Transcript Q1 2026

The quarterly trend shows a significant sequential and year-over-year acceleration, culminating in a record $34.9 billion in the most recent quarter.

Total Trailing Twelve Months (TTM) Capex: $22.6B + $21.4B + $24.2B + $34.9B = $103.1 billion

2. Management Guidance vs. Actuals

Management’s guidance for capital expenditures has evolved significantly over the past year, reflecting an acceleration in demand that surpassed initial expectations.

• Initial Guidance/Commentary: In the Q2 2025 earnings call, management provided an initial outlook for fiscal year 2026 (FY26), stating: “In FY26, we expect to continue investing against strong demand signals... However, the growth rate will be lower than FY25 and the mix of spend will begin to shift back to short-lived assets.” (MSFT Transcript Q2 2025). This guidance for moderating growth was reiterated in the Q4 2025 call.

• How Guidance Changed: The guidance held relatively steady until the most recent report.

  • In the Q2 2025 call, guidance for Q3 and Q4 was to “remain at similar levels as our Q2 spend” of $22.6 billion.

  • Q3 actuals came in slightly lower at $21.4 billion. Guidance for Q4 was updated to “increase on a sequential basis” (MSFT Transcript Q3 2025), which it did, coming in at $24.2 billion.

  • Guidance for Q1 FY26 was for spend to be “over $30 billion” (MSFT Transcript Q4 2025). The actual figure was a significantly higher $34.9 billion.

  • This outperformance and accelerating demand led to a major reversal in the full-year outlook.

• Current Full-Year Guidance: As of the Q1 2026 earnings call, the guidance for FY26 has been completely revised. CFO Amy Hood stated: “We now expect the FY ‘26 growth rate to be higher than FY ‘25“ (MSFT Transcript Q1 2026). This is a direct reversal of the previous expectation for moderating growth and signals a much more aggressive investment posture for the remainder of the fiscal year. Guidance for Q2 FY26 is for capex to “increase sequentially” from the record $34.9 billion in Q1.

3. Primary Drivers of Capex (The “Why”)

The documents make it clear that the primary driver for this elevated level of capital expenditure is overwhelming and accelerating demand for Microsoft’s cloud and AI infrastructure. Management consistently notes that demand is outpacing their available supply.

Strategic Drivers and Key Commentary:

• Meeting Cloud and AI Demand: The core rationale is to build capacity for Azure cloud services and, most critically, AI workloads. This includes building out data centers and acquiring servers (both CPUs and GPUs).

“Capital expenditures were $34.9 billion, driven by growing demand for our cloud and AI offerings.” (CFO Amy Hood, MSFT Transcript Q1 2026)

• Building a “Planet-Scale AI Factory”: CEO Satya Nadella has framed the investment in ambitious terms, reflecting a long-term strategic priority to build the foundational infrastructure for the AI era.

“We are seeing increasing demand and diffusion of our AI platform and family of copilots, which is fueling our investments across both capital and talent. When it comes to infrastructure, we’re building a planet-scale cloud and an AI factory...” (CEO Satya Nadella, MSFT Transcript Q1 2026) “We will increase our total AI capacity by over 80% this year and roughly double our total data center footprint over the next two years, reflecting the demand signals we see.” (CEO Satya Nadella, MSFT Transcript Q1 2026)

• Demand Exceeding Supply: A consistent theme throughout the past four quarters has been the constraint on growth due to a lack of available capacity.

“In our AI services, while we continue to bring data center capacity online as planned, demand is growing a bit faster. Therefore, we now expect to have some AI capacity constraints beyond June.” (CFO Amy Hood, MSFT Transcript Q3 2025) “While we’re accelerating the amount of capacity we’re bringing online, we will continue to balance Azure revenue growth with the growing needs across our first-party app and AI solutions... Therefore, we now expect to be capacity constrained through at least the end of our fiscal year.” (CFO Amy Hood, MSFT Transcript Q1 2026)

• Supporting Major Partnerships and Backlog: The spend is directly correlated with a massive and growing backlog of customer commitments, most notably from its partnership with OpenAI.

“Already, we have roughly 10x’d our investment, and OpenAI has contracted an incremental $250 billion of Azure services.” (CEO Satya Nadella, MSFT Transcript Q1 2026) “The remaining cloud in AI spend was primarily for servers, both CPUs and GPUs, to serve customers based on demand signals, including our customer-contracted backlog of $315 billion.” (CFO Amy Hood, MSFT Transcript Q3 2025)

Evolution of the Narrative: Over the past four quarters, the narrative has evolved from one of significant but managed investment to one of an all-out race to build capacity. While the core drivers (cloud, AI, data centers) have remained the same, the urgency and scale have intensified dramatically. The shift in full-year FY26 capex guidance from “moderating growth” to “higher growth than FY25” is the clearest evidence of this evolution, driven by demand signals “accelerating faster than we expected” (MSFT Transcript Q1 2026).

4. Segmental Capex Breakdown

The provided documents do not offer a quantitative breakdown of capital expenditures by the Productivity and Business Processes, Intelligent Cloud, and More Personal Computing segments.

However, management’s commentary synthesizes how these foundational investments support services across all three segments:

• Intelligent Cloud: This segment is the most direct beneficiary and user of the capital expenditure. The entire build-out of data centers, servers, and networking gear is to expand the capacity and capabilities of Azure. This supports everything from traditional cloud migrations (e.g., SAP and VMware workloads) to the massive Azure AI services that power third-party AI models and applications, such as those from OpenAI.

• Productivity and Business Processes: This segment relies on the underlying AI infrastructure to power its AI-infused services. Management commentary explicitly links capacity allocation to these products.

CFO Amy Hood noted that one of the top priorities for capacity is “the increasing usage and adoption and sales we’ve seen of Microsoft 365 Copilot and the usage of Copilot chat.” (MSFT Transcript Q1 2026) The investment in GPUs and CPUs directly enables the complex reasoning and generation capabilities of Copilots within Microsoft 365 and Dynamics 365.

• More Personal Computing: This segment also leverages the cloud infrastructure for key services.

CEO Satya Nadella noted in the Q4 2025 transcript: “We surpassed over 500 million hours of gameplay streamed via the cloud this year.” This streaming for Xbox Cloud Gaming is powered by the same Azure data centers being built out. Furthermore, AI features being integrated into Windows, such as Copilot, also rely on this cloud infrastructure to function.

In summary, while the assets physically reside within the Intelligent Cloud’s global infrastructure, they are a shared resource essential for delivering AI-powered innovation and cloud services across all three of Microsoft’s operating segments.

5. Future Outlook

Management has provided a clear, albeit revised, outlook for capital expenditures.

• Expected Trend for FY2026: Microsoft now expects capital expenditure growth in FY2026 to be higher than the growth in FY2025. This represents a significant acceleration. For the immediate next quarter (Q2 FY2026), capex is expected to increase sequentially from the record $34.9 billion spent in Q1.

• Stated Reasons for Future Outlook: The outlook is driven by the need to meet unprecedented demand and address ongoing capacity constraints.

  • Accelerating Demand Signals: The primary reason for the increased spending is that demand is growing faster than anticipated. CFO Amy Hood stated, “Accelerating demand and a growing RPO balance, we’re increasing our spend on GPUs and CPUs.” (MSFT Transcript Q1 2026).

  • Long-Term Data Center Build-Out: The investment is part of a multi-year strategy to fundamentally expand Microsoft’s global infrastructure footprint. CEO Satya Nadella stated the plan is to “roughly double our total data center footprint over the next two years, reflecting the demand signals we see.” (MSFT Transcript Q1 2026).

  • Ongoing Capacity Constraints: The company anticipates that even with this accelerated spending, it will remain supply-constrained for the foreseeable future, noting it expects to be “capacity constrained through at least the end of our fiscal year.” (MSFT Transcript Q1 2026).

6. Summary of Analysis

Microsoft is in the midst of a historic capital investment cycle, having spent $103.1 billion in capex over the trailing twelve months. This spending has dramatically accelerated, culminating in a record $34.9 billion in the most recent quarter. The primary driver for this massive expenditure is the unprecedented, accelerating demand for cloud and, particularly, AI infrastructure. Management has consistently reported that customer demand is outpacing their ability to bring new capacity online, creating a supply-constrained environment that they are aggressively investing to resolve. These investments are to build out a “planet-scale cloud and an AI factory,” which involves constructing new data centers globally and filling them with high-performance servers, including both CPUs and GPUs, to power everything from traditional cloud migrations to the most advanced AI models.

The narrative from management has intensified over the past year. What began as a plan for significant but moderating investment in fiscal year 2026 has transformed into a more urgent race to build. In its latest quarterly report, Microsoft reversed its prior guidance, now expecting capex growth in FY26 to be higher than in FY25, citing demand signals from bookings, usage, and a rapidly growing contracted backlog that are “accelerating faster than we expected.” While these investments are accounted for within the Intelligent Cloud segment, they are foundational to Microsoft’s entire strategy, powering AI-infused services across all segments, including Microsoft 365 Copilot, Dynamics 365, and Xbox Cloud Gaming. This spending, directly correlated to a massive backlog of nearly $400 billion in commercial commitments, represents a strategic, demand-driven push to build the fundamental infrastructure required to capture what management views as a generational shift in technology.

NVDA: Based on my analysis of the provided documents, here is a synthesis of NVIDIA’s revenue and demand trends for the 12 months preceding October 31, 2025.

1. Quarterly Revenue Trend (Last 12 Months)

Here is the Total Revenue and Data Center Revenue for each of the last four fiscal quarters.

• Q3 FY25 (ended October 27, 2024):

  • Total Revenue: $35,082 million

  • Data Center Revenue: $30,819 million

• Q4 FY25 (ended January 26, 2025):

  • Total Revenue: $39,330 million

  • Data Center Revenue: $35,600 million

• Q1 FY26 (ended April 27, 2025):

  • Total Revenue: $44,062 million

  • Data Center Revenue: $39,017 million

• Q2 FY26 (ended July 27, 2025):

  • Total Revenue: $46,743 million

  • Data Center Revenue: $42,752 million

Trailing Twelve Months (TTM) Totals:

• Total TTM Revenue: $165,217 million

• Total TTM Data Center Revenue: $148,188 million

Data Center Percentage of Total Revenue (Most Recent Quarter, Q2 FY26):

• Data Center represented 91.5% of Total Revenue.

2. Management Guidance vs. Actuals

• Initial Guidance: The earliest provided guidance was for Q4 FY25, given in the Q3 FY25 earnings call. Management’s outlook was for total revenue to be $37.5 billion, plus or minus 2%.

• Guidance Evolution: NVIDIA consistently beat its revenue guidance over the last year.

  • Q4 FY25: Actual revenue was $39.3 billion, exceeding the $37.5 billion outlook.

  • Q1 FY26: The guidance was for $43 billion. Actual revenue was $44.0 billion.

  • Q2 FY26: The guidance was for $45 billion. Actual revenue was $46.7 billion.

• Next Quarter Guidance: In the latest report (Q2 FY26 Earnings Call), management’s guidance for Q3 FY26 revenue is $54 billion, plus or minus 2%.

3. Primary Drivers of Demand (The “Why”)

Based on management’s commentary, the primary strategic driver for the massive growth in Data Center demand is the shift from simple generative AI to “reasoning and agentic AI.” These new, more sophisticated AI models require orders of magnitude more computation for both training and, critically, inference.

Key Commentary on Demand Drivers:

• Generative & Reasoning AI: The narrative has evolved from general generative AI to a focus on more complex tasks.

  • “Long-thinking reasoning AI can require 100x more compute per task compared to one-shot inferences.” (Q4 2025 Transcript)

  • “In a word, chatbots used to be one shot... Now the AI does research, it thinks and does a plan... the amount of computation necessary for one shot versus reasoning agentic AI models could be a 100 times, a thousand times, and potentially even more.” (Q2 2026 Transcript)

• Cloud Service Providers (CSPs): CSPs remain a foundational customer base, representing about half of Data Center revenue in Q4 FY25 and driving the initial large-scale deployments of the Blackwell platform.

  • “major hyperscalers are each deploying nearly 1,000 NBL72 racks, or 72,000 Blackwell GPUs per week, and are on track to further ramp output this quarter.” (Q1 2026 Transcript)

• Enterprise AI: The emergence of agentic AI is unlocking the enterprise market, as companies build custom AI agents and copilots for specific workflows.

  • “Enterprise AI is in full throttle. NVIDIA AI Enterprise, which includes NVIDIA NeMo and NIM microservices, is an operating platform of agentic AI.” (Q3 2025 Transcript)

• Sovereign AI: This has emerged as a major, distinct growth engine, with nations investing heavily in domestic AI infrastructure.

  • “Sovereign AI is on the rise as the nation’s ability to develop its own AI using domestic infrastructure, data, and talent presents a significant opportunity for NVIDIA Corporation... We are on track to achieve over $20 billion in Sovereign AI revenue this year, more than double that of last year.” (Q2 2026 Transcript)

• Physical AI / Robotics: Management views this as the next major wave, driven by foundation models that can understand the physical world.

  • “The age of physical AI has arrived, unlocking entirely new industries in robotics and industrial automation.” (Q2 2026 Transcript)

Narrative Evolution: Over the past four quarters, the narrative has sharpened significantly. It began with broad excitement around generative AI and inference. It has since evolved to a more specific and potent story centered on reasoning and agentic AI as a step-function change in compute demand. Concurrently, the customer story has broadened from a primary focus on CSPs to a multi-pillar strategy highlighting Sovereign AI, Enterprise AI, and Physical AI as massive, independent growth vectors.

4. Customer & Segment Demand Breakdown

• Hyperscalers / CSPs: Demand from this segment remains robust and foundational. They were consistently cited as representing roughly half of Data Center revenue and are the first to deploy new architectures like Blackwell at massive scale. Microsoft, Google, AWS, and Oracle are all mentioned as key partners deploying the latest technology.

• Other Customer Types:

  • Enterprise: Demand is accelerating, driven by the need for model fine-tuning and the deployment of agentic AI workflows. The launch of RTX Pro servers is aimed at making AI more accessible for standard enterprise IT environments, with management noting it is “poised to become a multibillion-dollar product line.” (Q2 2026 Transcript)

  • Sovereign AI: This has become a pillar of growth. Commentary highlights major initiatives and AI factory build-outs in India, Japan, the UK, and across Europe, with management tracking over $20 billion in revenue for the current fiscal year.

  • Automotive: This segment shows strong growth, driven by self-driving solutions. Revenue in Q2 FY26 was up 69% year-on-year, with management noting the new Thor SoC is their “most successful robotics and AV computer we have ever created”.

• AI Training vs. AI Inference Demand: The commentary overwhelmingly points to inference as the explosive growth driver. While training demand continues to grow, the computational needs of new reasoning and agentic models during inference are creating a surge that management believes will be a significant part of the future compute workload.

  • “We are witnessing a sharp jump in inference demand. OpenAI, Microsoft, and Google are seeing a step function leap in token generation. Microsoft processed over 100 trillion tokens in Q1, a fivefold increase on a year-over-year basis.” (Q1 2026 Transcript)

5. Future Outlook & Supply/Demand Dynamics

• Demand Trend for Next Year: Management expresses high confidence in continued strong demand.

  • “We have reasonable forecasts from our large customers for next year. A very, very significant forecast... the next several years, surely through the decade, we see just a really fast-growing, really significant growth opportunities ahead.” (CEO Jensen Huang, Q2 2026 Transcript)

  • “We see $3 to $4 trillion in AI infrastructure spend by the end of the decade.” (CFO Colette Kress, Q2 2026 Transcript)

• Supply vs. Demand Picture: The company is consistently supply-constrained for its newest and most powerful products.

  • “While demand greatly exceeds supply, we are on track to exceed our previous Blackwell revenue estimate...” (Q3 2025 Transcript)

  • “the buzz is everything sold out. H1 hers sold out. H2 hundreds are sold out... the demand is really, really high.” (CEO Jensen Huang, Q2 2026 Transcript)

• New Product Cycles: The company is executing on a rapid, annual product cadence which it views as a competitive advantage.

  • Hopper to Blackwell: The transition has been extremely fast, with the Blackwell platform described as “the fastest product ramp in our company’s history” (Q4 2025 Transcript). Blackwell is specifically architected for the new wave of reasoning AI, delivering a “50x increase in energy efficiency per token compared to Hopper” (Q2 2026 Transcript).

  • Blackwell to Rubin: The next platform, Rubin, is already in development. “The chips of the Rubin platform are in fab... Rubin remains on schedule for volume production next year.” (Q2 2026 Transcript). This annual cycle is designed to continually improve performance-per-watt, a critical factor for power-constrained data centers.

  • Spectrum-X: The company’s enhanced Ethernet solution is a major success, now at an “annualized revenue exceeding $10 billion.” (Q2 2026 Transcript)

6. Summary of Analysis

NVIDIA has experienced a period of explosive, record-breaking growth over the past twelve months, consistently surpassing its own ambitious guidance. Trailing-twelve-month revenue now exceeds $165 billion, with the Data Center segment accounting for over 91% of the total and driving the company’s trajectory. This unprecedented performance is fueled by the historic ramp of its new Blackwell architecture, which is being adopted at a faster pace than any product in the company’s history to meet surging demand for AI compute.

The primary catalyst for this demand is a fundamental shift in the nature of AI itself, moving beyond simple content generation to complex “reasoning and agentic AI.” Management consistently emphasizes that these sophisticated models require orders of magnitude more computational power for inference, creating a step-function increase in demand. This has broadened NVIDIA’s market from its core Cloud Service Provider customers to include rapidly accelerating, multi-billion-dollar segments like Sovereign AI, where nations are building their own AI capabilities, and Enterprise AI, where companies are deploying AI agents to revolutionize workflows. Demand for the company’s newest products continues to significantly outstrip supply, with management describing a “sold out” environment and citing very strong customer forecasts for the upcoming year.

Looking forward, NVIDIA is executing a relentless annual product innovation cycle, with its next-generation Rubin platform already in fabrication and on schedule for release next year. This strategy is designed to deliver consistent, order-of-magnitude improvements in performance and energy efficiency, which is critical for power-constrained data centers. By providing a full-stack, data-center-scale platform that addresses the entire AI lifecycle—from training to the exploding demand for reasoning-based inference—NVIDIA has positioned itself at the center of what management describes as a new industrial revolution with a multi-trillion-dollar market opportunity.

Recent AI CapEx News:

Hyperscaler Headlines

Recent earnings reports from Meta, Google, Microsoft, and Amazon have reaffirmed their commitment to massive spending on AI infrastructure, with all indicating significant increases in capital expenditures for 2025 and 2026. While specific new multi-billion dollar data center projects were not detailed in the last 14 days, the clear message from earnings calls is a continued and accelerating investment cycle to expand AI capabilities.

Google’s parent company, Alphabet, announced it was boosting its 2025 capital expenditure forecast to between $91 billion and $93 billion, a significant increase from earlier estimates, driven by the growth of its AI-powered cloud business. Microsoft’s Azure cloud business saw 40% growth, with the company signaling that AI data center capex will continue to increase year-over-year. Meta also raised its 2025 capex forecast to as high as $118 billion, citing the demands of heavier AI workloads. Amazon plans to continue increasing its data center spending through the remainder of the year and into 2025.

In a significant partnership, OpenAI has teamed up with Broadcom for custom AI silicon, aiming to reduce the cost of building out its AI infrastructure. This move mirrors Google’s success with its custom TPU chips, also developed with Broadcom, and signals a trend towards more specialized hardware to optimize AI workloads.

Supply Chain & Component Signals

The supply chain for AI components is running at full throttle, with key players accelerating their roadmaps and expanding capacity to meet insatiable demand.

NVIDIA is moving to a one-year release cadence for its next-generation platforms, with the Blackwell Ultra chips launching later this year, to be followed by the Vera Rubin platform in 2026 and Rubin Ultra in 2027. While there were some initial yield issues with the Blackwell architecture, these appear to have been resolved, with mass production on track. In a move to bolster supply chain resilience, NVIDIA CEO Jensen Huang has indicated the company could spend “several hundred billion” dollars on chips and electronics made in the U.S. over the next four years, leveraging TSMC’s new Arizona facility.

Advanced Micro Devices (AMD) is also speeding up its development cycle to compete with NVIDIA, with its MI325X GPU expected later this year, followed by successors in 2025 and 2026. Intel is looking to gain market share with its Gaudi 3 AI chips, which are positioned to compete with NVIDIA’s H100 and H200 GPUs.

The demand for advanced packaging remains a critical bottleneck. TSMC is aggressively expanding its CoWoS (Chip-on-Wafer-on-Substrate) packaging capacity to meet the demand from customers like NVIDIA and Broadcom. The company plans to establish 10 new factories worldwide next year, with a focus on advanced processes and packaging. TSMC’s CoWoS capacity is projected to increase significantly through 2026.

Broadcom is a major beneficiary of the AI build-out, with its AI-related semiconductor revenue soaring. The company has secured a significant $10 billion order for custom AI chips from a new client, reported to be OpenAI. Broadcom’s custom silicon (ASIC) business allows hyperscalers to develop optimized chips for their specific AI workloads, a trend that is gaining momentum.

Analysts’ Take

Analysts have been actively adjusting their ratings and price targets in response to the latest developments in the AI capex cycle.

Barclays raised its price target on NVIDIA to $200, citing strong supply chain demand and potential upside in the second half of the year. Oppenheimer also lifted its price target on Meta Platforms to $775, noting a stronger advertising backdrop and the potential to “unlock new business with AI.”

Truist upgraded AMD to a “Buy,” raising its price target to $213 and citing that customers are now considering AMD for large AI projects, not just as a secondary option to NVIDIA. The analyst believes AMD could capture about 10% of the GPU market.

However, there are also some notes of caution. Envision Research downgraded NVIDIA from “Buy” to “Hold,” suggesting that the positive news of resumed exports to China is now fully priced in. Julian Lin downgraded AMD from “Strong Buy” to “Neutral,” expressing concerns about the competitive gap with NVIDIA in the data center segment.

Executive Summary

The past two weeks have seen a strong reaffirmation of the massive AI infrastructure build-out, with hyperscalers signaling continued and accelerated capital expenditures for the foreseeable future. While the market is beginning to discriminate between companies based on their ability to monetize these investments, the overall trend of heavy spending on data centers and AI hardware remains firmly intact. The frantic pace of the build-out is creating a corresponding surge in demand throughout the supply chain, from GPUs and custom silicon to advanced packaging.

Key component suppliers are in a race to innovate and expand capacity. NVIDIA is accelerating its product roadmap to a one-year cycle, while competitors like AMD and Intel are also bringing new products to market. The critical constraint of advanced packaging is being addressed with significant new factory investments by TSMC. The rise of custom silicon, highlighted by the new partnership between OpenAI and Broadcom, points to a future of more specialized and efficient AI hardware. While analysts are overwhelmingly positive on the key players in this space, there is a growing awareness of competitive dynamics and the need for companies to demonstrate a clear return on their massive AI investments. The AI arms race is not just about building bigger data centers; it’s about building smarter, more efficient, and ultimately, more profitable ones.

As a note I’m utilizing some things I’m working on with “Clarity AI” to test out the capabilities. More info coming Soon(tm)

Subscribe now

CVNA:

Alt data per Clarity Markets

Expected unit numbers: ~158.7k (+46% YoY) (+11% QoQ)

This is the main driver really. Unit growth is absurd, forward estimates are too low and assume it falls off a cliff in 2026 when they’ve been comping +40-50% the entire year. Q4 is thus far like +42% YoY and they really just don’t stop growing.

ASP was something like $26k in Q3 which was higher sequentially. Q4 may be a bit lower. Hard to 1:1 translate that to GPU since there’s mix shift/etc but it’s hard to have significant downwards pressure on GPU with prices up.

Consumer credit has been a big concern for Twitter bears but it doesn’t really seem to be playing out. Down payment requirements have gone down two quarters in a row. Loan term lengths on average should be slightly down. There’s nothing that really points to some sudden deterioration in consumer credit. Volume has grown not slowed. Loan terms are relatively stable.

They called a bunch of their PIK 2028 debt a month ago. Not 100% sure how they executed that, they’ll likely share more soon. Not really a huge deal but they’ll save a bunch in interest expense regardless of what they did. One could argue using an ATM to pay off the debt is bad or something I guess if that’s what they did? Signals a lack of confidence? Maybe, but you also remove a lot of tail risk, so I can’t imagine that in isolation really means anything.

So yeah to summarize:

1. Units growing very fast and I’m guessing they guide to more sequential growth or similar level. Thus out years estimates are probably way too low

2. Margins seem relatively stable, maybe you get some slight movements either way but $7k+ GPU gives you a lot of flex and growth handles the rest

3. As a result, EBITDA is probably close-ish to expectations for Q3 and the stock I’d imagine just continuously grinds slowly higher as everyone comes to terms with the fact that they just won used car. They’ve almost caught up to Carmax on volumes recently, so yeah that’s kinda neat for them.

And here’s some AI stuff (note, maybe be some hiccups, particularly with news/stock price as I’m still fine tuning):

1. Main Revenue Segments & Performance

Carvana reports its revenue in three main segments. Over the past year, all segments have demonstrated significant growth, consistently outpacing the broader used auto market.

• Retail Vehicle Sales: This is the company’s core business of selling used cars to consumers. It is the largest contributor to revenue.

• Wholesale Sales and Revenues: This includes revenue from selling vehicles acquired from customers that are not sold through retail channels, often through its ADESA physical auction business.

• Other Sales and Revenues: This primarily consists of gains on the sale of auto loans Carvana originates, as well as revenue from sales of Vehicle Service Contracts (VSCs) and other ancillary products.

Performance Over the Past Year (YoY Growth)

Management Commentary: CEO Ernie Garcia stated in the Q2 2025 call that the company’s significant outperformance in growth “speaks clearly to the desirability of our offering and our team’s ongoing execution.” Management consistently attributes this success to a virtuous cycle: improving the customer offering (e.g., faster delivery times), which increases brand trust and awareness, which in turn drives scale and allows for greater inventory selection, further improving the customer offering.

2. Trend of Important Reported Metrics

Key financial metrics have shown dramatic improvement and record-setting performance over the last four reported quarters. The company has shifted from a focus on pure growth to profitable growth.

Key Metrics Trend (Q3 2024 - Q2 2025):

Significant One-Time Items:

• Q2 2025: Retail GPU included an approximately $100 benefit from tariff-related impacts, which management explained was due to pricing adjustments made during a period of pull-forward in demand following tariff announcements (Q2 2025 Transcript).

• Q4 2024: Adjusted EBITDA was negatively impacted by a $10 million “thank you” cash bonus to employees (Q4 2024 Transcript).

• Q3 2024: Other GPU included two positive one-time benefits: ~$150/unit from selling more loans than originated and ~$100/unit from favorable interest rate movements (Q3 2024 Transcript).

• Q2 2024 (Comparable Quarter): Other GPU had a positive impact of ~$100/unit from selling additional loans, which makes the YoY comparison for Q2 2025’s Other GPU less straightforward (Q2 2025 Transcript).

3. Comparison to Guidance

Carvana has consistently met or beaten its recent financial guidance.

• Q2 2025 Results vs. Q1 Guidance:

  • Guidance: Expected a “sequential increase in both retail units sold and adjusted EBITDA.”

  • Result: Met. Retail units grew from 134k to 143k, and Adjusted EBITDA grew from $488M to $601M.

• Q1 2025 Results vs. Q4 Guidance:

  • Guidance: Expected a “sequential increase in both retail units sold and adjusted EBITDA in Q1 2025.”

  • Result: Met. Retail units grew from 114k to 134k, and Adjusted EBITDA grew from $359M to $488M.

• FY 2024 Results vs. Q3 Guidance:

  • Guidance: Expected full-year 2024 Adjusted EBITDA to be “significantly above the high end” of the $1.0B-$1.2B range.

  • Result: Beat. Full-year 2024 Adjusted EBITDA came in at $1.38 billion.

4. Changes in Reported Metrics for Earnings

There have been no significant changes to the core KPIs Carvana reports. However, management has introduced new long-term targets and provided more context on new business lines.

• New Long-Term Target: In the Q1 2025 earnings call, CEO Ernie Garcia announced a new objective: “To grow to 3 million annual retail sales with 13.5% adjusted EBITDA margins in the next 5 to 10 years.“ This now frames the company’s long-term strategy and will likely be a key benchmark in future calls.

• “Retail Marketplace” Offering: Management has started discussing a “retail marketplace” business where they sell vehicles on behalf of commercial partners. While not a new KPI, it impacts financial reporting because, for these units, Carvana does not recognize the gross sales price as revenue. As this business grows, it will cause total revenue growth to lag behind total unit growth (Q3 2024 Transcript).

5. Significant Changes in 10-Q/10-K Filings

The filings do not indicate any new disclosures related to audits or fraud investigations. The most significant changes relate to strategic initiatives and balance sheet management.

• Acquisition of a Franchise Dealership: The Q1 2025 10-Q disclosed the acquisition of a dealership on February 28, 2025, for an “immaterial amount.” This is a notable strategic move, as it marks an experiment in the new car market, a potential expansion of Carvana’s Total Addressable Market (TAM).

• Balance Sheet Activities: Recent filings and calls have detailed significant efforts to strengthen the balance sheet. This includes:

  • Equity Issuance: Raising $924 million in Q4 2024 through an At-The-Market (ATM) program (Q4 2024 Transcript).

  • Debt Reduction: Repurchasing portions of its senior secured notes, including $120 million in Q4 2024 and $100 million in Q3 2024.

  • Management’s stated goal is to drive toward “investment-grade quality credit ratios over time” (Q4 2024 Transcript).

6. Recent Major Contracts or Client Lands

• New Auto Loan Buyer: In Q1 2025, management announced they had added a “new significant buyer“ for their originated auto loans. This diversifies their loan monetization channels beyond their existing partners like Ally and the securitization market, strengthening their financing platform (Q1 2025 Transcript).

• Retail Marketplace Partner: The company has a growing business with unnamed “large retail marketplace partners.” In Q2 2025, CFO Mark Jenkins noted a recent change in contract structure with one such partner, which resulted in Carvana holding that partner’s inventory on its balance sheet (Q2 2025 Transcript). This indicates a deepening of a significant commercial relationship.

7. Acquisitions Within the Past Year

• ADESA Integration (Ongoing): While the acquisition of ADESA occurred in May 2022, the ongoing integration of its 56 physical auction sites into Carvana’s reconditioning network is a critical factor impacting current performance. In 2024, six sites were integrated, with plans for another 10 in 2025 (Q4 2024 Transcript). Management states this is key to unlocking capacity for up to 3 million units per year with minimal CapEx, while also reducing logistics costs and delivery times.

• Franchise Dealership (February 2025): Carvana acquired one auto dealership in February 2025. The financial impact is currently “immaterial” and does not need to be factored into YoY revenue growth comparisons. However, its strategic importance is high, as it represents an exploration of the new vehicle market. CEO Ernie Garcia described it as an early-stage “experiment” (Q1 2025 Transcript).

1. Management Guidance

Based on the Q2 2025 earnings call, management has provided the following guidance for the upcoming quarter and full year:

• Retail Units Sold: A sequential increase in retail units sold is expected in Q3 compared to Q2.

• Advertising Spend: Management anticipates a larger sequential increase in advertising spending in Q3 compared to Q2.

• Full-Year Adjusted EBITDA: The company forecasts an adjusted EBITDA of $2.0 billion to $2.2 billion for the full year 2025.

2. Analyst Expectations for Q3 Earnings

Wall Street analysts are bullish on Carvana’s third-quarter results, which are scheduled to be announced on October 29, 2025. Consensus estimates from various analysts are:

• Earnings Per Share (EPS): Expected to be between $1.30 and $1.37, which would represent a year-over-year increase of over 100%. One firm, William Blair, has a higher estimate of $1.49 per share.

• Revenue: Projected to be in the range of $4.99 billion to $5.1 billion.

Overall, the consensus rating on Wall Street is a “Strong Buy” or “Moderate Buy”. JPMorgan recently increased its price target for Carvana to $490 from $425, anticipating that the company will deliver another “beat and raise” quarter.

3. Stock Price Performance (Last 90 Days)

Over the past 90 days, Carvana’s stock has experienced some volatility. After reaching a 52-week high of $413.34 on July 31, 2025, the stock has seen a decrease over the last 30 days. Despite this recent dip, the stock is up approximately 8% since the last earnings report in July and has risen nearly 80% year-to-date.

The recent sluggishness in the stock price has been attributed by some analysts to renewed fears about the health of consumer and auto credit, though some analysts believe these concerns are “overblown”.

1. Recent News and Fundamental Changes for Carvana

Carvana has experienced a period of positive momentum and significant business developments in the past 90 days. The company has been placed on an “upside 90-day catalyst watch” by a Citi analyst, citing a positive trend in retail unit sales. This momentum is reflected in their strong financial performance, with the company reporting record-breaking second-quarter earnings. Revenue reached $4.84 billion, a 41.9% year-over-year increase, and vehicle sales grew by 41% to over 143,000 units.

A key fundamental change for Carvana has been the successful integration of ADESA’s U.S. operations. This acquisition has strengthened Carvana’s logistics network, auction capabilities, and reconditioning infrastructure, contributing to improved operational efficiency. This has led to a 70% year-over-year increase in adjusted EBITDA in the second quarter. The company’s online-first model is capitalizing on a shift towards used cars, partly driven by rising tariffs on new vehicles. This has been a remarkable turnaround for a company that was facing bankruptcy concerns in late 2022.

2. Major Developments Impacting Carvana’s Upcoming Earnings

The most significant development for Carvana has been its record-breaking sales and revenue figures in the second quarter. This performance is expected to be a major topic of discussion in their upcoming earnings report. The increased demand for used vehicles, fueled by external factors like tariffs, has provided a significant tailwind for the company.

On the other hand, there have been substantial sales of company stock by insiders, including major shareholder Ernest C. Garcia II. While not a direct reflection of business operations, these transactions are often of interest to investors and may be a point of discussion.

3. Carvana’s News Compared to Competitors

Carvana’s recent news stands in stark contrast to that of its competitors, CarMax (KMX) and Vroom (VRM).

CarMax (KMX): CarMax is facing a challenging period. The company missed recent earnings and revenue expectations, with retail used car sales declining. In response, several analysts have downgraded their ratings for the company’s stock. The company is also being impacted by broader issues in the subprime auto market and has been forced to slash prices. This has led to a significant drop in the company’s value.

Vroom (VRM): Vroom’s situation is even more dire. The company has ceased all e-commerce operations and is in the process of liquidating its used vehicle inventory. This decision came after the company was unable to secure the necessary capital to continue its operations. Vroom has laid off 90% of its workforce and has a history of customer complaints and legal issues.

Trending Differences: The news trends for the three companies are heading in opposite directions:

• Carvana: Upward trend with strong growth, profitability, and positive market sentiment.

• CarMax: Downward trend with declining sales, profitability, and negative market sentiment.

• Vroom: Has effectively exited the online used car market.

GOOGL:

My general thesis on Google is really simple: The AI is really good, AI has not eaten search, it’s not very expensive.

That basic thesis has been a +60% or something since I bought a bunch. I also bought some leaps recently.

I really can’t stress enough how good AI is and how well positioned Google is. The focus has mostly been on consumer web UI’s and it’s like sure that’s nice but in 5-10 years every enterprise on earth is going to be slamming billions or trillions of tokens into models which is orders of magnitude more than where we are today. Google just so happens to:

1. Be a frontier model provider

2. Be a cloud provider

3. Be a chip designer

4. Have access to billions of consumers

5. Have contracts with basically every enterprise

They aren’t falling behind in AI (at least thus far). They’ve actually had the frontier model for the majority of 2025 and when Gemini 3.0 releases I imagine they’ll have it again. A cute little game they get to play as well is bundling Gemini into GCP more generally, as well as bundling GCP with TPU’s. Then if consumer AI is a hit well great, they have Search, Chrome and YouTube surfaces with billions of MAUs.

I also wouldn’t be surprised if they start drifting further into operating system land. As computer use agents become more of a thing from my understanding owning the OS becomes much more valuable. I’d be surprised if they didn’t have internal teams trying to figure out an AI native OS (MSFT as well).

Basically everything’s going to get a bit weird in 5-10 years and exact predictions are hard. Like what happens with SaaS? Great question! Nobody knows! What I do believe with pretty high confidence is that Google is probably one of the best packages to capitalize on that weirdness. Semis are a bubble? Cool Google gets them for cheaper and the compute is DEFINITELY not a bubble. Semis not a bubble? Cool Google has TPU’s and also should see benefits via GCP/Gemini regardless.

So to summarize:

1. I don’t care about current results, they’re probably fine

2. AI is really good you should be exposed

3. Google is a good way to do so

Anyways here’s some AI stuff as well.

1. Main Revenue Segments & Performance

Alphabet reports its revenues across three main segments: Google Services, Google Cloud, and Other Bets. Google Services is the largest and is further broken down.

Google Services: This is the core of the business, encompassing advertising and consumer services.

• Overall Performance: The segment has shown consistent, strong growth, posting a 12.1% year-over-year (YoY) increase in Q2 2025 to reach $85.2 billion.

• Breakdown & Management Commentary:

  • Google Search & other: This is the largest revenue driver. It has maintained double-digit growth, up 12.5% YoY in Q2 2025 to $56.9 billion. Management attributes this strength to broad-based performance across verticals, particularly retail and financial services. CEO Sundar Pichai has consistently emphasized that new AI features like AI Overviews and Circle to Search are increasing overall search usage and user satisfaction, with users asking “longer and more complex questions.”

  • YouTube ads: This has been a strong performer, growing 13% YoY in Q2 2025 to $9.8 billion. Growth is driven by both direct response and brand advertising. Management is pleased with the monetization of Shorts, noting in the Q2 call that in the U.S., “shorts now earn as much revenue per watch hour as traditional in-stream.” Connected TV (CTV) remains a major growth driver, with YouTube being the #1 streaming service by watch time in the U.S. for over two years.

  • Google Network: This is the only sub-segment that has consistently shrunk, declining 1.2% YoY in Q2 2025. This reflects a continued shift in advertising spend towards Google’s owned-and-operated properties like Search and YouTube.

  • Subscriptions, Platforms, and Devices: This is a high-growth area, accelerating to 20% YoY growth in Q2 2025 to $11.2 billion. Growth is primarily driven by subscriptions for YouTube (Premium, TV, Music) and Google One. The launch of the Pixel 9 series in Q3 2024 also contributed to growth in the past year.

Google Cloud: This segment continues to be a key growth engine.

• Overall Performance: Google Cloud has shown outstanding and accelerating growth, with revenue up 32% YoY in Q2 2025 to $13.6 billion. Its annual revenue run rate now exceeds $50 billion.

• Management Commentary: Management highlights massive customer demand driven by its comprehensive AI portfolio. Sundar Pichai noted in the Q2 call that the number of deals over $250 million doubled YoY and that they signed as many billion-dollar deals in the first half of 2025 as in all of 2024. The growth is attributed to GCP’s AI infrastructure (TPUs and GPUs), the Vertex AI platform, and Gemini models.

Other Bets: This segment includes Alphabet’s portfolio of emerging technologies like Waymo (autonomous driving).

• Overall Performance: Revenues are small and volatile, and the segment consistently operates at a loss (around $1.1-$1.2 billion per quarter).

• Management Commentary: Management is increasingly highlighting the progress of Waymo, which is now serving over 150,000 paid rides per week and has driven over 100 million autonomous miles. While not a significant financial contributor, it is framed as a long-term value creation opportunity.

2. Trend of Important Reported Metrics

Reported metrics have shown strong growth over the past year, though they have been affected by some significant one-time items.

Key Changes & Management Explanations:

• One-Time Windfall (Q1 2025): Net income was significantly boosted by $11.2 billion in “Other income,” which CFO Anat Ashkenazi explained was “primarily due to unrealized gain on our non-marketable equity securities related to our investment in a private company.” This was a one-time event that did not affect operating income.

• One-Time Expenses:

  • Q2 2025: Operating income was impacted by a $1.4 billion charge related to a “settlement in principle of certain legal matters.”

  • Q3 2024: Operating income was impacted by a $607 million charge for optimizing the company’s global office space.

• Operating Margin: Despite these charges and heavy investment in AI, operating margins have remained healthy, hovering around 32-34%. Management credits this to strong revenue growth and a continued focus on “reengineering our cost structure” and operating with efficiency.

3. Comparison of Results to Prior Guidance

Alphabet provides qualitative guidance. Over the past year, they have generally met or exceeded their stated expectations.

• CapEx: This has been a key focus. In Q1 2025, management guided for full-year 2025 CapEx of approximately $75 billion. In the Q2 2025 call, citing strong demand for Cloud and AI services, they raised this guidance significantly to approximately $85 billion. This is a critical indicator of their investment intensity and confidence in future growth.

• Revenue Outlook: Management has consistently guided for headwinds related to lapping strong performance from prior periods (e.g., financial services vertical strength, 2024 US election ad spend). However, growth has remained robust, suggesting they are successfully navigating these tougher comparisons.

• Cloud Constraints: Management has repeatedly stated that they are in a “tight demand-supply environment” for Google Cloud, with customer demand exceeding available capacity. The increased CapEx is a direct response to this. This suggests Cloud revenue growth could continue to be strong as new capacity comes online.

4. Changes in Reported Earnings Metrics/KPIs

There have been no changes to the formal segment reporting structure or financial KPIs in the 10-Q/10-K filings over the past year.

However, a significant change in the earnings calls is the introduction and consistent highlighting of new AI-related usage metrics. Management is using these to demonstrate user adoption and engagement with their new AI products. Key metrics mentioned recently include:

• AI Overviews monthly users (grew from 1B in Q3 2024 to over 2B in Q2 2025)

• Gemini app monthly active users (450M+ as of Q2 2025)

• Total tokens processed monthly across Google surfaces (doubled from 480T to 980T between May and July 2025)

• YouTube Shorts daily views (grew from 70B+ in Q3 2024 to 200B+ in Q2 2025)

5. Significant Changes in 10-Q/10-K Reporting

• Legal Proceedings: The filings consistently disclose significant and ongoing regulatory scrutiny globally, including the DOJ antitrust trial in the U.S. and actions from the European Commission. The most significant new item was the $1.4 billion legal settlement disclosed in the Q2 2025 10-Q. There have been no new disclosures of fraud investigations.

• Risk Factors: The risk factors are regularly updated to reflect the heightened competitive and regulatory environment related to AI. The core risks remain, but the emphasis on competition in AI, the cost of building and deploying models, and potential AI-related regulation has increased.

6. Recent Major Contracts or Client Lands

Management has frequently highlighted major enterprise customer wins, particularly for Google Cloud, to demonstrate market traction. Recent notable clients mentioned over the past year include:

• Financial Services: LVMH, DBS Bank, BBVA, PayPal, Citi, Hiscox

• Retail/Commerce: Wayfair, Target, Lowe’s, Mercado Libre

• Technology & Telecom: Salesforce, Snap, Roblox, Verizon, Vodafone

• Other Enterprise: Mercedes-Benz, Volkswagen, Capgemini, Deloitte, AstraZeneca

7. Recent Acquisitions and Their Impact

Alphabet has made two notable strategic acquisitions in the past year that you should be aware of:

• Wiz, Inc.: In Q1 2025, Alphabet announced its intent to acquire Wiz, a leading cloud security platform. The strategic rationale is to bolster Google Cloud’s multi-cloud security offerings. As this acquisition is integrated, it will contribute to Google Cloud’s revenue and potentially impact its operating expenses.

• Character.ai: Filings indicate this acquisition occurred in August 2024. As this happened over a year ago, its financial results are already included in the baseline for year-over-year comparisons. It was a strategic acquisition to strengthen their position in conversational AI. Its financial impact was likely not material enough to warrant a separate disclosure in the consolidated results.

1. Management Guidance for the Next Quarter

Based on the Q2 2025 earnings call, management provided the following guidance for the third quarter:

• Revenue: A potential tailwind to revenue is expected in Q3 at current spot rates, though exchange rate volatility could impact this.

• Advertising Revenue: Year-over-year comparisons for advertising revenues in the second half of 2025 will be negatively affected by lapping strong performance in financial services verticals in 2024 and significant spending on the U.S. election in the latter half of 2024, especially on YouTube.

• Google Cloud: A tight demand-supply environment is expected to continue into 2026.

• Capital Expenditures (CapEx): The full-year 2025 CapEx forecast has been increased to approximately $85 billion, up from the previous estimate of $75 billion. A further increase in CapEx is expected in 2026.

• Depreciation: The growth rate of depreciation is expected to accelerate in Q3.

• Headcount: A sequential increase in total headcount additions is anticipated, partly due to the hiring of new graduates.

• Expenses: Q3 expenses will be impacted by the August launch of the new Pixel family of products.

2. Analyst Expectations for Upcoming Earnings

Analysts expect Alphabet to report strong revenue growth for Q3 2025, with revenue forecasts in the range of $99.75 billion to $99.91 billion. This would represent a year-over-year growth of approximately 13.2%. Adjusted earnings per share (EPS) are expected to be around $2.26. Key areas of focus for investors will be the performance of Google Cloud, which is anticipated to grow by about 30%, and the growth of the Google Services division, which is expected to increase by around 10%.

3. Stock Price Performance Over the Past 90 Days

Over the past 90 days, GOOGL’s stock price has seen a significant increase. The stock closed at $192.86 on July 31, 2025, and rose to $267.47 by October 28, 2025, marking an increase of approximately 38.7%.

The primary drivers behind this upward trend include:

• Strong Financial Performance: The company reported strong Q2 2025 earnings, with double-digit growth in Search, YouTube, subscriptions, and Google Cloud.

• AI Momentum: Positive developments in AI, including the rollout of Gemini AI across core products, have been a key factor. AI-enabled features like AI Overviews and Circle to Search have improved user engagement in Search, and the Google Cloud’s AI suite has seen rising demand.

• Cloud Growth: Google Cloud is expanding at a nearly 30% clip, contributing to the positive sentiment.

• Favorable Legal Outcomes: A favorable ruling in the DOJ Search Commercial Agreement case has removed a major uncertainty for the company.

• Share Buybacks: A 7% reduction in total shares outstanding due to approximately $200 billion in share buybacks since 2022 has also supported the stock price.

1. GOOGL’s Business: Recent News

Recent news about Google’s business has been focused on Artificial Intelligence and the outcomes of legal challenges. The company is regaining its footing in AI-driven search after facing competition from rivals like ChatGPT, and this is helping to stabilize its advertising revenue. A significant development is the impending launch of Gemini 3, a new AI model that is expected to be a major upgrade and will be integrated with many of Google’s products.

A federal judge has ruled that Google’s payments to be the default search engine on smartphones violate antitrust law. While the ruling found Google’s agreements to be anticompetitive, the judge did not order a breakup of the company. In a separate case, a judge sided with online news publishers who allege that Google has unlawfully monopolized the digital advertising market.

2. Major Developments for GOOGL

The most significant developments for Google in the past 90 days are in AI and its cloud business.

• New Products: The rumored launch of Gemini 3, a powerful new AI model, is the most anticipated new product. It is expected to be a major competitor to other AI models and will be integrated into Google’s ecosystem, including Android and Workspace.

• New Major Clients: Google Cloud has secured a number of high-profile clients and partnerships:

  • Anthropic: The AI company is expanding its use of Google Cloud technologies in a deal reportedly worth tens of billions of dollars.

  • Meta: A deal valued at over $10 billion will see Meta use Google Cloud to power its infrastructure and AI workloads.

  • KPMG: The consulting firm is investing $100 million to develop AI-powered enterprise solutions using Google’s AI technology.

• Legal Developments: The antitrust ruling from the Department of Justice, while not resulting in a breakup of the company, will likely lead to changes in Google’s business practices, particularly concerning its default search engine agreements.

3. GOOGL vs. META vs. MSFT: A Comparative Look

The news for all three tech giants is dominated by the theme of Artificial Intelligence, but there are notable differences in their recent developments and business trends.

• GOOGL: Google’s narrative is a mix of AI advancement and regulatory challenges. The impending launch of Gemini 3 is a major potential catalyst, while the ongoing antitrust lawsuits present a significant headwind that is unique to Google among the three.

• META: Meta’s story is one of strong financial performance and a focus on integrating AI into its social media platforms. The company’s revenue has seen a significant year-over-year increase, and it is aggressively investing in AI to drive user engagement, particularly with its “Reels-first” strategy on Instagram.

• MSFT: Microsoft’s news is centered on the success of its AI and cloud strategies, particularly with its Azure platform and “Copilot” AI assistant. The company is seeing strong growth in its enterprise cloud services, and its gaming division, led by Xbox and Game Pass, continues to perform well. Microsoft’s focus is on the enterprise applications of AI, which is driving significant revenue growth.

In summary, while all three companies are heavily invested in AI, Google is also navigating significant legal challenges. Meta is focused on leveraging AI to enhance its core social media and advertising business, and Microsoft is capitalizing on the enterprise demand for AI-powered cloud services.

Anyways, the purpose of this article is not to sarcastically pat myself on the back, we’re here to learn and grow! Writing is always a great way to prompt some self reflection, so hopefully this is helpful for myself, and for those of you reading. We’ll get back to analytical content tomorrow if you’d like to subscribe. Cheers.

Subscribe now

Setting The Scene:

My background is one quite atypical for the buy-side, which I’ve mentioned before but we’ll briefly do so again. I graduated with a degree in math in 2021, and let’s just say Covid making my last two years of college online didn’t do a whole lot to give direction. In March of that year I started applying to basically any job that had “analyst” in the title as they all paid decently well and I didn’t really care what I was going to do as long as it never involved writing a proof again.

For whatever reason, I was hired to work for a family office, essentially tasked with monitoring their internal equities portfolio as well as manager allocations to ~10 long only funds. When I say family office to be clear it was one guy who made a bunch of money and instead of retiring decided to read a bunch of Buffett and hire college students every few years so he could have 1-2 people sanity check him (and handle paperwork). His eternally gracious wife for some reason tolerated this for the most part.

The blessing and curse of such a situation is that you are essentially forced to Figure It Out in regards to effectively doing your job. There were no endless nights working in IB honing my model formatting skills. There were not years of “rigorous” training in PE on valuing companies and analyzing operations. Just a Google Drive full of hedge fund letters and a goal of beating SPY by 3% annually.

With the help of my wonderful ex-coworker I managed to not get fired for the first 2 years as I learned, which also just so happened to coincide with a significant market crash, and total economic mayhem. In all honesty such a situation is arguably a blessing. The obvious reason being my “Official” track record starts in 2023. The less obvious one being that one of the best ways to even the playing field and outperform despite a lack of experience, is to work in a field where nobody has any idea what is going on.

Like if we look back objectively, it’s pretty hilarious how well everything went. The economy was suddenly ground to a halt, tanking markets. World health organizations were lying, incompetent, or both. Unruly populaces were locked down, unlocked down, locked down again. Then we just printed infinity dollars of stimulus money and hoped for the best. Suddenly the infinity dollars of stimulus started to create inflation, oops, so interest rates ballooned fast enough it killed numerous banks and had (some) people wondering if Schwab was liquid.

Then to cap off all of that covid fun, OpenAI drops ChatGPT and we just kinda moved on to “Stocks Always Go Up” AI edition. A global pandemic, credit crunches, absurd inflation, and SPY is up >100% in 5 years. Not a whole lot of people could have predicted that.

All that to say, I’d argue it was a pretty interesting time to get into public equities. Absurd highs, absurd lows, more absurd highs, all in such a concentrated period! Earlier this millennia markets at least had the respect to wait 7-10 years per cycle. While I of course am no historian on the topic, I’d say the job and markets themselves have changed quite dramatically over the past 5 years compared to prior periods. I do genuinely think there is still tremendous room for alpha basically irrespective of strategy, just take a look at META’s stock chart or quantum or BYND or OPEN or something if you want to argue markets are super efficient.

Separately, I do feel like I’m wired pretty well for money management. To date I haven’t withdrawn a cent from my PA despite the fact I can now technically afford the supercars and watches I was obsessed with in high school. The numbers are just numbers for the most part. I still felt like throwing up occasionally as I watched my annual salary get erased in 5 minutes post earnings every so often, but overall the nerves were honestly less than I had playing junior golf tournaments as a child.

I’m also much more concerned with “being right” than being a people pleaser, which led to the occasional hiccups with my old boss as I’ll happily argue until myself or the other party is convinced. Not “being right” in the sense of refusing to be wrong, but truly in the “if you know better please explain to me way” which is not exactly your role in your early and mid twenties, but alas. I do think it’s quite important to “manage” ones spine to not get overridden by thinks that make no sense, which can be quite frequent in markets.

And given all that, I’m not actively looking for a new role as an analyst/managing capital (for now?). So why?

Gardening Leave:

I’ve been asked a couple times about raising money for a fund. I am by no means an expert in the matter, but I do have a few thoughts.

Firstly I’d say that while the track record I have is absolutely phenomenal, it does have some tarnishes, both to potential investors and to me. The main issue is concentration. Running long only at 100% Net Exposure with 4-5 stocks is not something that typically gets allocated to. Like realistically as a percentage of the AUM in the industry, I’d imagine it’s miniscule. An allocator can always tell themselves it was a few lucky bets, not a proven system of generating excess returns. The whole obsession with proven systems means the largest pools of capital are actually managed by firms that underperform SPY.

Of course I am partly joking, the quants and multi-managers of the world do seem to be quite good at extracting value given the circumstances, just a bit too black box for my taste if I were to be dumping in billions.

The investments I made in the account I managed basically all worked out (besides CDLX, which gave tastes of glory to make the fall that much sweeter). At the same time, many many things worked out. It’s a fun exercise to go through your watchlist/old starter positions and see that despite all the narratives if you had just thrown a dart in 2023 it likely returned 100-200% or more since.

Carvana of course was the primary exception to this, which thus far has returned about 36x on the initial investment in that account, bless the Garcia’s and hopefully not jinxing anything. Carvana took up the majority of my time and I really lived and breathed Carvana for months in late 2022/early 2023. I honestly do not understand how managers can have 15, 20, 30+ stock portfolios that they feel they truly have an edge on. On the flip side, allocators and other market participants have no idea how I sleep at night with 4-5 positions, and currently ~3 in my PA.

I’ll get more into theory later, but the point is that my sense is allocators really wouldn’t be chomping at the bit to allocate to anything I’d fundraise for. I also feel some responsibility to prove to myself that continued outsized returns are repeatable before I took responsibility for large pools of capital. Experimenting with retirement funds seems dubious despite the significant upside of personal compensation.

Now assume I could theoretically find allocators to write me a check, what would that look like? I’m personally just not really the type of person to enjoy the monkey dance of raising capital. I’ve been on the allocator side as well and it’s truly absurdly hard to get confidence someone will do well if you don’t know them personally.

The natural conundrum any smart allocator will face is that there is no solved method of obtaining alpha.

Now this is rather trivially true, otherwise it wouldn’t be alpha, but really sit and think about it for a moment. The way the majority of dollars are allocated to managers is via check lists and investment committees and a whole lot of systems. So not only do allocators have to fly blindly and hope their systems are arbitrarily good at obtaining alpha, but it has to generate enough alpha to justify fees, time spent, and taxes (if applicable).

So yeah just get some confidence a random guy your employee met via a cap intro program will outperform SPY by 5%+ indefinitely.

I do all of that calculation in my head and just think “wow I would probably be really annoyed having conversations where we skirt around how absolutely absurdly hard that is”.

So instead I opted for gardening leave, aka tinkering around with SaaS and AI. It’s entirely possible I more formally explore equity management again in the future. As I said a large portion of my decision is basically me saying “hey can I really trust myself to do this well consistently” and time/iteration certainly solves that. And to insert some more sarcastic bragging that hopefully doesn’t jinx anything, at this rate my PA is growing 100% a year, so maybe I won’t need to be a monkey for seed capital soon!

More On Theory:

If you didn’t get roll your eyes hearing a 26 year old discuss an industry he participated in for a few years like an old man, now you can roll your eyes at said 26 year old discussing investing theory! Is it good theory? Or are past returns not indicative of future results? Fair warning, I do ramble here so feel free to skip.

An important thing to keep in mind when reading this article and others, is that everyone who’s done well with stocks thinks they are good at picking stocks. Even many who don’t do well just think they are unlucky. Part of the problem here is that there’s not really a good fundamental basis for what makes a good stock picker. Like someone who is “good” at golf we can just take a look at their handicap. Someone who is “good” at Chess we can take a look at their elo. Stock picking is much different, as the end result can be totally detached from the quality of work and ability.

CDLX for example has at times been a multi-bagger return, and at times been a 75% loss. At times it was legitimately just a very good bet based on informational advantage. Generally, it’s been awful to hold longer term. If the company turned itself around tomorrow and 100x’d in 5 years would everyone who sold this year be an idiot? Would those who bought it for $3 during a Twitter induced pump be genius stock pickers?

Over longer periods of course returns will normalize, but that’s a pretty awful way to determine current and future alpha, as perhaps the driver of alpha is temporary. Being early to AI was legitimately life changing for many NVDA investors for example, that doesn’t necessarily mean they are exceptional general stock pickers over longer periods, but their track record will basically always trounce the S&P due to that very good bet.

So back to theory - how can we consistently find differentiated positive results to hit the NVDA’s of the world repeatedly while also being cognizant that any systemic way to do so is implicitly impossible?

Before diving in I have two main beliefs that support everything else:

1. The world is essentially deterministic

2. Your model of it is wrong and you don’t know how

The “unknown unknowns” makes some people shake their head and just toss money in an index, which is quite honestly the best way to go for most, but that’s not why we’re here! We all want to outperform each other! To do so we just need to make a wrong model that’s right.

My basic hypothesis is that the best way to do so is play Show and Tell with everyone else.

Like yes, it is quite important to have emotional temperament, be able to open excel, be able to read, and what have you. Those skills however I would consider “table stakes” if our goal is to significantly outperform the market.

What’s interesting is that they weren’t always right? Like if we go back in time to 1930 or something. I’d imagine the average investor had no fucking idea what was going on, as did the average company. Like it’d be absurd to think the world has gotten drastically better at sports, literacy rates, academic discoveries, whatever, and not investing. In the 1930’s you probably couldn’t find a high school teaching calculus, yet nowadays we have kids reciting real analysis textbooks in elementary school. The men’s 100m sprint winner of the 1932 Olympics would not even be the best high school sprinter in the US these days.

The human race and thus markets are dynamic and continuously grinding towards higher performance, so I’d argue what’s more important than basically anything else is to continuously work to understand what everyone else is doing, because what you are doing being good or not depends heavily on them.

In simple terms, the number one thing to care about is what a stock’s price is, what expectations are, and why those expectations are believed.

This is actually a really hard question surprisingly! It’s also I imagine part of why AUM is moving increasingly towards the quant/pod model which is much more in tune with this idea.

I think it naturally kind of follows that in the earlier days of professional asset management all it took to be quite good was some fundamental analysis. I’d imagine 20-30 years ago audits were worse, the average investor was less informed, alternative data sources were far less common, computers weren’t arbitrarily sniping massive price inefficiencies, etc. Nowadays investors are more informed, computer pick up free money much more often, it’s easier to get reads into business performance and cover numerous companies.

In a non-systematic world, a system becomes alpha, then the system becomes standard, and front-running the system with data becomes alpha, and so on.

So for us lowly stock pickers, what do we do? Conceptually the game being played now I think is very rewarding for 9-18 month narratives. I believe this as I look at AUM flows moving towards quarterly/data release firms and repeatedly see very high volatility for “narrative violations”, both up and down.

Some examples recently are:

1. META dropping to $80 on CapEx/Metaverse/Apple changes

   1. The CapEx was clearly not just Metaverse

   2. Where else were companies going to buy ads? Newspapers?

2. GOOGL dropping to $140 on “AI loser fears”

   1. They were trivially performing very well with enterprise model capabilities

   2. TPU’s are clearly very good

   3. ChatGPT was obviously being used for queries that didn’t have high overlap with Search monetization

In these cases you performed exceptionally well with a very very basic thesis, because the narrative was a near quarters data narrative that seemed totally out of sync with the underlying technology.

Since price depends on what other people believe, I think it also follows that stocks/industries that don’t fit the “norm” will have higher outcome variance, positive or negative. Financial markets clearly underestimated AI, and arguably continue to do so. Having a basic understanding of that area has been a massive alpha pool. Similar to SaaS being poorly modeled earlier this decade, or the internet being massively overvalued in the 90s. As the world continously trends towards “data enabled” understanding, being able to contextualize can be alpha.

How does one contextualize? Depends on industry and is very hard. This is another part of why I’m taking a swing at a non-investing focused role. My conviction in Google was gained because I had time to play with their products and not track 30 arbitrary stocks each day. If Google was more private with their models and not giving them away basically for free it’d be harder to tinker with. CDLX for example might land a massive new partner tomorrow and we’d just not know until it’s explicitly stated in an 8-K. This is interesting because intuitively mega caps would be more well covered and harder to unlock alpha in than small caps right? So it’s very difficult to truly know ahead of time in a systematic way how to really properly understand a given company.

Carvana was a fun one because bears were and still are very vocal about their thesis, which was clearly outright incorrect. Of course random bears on Twitter aren’t the majority of short volume, but it’s helpful to understand opposing arguments and I sort of obsessively sought them out. Battle ground stocks are honestly easier for me to own because of the clear bear thesis. It’s much easier to work backwards showing something false isn’t true, than asserting a company has the mandate of heaven to grow bottoms up. I always just thought “if Carvana can have slightly down units on insane price taking, longer delivery times, macro struggles, and incessant negative headlines, I don’t know what they will grow as that normalizes, but it’s much higher”

Going purely on relative momentum of course can be bad and is part of why CDLX did so poorly for me. The idea was “the product is growing X% and getting better with no real alternatives so it should grow more”. What this failed to consider was what their customers were thinking, where they were basically happy to grow budget in the new fun channel that was ramping from volume growth, and despite the product getting better over time it wasn’t a “hot thing” anymore to be doing card linked offers, so budgets were moved (plus mismanagement, etc).

In really simple terms if I look through the investments that have done well for me like GOOGL META CVNA BLND NTES TPR etc the core theme really something like “very good product has temporary narrative violation that will be quickly resolved or maybe they even accelerate”. The issue with CDLX is the narrative violations were never solved, they just compounded, and honestly the product wasn’t really that great from the eyes of advertisers irrespective of things like “ROAS” (partly why I have a preference for B2C investing).

I’d imagine this style loses some of its edge as AI gets better at picking up sentiment and such from a wider lens. It will likely also be relatively harder for companies to bounce back from narrative violations in tighter financial conditions.

Overall though, this section is getting very long and a bit of a ramble, just wanted to note some thoughts down on the ever elusive topic of investing theory before going back to just posting some raw analysis. We’ll see what works throughout the rest of 2025 and into 2026 and I’ll keep everyone updated, fear not.

Next:

Enough long introspective stuff. Tomorrow we get >100% of my net worth reporting earnings as CVNA/GOOGL represent ~95% of my account which is ~120% net long.

I’ll post an update on those tomorrow with the help of some Clarity AI features that are hopefully publicly scalable soon. More on that later as well. Cheers.

Share

Clarity AI:

In my last article I shared I would be focusing on building AI tools for financial research, and Clarity AI is the very first iteration of that effort. While there’s a lot of focus in equity markets on what companies can do to increase the quality of AI model output, there isn’t much focus on what users can do. I think this is a material disconnect, as you can actually squeeze a lot of juice out of current model capabilities with some clever usage. Given how volatile this year has been, that extra juice can make a huge difference. I also know that basically every fund I’ve spoken with this year is desperately trying to integrate AI into their research process, so I’d like to assist.

How’s it work and why’s it different?

The initial iteration of Clarity AI (link here) looks like this:

Before diving into each component, I want to highlight the overarching goal here.

With Clarity AI I want to abstract away your busy work with research. I want to proactively grab the files you need, the data you need, and send you the high impact research that makes a difference.

Or in cheeky terms, productizing myself.

To do this, I do the following:

1. Finding relevant files and data so the model can answer your question can be a chore, so I handle that for you.

   1. Currently this is stock price data, financial ratios, historical earnings transcripts, 8-K’s, 10-Q’s, and 10-K’s for the Nasdaq 100.

   2. More tickers and more data integrations are planned

2. You hire analysts for proactivity, not reactivity. With Scheduled Queries you can create future dated analysis that keeps you in the loop.

   1. Want a daily news summary on 100 stocks but don’t want to have 100 chats every day? Schedule a daily recurring query.

   2. Too many earnings calls for one day? Schedule a future query that will automatically grab the quarterly filing and earnings transcript and send you a report.

   3. Forget your anniversary often? You can schedule a query for that too.

3. Not everyone wants to toil with model settings to optimize outputs, so I made Workflows where you can browse and test pre-made queries.

   1. If you are a pro prompt engineer you can create and publicize workflows so other users can admire your genius

   2. Clarity will also make our own Workflow templates that you can either subscribe to or adjust for your own questions.

Now as a disclaimer, this version was released today, so totally possible there are bugs. I’d greatly appreciate if you DM me or email me (Indra@Clarity-Markets.com) if you have any issues. It’s also very early on data integrations and ticker expansion, so I have it totally free as I’d like feedback on what tickers/data/features you think would be impactful.

So if you are a new user I’d recommend going to Workflows and browsing some templates I uploaded for . You can then edit those, try them out in chat, or just subscribe. If you’re brave and want to make public workflow contributions, that’d be cool too. It won’t show your name or publicize files you upload, so no pressure.

Currently you can set up queries to happen on a daily/weekly/monthly/quarterly basis. You can also just enter a ticker and have it pull the scheduled earnings date for any future queries related to earnings. In theory if it works properly, you can also have this repeat for future earnings. I’d like to add more event based scheduling such as price movements, 8-K’s, Form-4’s, 13F’s, etc. It’d also be fun to have things like “New CEO LinkedIn post” or “Management profile removed from website with no 8-K” for those quirky SMID caps that like to leak alpha.

By entering a ticker you can have the tool grab SEC filings/price data/earnings transcripts to add to context. You can also provide URL’s which the model can browse (some are blocked, nothing I can do on my end) or your own files like research templates or proprietary info to augment the response.

For advanced users you can play around with model settings that are abstracted away in the web client you’d find at ChatGPT.com or Gemini.com. Google Search and URL Context are very powerful tools if the info you are looking for is recent. You can also swap between Model types (currently Gemini 2.5 Pro and Flash) depending on your needs. Temperature let’s you control how “deterministic” a response is. Looking for daily news flow? Turn it up and let the model run. Looking for a specific number from a doc? Turn it down so the model doesn’t overthink.

In Workflows you can browse your own self-created Workflows or public ones. I’ve added tags and basic search capabilities here so you can track down what you’re looking for. Public workflows will either state a source of Clarity or User so don’t fret getting outed for being into mining stocks. Some major to-do’s here for me are adding more templates beyond the few CVNA ones I have up now, as well as better management of workflows so you can manage your subscriptions and filter as more content is added.

The chat function exists so you can either just chat with the robot as you normally would on say a Gemini.com or ChatGPT.com, but also so you can test out Workflows before subscribing. The UI here can be a bit messy as of now, but it supports the same tools and context management as queries. It also saves your history if you ever need to go back to something.

That’s a wrap for the basics of Clarity AI. I’d love to see users engaging with the platform and creating Workflows. If creation isn’t your thing give the public workflows a browse to see if anything catches your eye. The function that finds podcasts for a given ticker is one of my favorite niche use cases.

In the future I’d like to essentially allow you to upload a portfolio and automatically curate a list of research flows, which would include proprietary data integrations like alt data, sell-side estimates, social media sentiment, and more. My hypothesis with AI is that basic chat isn’t the best use case for enterprise workflows where there’s a lot of recurring work and requisite data integrations, so I’m trying to handle that for you, and empower users to stay on the cutting edge of model outputs.

Which brings us to my industry AI ramblings.

How will AI evolve for office work?

Pretty much every week at this point has dozens of AI press releases that leaves investor heads spinning. AMD and OpenAI just casually moving around tens of billions in future dated commitments. Everyone and their mother announcing some kind of MCP thing or Agent thing or what have you. During such times it’s important to ground and remember the plot, which has basically two components:

1. We need 100x+ more compute/memory/power/etc. There is a price at which those builds aren’t accretive, but they absolutely will be used. Token usage will be orders of magnitude higher in 5-10 years as enterprise use cases scale. Just testing Clarity AI work flows over the past 1-2 months as an individual used more tokens than probably 99.9% of users, and I haven’t even scratched the surface of data integrations.

2. There is no standard for how this stuff will be used and everyone is flying blind.

These two points can cause a lot of disconnect, where it’s like “wait, how are you so certain we will need so much more if we don’t even know what to do with it”. Queue the bubble talk.

Basically even Fortune 500’s are in the same boat as me when it comes to creating AI workflows. We’ve never seen technologically change so fast and propagate so fast. Like a year ago the majority of queries I’m using now would be impossible. Even 6 months ago on early Gemini 2.5 releases they hadn’t added as many tools/memory capabilities so queries were unstable. How many large companies are able to rapidly shift and have dynamic development schedules? By the time most have scoped a project, there’s a new model drop with new capabilities. Add on the fact the model labs also had no set structure for how to interact with models (the API has changed a couple times which is a headache).

Everything is stabilizing and will continue to stabilize now that frontier models are so good. OpenAI and Google will see more focus on product enablement than pure capabilities racing, at least in text. So basically, we are entering a world of tooling. The compute will absolutely be needed, and the next 6-18 months will show us where.

My thoughts on where:

As I said above, text is really not a great way to communicate with AI, especially if we suddenly give chat access to literally everyone. Early adopters of AI based customer support quickly found this out quite quickly. I’d wager 80% of customer support volume for a given company comes from ~5-10% of users. This also likely translates to internal AI based chat tools having issues sent to IT/engineering.

My mother for example has essentially no understanding of computers. I have been asked to come over and help her since the internet is out. When I get there, the internet is not out, she just opened the wrong Chrome profile and couldn’t login to Ancestry.com. I have been told the printer spontaneously stopped working, then gone over and see she accidentally unplugged the power code, and put it back into a random hole in the back. My mom had a full ride college scholarship which she turned into a successful career as a CFA/CFP. She still has absolutely no idea how to even begin asking the right questions about anything computer related. What would help her tremendously, is if the computer could be proactive, and help her within the mediums she knows.

If we assumed for a second that I just created an AI tool with operating system access, on her computer, I could automatically schedule an email/text which I know she can read, and that email/text could tell her the Chrome profile was switched and how to switch it back, it could tell her if the internet was really out, etc.

This is the future of SaaS in my view. There will be tools that automatically read your email, your Slack, your Google Drive, and they will automatically perform tasks, ask you to perform tasks, send you info to perform tasks, and more.

Consumers that can’t understand credit cards? Wire everything up to an LLM that auto prioritizes higher APR balances for payment. Automatically does a balance transfer.

Fucked up a Github merge? Their agent automatically handles the revert and suggests code changes to fix. Legal uploaded a new file to BOX? Sends relevant stake holders a message/email. Twilio can automatically transform your middle managers ramblings on Slack into a comms campaign.

Proactivity that fits into natural workflows is the future in my view. MCP and all is a first step at giving AI access to these kinds of tools, but really the biggest hurdle here is likely going to be two things:

1. Auth

2. Trust

By Auth I mean actually ensuring everything along the AI chain is authenticated properly. Like with Clarity if I want sell-side, maybe you have a CapIQ sub. It’s totally possible, just not commonplace, that we could take your CapIQ details, have Clarity say “Hello, here is my CapIQ sub” to CapIQ, and return that info to only you. Of course what if I’m just pretending to be Mr.CapIQSub@hedgefund.com? There isn’t a great standard for that. Perhaps the best thus far is Google SSO which Clarity and thousands of other platforms support, so watch out for Gemini 3.0 and their agent capabilities…

With trust I mean trust in the model to actually do anything I outlined well. We are getting there, but again enterprises are not designed to rapidly iterate. We have current era models doing state of the art math, but don’t trust them to summarize our emails. Give it 18-36 months for senior stake holders to start seeing what the next 6-18 months enable. That’s when we’ll really be off to the races.

I could go on, but in short I think SaaS will basically be determined by whoever can come up with proactive flows to enable user productivity, and make them comfortable for users to use. These will become extremely sticky to the point that even if model providers come along eventually with a solution, it could be a headache to rip out.

So yes, please go try Clarity AI, and maybe it becomes a sticky and productive tool for you.

Cheers.

Subscribe now

To be clear, this article will be about AI, not pop psych. Sorry to disappoint.

Roughly five months ago I put ~10% of my money into a ~$2T market cap company. Five months later it has appreciated ~60%. That company is Google, and I still believe it, and perhaps the entire market, are drastically mispriced. This article explores why.

Hypothesis:

I’ve been working closely with AI tools for a variety of tasks over the past 12 months. This experience has lead me to believe the following:

1. The average investor has a poor understanding of how models work

2. The average investor drastically under estimates model capabilities

The natural conclusion if both of those are true is that many sections of public markets are obscenely mispriced. So let’s explore.

How Models Work Basically:

As a note, everything I say is purely my on conclusions from observing and interacting with models. I (unfortunately) do not have insider contacts at any frontier labs. I don’t have proprietary expert networks, I’m just taking the capabilities I have observed to their logical conclusion.

We are actually going to stay relatively high level here if you were worried about the focus being on the math or code behind the veil. Our focus is essentially going to be “how do we evaluate input and output quality, and what are the consequences if qualify of inputs and outputs changes”. Like if a model can correctly answer “What is 2+2” that’s a neat technological feat a few years ago, but not particularly useful given the many other ways to obtain that information. All we care about is the net economic impact and disruption as model outputs get arbitrarily better. So how good are they?

To start, I would argue that the key to understanding models is to understand ourselves. How do us humans interact with the world? We have a central processing unit in our brain that receives stimulus. It responds to stimulus and sends out signals to a variety of body parts to execute different functions. A brain in a jar is actually a very useless hunk of flesh. Similarly, a body without an active brain is entirely functionless.

The single most important insight you can have about models is that they work basically the exact same way.

Essentially a model is a core logic engine, this is what all the money and time is spent training with weights and reward functions and technical jargon. The product you actually interact with when you ask ChatGPT or Gemini a question is NOT that central processing unit. It is the processing power filtered through a variety of tools.

For example, let say I ask you “whats the price of NVDA”. Your responses will vary. Some may follow it closely and know its roughly $180 recently, so they say ~$180. Some maybe haven’t checked in 6 months so they say ~$130. Some may have no idea and say no idea. Some may look it up and quote $187.68 or whatever the current price is as you read this article.

Let’s ask Gemini the same question in AI studio, where I have express control over a variety of model settings.

This of course is entirely worthless output. If I put a gun to the head of any reader and demanded they find me the current stock price of NVDA, they would pull out their phone and quote it within a minute. If models are so smart, why can’t they perform this basic task?

Well suppose I put a gun to my dear readers head (I would never) and demanded they tell me the current price without using the internet. You would take a wild guess and say that is extremely unfair. Models are much the same way. I can figuratively put a gun to the head of the model, it will never answer this question correctly. Let’s try it.

Much like you dear reader, a model cannot divine information irrespective of how smart it is. Demanding it produce information that it physically cannot produce is a common occurrence that causes significant misconceptions around capabilities. The model itself is simply a logic engine. What I will argue however, is that as tooling gets added, it is a much much much better logic engine than you think.

Let’s even the playing field and ask the model the same question, but enable access to Google search.

And there we have it. By enabling access to the Google search tool, I can now have an intelligent slave that tells me the stock price of NVDA.

The tooling build out is still extremely early. These frontier labs have essentially put an Einstein brain into the body of a deformed toddler. It does not have the proper tools to execute the tasks it believes it should do for most job functions.

Model Capabilities

The stock price of NVDA is an easy query. What if I tasked you with extracting the 3 financial statements from NVDA’s most recent 10-K? It’s not a particularly difficult task, it only has a few component steps.

1. Go grab the SEC filing

2. Find the financial statements

3. Open Excel

4. Insert the information into excel

A high school intern can easily complete this task. Models by default cannot interact with xlsx at all.

Yet I created my own Python script that can execute that task utilizing the Gemini API. It can extract those values in 5-10 minutes for the costs of a few cents, drastically lower cost than your time I’d hope. I did so by creating software that shifted the inputs and outputs from xlsx to a form the model could interact with, so it could use its super human pattern recognition for extremely fast data extraction.

Let’s explore a question that requires no tooling, no fancy tasks, just plain old math.

You have 30 seconds, please find the integral of the following:

(sinx+cosx)^3

I have a math degree and have no idea where to even start without a refresher. The model however has near perfect recall and trivially solves this problem in different ways in 30 seconds flat with a tutorial to boot. (the formatting is a bit wonky for the solution in AI studio, my apologies)

If you aren’t yet convinced, fear not, we will now venture into the world of coding for more examples.

Frontier labs focusing on developing their models code capabilities makes a ton of sense. It provides a variety of benefits and is a great place to build.

1. Code is a very high skill product, so coding assistants are extremely valuable. A trivial look into the market caps of start ups in the space makes this obvious.

2. The frontier labs all code, so coding assistants enable them to just work faster

3. For the most part, code either works or it doesn’t. This simple property of having objective right/wrong answers makes training much easier.

4. Tooling requirements are much simpler than lots of other knowledge work.

5. An absurd amount of high quality public training data or access to centralized private repositories like Gitlab/Github. Much other knowledge work is silo’d internally at firms.

So how good are models at code and how do tools impact that?

As an example, ~4-5 months ago when Gemini 2.5 models were quite new, I added the Gemini coding assistant to my visual studio. At the time, if I added more than a few dozen lines of code to the prompt, the assistant would run out of memory and not respond to my prompt. It was functionally worthless compared to Cursor or Copilot.

Since then, magic has happened. Google added higher memory limits and compute limits, allowing drastically larger context. Google added the capability for the model to automatically read files within the repository. For those not in the know, many “coding” projects have dozens or hundreds of separate files, often with information from one file being used in numerous others. Previously users had to manually tag each file into the context window, the assistant now intelligently determines connections. Google added an “agent mode” which is essentially allowing the model to have a chain of thought and string together actions.

As an exercise, I downloaded a fresh install of Visual Studio Code on my laptop while writing this blog. You can do so as well, it’s free, so feel free to follow along. I then added the Gemini Code Assist extension.

I then gave it a simple instruction: create a simple javascript game

This is essentially the same task that one might get in a Computer Science course during their say 2nd semester. Not to brag or anything but I was the whole Presidential Scholar National Merit something something. I would spend 8+ hours on my CS projects and did quite well in those classes.

Here’s what the model spit out in about 30 seconds.

A full working game of Tic-Tac-Toe you can play in your browser. Three separate interconnected files, ~200 lines of code, 30 seconds, a handful of cents of compute usage.

That is orders of magnitude faster and cheaper than any human.

Let’s spice it up if Tic-Tac-Toe isn’t your thing. I asked Gemini for a prompt to show-case agentic capabilities and it suggested the following:

Which after about a minute of work gets us:

An interactive product catalog with a locally hosted “back-end”. It created 5 files, installed required dependencies automatically, wrote hundreds of lines of code, again for a few cents and less than a minute.

This is all one shot, I did not run these multiple times, I am writing this blog and copy-pasting screen shots as I approve changes the agent makes live.

Of course the UI here is pretty basic and a shopping experience that looked like that would fail. So let’s try a full re-write of the front end using modern tools, make the model actually work a bit. I’m going to skip some steps here as following along will get difficult. We’ll need to install Node.js manually as the model cannot (yet) and change some Windows system settings to enable running the updated store.

After about 10 minutes we get the following.

Now for the sake of exercise I’m not going to actually go grab product images and format out the shopping experience, but this version is very close to a basic e-commerce site (sans all the external integrations). It would obviously be better if say Gemini Code Assist could query Gemini image gen or Gemini video gen to create files, but that tooling hasn’t been fleshed out.

It did however create about 20 different files, import dozens of different tools, set up a robots.txt file, a .gitignore file (for Github purposes) and write hundreds of lines of codes that can be further refined with natural language.

The financial equivalent would be your analyst returning you a fully functional DCF with default excel formatting and hard coded live stock price/forward estimates because he didn’t know how to import your Bloomberg data.

That would be a pretty good one shot from a junior in 10 minutes.

Additionally, at this point I’m convinced any moderately intelligent human can simply execute AI instructions to create full-stack applications. Linking the basic “site” above to an actual database for products, an actual host so it’s web accessible, and getting actual product images can all be easily done by simply asking the AI how to do so. It knows the correct answers, it simply doesn’t have the tools to execute those answers itself yet.

So Google, OpenAI, Anthropic, and the rest, are month by month, continuing to add tools to that malformed Einstein baby they created. The results will continue to impress, as the core logic engine is absurd.

After all, if a specialized version of Gemini can achieve IMO Gold already there really isn’t much hope for us meatbags once it gets its hands on full access to a computer. It’s an Excel and Bloomberg training session away from replacing half the finance industry’s output.

What’s All This Mean:

I have a few forward looking ideas. If we subscribe to the fact that models already have “brains” on par or better with the vast majority of the American workforce, it has some large implications.

1. Memory and compute are going to continue having demand that is a straight line up. Simply providing the same model with the same tools longer time to compute and more information to go through dramatically improves performance. We are currently heavily rationed on that front, artificially limiting use cases

   1. A natural result of this as well is that every physical component of that value chain will continue to grow incessantly. Power, fiber, NVDA, TSM, INTC, etc.

2. The world of tools and integrations will become increasingly relevant. A large part of this will be from the model providers themselves, however all established companies with exposure to the internet will need to consider their place in the eco-system. Start ups have and will continue to emerge that focus on creating tools to enable the models.

   1. How does a model shop for products? What do Visa/Mastercard need to do?

   2. How does a model do X knowledge work task? Excel needs to make their software agent friendly. Workday, Salesforce, Hubspot, etc, all of these SaaS companies will be working to enable AI integrations

3. Private data becomes infinitely more valuable. We’ve seen inklings of this with Google paying Reddit for training data, but inference is still early. All knowledge work is essentially some function applied to internal data to produce a service. Model providers and others will crave access to API’s and integrations to get the one-up on each other

   1. Imagine Indeed and Ziprecruiter. Indeed gets unfettered access to LinkedIn profiles via API. Ziprecruiter does not. Who dies?

      1. Imagine Microsoft doesn’t let anyone have access and expands here themselves? Do competitors have enough user data to compete?

   2. Suppose OpenAI for whatever reason was banned from accessing SEC filings and stock price data, but Gemini was not. OpenAI would become worse for many financial use cases irrespective of model capabilities

   3. How will companies balance opening themselves to model providers with the fact that their data is their business?

4. Labor will need to look different in the future. If all of your time spent reading earnings calls/modeling is suddenly compressed by 50-80%, where do you spend that time? What does that enable? How does that look for other roles? How does that continue to develop as we continually improve AI to meet those new work flows?

   1. Do you cover 40 stocks instead of 15? 100 stocks?

   2. Do you spend more time talking to management?

   3. More time doing field research?

   4. More time developing custom alternative data?

5. The amount of change required and the speed at which it is required is unprecedented. There is no standard on “How to utilize AI”. Every senior person at a firm got there prior to AI. Their juniors are now coming into a world full of it. How does that develop?

   1. An IB analyst spitting out a deck a day 10 years ago would be super human. Now it’s guaranteed to be AI generated. How do you determine who’s good at their job? How do you evaluate output that isn’t necessarily human reviewed? How do juniors get better at creating with AI?

   2. Updates happen so fast that Gemini is actually bad at helping you integrate itself. Asking Gemini to assist you in implementing the Gemini API into a program by default utilizes deprecated models and versions because it’s seen so many updates since the training data.

I could go on, but we’ll save some content for future articles. For now, I’ll briefly discuss what I’m doing with this in mind.

What To Do:

I’m essentially doing two things.

1. I significantly added to my Google position and will likely continue to add to similar positions such as INTC. I’ve taken short term bets on MU/ASML/TSM/NVDA this year and will likely continue to do so. Google has a host of advantages that are too long to list here.

2. I’m dedicating the next 6 months or so of my life to building out tooling to enable models to be more productive with financial work flows. If they were nearly as capable in finance as they are in code, the market caps of all of these AI exposed companies would likely 2x or more over night. I’ll update more on this once I get to a production ready build.

And the bonus thing is I’ll write about all of it occasionally. I will say having been in the investing seat and now the development seat, I was definitely off base in many of my assumptions in the tech and software area prior to actually working with them directly. Given how good AI is in that arena, I highly encourage you to try it out as well.

Cheers.

Subscribe now

A step in the research process for many is of course finding historical KPI’s and financials and trying to get a rough sense for something like the following:

1. How a company makes money

2. How much money it makes

3. At what margins

4. How that has trended over time

There are a lot of really nice tools for this like AlphaSense’s doc search+Canalyst or free/cheap solutions like Tikr/Quart/Yahoo finance. Typically though to actually get everything in an excel file including not just financials, but KPI disclosures, you’d need a Canalyst subscription, an intern, or to do it yourself.

Thanks to the complete wizards over at Google, this is no longer the case!

Over the past week I’ve been iterating on some code that basically reads a model, then auto-completes it using the appropriate filings. This initial test was for CVNA, but it’s scalable to basically any publicly listed company and even works pretty well for international filings given similar disclosures.

A link to that sheet (the Google sheet version, thanks Substack) can be found here: Link

It currently takes about 15 minutes to fully parse a filing and spit out an updated model, which I’m rather confident is faster than just about any human, and also drastically cheaper!

I’m writing this article because I think that’s a really cool thing, in addition to the other stuff I’m working on. We’re starting to get to the stage of trialing our AI workflows where we probably work closely with 5-10 funds/investors to rapidly iterate and scale out coverage on their names before we get to covering everything publicly listed.

We want to have a product where you can

1. Have access to SEC filings/earnings transcripts/relevant news articles as well as AI tools to assist with analyzing those.

2. Have access to model generations that can be the template for valuation work or used to quickly spin up on a name

3. Utilize our AI generations to find data sources/information sources on industry KPI’s/catalysts/blogs/podcasts/etc. Stuff like Manheim vehicle index and industry reports, podcasts with C-Suite, Substack writers who cover the stock, and more

4. Receive daily/weekly updates on developments at the company, alt data, macro factors, competitor earnings, etc

5. Understand sentiment over time for key business arms

6. How narratives have changed around the stock over the past few years and what events led to those shifts. Why was CVNA down 99% and then up 100x? Great question!

7. So much more.

A lot of that functionality is already live and we’re continuously improving on it. If you’re interested in working with us would love to chat. The models are slowly becoming absurd productivity unlocks, allowing much more depth and breadth of coverage. If that sounds useful to you, shoot an email to Kyle@Clarity-Markets.com or shoot me a DM on Twitter.

Cheers!

Subscribe now

In short, this means you can now start predicting a variety of company KPIs and other financial news to compete for cash prizes.

When you create an account you start with 1,000 “Clarity coin” which you can trade in our markets. Each quarter end we will tally up everyones performance and distribute cash rewards to those with the most profit.

This first quarter the reward is $1,000 to whoever gets first place, which we hope to scale up meaningfully going forward.

Additionally, if you are interested in asking questions to our predictors, we’d be happy to pilot your questions! Our goal is to utilize prediction markets for high accuracy estimates on any metric imaginable. Some examples:

The usual: Macro data, stock performance, and AI.

The new and unique: We are also exploring markets related to specific KPIs and legal outcomes. Have a question about a stock you own? Feel free to reach out.

Our goal is to make a platform you want to use, so please feel free to reach out with any feedback/bugs. A question sucks? Let us know. You think a question would be interesting? Let us know.

We’ll keep iterating here to hopefully be worth your precious time each day! Coming soon are leaderboards to track your progress, feed improvements, more questions, better info to jump start your predictions, referral rewards, etc.

We’re also going to be testing out how different AI models perform which we’ll add to the leaderboards. Think you're better than Gemini? Prove it.

Cheers!

Given it’s a market holiday I figured you may be bored after the exciting past week, so time for another fully AI written article!

What I’m doing is essentially creating “AI agents” that follow a workflow checklist to create detailed modular investment research. What does that mean? First let’s talk some brief context.

Some of you may have heard of Rogo or similar chatbot focused tools. My opinion is that those can be very bad, because prompting models and context loading is still very much required. A very common complaint with chatbot focused models is hallucinations or misinterpretations of instructions. These are largely avoidable problems, but prompting skills are not something the average analyst/PM is spending a ton of time on. Additionally, actual good prompting for longer form tasks is often multi-step and requires massaging of context, not fun things to do for humans.

The solution in my view is to basically read the minds of analysts and have AI copy their workflows to rapidly accelerate content consumption. Basically skip the prompting, proactively create and deliver reports. Like sell-side, but many orders of magnitude cheaper and faster.

The text below is the example output of my “AI agent” for 10-K analysis, run on Carvana’s filings. I tried to define at a high level a bunch of topics like “What does the company do” “How does the company make money”, “What is the track record of capital allocation”, “What has changed since the prior year”. Essentially the same things I’d be looking at myself. What’s cool is I can chunk these out separately. You already know what the company does and don’t need an overview? Sure thing, easy to not include. You want to know changes in the “Risks” section since last year? That can be delivered before your analyst even opens the document. Cool stuff!

Our website will soon have these types of reports in a “closed alpha” type deal covering CVNA/Mag-7/LULU. Once we’re happy with those, ideally we can cover pretty much anything that’s listed publicly. The goal is to provide analysis of filings, related industry reports, earnings calls, alternative data, catalysts, expert calls, etc.

Ambitious, still very early, but I’m excited. If you’re interested in following along, feel free to subscribe to this blog. If you’d like to alpha test, feel free to shoot me a DM or reply to this email. Cheers.

Subscribe now

High Level Summary:

The Bull Case (Potential Long Position)

A long position in Carvana would be supported by the company's significant financial turnaround, demonstrated operational leverage, and large market opportunity.

• Financial Turnaround and Profitable Growth: After a challenging period, Carvana shifted its focus to "profitable growth" in 2024. This is evidenced by a 33.1% increase in retail units sold and a 26.9% increase in total revenue for 2024. Profitability has improved substantially, with net income growing to $404 million in 2024 from a net loss of $2.894 billion in 2022.

• Margin Expansion and Operational Leverage: Key profitability metrics have expanded significantly. Total gross profit per unit (GPU) reached $6,908 in 2024, up from $3,022 in 2022. The company has also demonstrated operating leverage, with Selling, General & Administrative (SG&A) costs per retail unit decreasing from $5,741 in 2023 to $4,501 in 2024.

• Competitive Moat and Infrastructure: Carvana believes its "Purpose-Built Vertically Integrated E-commerce Platform" provides a significant cost advantage over traditional dealers. This is supported by proprietary technology, including 41 issued U.S. patents , a national logistics network, and scaled infrastructure with the capacity to recondition over 1 million vehicles annually.

• Brand and Market Opportunity: The company benefits from strong brand recognition, driven by its unique car vending machines and a high average customer satisfaction rating of 4.7 out of 5 stars. The U.S. used car market is large and highly fragmented, with the top 10 retailers accounting for less than 10% of market share, presenting a substantial opportunity for disruption and growth.

• Strengthened Balance Sheet: Carvana has improved its financial position by reducing total debt to $5.5 billion in 2024 from $6.0 billion in 2023 and increasing its cash and cash equivalents to $1.716 billion.

The Bear Case (Potential Short Position)

A short position would focus on risks related to corporate governance, substantial debt, intense competition, and a history of financial vulnerability.

• Corporate Governance and Related-Party Risks: The Garcia Parties control a majority of the voting power, making Carvana a "controlled company" exempt from certain NYSE governance requirements. There is a significant business relationship with DriveTime, an entity also controlled by the Garcia Parties, for critical services like VSC administration and facility leases. These arrangements were not negotiated at arm's length, creating potential conflicts of interest.

• Substantial Indebtedness: As of December 31, 2024, the company had approximately $5.5 billion in total outstanding debt. This level of debt could affect financial flexibility, requires significant cash flow for service payments, and contains restrictive covenants that could limit future actions.

• Dependence on Volatile Markets: The business is highly dependent on the sale of retail vehicles and the sale of automotive finance receivables it originates, making it susceptible to downturns in the used car market, rising interest rates, and changes in the capital markets. The company has a history of significant net losses, and sustained long-term profitability is not guaranteed.

• Intense Competition and Legal Scrutiny: Carvana operates in a highly fragmented and competitive industry, facing pressure from traditional dealers like CarMax and online marketplaces. The company is also subject to numerous legal proceedings and government inquiries, including from various state Attorneys General, regarding its business practices.

• Equity Dilution and Stock Volatility: The market price of Carvana's stock has been extremely volatile. The company has actively used its "at-the-market" (ATM) program to sell stock, raising $1.3 billion in 2024 but also causing dilution for existing shareholders.

Business Overview:

Carvana Co. (NYSE: CVNA) is a leading e-commerce platform dedicated to revolutionizing the used car market by simplifying the processes of buying and selling vehicles. The company's vertically integrated business model combines a comprehensive online shopping experience with a sophisticated in-house supply chain, aiming to provide customers with a broad selection, competitive value, and transparent pricing in a low-pressure environment.

Core Business Operations:

Carvana's business model is structured to manage the entire vehicle transaction cycle, from acquisition to after-sale support, primarily through its digital platform.

Vehicle Acquisition and Preparation:

• Sourcing: Carvana acquires its inventory of used vehicles directly from customers, as well as from wholesale auctions and other suppliers. This direct sourcing from customers helps to eliminate auction fees and diversify their vehicle offerings.

• Inspection and Reconditioning: Upon acquisition, vehicles are transported to one of Carvana's Inspection and Reconditioning Centers (IRCs). Each vehicle undergoes a thorough inspection of its features, brakes, tires, and cosmetic appearance. Vehicles that meet Carvana's uniform standards are reconditioned, while those that do not are sold through wholesale channels.

Sales and Fulfillment:

• Online Platform: The company's mobile-optimized website, carvana.com, serves as the primary retail channel, featuring a vast inventory of vehicles. A key feature is the patented 360-degree virtual vehicle tour, which uses high-definition photography to display both the interior and exterior of the actual vehicle, including any imperfections.

• Financing and Ancillary Products: Carvana offers integrated financing solutions through a proprietary platform that provides customers with pre-qualified terms without impacting their credit score until a purchase is finalized. In addition to vehicle sales, the company generates revenue from complementary products such as Vehicle Service Contracts (VSCs), Guaranteed Asset Protection (GAP) waivers, and auto insurance offered in partnership with Root, Inc..

• Nationwide Logistics and Fulfillment: A proprietary logistics network, built on a "hub and spoke" model, enables efficient transportation of vehicles across the country. Customers can opt for home delivery via a branded Carvana hauler or choose to pick up their vehicle from one of the company's distinctive, patented car vending machines.

Customer-Centric Approach: Carvana emphasizes a simple and transparent customer experience, backed by a seven-day return policy and dedicated post-sale support from customer advocates. This focus on customer satisfaction has resulted in high average customer ratings and fosters repeat business and referrals.

The company's primary strategy revolves around increasing retail vehicle unit sales, which drives revenue from vehicle transactions, financing, and the sale of complementary products and services. By leveraging technology and a vertically integrated model, Carvana aims to achieve significant operating leverage and disrupt the traditionally fragmented automotive retail industry.

Based on Carvana's 2024 annual filing, the company's revenue is primarily categorized into three main streams: retail vehicle sales, wholesale sales and revenues, and other sales and revenues. For the fiscal year ending December 31, 2024, Carvana's total revenue was approximately $13.67 billion.

How They Make Money:

Revenue Breakdown by Category

A detailed breakdown of Carvana's revenue for the year ended December 31, 2024, is as follows:

Table data sourced from Carvana's 2024 Annual Report.

Description of Revenue Categories

• Retail Vehicle Sales: This is Carvana's largest revenue source, generated from the sale of used vehicles to customers through its online platform. This category also includes shipping and delivery fees.

• Wholesale Sales and Revenues: This revenue is generated from selling vehicles acquired from customers to wholesalers. It also includes revenue from the sale of wholesale marketplace units by non-Carvana sellers through its wholesale platform, which includes auction fees and related service revenues. The acquisition of ADESA in 2022 has significantly contributed to this segment by expanding Carvana's wholesale auction capabilities.

• Other Sales and Revenues: This category primarily consists of income from:

  • Gains on the sales of loans: Generated from originating and then selling automotive finance receivables to third-party finance partners or through securitization transactions.

  • Commissions on complementary products: This includes commissions on Vehicle Service Contracts (VSCs), Guaranteed Asset Protection (GAP) waiver coverage, and auto insurance sold through its platform. Carvana has a partnership with Root, Inc. for its auto insurance offerings. A significant portion of VSC commissions are from an agreement with DriveTime.

Subsidiary and Brand Information

• ADESA: Acquired in May 2022, ADESA's U.S. physical auction business is a key part of Carvana's wholesale operations. For the year ended December 31, 2023, ADESA operations contributed approximately $856 million to wholesale sales and revenues.

• DriveTime: While a separate entity, Carvana has a significant business relationship with DriveTime, a related party. This includes a master dealer agreement where Carvana earns commission on VSCs administered by DriveTime.

• Root, Inc.: Carvana has a partnership with this online insurance company to offer integrated auto insurance to its customers. Revenue from this partnership is included in "Other Sales and Revenues".

It is important to note that all of Carvana's revenue is generated within the United States.

History of Revenue and Margins:

Based on Carvana's (CVNA) 2024 annual report, the company has seen significant shifts in revenue and margins, driven by a combination of fluctuating sales volume, strategic shifts in pricing, and a strong focus on operational efficiency.

Revenue Performance

Carvana's total revenue has experienced volatility in recent years. After reaching a peak in 2022, revenue declined in 2023 before rebounding in 2024.

• 2022: Total revenue was $13.6 billion.

• 2023: Total revenue decreased to $10.8 billion.

• 2024: Total revenue increased to $13.7 billion.

The primary driver of these revenue changes has been the volume of retail units sold. The increase in revenue in 2024 was primarily due to a 33.1% increase in retail vehicles sold, from 312,847 in 2023 to 416,348 in 2024. This increase in volume was partially offset by a decrease in the average retail revenue per unit sold, which fell from $24,018 in 2023 to $23,252 in 2024. This decline in average selling price was partly due to a higher proportion of retail marketplace units, which are recognized on a net revenue basis.

The acquisition of ADESA's U.S. physical auction business in May 2022 also impacted revenue, significantly expanding Carvana's infrastructure and enhancing its wholesale operations. This is reflected in the growth of wholesale sales and revenues, which increased to $2.8 billion in 2024 from $2.5 billion in 2023, driven by a greater number of wholesale units sold.

Margin Expansion

Carvana has demonstrated significant improvement in its profitability, with both gross and net margins expanding considerably.

• Total Gross Profit per Unit: $3,022 in 2022, $5,511 in 2023, and $6,908 in 2024.

• Net Income Margin: -21.3% in 2022, 1.4% in 2023, and 3.0% in 2024.

• Adjusted EBITDA Margin: -7.7% in 2022, 3.1% in 2023, and 10.1% in 2024.

The expansion in margins can be attributed to several factors, including lower vehicle acquisition costs, improved operational efficiencies, and operating leverage.

Gross Margin Improvement: The increase in retail vehicle gross profit per unit to $3,312 in 2024 from $2,385 in 2023 was a key driver of overall margin expansion. This was achieved through:

• Lower Raw Material Costs: The company experienced lower vehicle acquisition costs relative to sales prices. A key part of its strategy is to increase the number of vehicles purchased directly from customers, which are, on average, more profitable than those acquired at auction.

• Reduced Labor and Operational Costs: Carvana reported lower reconditioning and inbound transport costs on vehicles sold during 2024. The company aims to better utilize its existing inspection and reconditioning infrastructure, which has the capacity to handle over 1 million vehicles per year.

• Faster Inventory Turn: A reduction in the average number of days to sell a vehicle also contributed to higher gross profit by minimizing vehicle price reductions due to depreciation.

Operating Leverage: While total Selling, General, and Administrative (SG&A) expenses increased to $1.9 billion in 2024 from $1.8 billion in 2023 due to higher sales volume, the company demonstrated improved operating leverage. Total SG&A expenses per retail unit decreased from $5,741 in 2023 to $4,501 in 2024. This indicates that the company's centralized online sales model and existing infrastructure are allowing it to scale sales more efficiently without a proportional increase in overhead costs.

Key Performance Indicators

Company-Defined Key Operating Metrics

Carvana regularly reviews several key metrics to evaluate its business, measure progress, and make strategic decisions. The company states that its most important long-term measure of performance is the number of retail units sold.

Retail Units Sold

This is defined as the number of vehicles sold to customers in a given period, net of returns under the company's seven-day return policy. Carvana views this as a key measure of growth, as it is the primary driver of revenue and enables multiple complementary income streams.

Trend:

• 2014: 2,105

• 2015: 6,523

• 2016: 18,761

• 2017: 44,252

• 2018: 94,108

• 2019: 177,549

• 2020: 244,111

• 2021: 425,237

• 2022: 412,296

• 2023: 312,847

• 2024: 416,348

Total Gross Profit per Unit

This metric is defined as the aggregate gross profit in a given period divided by the retail units sold in that period. It includes gross profit from retail vehicle sales, gains on loan sales, commissions on complementary products, and gross profit from wholesale vehicle sales.

Trend:

• 2014: $ (201)

• 2015: $ 206

• 2016: $ 1,023

• 2017: $ 1,539

• 2018: $ 2,090

• 2019: $ 2,852

• 2020: $ 3,253

• 2021: $ 4,537

• 2022: $ 3,022

• 2023: $ 5,511

• 2024: $ 6,908

Total Gross Profit per Unit, non-GAAP

This is defined as the aggregate non-GAAP gross profit in a given period divided by retail units sold. Non-GAAP gross profit adds back depreciation, amortization, and restructuring expenses in cost of sales, and share-based compensation, while subtracting revenue related to Root Warrants.

Trend:

• 2014: $ (201)

• 2015: $ 206

• 2016: $ 1,013

• 2017: $ 1,537

• 2018: $ 2,136

• 2019: $ 2,878

• 2020: $ 3,298

• 2021: $ 4,593

• 2022: $ 3,337

• 2023: $ 5,984

• 2024: $ 7,196

Average Monthly Unique Visitors

This is defined as an individual who has visited Carvana's website or mobile application within a calendar month. The company views this as a key indicator of its brand strength and the effectiveness of its advertising campaigns.

Trend:

• 2023: 15,819 (in thousands)

• 2024: 17,248 (in thousands)

Total Website Units

This is the number of vehicles listed on the company's website on the last day of a reporting period. This metric is a key measure of growth as it indicates the selection available to consumers and the company's ability to scale its vehicle purchasing and reconditioning operations.

Trend:

• 2023: 33,075

• 2024: 53,360

Other Key Business Metrics

The annual report also provides historical data for other financial and operational metrics that are crucial to understanding the business's performance.

Total Revenue

This represents the total net sales and operating revenues generated by the company.

Trend (in millions):

• 2014: $42

• 2015: $130

• 2016: $365

• 2017: $859

• 2018: $1,955

• 2019: $3,940

• 2020: $5,587

• 2021: $12,814

• 2022: $13,604

• 2023: $10,771

• 2024: $13,673

Adjusted EBITDA Margin

Adjusted EBITDA is a non-GAAP measure defined as net income (loss) adjusted for items such as interest expense, taxes, depreciation, amortization, and other specific expenses or gains. The margin is Adjusted EBITDA as a percentage of total revenues.

Trend:

• 2014: (31.0)%

• 2015: (24.6)%

• 2016: (23.0)%

• 2017: (16.1)%

• 2018: (9.2)%

• 2019: (5.2)%

• 2020: (4.0)%

• 2021: 0.5%

• 2022: (7.7)%

• 2023: 3.1%

• 2024: 10.1%

Carvana IRC & Integrated ADESA Sites

This metric reflects the growth of the company's physical infrastructure, including Inspection and Reconditioning Centers (IRCs) and integrated ADESA auction sites.

Trend (Total Sites):

• 2014: 2

• 2015: 3

• 2016: 3

• 2017: 4

• 2018: 5

• 2019: 7

• 2020: 11

• 2021: 14

• 2022: 17

• 2023: 17

• 2024: 24 (18 Carvana IRCs and 6 Integrated ADESA Sites)

Competition and Moat:

Main Competitors

Carvana operates in the highly fragmented and competitive used car marketplace. The company categorizes its current and future competitors as follows:

• Traditional Used Vehicle Dealerships: This includes major players like CarMax.

• Internet and Online Automotive Sites: A broad category that includes Amazon, AutoTrader, Cars.com, Carfax, CarGurus, eBay Motors, Edmunds, Google, KBB, and TrueCar.

• New and Used Vehicle Dealers: This includes dealers that have their own e-commerce business or online platforms.

• Automobile Manufacturers: Companies such as Ford, General Motors, Toyota, Volkswagen, Tesla, Rivian, and Lucid are listed as potential competitors.

• Offline Services: This includes membership-based car-buying services like the Costco Auto Program.

• Other Marketplaces: This includes platforms that could compete with Carvana's own wholesale marketplace program.

• Private Transactions: The company also competes with privately negotiated vehicle sales between individuals.

Carvana's Competitive Moat: Why The Company Believes It Can Win

Carvana states that its primary competitive advantages stem from its "Purpose-Built Vertically Integrated E-commerce Platform". This model is designed to provide a better customer experience while operating at a lower cost than traditional dealerships. The company's competitive moat is built on the following pillars:

• Cost Advantage: Carvana believes its proprietary technology and vertically integrated business model result in a "significantly lower variable cost structure versus traditional dealerships". By controlling all critical operations—from vehicle acquisition to reconditioning, financing, and logistics—the company aims to operate more efficiently. Its scaled infrastructure has the capacity to inspect and recondition more than one million cars per year at full utilization.
  
  

• Network Effects: The business benefits from powerful network effects driven by scale. Carvana offers customers a "nationally pooled inventory of over 53,000 high-quality used vehicles". This large selection increases the probability that a customer will find their desired vehicle, which in turn attracts more customers to the platform.

• Proprietary Technology and Patents: Technology is central to Carvana's operations.

  • The company holds 41 issued U.S. patents covering its vending machine technology, photo technology, website user interface, and artificial intelligence, among others.

  • Its patented, automated photo technology creates a 360-degree virtual tour of each vehicle, which it believes generates the trust needed for customers to buy a car online.

  • A proprietary financing platform, trained on over a decade of Carvana's loan data, provides customers with nearly instantaneous and flexible credit terms without impacting their credit score until a purchase is finalized.

• Brand and Differentiated Experience: Carvana focuses on creating a "transparent and simplified solution to car buying".

  • It offers a distinctive fulfillment experience through home delivery on a branded hauler or pickup at one of its patented, multi-story car vending machines.

  • The company believes these vending machines provide a unique pickup experience that develops brand awareness.

  • A seven-day return policy and a high customer satisfaction rating (averaging 4.7 out of 5.0 stars from over 215,000 surveys) are key to building brand trust and generating referrals.

Forward Strategy:

Carvana's 2024 Annual Report outlines a multi-faceted strategy centered on achieving profitable growth by focusing on innovation, deeper penetration into existing markets, and developing new products, while also keeping acquisitions as a potential tool.

Overall Strategy: Profitable Growth

After a period of focusing on operational efficiency and profitability in 2022 and 2023, management states that in 2024 they shifted "towards the long-term phase of driving profitable growth." This strategy aims to balance scaling the business with enhancing customer experiences and improving efficiency. The foundation of this strategy remains increasing retail vehicle unit sales, which drives multiple revenue streams.

In his letter to shareholders, CEO Ernie Garcia emphasizes that the company will remain "hungry and foolish," indicating a continued commitment to ambitious and aggressive goals. The plan is to build a business at a scale not yet seen in the automotive industry by delivering the best customer experiences.

Key Strategic Pillars:

Innovation and Technology

A core component of Carvana's strategy is to innovate and extend its technological leadership. Management intends to make significant investments in technology to improve and expand its customer offerings, believing the complexity of auto retail provides substantial opportunities for differentiation.

Key areas of innovation include:

• New Products and Services: Carvana plans to leverage its e-commerce and logistics platform to introduce new complementary products and services, such as different insurance products, leasing, or maintenance.

• Process Improvement: The company states it will "relentlessly focus on creatively finding solutions for our customers, building them, and continually improving them." This includes constantly improving how they predict customer demand and optimize vehicle pricing.

• Technology Leadership: Carvana highlights its custom-built technology portfolio, from its inventory management system and patented photo technology to its proprietary financing platform, as a key competitive advantage that it plans to build upon.

Market Expansion and Penetration

Instead of rapidly expanding into new geographic markets, the company's strategy focuses on increasing its market share within its existing footprint.

• Deeper Market Penetration: A primary strategy is to "Increase Sales Through Further Penetration of Our Existing Markets." This will be achieved by actively building brand awareness and improving operational speed and efficiency in the markets they already serve.

• Leveraging Existing Infrastructure: Management notes that it has the physical infrastructure, including Inspection and Reconditioning Centers (IRCs), with the capacity to recondition over one million vehicles per year, which will support continued growth.

Acquisitions

While not listed as a primary focus in its "Business and Growth Strategies" section, the report indicates that acquisitions remain a potential component of Carvana's broader strategy.

• The company acknowledges it "may continue to acquire other companies or technologies."

• The 2022 acquisition of ADESA is cited as a key strategic move that provided significant infrastructure for future growth, with 56 additional locations that could be developed to increase reconditioning capacity.

Capital Allocation:

Carvana's Capital Allocation: A Look at Historical Investments and Financing

Carvana's historical capital allocation strategy has been heavily geared towards fueling rapid growth, expanding its physical infrastructure, and managing its debt. A cornerstone of this strategy was the significant acquisition of ADESA's U.S. physical auction business, alongside substantial debt and equity financing activities.

Major Acquisition: ADESA U.S.

A pivotal moment in Carvana's recent history was the acquisition of ADESA's U.S. physical auction business on May 9, 2022.

• Purchase Price: Carvana acquired ADESA for approximately $2.2 billion in cash.

• Strategic Assets: The deal brought 56 auction locations across the United States under Carvana's control, significantly expanding its physical footprint with 6.5 million square feet of buildings on over 4,000 acres of land. This acquisition was intended to enhance Carvana's customer offering by providing a broader selection of vehicles and enabling faster delivery times.

• Financing the Deal: The acquisition was financed through the proceeds from the issuance and sale of its 2030 Senior Unsecured Notes.

• Goodwill: The transaction resulted in the recognition of $838 million in goodwill, which is tax-deductible.

Debt and Equity Financing

Carvana has actively utilized both debt and equity markets to fund its operations and strategic initiatives. As of December 31, 2024, the company's total outstanding debt was approximately $5.5 billion.

Debt Instruments:

• Senior Secured Notes: The bulk of Carvana's debt is in Senior Secured Notes with varying maturity dates and interest terms.

  • 2028 Senior Secured Notes: $611 million outstanding with a 9% cash interest rate after an initial PIK (Payment-In-Kind) period.

  • 2030 Senior Secured Notes: $1,660 million outstanding with a 9% cash interest rate following a PIK and toggle interest period.

  • 2031 Senior Secured Notes: $1,982 million outstanding with a 9% cash interest rate after a PIK period.

• Senior Unsecured Notes: As of year-end 2024, Carvana had $205 million in outstanding Senior Unsecured Notes with interest rates ranging from 4.875% to 10.250%.

• Debt Exchange: In a significant move to manage its liabilities, Carvana executed an exchange offer in September 2023, swapping $5.5 billion of Senior Unsecured Notes for $4.2 billion of new Senior Secured Notes and $341 million in cash.

• Asset-Based Financing: The company also relies on various asset-based financing facilities:

  • A $1.5 billion floor plan facility to finance its vehicle inventory.

  • $4.2 billion in total commitment across several finance receivable facilities.

  • $482 million in real estate financing through sale and leaseback arrangements as of the end of 2024.

Equity Issuances:

• At-the-Market (ATM) Program: In the fiscal year 2024, Carvana raised $1.3 billion in gross proceeds by issuing 6.8 million shares of its Class A common stock through its ATM program at a weighted-average price of $186.56 per share.

• Private Placement: In August 2023, entities associated with the Garcia family made a $126 million equity investment in the company, the proceeds of which were used to partially fund a tender offer for some of the company's senior notes.

Overall Capital Deployment

Carvana's capital has been primarily deployed to:

• Fuel Growth and Expansion: This includes entering new markets and funding strategic moves like the ADESA acquisition.

• Enhance Operational Efficiency: Facing market shifts in 2022 and 2023, the company focused on profitability by improving its operational cost structure.

• Manage and Restructure Debt: The company has actively managed its debt profile through exchanges and repurchases.

• Support General Corporate Needs: Proceeds from financings are also allocated to working capital and other general corporate purposes

Risks and Changes In Risk:

Carvana's 2024 annual report outlines a wide array of risks to its business, with several notable changes and additions compared to its 2023 filing, reflecting an evolving automotive and economic landscape. The primary risks revolve around macroeconomic conditions, competition, operational execution, and reliance on technology and data.

Summary of Risks to Carvana's Business (2024)

Carvana's business is subject to numerous risks that can be broadly categorized as follows:

Business and Economic Risks:

• Macroeconomic Conditions: The company's performance is sensitive to the broader automotive ecosystem and economic trends, including consumer demand, supply chain issues, inflation, and interest rates that affect vehicle affordability.

• Competition: The used car market is highly competitive and fragmented, with pressure from traditional dealerships, online marketplaces like Amazon, and new and used vehicle dealers with e-commerce platforms.

• Inventory Management: Acquiring desirable used vehicles at favorable prices and expeditiously selling that inventory are crucial. Failure to do so can lead to depreciation and over-supply risks.

• Growth and Profitability: After a period of rapid growth, the company has shifted its focus to profitability and operational efficiency. There is no guarantee of returning to historical growth rates, and failure to manage growth effectively could harm the business.

• Brand and Reputation: Maintaining a positive reputation for integrity and customer service is critical, and failure to do so could impact customer trust and demand.

• Reliance on Key Relationships: The company maintains a business relationship with DriveTime and its affiliates for services such as the administration of vehicle service contracts (VSCs).

Operational and Regulatory Risks:

• Regulatory Environment: Carvana operates in highly regulated industries and is subject to a wide range of federal, state, and local laws. Changes in these regulations or failure to comply could result in penalties and operational disruptions.

• Logistics: The business relies on a combination of internal and external logistics to transport its vehicle inventory, exposing it to risks associated with the transportation industry.

• Legal Proceedings: Carvana is subject to various legal proceedings, claims, and investigations which could have a material adverse effect on the business.

Technology and Cybersecurity Risks:

• Cybersecurity and Data Privacy: The company collects, processes, and stores sensitive personal information, making it vulnerable to security breaches and subject to evolving privacy laws. The use of artificial intelligence (AI) also introduces new risks.

• Technology Reliance: The business is highly dependent on its technology networks and infrastructure, including its website and mobile applications. Disruptions could harm the business.

• Intellectual Property: Protection of proprietary technology, such as algorithms and its vending machine design, is crucial to maintaining a competitive advantage.

Financial and Structural Risks:

• Indebtedness: Carvana has substantial indebtedness, which could affect its financial flexibility and ability to incur additional debt.

• Automotive Finance Receivables: A significant portion of the company's gross profit depends on the sale of automotive finance receivables it originates.

• Organizational Structure: As a holding company, Carvana's principal asset is its interest in Carvana Group, and it depends on distributions from this entity to pay taxes and expenses. Conflicts of interest could arise between stockholders and LLC Unitholders.

Changes in Risk Disclosures from 2023 to 2024

While many of the core risks remained consistent between the 2023 and 2024 annual reports, there were several key changes and additions in the 2024 filing:

• Artificial Intelligence (AI): A notable addition to the 2024 report is the explicit mention of risks associated with the use of AI. The report highlights potential cybersecurity, data privacy, and labor and employment risks stemming from both third-party generative AI platforms and internal software. It also notes the risk of inadvertent disclosure of proprietary information through AI tools.

• Refined Macroeconomic Risks: The 2024 report provides more specific examples of macroeconomic challenges, such as the impacts of inflationary pressures on costs and scarcity of certain products which have affected vehicle prices and consumer demand.

• Emphasis on Profitability Initiatives: Both reports mention a shift in focus to profitability, but the 2024 report elaborates on how this shift impacts various aspects of the business, including a move towards a long-term phase of driving profitable growth.

• Cybersecurity Disclosures: The 2024 report includes a more detailed disclosure on cybersecurity risk management, governance, and strategy, in line with new SEC rules. This includes the role of the Board's Audit Committee and the Chief Information Security Officer (CISO).

• Tax Asset Preservation Plan: The 2023 report detailed a Tax Asset Preservation Plan to protect NOL carryforwards. The 2024 filing notes the termination of this plan in June 2024, as it was determined to no longer be necessary.

• Short Seller and "Short Squeeze" Risks: The 2023 report included a specific risk factor related to manipulative practices by short sellers and the potential for a "short squeeze" to cause extreme price volatility. While still a general market risk, this is not highlighted with its own specific subsection in the main risk factor summary of the 2024 report.

• Updated Financial Figures: All financial figures related to debt, revenue, and profitability have been updated to reflect the full fiscal year 2024 results, showing changes in the company's financial position and performance. For instance, total debt as of December 31, 2024, was $5.5 billion, down from $6.0 billion at the end of 2023.

In conclusion, while the fundamental risks to Carvana's business remain largely the same, the 2024 annual report reflects a company navigating a challenging economic environment with a heightened focus on profitability and grappling with new technological risks like those posed by AI.

Key Partners, Suppliers, and Customers:

Carvana (CVNA) has key relationship risks and dependencies with several partners and suppliers that are critical to its operations, particularly DriveTime Automotive Group, Inc., and its financing partners like Ally Financial.

Key Relationship Risks and Dependencies

DriveTime Automotive Group, Inc. Carvana has a significant and multifaceted business relationship with DriveTime, a related party controlled by Carvana's controlling stockholders, the Garcia Parties. The risk factor section explicitly states that the arrangements with DriveTime cannot be assumed to have been negotiated at arm's length and there is no assurance the arrangements will continue on similar terms, if at all.

Key dependencies on DriveTime include:

• Vehicle Service and Warranty Administration: DriveTime administers the Vehicle Service Contracts (VSCs) that Carvana sells to its customers. Carvana earns a commission on these sales and is entitled to profit-sharing revenues based on performance. DriveTime also administers the limited warranty provided with every vehicle purchase.

• Servicing of Finance Receivables: DriveTime performs ongoing servicing and administrative functions for the automotive finance receivables that Carvana originates.

• Real Estate Leases: Carvana leases critical facilities, including inspection and reconditioning centers in Texas, New Jersey, and Georgia, from DriveTime.

• Wholesale Vehicle Transactions: DriveTime purchases wholesale vehicles from Carvana through competitive auctions.

Financing Partners (Notably Ally Financial) Carvana's business model is heavily dependent on its ability to sell the automotive finance receivables it originates, as this accounts for a substantial portion of its gross profit.

• Ally Bank and Ally Financial Inc. (Ally Parties): Carvana has a Master Purchase and Sale Agreement (MPSA) with the Ally Parties to sell a significant volume of its finance receivables.

  • Under the MPSA, Carvana sold $3.0 billion of finance receivables in 2024 and $3.6 billion in 2023.

  • The agreement was amended to reestablish a commitment by the Ally Parties to purchase up to $4.0 billion of finance receivables between January 2025 and January 2026.

• Securitization and Capital Markets: Beyond specific partners, Carvana relies on its access to capital markets to sell receivables through securitization transactions. A material decline in access to these markets at competitive rates could harm the business.

Vehicle and Parts Suppliers Carvana's ability to operate is dependent on acquiring desirable vehicles and the parts needed to recondition them.

• Vehicle Sourcing: The company acquires its inventory from various sources, including customers, used car auctions, and wholesale used vehicle suppliers. A reduction in the availability or access to these inventory sources could have a material adverse effect on business and sales.

• Inventory Financing: The company relies on agreements with third parties, such as the Floor Plan Facility with the Ally Parties, to finance its vehicle inventory purchases. A failure to extend these agreements could reduce inventory supply.

Insurance Partner (Root, Inc.) Carvana has partnered with Root, Inc. to provide an integrated auto insurance solution on its e-commerce platform. While this provides a complementary revenue stream, Carvana's influence over Root is limited, and its inability to control Root's operations could result in receiving less benefit than expected from the commercial relationship.

Technology and Marketing Partners Carvana has a broad dependency on a variety of external technology and marketing services.

• Traffic and Lead Generation: The company depends on internet search engines, lead generators, social networking sites, and vehicle listing sites to drive traffic to its website.

• Third-Party Technology: Carvana relies on third-party technology for functions such as its website's network infrastructure and customer interaction software. Its technology footprint is significantly hosted in third-party data center facilities it does not control.

Debt and Funding:

Carvana (CVNA) funds its operations through cash generated from sales, financing activities, and has undertaken specific measures to manage its debt and liquidity.

Funding of Operations and Financial Health

How Operations Are Funded Carvana generates cash from several primary sources:

• The sale of retail and wholesale vehicles.

• The sale of auto loans it originates for customers.

• Sales commissions on complementary products like Vehicle Service Contracts (VSCs) and GAP waiver coverage.

• Financing activities, including short-term revolving credit facilities, real estate and equipment financing, and the issuance of debt and equity securities.

In 2024, the company utilized an "at-the-market offering" (ATM) program, issuing 6.8 million shares of Class A common stock for gross proceeds of $1.3 billion, which is being used for general corporate purposes.

Operational Cash Flow The core business's ability to self-fund, as measured by cash flow from operations, has shown significant improvement.

• 2024: Net cash provided by operating activities was $918 million. The increase from the prior year was primarily due to improved operating results and a $274 million reduction in cash interest paid, partially offset by increased inventory acquisitions.

• 2023: Net cash provided by operating activities was $803 million.

• 2022: Net cash used in operating activities was ($1,324) million.

This trend indicates a significant turnaround from a large cash deficit in 2022 to generating substantial positive cash flow from its core operations in 2023 and 2024.

Cash Position and Liquidity Management believes that its current working capital, cash flows from operations, and available financing arrangements will be sufficient to fund operations for at least the next 12 months.

As of December 31, 2024, Carvana's available liquidity resources included:

• Cash and cash equivalents: $1,716 million.

• Availability under short-term revolving facilities: $1,879 million.

• Total committed liquidity resources available: $3,595 million.

The company's total liquidity potential, which also includes additional secured debt capacity and unpledged assets, was $5.7 billion at the end of 2024, a significant increase from $3.1 billion at the end of 2023.

Debt Structure and Ability to Pay

As of December 31, 2024, Carvana was in compliance with all its debt covenants. The company actively managed its debt profile in 2024 by repurchasing and cancelling $370 million and redeeming $100 million of its 2028 Senior Secured Notes.

The total outstanding principal amount of indebtedness was $5.5 billion as of December 31, 2024. A detailed breakdown of the debt is as follows:

Asset-Based Financing ($0.9 billion total)

• Floor Plan Facility: Used to finance vehicle inventory.

  • Amount Outstanding: $67 million.

  • Terms: Interest at a prime rate plus a spread; secured by vehicle inventory and other assets. Matures in April 2025.

• Finance Receivable Facilities: Used to fund auto loans before they are sold.

  • Amount Outstanding: $0 (down from $555 million in 2023).

  • Terms: Various revolving facilities with maturities extending to 2025 and 2026.

• Financing of Beneficial Interests in Securitizations:

  • Amount Outstanding: $354 million.

  • Terms: Secured by retained interests in securitizations with expected repurchases from October 2025 to December 2032.

• Real Estate Financing:

  • Amount Outstanding: $485 million.

  • Terms: From sale and leaseback transactions accounted for as financing, with initial terms of 20 to 25 years.

Senior Notes ($4.6 billion total)

• Senior Secured Notes:

  • Amount Outstanding: $4.4 billion (including $105 million of accrued payment-in-kind interest).

  • Terms: Interest is payable semi-annually. For the first year, interest was paid-in-kind (PIK), adding to the principal balance. For the 2028 and 2030 notes, the company has switched to cash interest payments as of February 2025.

    • 2028 Notes ($0.6B): 12% PIK, now 9% cash interest.

    • 2030 Notes ($1.7B): 13% PIK, now 9%/11% cash/PIK toggle.

    • 2031 Notes ($2.0B): 14% PIK, now --/14% cash/PIK toggle.

• Senior Unsecured Notes:

  • Amount Outstanding: $205 million.

  • Terms: Fixed interest rates ranging from 4.875% to 10.250%.

Debt Maturities The aggregate principal maturities for the company's long-term debt as of December 31, 2024, are as follows:

• 2025: $235 million.

• 2026: $98 million.

• 2027: $94 million.

• 2028: $670 million.

• 2029: $43 million.

• Thereafter: $4,157 million.

Seasonality:

Carvana's business is subject to seasonal fluctuations that affect both vehicle sales and prices.

Sales Volume Seasonality:

• Peak Season: Used vehicle sales typically peak late in the first calendar quarter. This timing coincides with the issuance of federal government tax refunds.

• Low Season: Sales generally diminish throughout the rest of the year, with the fourth calendar quarter expected to have the lowest relative level of sales.

• Company Growth vs. Seasonality: Due to its history of rapid growth, Carvana's overall sales patterns have not always mirrored the general seasonality of the industry. However, as the business matures, the company expects its results to become more reflective of typical market seasonality.

Vehicle Price Seasonality:

• Used vehicle prices also demonstrate seasonal trends, primarily related to the rate of depreciation.

• Vehicles generally depreciate at a faster rate during the fourth and first quarters of the year.

• The rate of depreciation is typically slower in the second and third quarters.

Equity Dilution:

Carvana (CVNA) Class A common share count has increased significantly from the end of 2021 to the end of 2024, primarily driven by equity sales and stock-based compensation. The total number of Class A common stock shares outstanding grew from 89,930,000 at the end of 2021 to 133,271,000 at the end of 2024.

Here is a breakdown of the changes in Carvana's share count over the last three years:

Change in Share Count (2022-2024)

Breakdown of Share Count Changes:

Equity Sales: Carvana has actively raised capital through the sale of its Class A common stock.

• 2024: 6,826,000 shares were issued through its "at-the-market offering" (ATM) program, generating gross proceeds of $1.3 billion.

• 2023: 7,157,000 shares were issued via the ATM program for gross proceeds of $336 million.

• 2022: 15,625,000 shares were issued through equity offerings, raising $1.227 billion net of expenses.

Stock-Based Compensation: A significant portion of the increase in share count is attributable to compensation for employees and directors.

• 2024: A net of 5,255,000 shares were issued, comprising 4,834,000 shares to settle vested restricted stock units (RSUs) and 440,000 shares from the exercise of stock options, offset by 19,000 forfeited or surrendered shares.

• 2023: A net of 1,044,000 shares were issued, including 1,057,000 for vested RSUs and 17,000 from exercised options, less 30,000 forfeited shares.

• 2022: A net of 556,000 shares were issued, which included 390,000 for RSUs, 88,000 for options exercised, and 86,000 under the Employee Stock Purchase Plan (ESPP), offset by 8,000 forfeited shares.

Executive Compensation: The annual report does not provide a specific breakdown of stock-based compensation between executives and other employees. This information is typically found in the company's definitive proxy statement, which is not included in the provided document. However, the report does mention that performance-based RSUs are granted to certain executive employees, indicating they are a component of the overall stock-based compensation figures.

Acquisitions: The provided 2024 annual report does not indicate any significant acquisitions paid for with company stock during the 2022-2024 period. The major acquisition of ADESA's U.S. physical auction business in May 2022 was funded with approximately $2.2 billion in cash, which was raised through the issuance of debt, not equity.

Exchanges of LLC Units: Carvana has a corporate structure where holders of LLC Units in Carvana Group can exchange their units for shares of Class A common stock of Carvana Co..

• In 2024, 8,685,000 LLC Units were exchanged for 6,923,000 shares of Class A common stock.

• In 2023, 34,000 LLC Units were exchanged for 31,000 shares of Class A common stock.

• In 2022, the exchange of LLC units resulted in the issuance of 46,000 shares of Class A common stock

Non-GAAP Adjustments:

Based on Carvana's 2024 Annual Report, the company reports several non-GAAP (Generally Accepted Accounting Principles) financial metrics to supplement its consolidated financial statements . Carvana's management uses these metrics to measure the operating performance of the business as a whole and believes they are useful to investors because they exclude certain financial, capital structure, and non-cash items that may not directly reflect core operations .

Here is a summary of the non-GAAP metrics, their adjustments, and the logic provided for their use:

Adjusted EBITDA and Adjusted EBITDA Margin

• Definition:

  • Adjusted EBITDA is defined as net income (or loss) with several adjustments .

  • Adjusted EBITDA Margin is Adjusted EBITDA shown as a percentage of total revenues .

• Logic for Use: Carvana uses this metric to measure its operating performance . The company believes it is a useful measure for investors because it excludes certain items that are not indicative of recurring operations, allowing for more effective period-over-period evaluation and comparison with competitors .

• Adjustments from Net Income (Loss): To calculate Adjusted EBITDA, the following items are added back to or subtracted from Net Income (Loss) :

  • Additions:

    • Income tax (benefit) provision

    • Interest expense

    • Depreciation and amortization expense (from both cost of sales and SG&A)

    • Share-based compensation expense (from both cost of sales and SG&A)

    • Goodwill impairment

    • Restructuring expense

    • Loss on debt extinguishment

    • Other operating expense, net

  • Subtractions:

    • Gain on debt extinguishment

    • Revenue related to Root Warrants

    • Other (income) expense, net

Gross Profit, non-GAAP and Total Gross Profit per Retail Unit, non-GAAP

• Definition:

  • Gross Profit, non-GAAP is defined as GAAP gross profit with adjustments for certain non-cash items and restructuring costs included in the cost of sales .

  • Total Gross Profit per Retail Unit, non-GAAP is the non-GAAP Gross Profit divided by the number of retail vehicle units sold .

• Logic for Use: This metric is used to measure operating performance relative to retail unit sales . Management believes excluding certain non-cash items provides a clearer view of the core operational profitability per vehicle sold .

• Adjustments from Gross Profit: The calculation for Gross Profit, non-GAAP involves the following adjustments to GAAP Gross Profit :

  • Additions:

    • Depreciation and amortization expense in cost of sales

    • Share-based compensation expense in cost of sales

    • Restructuring expense in cost of sales

  • Subtraction:

    • Revenue related to Root Warrants

SG&A Expenses, non-GAAP and Total SG&A Expenses per Retail Unit, non-GAAP

• Definition:

  • SG&A Expenses, non-GAAP is defined as GAAP Selling, General, and Administrative (SG&A) expenses, excluding depreciation and amortization, share-based compensation, and restructuring costs .

  • Total SG&A Expenses per Retail Unit, non-GAAP is SG&A expenses, non-GAAP divided by the number of retail vehicle units sold .

• Logic for Use: These measures help evaluate the core operating expenses of the business relative to retail sales volume, excluding certain non-cash and non-recurring items .

• Adjustments from SG&A Expenses: The calculation for SG&A Expenses, non-GAAP involves the following adjustments to GAAP SG&A Expenses :

  • Subtractions:

    • Depreciation and amortization expense in SG&A

    • Share-based compensation expense in SG&A

    • Restructuring expense in SG&A

Carvana cautions that these non-GAAP measures may not be comparable to similarly titled metrics provided by other companies due to potential differences in calculation methods .

Changes From Prior Year:

Based on a comparison of Carvana's 2023 and 2024 annual reports, here are the most significant changes:

Strategic and Leadership Tone

A notable shift occurred in the company's strategic focus and the tone of its leadership.

• From Survival to Growth: The 2023 report emphasized resilience and the successful navigation of significant hardship, focusing on achieving profitability through operational efficiency and expense reduction. The 2024 report signals a strategic shift from survival to pursuing "profitable growth".

• Return to "Hungry and Foolish": The 2024 CEO letter embraces the "hungry and foolish" ambition that characterized Carvana's early days, acknowledging that this approach led to challenges in 2022 but is essential to its mission of large-scale disruption. This contrasts with the 2023 letter's focus on the "resilient people" who weathered the company's stock drop.

Financial Performance and Position

Carvana's financial results showed a significant positive turn, moving from stabilization to growth.

• Resumption of Growth: After a sharp decline in 2023, retail units sold increased by 33.1% in 2024, rising from 312,847 to 416,348. Total revenue followed, growing 26.9% to $13.673 billion in 2024 from $10.771 billion in 2023.

• Improved Profitability: The company built upon its 2023 profitability.

  • Net Income: Increased to $404 million in 2024, up from $150 million in 2023.

  • Gross Profit Per Unit (GPU): Total GPU, a key metric, rose to $6,908 in 2024 from $5,511 in 2023. Retail vehicle GPU specifically increased to $3,312 from $2,385.

  • Adjusted EBITDA: More than quadrupled, reaching $1.378 billion in 2024 compared to $339 million in 2023.

• Strengthened Balance Sheet:

  • Debt Reduction: Total debt decreased from approximately $6.0 billion in 2023 to $5.5 billion in 2024, partly due to the repurchase and redemption of $470 million of its Senior Secured Notes. This followed a major debt exchange in 2023 that had already significantly lowered the debt burden.

  • Increased Liquidity: Cash and cash equivalents grew substantially to $1.716 billion at the end of 2024, up from $530 million at the end of 2023.

Operational Scale

The company's operational footprint expanded significantly in 2024 after contracting in 2023.

• Inventory Growth: Total vehicles available on its website grew to over 53,000 by the end of 2024, a sharp increase from approximately 33,000 at the end of 2023.

• Employee Headcount: Total full-time and part-time employees increased to over 17,400 in 2024 from over 13,700 in 2023.

• Infrastructure Integration: Carvana advanced the integration of its ADESA auction sites, converting six of them to provide Inspection and Reconditioning Center (IRC) capabilities, bringing the total number of production sites to 24.

Technology and Risk

The 2024 report reflects the increasing role of Artificial Intelligence in Carvana's operations and the associated risks.

• Intellectual Property: The company reported holding 41 issued U.S. patents as of December 31, 2024, up from 32 in the prior year. The 2024 report specifically notes that patents now cover artificial intelligence technology.

• New Risk Factors: The 2024 report introduces risks associated with rapidly advancing generative AI technology, noting its uncertain impact and potential to introduce new vulnerabilities.

Red Flags:

Here is a list of potential red flags for investors:

Financial Performance and Profitability

• History of Net Losses: The company has a history of significant net losses, reporting a net loss of $2.894 billion in 2022, although this narrowed to a net income of $150 million in 2023 and $404 million in 2024. A history of substantial losses can indicate challenges in achieving sustained profitability.

• Dependence on Vehicle Sales for Revenue: A majority of Carvana's revenue is driven by retail vehicle unit sales. This makes the company highly susceptible to fluctuations in the used car market and consumer demand.

• Sensitivity to Vehicle Prices: The business is sensitive to changes in the prices of new and used vehicles. An increase in used vehicle prices could make them less affordable for customers, while a decrease could reduce the company's gross profit on inventory.

• Fluctuating Gross Profit Per Unit: While total gross profit per unit increased to $6,908 in 2024 from $5,511 in 2023, this key metric can fluctuate based on inventory acquisition costs, reconditioning expenses, and vehicle depreciation. The company's ability to maintain and grow this metric is crucial for long-term profitability.

Debt, Liquidity, and Capital Resources

• Substantial Indebtedness: As of December 31, 2024, Carvana had a significant amount of outstanding debt, including $4.4 billion in Senior Secured Notes and $205 million in Senior Unsecured Notes. This substantial debt could affect financial flexibility and requires a significant portion of cash flow for debt service payments.

• Dependence on External Financing: The company relies on the sale of automotive finance receivables for a substantial portion of its gross profit. Access to capital markets at competitive rates is critical, and any decline could harm the business. Carvana also uses an "at-the-market" (ATM) program to sell stock for capital.

• Restrictive Debt Covenants: The indentures governing the company's Senior Secured Notes contain restrictive covenants that could limit its ability to incur additional debt, pay dividends, or make certain investments.

• Risk of Additional Capital Needs: Carvana states it may require additional capital to fund its objectives, and this capital may not be available on acceptable terms, or at all.

Operational and Market Risks

• Highly Competitive Industry: The used car market is highly fragmented and competitive. Carvana faces competition from traditional dealerships like CarMax, online marketplaces like AutoTrader and Cars.com, and even major companies like Amazon.

• Macroeconomic Headwinds: The business is affected by broader economic conditions, including inflation, recessions, and consumer disposable income. The report notes that changes in the economy, market, and industry in 2022 and 2023 prompted a shift in focus to driving profitability over growth.

• Inventory Acquisition and Management: The company's success depends on its ability to acquire desirable used vehicles at competitive prices. Obstacles to acquiring sufficient inventory could materially harm the business.

• Cybersecurity and Data Privacy Risks: Carvana collects and stores sensitive customer data, including personally identifiable information and financial details. A breach of cybersecurity measures could lead to financial losses, reputational damage, and legal liabilities.

Corporate Governance and Related-Party Transactions

• Controlled Company Status: The Garcia Parties control a majority of the voting power of Carvana's capital stock, making it a "controlled company" under NYSE rules. This exempts Carvana from certain corporate governance requirements, such as having a majority of independent directors. This concentration of control could lead to decisions that may not align with the interests of other stockholders.

• Relationship with DriveTime: Carvana has a significant business relationship with DriveTime, an entity also controlled by the Garcia Parties. This includes agreements for vehicle servicing, leases, and administration of Vehicle Service Contracts (VSCs). These arrangements were not negotiated at arm's length, and their terms could change or be terminated, potentially impacting Carvana's operations and costs.

• Tax Receivable Agreement (TRA): Carvana has a Tax Receivable Agreement with its LLC Unitholders that requires substantial cash payments to them in respect of certain tax benefits. These payments could reduce cash flow available for reinvestment in the business. As of December 31, 2024, the TRA liability was $82 million.

Legal and Regulatory Scrutiny

• Extensive Regulation: Carvana operates in a highly regulated industry, subject to federal, state, and local laws governing vehicle sales, financing, advertising, and data privacy.

• Legal Proceedings and Investigations: The company is involved in various legal proceedings, claims, and government inquiries. This includes lawsuits related to securities laws, consumer protection, and the timely delivery of vehicle titles and registrations. State Attorneys General have also conducted inquiries into the company's practices.

And for posterity:

Cvna Proxy Analysis

547KB ∙ PDF file

Download

Download

Going through a 76 page proxy by hand can be a quite lengthy process (it took me basically a whole workday to manually write a sample output). My goal is to essentially automate the vast majority of this so the imagine workflow could be something like:

1. You read through the ~2 page summary that covers the basics of compensation, insider transactions, any key terms, etc.

2. Any additional information is split into defined sections you can easily look further into.

For example we can section out all outstanding equity awards, realized awards from 2024, the total value of all outstanding equity awards (if for whatever reason CVNA was going to get acquired?), the share structure, etc. I want to empower users to be able to ping around to whatever section they need a double click on, while having powerful high level overviews as well.

The plan is to make everything I do as modular as possible which has a plethora of benefits

1. If you want info/data it’s much easier to find

2. I can allow you to subscribe to or follow only specific modules you’re interested in. Don’t care about share structure but want an email update on equity awards and targets? I got you (soon tm)

3. Hallucinations near 0? As far as I know the document has 0 incorrect values despite being ~30 pages of AI generated content. Trying to 1 shot the report or interacting via extended chat windows has a MUCH higher risk in my experience.

   1. If you find a hallucination, let me know!

4. Comparisons across competitors/similar companies can be done on a relatively like for like basis

5. We can turn stale documents into living documents. We could say update insider ownership with SEC filings, regenerate any sections of the report that are impacted, etc.

It’d be really cool if I can basically get it to the point that anything you want to know about a proxy, or about recent filings, or about 10-K’s, is being automatically generated. Then just keep moving the needle towards investor presentations, expert calls, alternative data, etc.

If I can compress becoming an expert on a company from a month+ to a couple days of reading, that opens up a ton of possibilities to find even more neat angles of optimization, or just time to hang out at the beach.

Anyways, I’ll likely end up adding some more stuff that’s visual and shows historical comps for a variety of metrics over the coming weeks as I hammer out version 0’s of everything. I’m likely leaning towards the end UI for these being via the Clarity website or email notifs since Substack doesn’t play nice with dynamic charts. The Substack would then maybe act as a “Best Hits” in terms of where I’m seeing data moves and such.

Brief CVNA/KMX/CDLX Note:

Currently travelling, will do my best to automate some Twitter delivery of free CVNA/KMX/CDLX data in the next week or so, keep an eye out.

None of it will be written by me, so doing a small public stress test of the consistency. Cheers.

Subscribe now

In the future I’m thinking of providing more graphs/visuals as well as expanding on market expectations in the catalysts section. It’d likely be helpful for example to have job postings tracking beyond just WoW, YTD 2yr yield and comps vs priors, etc. The article mentions S&P Global prelim ratings on their ABS transaction, so some analysis would likely be nice there as another example. If there’s anything that would be particularly helpful, feel free to let me know.

Unfortunately this coverage of most recent week sales will not be something I’m freely updating consistently as it includes alternative data from our Clarity coverage. Not quite sure yet how exactly we end up structuring everything, we may end up having 1 month trailing reports be free similar to our data coverage, there may be some small fee, tbd.

The goal here is to provide a ton of high quality content covering anything from weekly and daily metrics to higher level earnings reviews, proxy analysis, 10K reviews, etc. In the coming weeks you will very likely be seeing things like that for CVNA posted here. We’re tentatively going to do similar things for Google and Tesla next. If you have any feedback on any of these posts, please do reach out.

I want to put a professional analyst in your pocket for any ticker you can imagine, but baby steps.

Subscribe now

Carvana (CVNA) Data Weekly Analysis: 5/26/2025 - 6/1/2025

• Trailing Week Sales: 10,720 (-287, -3% WoW) (+3,068, +40% YoY)

• Trailing Week Orders: 13,971 (-986, -7% WoW)

• Trailing Week Listings: 10,365

• Q2 Unit Run Rate: 142,827 (+41% YoY)

• Inventory as of 6/1/2025: 50,140 (-1,898 WoW)

• Average Inventory Price 6/1/2025: $26,580 (+$200 WoW)

• Trailing Week ASP: $25,811 (+$127 WoW)

• Average Delivery Estimate: 4.40 days (-0.15 WoW)

Used Car Data Weekly Analysis: 5/26-6/1/2025

• Carvana Average Inventory: 51,434

• CarMax Average Inventory: 73,018

• Unweighted Mix Adjusted Price Difference (CVNA vs KMX): $-783.50

• CVNA Mix Adjusted Weighted Price Difference: $-1,492.53

• KMX Mix Adjusted Weighted Price Difference: $-999.98

• Carvana Top Sold Vehicle: Chevrolet Malibu (4.2%)

• CarMax Top Sold Vehicle: Honda Civic (2.7%)

• Carvana Top Inventory Vehicle: Chevrolet Malibu (7.6%)

• CarMax Top Inventory Vehicle: Nissan Rogue (2.3%)

• Carvana Inventory Average Vehicle Year: 2020.44

• CarMax Inventory Average Vehicle Year: 2020.21

Carvana (CVNA) Other KPI’s: 5/26/2025-6/1/2025

• 6/1/2025 2yr Yield: 3.898% (-9.5 bps WoW)

• 6/1/2025 CVNA Job Postings: 1201 (-48 WoW)

• 6/1/2025 Average Recommendation: Moderate Buy

• 6/1/2025 Average Target Price: $275.65

Daily News: 5/26/2025-6/1/2025

Summary: The week for Carvana was marked by substantial insider stock sales from top executives and a major shareholder, all conducted under pre-arranged Rule 10b5-1 trading plans. Despite the heavy selling volume, Carvana's stock price continued its upward trend, closing the week higher. Operationally, the company announced the expansion of its same-day delivery service to the Denver area. The financial landscape saw a favorable dip in the 2-year Treasury yield. A notable market development was the launch of a new 2X leveraged ETF tracking Carvana (CVNX), signaling significant trader interest. Broader industry news included concerns about OEM production due to potential mineral shortages and ongoing discussions about the impact of tariffs on vehicle prices.

Monday, May 26, 2025 (Memorial Day)

• SEC Filings: None reported due to the U.S. market holiday.

• Stock Price (CVNA): U.S. markets were closed.

• 2y Treasury News: U.S. Treasury markets were closed.

• Other Relevant News: Volvo Cars announced job cuts, citing high costs, a slowdown in EV demand, and uncertainty around trade tariffs.

Tuesday, May 27, 2025

• SEC Filings: No new SEC filings directly pertaining to Carvana were prominently reported.

• Stock Price (CVNA):

  • Open: $316.90

  • High: $323.13

  • Low: $313.00

  • Close: $320.31

  • Volume: 3,510,000

• 2y Treasury News: The 2-year Treasury yield was reported at 3.96%.

• Sell Side News: A Nasdaq article analyzed Carvana's stock surge, pointing to strong Q1 results, cost-cutting measures, and potential benefits from tariffs on new vehicles.

• Company News/Press Releases: Carvana announced that CFO Mark Jenkins would present at the William Blair 45th Annual Growth Stock Conference on June 4, 2025.

Wednesday, May 28, 2025

• SEC Filings:

  • A Form 4 was filed for Chief Product Officer Daniel J. Gill, reporting the sale of 32,500 shares for approximately $10.4 million.

  • A Form 144 was filed by Officer Paul Breaux, indicating an intent to sell 20,000 shares.

  • A Form 144 was filed by the Daniel & Lian Gill Family Trust, signaling an intent to sell 32,500 shares.

• Stock Price (CVNA):

  • Open: $323.57

  • High: $323.57

  • Low: $315.66

  • Close: $319.65

  • Volume: 2,100,000

• 2y Treasury News: The 2-year yield rose, with reports showing it at 3.96% to 4.009%.

• Company News/Press Releases: Carvana announced the expansion of its same-day vehicle delivery service to the Denver area.

Thursday, May 29, 2025

• SEC Filings: A Form 4 was filed by CEO Ernest C. Garcia III detailing the sale of 10,000 shares at an average price of $318.55, for a total of approximately $3.185 million, as part of a Rule 10b5-1 trading plan.

• Stock Price (CVNA):

  • Open: $318.08

  • High: $327.75

  • Low: $316.19

  • Close: $327.16

  • Volume: 4,460,000

• 2y Treasury News: The yield fell to 3.92%, its largest single-day drop since May 15.

• Other Relevant News: Defiance ETFs launched CVNX, the Defiance Daily Target 2X Long CVNA ETF, designed to provide 2X leveraged daily exposure to Carvana's stock.

Friday, May 30, 2025

• SEC Filings: A heavy day for insider transaction disclosures, all under Rule 10b5-1 plans:

  • CFO Mark W. Jenkins sold 51,380 shares at an average price of $326.03, totaling approximately $16.75 million.

  • CEO Ernest C. Garcia III reported further sales, including 3,086 shares valued at approximately $1 million.

  • 10% shareholder Ernest C. Garcia II sold 50,000 shares at a weighted average price of $325.1554, for a total of approximately $16.26 million.

  • A Form 144 was filed by Ernest C. Garcia II, indicating a proposal to sell 500,000 shares valued at approximately $159.8 million at the time of filing.

• Stock Price (CVNA):

  • Open: $326.24

  • High: $331.97

  • Low: $316.41

  • Close: $331.30

  • Volume: 3,140,000

• 2y Treasury News: The yield closed the week at 3.89%.

• Other Relevant News: The University of Michigan's final May Consumer Sentiment Index was unchanged from April at 52.2.

Saturday, May 31, 2025 & Sunday, June 1, 2025

• Other Relevant News: Weekend financial news coverage analyzed the significant insider selling activity at Carvana that was disclosed during the week.

Upcoming Potential Catalysts

• Earnings:

  • Carvana's Q2 2025 earnings are expected around July 30, 2025.

  • The consensus EPS estimate for Q2 is $1.11.

  • Competitor earnings reports: CarMax (KMX) around June 20, Lithia Motors (LAD) around July 16, and AutoNation (AN) around July 23-30.

• Conferences:

  • CFO Mark Jenkins is scheduled to present at the William Blair 45th Annual Growth Stock Conference on June 4, 2025.

• Macroeconomic Data:

  • Employment Situation (Jobs Report): May data on June 6, 2025.

  • Consumer Price Index (CPI): May data on June 11, 2025.

  • Producer Price Index (PPI): May data on June 12, 2025.

  • U.S. Retail Sales: May data on June 17, 2025.

  • FOMC Meeting: June 17-18, 2025.

• Industry Releases:

  • Manheim Used Vehicle Value Index: May data is expected around June 6, 2025.

• Other Potential Catalysts:

  • Ongoing developments regarding U.S. automotive tariffs.

  • Q2 2025 13-F filings disclosing institutional holdings are due by mid-August 2025.

News Mentioning CVNA 5/26/2025-6/1/2025

• Carvana Press Release (5/28/2025): The company officially announced the extension of its same-day vehicle delivery service to customers in the greater Denver metropolitan area. The release highlighted how the operational enhancement leverages Carvana's regional logistics and reconditioning capabilities to improve convenience for local buyers and sellers.
  
  

• GlobeNewswire (5/29/2025): A press release announced the launch of a new exchange-traded fund, CVNX, by Defiance ETFs. The fund is designed to provide twice the daily leveraged exposure to Carvana's stock, signaling significant market interest and potentially leading to increased trading activity.
  
  

• S&P Global (6/4/2025): In an announcement just after the reporting week, S&P Global Ratings assigned preliminary ratings to Carvana Auto Receivables Trust 2025-P2. This development points to Carvana's continued access to the asset-backed securities (ABS) market, a vital funding source for its auto loan business.

Hedge Fund CVNA Transactions 5/26/2025-6/1/2025

• No new Form 13F or 13G filings from major hedge funds were reported during the week.

• News reports on May 26 highlighted that smaller institutions, such as Brucke Financial Inc., had initiated or increased positions in Carvana during Q4 2024.

• The most significant institutional product development was the launch of the Defiance Daily Target 2X Long CVNA ETF (CVNX), reflecting market demand for trading instruments tied to the stock.

Additional News

• OEM Updates & EV Market Shifts: Reports indicated that Suzuki and Ford faced production halts in late May due to shortages of components linked to China's restrictions on rare earth mineral exports. This could reduce new car availability, potentially increasing demand for used vehicles. Competitor CarMax announced new initiatives to enhance the used EV buying experience.

• Tariff Landscape: A Yale Budget Lab report suggested that existing and potential tariffs could increase average new car prices by about 5%. Higher new car prices can make used cars a more attractive alternative, potentially boosting Carvana's sales but also increasing its vehicle acquisition costs.

• Used Car Market Trends: Cox Automotive reported that while retail sales remained strong year-over-year, momentum softened in May, with used vehicle prices trending downward. Conversely, an Edmunds Q1 2025 report noted that structural supply constraints for lightly used vehicles were supporting higher prices for late-model cars.

If the title sounds like hyperbole, I assure you it is if anything, understated.

Yesterday I pushed a blog post covering Carvana from 5/12/2025 - 5/18/2025. You can find a link here.

Not a single piece of text in that report was written by me! Absolutely 0% of the content was human generated, 100% Gemini.

That may be obvious for parts of it that reference shitty Zacks articles, but that report is the worst the AI will be, which is already better at extracting good info than I was when I started my job! My goal is to automate such coverage, continue to improve it with more data sources, and expand coverage to additional stocks.

My end vision here is basically having automated weekly reports covering basically every “check box” type of weekly information you’d want for any stock you can think of. SEC filings, news articles, job postings, upcoming catalysts and market consensus on those, changes in sell-side estimates, the CEO liking a partners LinkedIn post ahead of an announcement, etc. You want alt data coverage? Let’s make it cover alt data.

Cherry on top, I want to let you chat with it too! Essentially a fully automated junior analyst (minus the modeling capabilities, but Excel plug-ins can happen down the line). That’s part of the goal with which is where I’ll likely be posting articles in the future related to this type of coverage. It will also naturally be a work in progress as I expand here. I need to built out the data collection side, scale up the automated coverage side, wrangle the robot overlord to my will, etc.

It’s very likely some of this ends up being paywalled. If our goal is to basically create a junior analyst that can cover hundreds of stocks, I hope a tiny fraction of that value is a worthy payment. The goal will be to make it feel like a total steal.

The whole journey is hopefully quite informative to you, the reader. It most definitely has been very informative for me, the person messing with models 8-12 hours a day. As it so happens, if you run in to Gemini and ask it to give you a report on how many cars CVNA sold last week, it will produce total junk, and the formatting won’t be consistent, and numbers may be fake.

So below we’ll explore some of my model learnings, and feel free to subscribe here or on the Clarity blog for future updates. Cheers.

Subscribe now

The Basics Of Models:

The key to getting good model output is understanding how models work

Models can be broken down into a few key parts, roughly as so

1. Model capability

2. Data sources

3. Prompt

Below we are going to take a look at how each of these work, and how you can interact with them to produce viable output.

Model Capability:

This is the variable you cannot control (besides picking different models in ChatGPT/Gemini). Roughly speaking, model capability has improved to an absurd degree. Probably the most visceral example is of Will Smith eating spaghetti, comparing the viral early examples from 2023 with more current capabilities, that as of now are not even state of the art anymore!

Reddit users have started making full music videos of surprising quality. Some short clips are starting to look basically indistinguishable from reality. It’s phenomenal really, I heavily encourage keeping an eye on it.

Benchmarking text output of course is less intuitive, but I think we’re safely at the point where the logic skills of AI are better than almost all fresh college graduates. Google has made models with math skills on par with the top 0.01% already, so like, good luck to us humans I guess?

Of course logic isn’t everything. Some fields like programming/math tend to have objective truths, lots of publicly available training data, not a lot of missing context. There isn’t some big “Encyclopedia of High Level Math” locked in the vault of American Math Inc. or something. Whereas say with sales, companies like Salesforce basically operate as big repositories of somewhat proprietary lead sources. Everyone isn’t nicely uploading their work phone, email, SaaS requirements, etc on to a clean searchable profile. It’s a bit messy.

As a result it doesn’t really matter how good the model is. If you were born to sell B2B SaaS ask it to find 100 sales leads it will likely hallucinate quite heavily. If you ask it model CVNA’s future earnings, it will either do a linear regression on past earnings, quote a Zack’s article, or just make something up. The amount of soft inputs that go into something like that is extremely complex, so logic doesn’t quite matter. It needs data sources, it needs context on the data sources, it needs to be directed.

So let’s explore those.

Data Sources:

As you will notice, default Gemini will as of now, never give you good info on how many cars Carvana sold last week. Models aren’t great at making 10 leaps of logic, spinning up some code, spinning up a db, running code, doing math, and spitting you out an answer in 30 seconds. Some stuff still requires humans to go gather because it’s not direct, it’s messy, it has nuance.

To avoid the model hallucinating here, we can simply input the data ourselves! That’s what we’re doing now with Clarity, we are taking our websites data sources and feeding them into a model to produce commentary. Like I said, models are extremely good at logic. It’s quite able to take well labeled info from a database or spreadsheet and spit out a version in human readable text. It took a bit of wrangling variable names and such for consistency, but I was able to get the model to pull the data quite consistently.

Going forward, my goal is to hook up more and more data sources. Some of these may be job postings for example, or prediction market odds for macro event catalysts. In the future once we get enough money, I’d like to hook up credit card data, FactSet/Bloomberg, options chains, factor data, LinkedIn posts, social media statistics, expert networks, etc.

Say you were interested in how Wayfair was handling tariffs. It’d be pretty cool to see something like:

Week of 6/1/2025:

• Credit Card Data Sales: +3% YoY vs +1% YoY for the week of 5/25/2025

• Credit Card Data Sales QTD: -1% YoY

• Twitter Followers: 111,111 (+113, 0.1% WoW) (+5,843, 0.5% YoY)

• Instagram Followers: 2,193,129 (+10,000, 0.5% WoW) (+193,129, 10% YoY)

• Analyst Consensus Q2 2025 EPS: $1.15 (+$0.05 WoW)

  • New “Buy” note from JPM talking about blah blah blah

• 1 LinkedIn post by CEO Niraj Shaj discussing tariff impacts on home decor

  • Link

• 5 Expert Network Calls

  • Former Wayfair Logistics Manager

    • Discusses potential tariff impacts on Gross Margin: Link

  • etc

• SEC Filings:

  • Form 4: Insider Buy by CFO Kate Gulliver for 10,000 shares at $25.00 for a total of $250,000

    • Total ownership: $1,000,000

  • Form 4: Insider Buy by CEO Niraj Shah for 10,000 shares at $25.00 for a total of $250,000

    • Total ownership: $1,000,000

  • 8K: Wayfair Announces New Factory Opening in India

    • Link

You get the point.

In theory something that like is totally possible, dare I say easily doable, when hooked up to the right data sources. It’s also effectively a large part of what an analyst may work on during a week, which can leave more time available for the work with a bit more flair. Another added benefit is that maybe I don’t have a great pipeline for Wayfair, I’m new to it. Instead of building out everything from scratch, I can have a repository of analysis and news and build from there.

Overall, very very bullish about the potential here for AI. That Wayfair example? With pre-collected data the actual report could be generated for fractions of a cent. Of course bundling Tegus/Bloomberg/Yipit is hundreds of thousands of dollars per year, or with raw data sources potentially millions, but the added cost for the output is basically a rounding error compared to a human. It’s absurd!

A primary concern of course here is hallucinations, you may still be doubting me on the efficacy. After all, you may see a viral tweet like the below and properly note the output is totally non-sensical, so why am I so excited?

Essentially, you should absolutely never do what is happening in that prompt. Models are still very bad at taking extremely complex asks, that don’t have pre-loaded data sources, and spitting out something good. There’s a gigantic difference between ripping a “what was mag7 capex last year” and a proper prompt.

Here’s Gemini providing a ton of totally wrong answers to that prompt: https://g.co/gemini/share/06eaa869784b

The reason here is data and the prompt. The way these models work is they basically follow a chain of reasoning using probabilities to turn your prompt into an output. The longer the chain of reasoning, or the wider the probabilities, will increasingly produce bad results. If the model has a 50% chance to get lost, and you ask it a 10 step problem, you’ll get a correct answer maybe once every thousand prompts! When interacting with a model, you want to reduce the probability of getting lost as much as possible, as well as reducing the opportunities to get lost.

The META CapEx above was wrong, so let’s try it again, but this time simply copy paste their Cash Flow Statement from Yahoo Finance

https://g.co/gemini/share/aac5ea638d8f

We get it right! I didn’t need a fancy solution or anything, just a raw copy pasted cash flow statement. Like I said, the model is very very good at logic, but it has no idea if “capex” means it should go find the actual SEC filing or just Bob’s Trade Blog as a source.

If we really wanted to chart CapEx over the last 5 years and next year, it would be trivial to do so by loading the prompts with the relevant filings, loading them will sell-side estimates, and basically just leaving the model to parse and graph. It’s really really good at well defined math!

The downside here being that it’s not sexy to go grab a bunch of filings and plug them in vs just having the model one shot it. Having to go find sources for your questions? Might as well do it yourself.

I agree, LLM’s are not quite good enough at contextual understanding to do random bespoke work. Thankfully, I’d say the majority of finance work is not bespoke! It’s parsing well defined news flows and data, and using that to do bespoke work! This makes your data sources and integration imperative to finding value from LLM’s. My goal is to basically read the markets mind for every stock, and aggregate data coverage for everything relevant. Hopefully as I do that, I can then use all of that work as a grounding “database” for bespoke stuff. Work in progress.

On to prompting.

Prompt:

There’s a lot that can be said about prompts, but we’ll focus on stuff primarily useful to the financial research use case.

The first thing to know is that prompts depend on the model. Deep research for example is very very good at finding info from a ton of different links. It uses an absurd amount of compute to aggregate an absurd amount of information. If you try to prompt “what color is the sky” it will look through dozens of links and casually write you an 18 page report. Naturally there are a ton of steps here, and as we discussed each step has variance, so getting a consistent output is very hard. If I run the same prompt 10 times, it may cite Nasa all 10 times, but what it cites, how it says it, how its formatted, all of that will change.

Thus if you walk into deep research and ask it to write a research note in a certain format, it will fail basically 100% of the time (believe me, I’ve tried).

What’s very helpful is to break up anything you’d like an LLM to do into parts. You can see this in the blog. There are numerous “sections” and each was generated differently.

For example, if I want analysis on how many vehicles Carvana sold last week, and I have a database with that information, I really don’t need Gemini to do a google search. I don’t need it to find 500 sources. I need it to run some basic logic on a provided dataset and spit out text. Thus I simply provided it with a format from prior work, and asked it to generate for that week. All in all, it cost about $0.0001.

Now if I want it to find and summarize general news? That requires a bit more discretion. In that case, having Gemini run through 500 sources and try to establish relevance to CVNA can make sense. The prompt likely needs to contain some idea of what you are looking for, some whitelisting or blacklisting of sources (the blog for example contains a bunch of Zack’s slop, oops), a time frame, etc. It’s also very helpful to load sample reports into the prompt as context for what kind information is useful. I hand wrote a report for a different week, which I fed into the model as an idea in terms of highlighting SEC filings, interest rates, press releases, general automotive industry news, etc.

The output of that report of course will be almost totally random in terms of format and to some extent content. What’s cool is that because this is vastly cheaper than a human going through hundreds of sources, we can do the neat trick of running it a few times! Given the non-deterministic nature of the models, this can also dig up new information that may have been missed on a prior run. For example, one run nicely flagged that CVNA’s ABS trusts release reports on the 15th of May. Combing it with another report led to the final blog having that information, whereas a singular run missed!

On the format side, again the models are great at logic. What we can do is simply provide a sample report to the model, and ask it to take the output files of Deep Research and condense into the format of the sample report! Likely will need some fine tuning here in terms of what makes the cut as again the model doesn’t know if Zack’s is better than the WSJ to include in news, but for a v0 it’s quite good. It did not take much effort to get the formatting side nailed down, and going forward each completed article can be fed in as context to create new articles, reinforcing the “quality”

So to summarize:

1. Understand the model you’re working with and it’s capabilities/tendencies

2. Feed the model examples

3. Attempt to tailor context and provide rules/weighting on info sources

4. Magic

Fin:

As always, hopefully this article was helpful. I’d like to expand coverage as much as possible while also ensuring the quality is something I’d actually be interested in. Not trying to have the model draw lines on a chart and pretend it’s the work of Einstein.

That said, irrespective of how well I can figure this out, I do think very large portions of the analyst job in terms of time spent will become redundant much faster than imagined. I’d rather curate a task to Gemini at this point than higher the majority of college grads. Additionally, these models are about 3 years old. It takes us humans about 22 years to be considered vaguely employable, probably another 3-5 to be value generating, and we aren’t always awake and available. We have families, hobbies, maybe we leave jobs. Give it an extra ~50 years on average and we stop existing altogether. Tough competition!

The utilization of mass scale intelligent compute is in it’s infancy and it’s already a multi-trillion dollar endeavor. We’re going to start seeing mass market AI music videos, AI cartoons, AI video games, AI tutors, AI teachers, AI analysts, AI AI consultants, AI influencers, AI porn, etc.

The vast majority of tasks on this earth are able to be defined. Software was already eating the world, AI is an even further step change.

So anyways, I bought some Google a few weeks ago. In a decade hopefully Gemini looks upon me favorably for that.

Cheers.

Subscribe now

I'm trying out a new weekly format to bring you the latest on Carvana. Let me know if you enjoy this new approach by dropping a comment below or replying to this email. Your feedback is much appreciated!

Subscribe now

Carvana (CVNA) Data Weekly Analysis: 5/12-5/18/2025

• Trailing Week Sales: 10,700 (-4, -0.04% WoW) (+2,884, +36.89% YoY)

• Trailing Week Orders: 15,320 (+1,902, +14.17% WoW)

• Trailing Week Listings: 13,991

• Q2 Unit Run Rate: 143,396 (+41% YoY)

• Inventory as of 5/18/2025: 52,252 (+1,082 WoW)

• Average Inventory Price 5/18/2025: $26,313 (+$236 WoW)

• Trailing Week ASP: $25,646 (+$142 WoW)

Carvana demonstrated a mixed but generally positive operational picture in the week ending May 18, 2025. While trailing week sales saw a marginal dip of 0.04% week-over-week to 10,700 units, they remained significantly elevated year-over-year, up 36.89%.

A standout metric was the surge in trailing week orders, which jumped 14.17% from the previous week to 15,320. Q2 unit run rate of 143,396, indicates a 41% increase compared to the same quarter last year thus far.

Inventory levels also grew, with 52,252 units available as of May 18th, an increase of 1,082 units week-over-week. Alongside this inventory growth, average prices continued to climb. The average inventory price rose by $236 to $26,313, while the average selling price (ASP) for the trailing week increased by $142 to $25,646. The number of new listings for the week stood at 13,991.

Overall, Carvana saw slightly lower year-over-year unit growth, but improving demand from orders and inventory build, implying more consumer confidence.

Used Car Data Weekly Analysis: 5/12-5/18/2025

• Carvana Average Inventory: 52,807.5

• CarMax Average Inventory: 76,902.17

• Unweighted Mix Adjusted Price Difference (CVNA vs KMX): -$615.60

• CVNA Mix Adjusted Weighted Price Difference: -$1167.17

• KMX Mix Adjusted Weighted Price Difference: -$810.88

• Carvana Top Sold Vehicles:

  • Make: Chevrolet (11.19%)

  • Model: Tesla Model 3 (2.48%)

• CarMax Top Sold Vehicles:

  • Make: Toyota (12.00%)

  • Model: Honda Civic (2.48%)

• Carvana Top Inventory Vehicles:

  • Make: Chevrolet (11.04%)

  • Model: Ram 1500 Crew Cab (1.73%)

• CarMax Top Inventory Vehicle:

  • Make: Toyota (11.23%)

  • Model: Nissan Rogue (2.22%)

• Carvana Inventory Average Vehicle Year: 2020.24

• CarMax Inventory Average Vehicle Year: 2020.15

The average vehicle year for both retailers was remarkably similar, with Carvana at 2020.24 and CarMax at 2020.15, indicating both are offering relatively recent model years to consumers.

In terms of pricing, Carvana's vehicles were listed lower on average. The unweighted mix-adjusted price difference showed Carvana's listings priced $615.60 below CarMax's. When applying a weighted mix adjustment, this gap widened, with Carvana's prices $1167.17 lower, while CarMax's mix-adjusted weighted price was $810.88 higher than at Carvana. This suggests Carvana is positioning itself as a more aggressively priced option while also maintaining better margins.

Consumer preferences, as reflected in top-selling vehicles, showed some overlap and some distinctions. Chevrolet was the top-selling make for Carvana (11.19%), while Toyota led at CarMax (12.00%). Interestingly, the Tesla Model 3 was Carvana's top-selling model (2.48%), mirroring the Honda Civic's popularity at CarMax (2.48%).

Looking at inventory composition, Chevrolet also topped Carvana's stock (11.04%), with the Ram 1500 Crew Cab being its most stocked model (1.73%). Toyota was the most prevalent make in CarMax's inventory (11.23%), and the Nissan Rogue was its most common model (2.22%).

Carvana (CVNA) Other KPI’s: 5/12-5/18/2025

• 5/16/2025 2yr Yield: 3.98% (+10bps WoW from May 9, 2025)

• 5/18/2025 CVNA Job Postings: 1196 (+42 WoW from May 9, 2025)

• 5/18/2025 Sell-Side Average Recommendation (Zacks data as of May 17-23): Average Brokerage Recommendation (ABR) of 1.87 (between Strong Buy and Buy).

• 5/18/2025 Sell-Side Average Target Price (Zacks data as of May 17-23): Approximately $305.31.

Daily News:

Summary:

The week of May 12-18, 2025, for Carvana was characterized by significant insider stock sales (including by the CEO and other executives) and Q1-end holding reductions by institutional investors like Millennium Management and CAS Investment Partners. This contrasted with continued analyst optimism, notably a price target increase from Morgan Stanley. Operationally, Carvana extended its PPA Tour sponsorship, unveiled a special NASCAR paint scheme for Jimmie Johnson, and launched same-day delivery in Denver. The stock price exhibited volatility but closed the week slightly higher. The broader environment saw a temporary US-China trade tariff truce (though key auto tariffs remained) and a Moody's downgrade of the U.S. sovereign credit rating.

Monday, May 12, 2025

• SEC Filings:

• Form 4 by Michael E Maroone (Director): Disclosed sale of 9,000 shares on May 9 at various prices ($276.89 - $279.06), totaling approximately $2.46 million.

• Stock Price (CVNA):

• Open: $280.56

• High: $299.00

• Low: $278.49

• Close: $293.06

• Volume: 5,890,000

• 2y Treasury News:

• The 2-year yield rose 0.047 percentage points.

• Sell Side News:

• None specific to Carvana found for this day.

• Other Relevant News:

• US-China 90-day trade tariff truce announced (effective May 14), but the 25% US tariff on imported vehicles and auto parts remained.

• Toyota teased a new model reveal for May 20.

Tuesday, May 13, 2025

• SEC Filings:

• Form 4 by Paul W. Breaux (Officer): Sale of 30,000 shares on May 13 at $298.00, totaling $8.94 million (10b5-1 plan).

• Form 4 by Ira J. Platt (Director): Disclosed distribution of 24,959 shares from a trust on May 13.

• Stock Price (CVNA):

• Open: $297.28

• High: $311.88

• Low: $296.68

• Close: $305.52

• Volume: 6,480,000

• 2y Treasury News:

• Yields fluctuated; general news highlighted US Treasury yields climbing mid-week due to Fed minutes concerns.

• Sell Side News:

• Zacks published "Brokers Suggest Investing in Carvana (CVNA)".

• Company News/Press Releases:

• Carvana announced a multi-year extension of its title sponsorship with the Professional Pickleball Association (PPA) Tour.

• Other Relevant News:

• Nasdaq.com article recapped CVNA insider trades, hedge fund activity, and analyst ratings.

• MarketBeat reported Ensign Peak Advisors increased its CVNA stake in Q4 2024 (news published May 13).

Wednesday, May 14, 2025

• SEC Filings:

• Form 4 by Ernest C. Garcia III (CEO): Sale of 10,000 shares on May 14 (various prices ~$302-$304.50), totaling approx. $3.04 million (10b5-1 plan; filed May 16).

• Form 4 by J Danforth Quayle (Director): Sale of 11,614 shares on May 14 at avg $304.24, totaling $3.53 million; also exercised options (filed May 15).

• Stock Price (CVNA):

• Open: $302.64

• High: $304.01

• Low: $293.73

• Close: $297.40

• Volume: 3,290,000

• 2y Treasury News:

• The 2-year yield was reported at its highest since this date when it later rose to 4.016% on May 21.

• Sell Side News:

• Moomoo.com reported on various analyst ratings and price target adjustments around the Q1 earnings period.

• Company News/Press Releases:

• Carvana unveiled Jimmie Johnson's 700th start tribute paint scheme for the Coca-Cola 600.

• Carvana launched same-day vehicle delivery in the greater Denver area.

• Other Relevant News:

• Zacks.com published multiple articles on CVNA's stock momentum and 52-week high.

• US-China tariff truce effective.

Thursday, May 15, 2025

SEC Filings Disclosed:

• Form 4 by Ernest C. Garcia III (CEO): Sale of 10,000 shares on May 15 (various prices ~$302-$304), totaling approx. $3.03 million (10b5-1 plan; filed May 16).

• Form 13F-HR by Millennium Management LLC: Disclosed ownership of 301,573 CVNA shares as of March 31, 2025, a 27% decrease from Q4 2024.

• Schedule 13G/A by CAS Investment Partners, LLC: Disclosed ownership of 5,685,029 CVNA shares (4.2%) as of March 31, 2025, an 11.91% decrease.

• Form ABS-EE (Exhibits Submission for Asset-Backed Securities Offerings) filed by Carvana Auto Receivables Trust 2024-P4.⁵⁴

• Form 10-D (Asset-Backed Issuer Distribution Report) filed by Carvana Auto Receivables Trust 2024-P3.

Stock Price (CVNA):

• Open: $299.75

• High: $304.14

• Low: $294.22

• Close: $299.89

• Volume: 3,680,000

2y Treasury News:

• The 2-year yield reported to have closed above 4% for the first time since late March.

Sell Side News:

• Morgan Stanley maintained "Overweight" rating, raised price target from $280 to $290.

• Nasdaq.com article: "Carvana (CVNA): Should You Buy It Now Given Wall Street Analysts Look Optimistic?".

• Company News/Press Releases:

• Carvana Auto Receivables Trust 2024-P2 filed Form 10-D.

• Other Relevant News:

• University of Michigan Consumer Sentiment (Preliminary May): 50.8 (Forecast: 53.1).

• April Retail Sales data released.

Friday, May 16, 2025

SEC Filings Disclosed:

• Form 4 by Ryan S. Keeton (Officer): Sale of 13,137 shares on May 16 at $300.00, totaling approx. $3.94 million (10b5-1 plan).

• Form 4 by Ernest C. Garcia III (CEO): Sale of 10,000 shares on May 16 (avg. price ~$300.15), totaling approx. $3.00 million (10b5-1 plan; filed May 20).

• CapEdge listed Form 4 filings for Thomas Taira, Mark W. Jenkins, Daniel J. Gill, and Benjamin E. Huston.

Stock Price (CVNA):

• Open: $294.50

• High: $306.98

• Low: $293.06

• Close: $305.21

• Volume: 3,760,000

2y Treasury News:

• The 2-year Treasury note ended the week at 3.98%.

Sell Side News:

• Nasdaq.com article noted a "Moderate Buy" consensus for CVNA.

Other Relevant News:

• AInvest article: "Carvana's Insider Sell-Offs Amid Rally: Opportunity or Warning?".

• Moody's downgraded U.S. sovereign rating from Aaa to Aa1.

Saturday, May 17, 2025 & Sunday, May 18, 2025

SEC Filings: N/A

Press Releases: N/A

Stock Price: N/A

2y Treasury News: N/A

Other Relevant News:

• Nasdaq.com (May 17/18): "Will Carvana Stock Keep Soaring to New Highs or is the Rally Overdone?".

• Zacks.com (May 17): "Here's Why Carvana (CVNA) is a Strong Growth Stock".

Upcoming Potential Catalysts (As of May 18, 2025):

Earnings:

• Date: Estimated July 21-30, 2025 (some sources point to July 30).

• Q2 2025 EPS Estimate as of mid-May: ~$1.00 - $1.10. (Nasdaq.com data as of May 26 showed consensus EPS forecast for Q2 2025 increased to $1.11).

• Q2 2025 EPS Upward Revisions in last month (Nasdaq data as of May 26): 4.

• Q2 2025 EPS Downward Revisions in last month (Nasdaq data as of May 26): 0.

• Q2 2025 Retail Units Sold Guidance: Carvana expects a sequential increase from Q1's 133,898 units. One source projected 138,000 units. Based on Clarity Markets data, run rate Q2 units imply 143,396 unit sales

Relevant Industry Earnings:

• KMX: CarMax Inc. estimated reporting date June 20, 2025.

• LAD: Lithia Motors estimated reporting period July 25-31, 2025.

• AN: AutoNation estimated reporting period July 23-30, 2025.

• ALLY: Ally Financial estimated reporting period July 16-21, 2025.

Conferences:

• Carvana to Present at William Blair 45th Annual Growth Stock Conference: CFO Mark Jenkins, Wednesday, June 4, 2025.

Macro:

• FOMC Meetings: June 17-18, 2025; July 29-30, 2025.

• US Consumer Price Index (CPI): May data due June 11, 2025; June data due July 15, 2025.

• US Producer Price Index (PPI): May data due June 12, 2025; June data due July 16, 2025.

• U.S. Retail Sales: May data due June 17, 2025; June data due July 17, 2025.

• University of Michigan Consumer Sentiment: Final May data on May 30, 2025; Preliminary June data on June 13, 2025.

• Employment Situation (Jobs Report): May data due June 6, 2025; June data due July 3, 2025.

• Gross Domestic Product (GDP): Q1 2025 (Second Estimate) on May 29, 2025; Q1 2025 (Third Estimate) on June 26, 2025; Q2 2025 (Advance Estimate) on July 30, 2025.

Industry Releases:

• Manheim Used Vehicle Value Index: Data for May typically released in early June; June data in early July.

Other Catalysts:

• Automotive Tariff decisions (e.g., US-China tariff truce expiry ~Aug 10-12, 2025).

• Significant macro releases, 13-F releases.

News Mentioning CVNA 5/12-5/18/2025:

• Cabot Top Ten Trader (May 27 issue, reflecting on prior weeks): Featured CVNA with a buy range of $290-$303.

• Nasdaq.com (May 13): "$CVNA stock is up 5% today. Here's what we see in our data" - recapped trading, insider activity, hedge funds, and analyst ratings.

• Nasdaq.com (May 17-18): "Will Carvana Stock Keep Soaring to New Highs or is the Rally Overdone?" - discussed YTD gains, Q1 results, and Zacks Rank.

• Zacks.com (May 13): "Brokers Suggest Investing in Carvana (CVNA)" - highlighted favorable ABR.

• Zacks.com (May 14): Multiple articles on CVNA's stock momentum and 52-week high.

• Zacks.com (May 17): "Here's Why Carvana (CVNA) is a Strong Growth Stock".

• AInvest (May 16): "Carvana's Insider Sell-Offs Amid Rally: Opportunity or Warning?" - analyzed insider selling, valuation, debt, and risks.

• Sports Business Journal / Auto Remarketing (May 13-16): Covered Carvana's PPA Tour title sponsorship extension.

Hedge Fund CVNA Transactions (Filings Reported May 12-18, 2025):

• Millennium Management LLC: Filed 13F-HR on May 15, disclosing ownership of 301,573 shares as of March 31, 2025, a 27% decrease (111,528 shares) from Q4 2024.

• CAS Investment Partners, LLC: Filed Schedule 13G/A on May 15, reporting ownership of 5,685,029 shares (4.20%) as of March 31, 2025, an 11.91% decrease from its prior filing.

• Ensign Peak Advisors Inc.: News reported on May 13 that the firm had increased its CVNA position by 32.9% during Q4 2024, holding 41,230 shares.

• Other Mentions (from Nasdaq.com, reflecting Q1 or earlier activity): Morgan Stanley reportedly removed 3.8M shares (-72.8%) in Q1 2025. Lone Pine Capital LLC added 2.3M shares (new position) in Q1 2025. Thrive Capital Management, LLC added 2.2M shares (new position) in Q4 2024.

Any Additional News:

• OEM Updates & EV Market Shifts: Ford continued focus on hybrids; GM L87 engine recall discussions; Toyota teased new RAV4.

• Relevancy: OEM strategies impact new/used car supply and demand.

• Tariff Landscape: US-China trade truce (eff. May 14) reduced many tariffs but kept the 25% U.S. tariff on imported vehicles/parts.

• Relevancy: Auto tariffs can keep new imported vehicle prices elevated, potentially bolstering used vehicle demand.

• Macroeconomic Factors: Moody's downgraded U.S. sovereign credit rating to Aa1 (May 16); April CPI showed YoY inflation at 2.3%; Preliminary May Consumer Sentiment dipped.

• Relevancy: Impacts borrowing costs, consumer spending, and overall market uncertainty.

I’d like to spart spitting out some more general investing blogs over the next couple weeks/months as I’ve mentioned on Twitter. Call this blog the start of that. For name specific stuff you can read my blog and Twitter where I’ll start posting updates on free stuff. This article will naturally touch on those, but they won’t necessarily be the focus.

I also will definitely touch more on AI in the future. I can’t stress enough how impressive Gemini is (or any model really) compared to where we were a few years ago. I also chucked some money at for posterity’s sake.

Anyways, feel free to subscribe and such. The format here is a bit all over the place, but perhaps the reading experience is helpful in jogging your own thought process. Cheers.

Subscribe now

Defining Mistakes:

My background is primarily Long Only equity investing. Long term, embrace volatility, blah blah, I’ll stop short of “private equity in public markets”.

I say this to provide context on what I mean by “mistake”. I’m not talking about “I entered $509.1 instead of $50.91” or “My Newtonian-Lagrangian asymptote pi kappa phi was off by 7 Dusseldorf’s within 5 picoseconds on that trade and we lost .00001 bps”.

In this context, by mistake we’re talking about mistakes predicting differentiated future business trajectory 5 years out from an outside perspective. Really simple stuff in all honesty.

In this area, even defining mistakes can be a challenge. To be realistic, there are a bunch of constraints on a fundamental equity analyst/PM. You only have so many hours you can feasibly dedicate to research, so many dollars you can spend on research, and basically unlimited knowledge to collect. Many public companies will have dozens of internal teams working to create simple financial projections that can be totally wrong even with internal data. Our industry hires 20 somethings to outperform their work with way less info!

Personally, I am very lenient when it comes to what constitutes a mistake and dislike the results bias from allocators. For example, it would have been rather ridiculous for an asset manager to be ridiculed on 9/17/2001 for not foreseeing the recent market drop. It of course hinders their results in that moment, and the job is to make money, but some level of detaching outcome and mistake is clearly warranted.

Yet at the same time if you for whatever reason happened to both manage money and have a good relationship with Al-Qaeda, you could have made generational returns (ignoring the future indictment of course). The conditional probability of two Boeing 767’s hitting the twin towers that day would have very different answers depending on who you kept as friends.

Now of course, 9/11 was an extremely tragic one-off terror attack that permanently changed aviation safety. The reaction was not ridiculing asset managers nor focused on asset managers at all, it was largely a nation coming together to offer condolences and help where necessary and focusing on not letting it happen again.

Equity research is clearly much lower stakes, but I use the extreme to highlight that even very very bad outcomes, arguably one of the worst events in the history of our country, can occur with little justifiable ability to sling around blame even if theoretically the outcome was knowable.

So what would I call a mistake? Let’s do a non-exhaustive list:

1. Missing table stakes information.

   1. Say your model has 10% growth for CVNA units in Q1 heading into earnings even if that is practically impossible based on alt-data. You can’t pull the “unknowables” card if clearly everybody else knew.

2. Being overly emotional.

   1. If the market makes you feel some type of way and you blow out at the bottom or over indulge at the top, that can be extremely value destructive.

3. Optimizing for things other than returns.

   1. You’re a fund and you buy some avant-garde thesis vs a higher underwritten return in say MSFT just so you look differentiated for LP’s. Or you’re an individual and you buy GME so you can chat with your buddies about giving it to the “hedgies”.

4. Limiting your opportunity set.

   1. If you only did “predictable future cashflows” stocks, you probably underperformed the indices over the past couple decades as we witnessed almost all returns come from like 10 tech stocks.

5. On the flip side, straying from your lane.

   1. There are also plenty of people and firms that really had no idea what they were doing, but hopped on hot new trends. Sometimes it works, like AI/crypto. Sometimes it’s hilariously value destructive like 3D printing. It’s good to know where you can realistically provide good analysis, but it’s a balancing act with not getting stale.

6. Trusting the wrong people.

   1. This is the most nebulous of the above. You can’t get inside someone’s head, but it’s good to occasionally take a look back and see if your judgement of people has been directionally correct. Was a CEO truly a visionary or just a good liar?

7. General judgement errors and logical consistency.

   1. If there truly were a bunch of unknowable’s, what’s the track record of your judgement with those? Is there a way to pin down what went wrong? Why? This is sort of the catch-all for bad outcomes that flow through the above without a clear error.

I like to have this list as a way to mentally benchmark everything I own and the outcomes, it even extends far beyond equity investing. Each mistake is different, the ways you handle mistakes differ, but having some general buckets makes it easier to parse when you don’t always know what the mistake even was. “Why did my investment underperform” can be entirely unrelated to “Why did this business underperform”. Say you loaded up on physical retailers in November 2019. You got totally and completely fucked. The business underperformed because of Covid, but your investment likely underperformed (assuming you held) because you got something wrong with modeling the risk there in a new environment. The business couldn’t have avoided covid, but you personally have much more flexibility in public markets than say Party City did to entirely change their business model while heavily levered.

Running your own outcomes through the list above hopefully creates some more intellectual honesty about where process improvements can happen. We don’t want to place blame for unknowable’s, but we also don’t want them to be a scapegoat.

Why Opine On Mistakes:

My background includes a bit of work on the allocator side of things not just the direct investing side, so I’m likely a bit more indexed towards the “theory” here. Like anything that’s a highly competitive multi-player field, I believe that to be good at investing requires constant iteration and learning. Analyzing mistakes happens to be quite good for that! You can also obsessively try to not make mistakes and still get bad outcomes, which is probably the best time to take a step back and analyze the situation for potential lessons.

With myself for example, I have had one position that has caused meaningful drag on performance in the last few years. I easily liquidate it, continue on with my life, quote a high hit rate, say things were unknowable, and move on. That is certainly better for marketing, and perhaps even a good approach overall, but instead here we are, trying to do better.

The Flow Chart Of Mistakes and CDLX:

Let’s do a quick case study using my mistake list and CDLX. To be clear ahead of time, I still own CDLX and don’t hate the stock, but I think at this point it may warrant a “mistake” flag even if I’m not yet 100% sure. I do this to be illustrative of the process, condense my own thoughts, and perhaps provide a learning opportunity. So let’s take a self-deprecating journey together using my list from above:

0. Preview:

   I first purchased CDLX in I believe late 2021 or 2022. I wrote extensively about the stock here and on Twitter. It has varied from a small starter position, to at peak ~50% of my capital, to currently 5-10% given on the day. I’ve added, I’ve trimmed, I’ve seen 3 CEO’s, I’ve had it be up 3-4x, I’ve had it be down >75%. Really the been around the block a few times in a shockingly short timeframe. This will be a very condensed analysis of how things have gone and my perspective.

1. Missing table stakes information.

   The first mistake I always consider is if I just missed something. Especially with small caps, you can have days when stocks just arbitrarily go up and down 5-10%. Combined with the rather common semi-neurotic boss, it’s easy to stress out that the market knows something and you fucked up. What I can say with CDLX, is that I don’t believe I was ever behind the curve on information. Due to my own research, my network, calls with management, the work of , expert calls, 1P data collection, etc, I’d like to imagine I was rather informed on the actual established “facts” of the business as well as one could from the outside.

   That’s not to say I knew everything I wanted to know, as CDLX is perhaps the most opaque stock I’ve had the pleasure of analyzing thus far. The secrecy of banks combined with general small cap inefficient IR leads to a company that can’t even name their own clients. We have no disclosures on activation rates, MAU’s has been redefined twice and none of the three versions are particularly good, economics per client are directional guesses at best, ROAS is drip-fed by agency LinkedIn posts, etc. There’s a ton of missing info, but none of it is table stakes.

   So I’d tentatively say that my actual fact finding with CDLX was done decently well. Of course maybe it becomes obvious in the future I did something clearly wrong, but as of now I’d say the judgement leans in my favor.

2. Being overly emotional.

   If you ask my ex-girlfriends, my problem here is typically the opposite! I do think it’s natural to get an emotional link with any position, especially if it’s volatile. It does suck for something to erase value, it is great when something creates value. The trick is to minimize the extent to which that has an effect on investing returns.

   I am quite transaction light in the account I manage, so I wouldn’t say emotions really have much impact on the returns there. My PA is more transaction heavy, more volatile, but has ended up in a pretty similar spot return wise. Doing a little PA trading and arguing online can be an outlet for when you feel some type of way, but leave the legitimate transactions to times when the headspace isn’t clouded. I’d like to think I did a decent job of that.

3. Optimizing for things other than returns.

   Thankfully I’m just some guy. I was managing internal capital and beholden to a boss, not numerous external LP’s and fundraising needs. I definitely made him quite unhappy at times for a variety of reasons, but a little bit of an adversarial relationship when managing funds for someone can be healthy to keep focus. This does become a very hard problem if you do need to fundraise for survival, so it can be tough to call it a mistake, but when our lens is simply “are you doing what’s best for returns”, we have to be a bit ruthless. If I end up in a fundraising position at some point, I’d hope to not fold on this principle, but at the end of the day you need money to manage money.

   From a personal perspective however, there’s certainly some conflicts when you start publicly writing about stocks. In a world where CDLX was say $40 currently, not only would I have a lot more money, but I’d also likely have a lot more followers, fewer upset DM’s, etc. When publicly giving opinions on the future of anything, those opinions will come back to help or hurt you. Your reputation becomes leveraged to the opinion. This has the opportunity to produce some very unideal doubling down, which can not only be destructive of your reputation, but more importantly your returns.

   I’d like to again say I handled this somewhat well. I don’t think my writing has had a material impact on my portfolio management, if anything it helps with thought crystallization which can make you think harder about your assumptions. Of course I can’t say definitively as I’m the guy who wrote about the stock. What would I have done if I never wrote about it? Impossible to say for sure. I have added, I have trimmed, generally because I think the stock is cheaper/more expensive relative to other opportunities. My Twitter/Substack following was never the prime consideration at least consciously. So we’ll tentatively call it a pass.

4. Limiting your opportunity set.

   This mistake isn’t really applicable for a stock that you actually invested in, more of an error of omission type deal. That said, there is an opportunity cost to every second of your time. For example, I’d say on average, each minute of research on a small cap is more likely to produce marginal insights vs each minute on a megacap. CDLX was definitely one where at the time incremental work felt productive as things like SBUX temporarily returning was visible via app tracking and saw the stock up something like 15% over the course of the day as the news spread. Little nuggets of differentiated information were obtainable, but was it worth it? For small cap differentiated stuff you have to be checking niche channels like singular apps, personal LinkedIn’s, etc. It isn’t generally cross productive with your other holdings. It’s also very hard to formalize by nature.

   From the lens of reviewing my research process, I definitely could have been more time efficient. Like in hindsight if I threw a dart at my watchlist from 2022/2023 everything is up astronomically. Many things I briefly owned heavily outperformed CDLX. Would more time spent on those have been value accretive? Very possibly. This realization is part of why I’m trying to build something that makes it easier to aggregate niche research, but WIP.

   Lesson learned? Going forward try to be much more tech enabled with aggregating weird info flows. Not always possible, but good to keep an eye out especially with how fast AI changes.

5. On the flip side, straying from your lane.

   To be completely honest, I don’t think CDLX was much of a stray. Much of my investing has been in things that are some mix of finance/tech/consumer. My largest position is an online car retailer that packages up loans. The portfolio I covered was almost entirely SaaS or consumer facing tech, with revenue largely coming from advertising for the latter. I really liked META, hated SNAP, thought TTD was too expensive, etc. CDLX was not pulled out of a hat one day, it was a somewhat natural fit at least in terms of understanding what could be understood about the business. I never felt like anything was too complicated to understand.

   That being said, I do think I leaned a bit on the fact other people I believe to be quite smart also owned the stock. Austin Swanson, Cliff Sosin, numerous anon twitter accounts, etc. Not to a degree that I would at all say I didn’t “own” the position, but I’m definitely younger, less experienced, and VASTLY poorer than some of those guys. I’m not exactly the most respectful of authority, but it was definitely a comforting factor that numerous smart guys believed in the story. I think this can even be helpful in terms of figuring out if the market knows something you don’t. If everyone has a pretty similar idea of how things work, then you likely aren’t fucking up Mistake #1, which is a pretty good spot to be. I’d tentatively say the networking here was additive even if I may have sold without it. On net I’d say it made me more aware of the known truths, and what I do from there is up to me.

6. Trusting the wrong people.

   This is where I think we start to get interesting and have to diagnose a bit. The difference between the world now and when I first entered CDLX is something like.

   1. CLO’s are much more common, with Chase launching their own and AmEx adding CDLX in to their network

   2. The advertising world has generally ramped significantly post-covid with META and GOOGL materially larger than a few years ago. The economy has grown and with it digital advertising

   3. Chase saw numerous years of double digit growth and showed the power for the new ad UX combined with bank push to create large growth for CDLX.

   4. CDLX integrations with partners have lead to better proving out of ROAS

   5. Despite the above, BofA is pseudo dropping the contract and explored different vendors. Chase pulled some volume in house. CDLX is smaller than it was a few years ago.

   So why has CDLX underperformed the sort of general market of both digital advertising as well as CLO’s? Why have new entrants taken up some of their space?

   The exact reasons here are still not 100% clear, nor may they ever be, but I have some guesses. Let’s highlight each

   1. Tech development when you don’t control distribution and have 20 different versions in highly regulated spaces with unique security audits is probably really hard. CDLX will naturally take longer, with higher OpEx, to do rollouts relative to those that control distribution. They are also beholden to tech dinosaurs that operate businesses on the opposite end of the spectrum from “move fast and break things”

   2. Given a lack of distribution control and not owning the data, sharing results naturally has some “trust me bro”. Numerous expert calls have highlighted disbelief of CDLX ROAS given the platform is niche and thus not “feelable” in results like a META or GOOGL. Even if CDLX is say providing $1m in value, is that meaningful enough to survive general corporate politics and turnover within advertisers?

   3. Numerous initiatives have been a total waste of time at best, or obscenely value destructive at worst. Dosh was acquired for ~3x the current market cap and no longer even exists! Bridg was even more destructive. The push for self-serve was dropped after a year of wasted effort. Product level offers were promised ~5 years before a single one went live. Local offers was a total bust. Fintech partnerships amounted to nothing, etc.

   4. As a result of the above, we’ve seen management musical chairs and some partner loss for banks that inexplicably make their offering worse instead of modernizing and dealing with CDLX.

   That’s a very watered down version of events, but it’s my basic view of how things developed with CDLX. The general market has left them in the dust unfortunately. The question for our purposes here becomes, how much of that is on the people?

   I truly think the biggest problems in hindsight really were the lack of distribution control and lack of technological control. This is an n of 1 company, there isn’t a plethora of deeply integrated transaction data advertisers. Going in I knew it was a struggle and probably would continue to be, but would have the people really changed that? I think it’s possible, but hard to prove of course. I wouldn’t say CDLX has been dealing with A players, but one would hope the businesses they invest in don’t require A players, as getting A players in the small/mid cap space can be quite difficult, especially if they have to work with banks.

   Where I think CDLX has really struggled on the people side is the insane value destruction and opacity. I definitely will have a higher quality bar when evaluating M&A activity going forward. I definitely won’t be playing lawsuit chicken where we say everything’s fine then arbitrarily settle with no justification. I will absolutely have a higher bar for transparency when decisions are made that are contrary to what I believe should be happening.

   For example, when CVNA was tanking in 2021/2022 my initial models had them cutting units to reduce OpEx and weather through a demand crunch. They did not do this, and instead opted towards growing their way out of the hole to some extent. This didn’t match my thesis going in, but I could see in the alt data and the models that such a strategy could make sense, that they were making headcount reductions without imploding the business model, etc. With CDLX the opacity mixed with the messiness created some tough problems, and it’s hard to have faith the tough problems are being solved when nothing is communicated.

   I’m certainly sympathetic to their plight given the space they work in, and I do believe current management is a better fit, but I think it’s clear in hindsight that Lynne didn’t really know what was going on under the hood, that Karim didn’t have a good idea what he was signing up for, and that the communication through it all was lacking.

   The biggest lesson here is probably that if a CEO doesn’t know how their own business works and I can’t externally prove it out, then it will also be hard to maintain and grow partners/customers. That said, it’s a tricky game because stocks are cheapest when everyone thinks the management is dumb as bricks or criminals. Look at META, look at CVNA, look at TSLA, etc. My bar will simply a bit higher in terms of balancing transparency and opacity.

7. General judgement errors and logical consistency.

   Thus in all my hubris I’ve determined I did not largely make mistakes above, I simply had bad judgement!

   With CDLX, what it really comes down to performance wise is a couple things. First off is that despite being directionally correct about CLO’s scaling over time and Chase likely growing >20% per year by my guess, the business has not seen that output. This is largely due to losing BofA volume, very slow ramps at partners other than Chase, and a touch due to Chase in-sourcing.

   Like to be clear, management was initially telling us the new ad UI/UX would be rolling out starting in 2022. We are now in 2025 and not a single old partner besides Chase is there. US Bank has had the new ad server since 2020 or 2021 and never updated to the new UI. BofA simply never updated to the new ad server, let alone any UI changes. Simply adding some pictures and marginally better ad delivery is an endeavor that has taken ~5 years to get live for one partner. CDLX could have alternatively never acquired Bridg for product level, never acquired Dosh for UI/UX acquihire, fired everyone not just working in product maintenance, and it would likely have 10x-20x the market cap just from eliminating the value destruction.

   The basic idea was like “you got these banks to sign up and share transaction data, surely getting them to add some pictures won’t be that bad” and here we are. I DRASTICALLY underestimated how much of a pain the bank tech issue would be, and I was highlighting it in my initial pitches! Hell AmEx was announced over a year ago and we only started getting some offers a year later, still waiting on a full rollout! I’ve heard that CDLX even offered to loan BofA engineers to make the tech transition happen and instead BofA just degraded their consumer experience because tech is too hard. The real lesson here with the banks is that they all absolutely suck besides Chase and AmEx, which if you had invested in them instead you’d be far better off!

   Beyond my error with the bank tech, I also came into this expecting advertisers to be a bit more rational. It’s a bit of an alien view for me to have CDLX quote you a ROAS, have Nielsen quote you a ROAS, then you turn around and say “I don’t believe you”. I can understand the limitations of the CDLX model that can create that inclination where others don’t have it, but the idea that banks are providing transaction data to a company defrauding you just never jived. The idea was that AmEx and other partners would add volume to make results “feelable”, but the issues with BofA and other partners never scaling into new tech made that idea moot. The idea was that if CDLX was growing >30% a year, surely there was momentum there and belief in the product. Instead it seems like the growth was just because it was a shiny new thing, and once times got a bit tighter, the shiny new toy wasn’t so shiny and everyone retreated to META and GOOGL. My error here was assuming advertisers were playing the game I had defined in my head of turning advertising budget into revenue at the best rates possible. Instead, they were turning advertising budget into job security and herding. One the new toy lost it’s luster, why come back even if it was now a better product than when you were spending on it before?

   As a result of the above, I got pump faked not once, but twice on CDLX. Two times I was up hundreds of percent, and currently it’s down ~60-70%. Drastic differences in IRR! I was indeed correct that the opacity could create very large mispricing, but likely should have had a better model to deal with that. It’s also entirely possible to stock does extremely well from here, because again opacity creates very large mispricing, but it’s also possible it doesn’t and I locked up capital in something where had it been invested elsewhere it would be worth 3-5x or more. I’d tentatively say that a large difference between CVNA and CDLX was that with CVNA I could tell that the business was executing and momentum was moving in their favor, even after big headlines like debt restructuring and such passed. With CDLX we’d get some nice snazzy headlines, tie ourselves in knots trying to figure them out while management said “we can’t comment on that”, then get rug pulled on the disappointing results.

   It’s hard to take definitive lessons from bespoke situations, but I’d guess it’s likely beneficial in the future to more seriously consider the sustainability of quarter to quarter narratives. This is somewhat antithetical to the “long term volatility embracing” approach, but just because something sounds great long term doesn’t mean the narrative power will hold before it’s realized. CVNA had the benefit of beating sell-side expects for nearly 3 years straight. CDLX was a black box of results with product timelines measured in years. Even BLND for example we could see the rate of home purchases and just plug in BLND taking price. If I’m taking positions in things based on over-done narratives to the downside, I need to be prepared to exit positions based on over-done narratives to the upside. Obviously harder said than done, but it’s a process.

Fin:

The goal here was to illuminate my thought process around making investing mistakes using an example from things I’ve owned. If you’re unfamiliar with the name it will likely fail to resonate, but hopefully the general ideas can be worth the skim.

Part of the purpose of this blog is to condense my thoughts and just write stuff, and the theory of investing is a big share of the mental work whereas much of the writing is the stock specific output. I think mistakes are just generally under covered and discussed, and reading through other people’s learnings can be quite helpful to me, so maybe this was helpful to you.

Feel free to subscribe or share as you wish. Cheers.

Subscribe now

Clarity Markets Newsletter

Dive into the intersection of prediction markets and deep public equity research

This article is a brief snippet based on our website tracking of CVNA which reports earnings after the close on 5/7/25.

CVNA Pre-Earnings Outlook:

• Q1 Used Retail Units: 135,300 units sold, up 47% year-over-year. Note that actual results may vary by 2–3% due to adjustments.

• Q2 Used Retail Units: On pace to reach approximately 145,000 units, representing a 43% increase year-over-year. Earlier in the quarter was high 40’s, recent days have been stronger after a slight slow down through April.

• Average Selling Price (ASP): ASP declined slightly in Q1 on a sequential basis, though reported results are affected by partner inventory accounting making exact numbers hard to guess. Tax season typically skews pricing lower. Our Car Data reports provide deeper insights into pricing trends and inventory mix impacts. Q2 ASPs are trending higher, with bumps due to tariffs earlier in April that alleviated slightly through the month. Current higher ASP’s are due to a shift towards more premium inventory, with CVNA seeing large discounts relative to KMX on a mix-adjusted basis.

• Inventory Levels: Inventory slightly declined in Q1, starting at 50.7k units and ending at 49.8k. Early Q2 data shows a modest build, with inventory currently around 51k units.

• Financing Trends: CVNA’s baseline APR estimate began Q1 at 9.04%, rose to 9.95% by quarter’s end, and remains steady. Estimated down payments increased in April but have fallen in recent days.

KMX Data Coverage Now Available:

We’ve recently launched data tracking for KMX, with coverage of key metrics similar to CVNA. Currently, there’s about a two-week reporting lag, but we plan to tighten this over time. A free sample report is available on our site. Pricing matches your existing CVNA subscription, feel free to reach out if interested. A subscription to KMX will also provide a comped subscription to our Used Car data.

Used Car Data Enhancements:

We just launched our Used Car data which includes breakdowns by make, model, trim, and year for both inventory and sold units. It also features mix-adjusted price comparisons between CVNA and KMX. KPI add-ons like unique SKU counts can be included as we iterate on the coverage. Pricing is 50% of the CVNA data plan or included at no charge with a KMX subscription. Files through April are available for free via the site as a sample. Future coverage will require subscription.

Upcoming: CARG Inventory Tracking

We’re actively working on full inventory coverage for CARG. Once complete, this data will be incorporated into the used car dataset to provide a larger picture of market trends and comp KMX/CVNA to smaller dealers.

Hopefully this email is helpful to those following CVNA earnings this evening. Cheers.

Subscribe now

Below I’ve cross posted the most recent blog from my Clarity Markets account exploring prediction markets and finance in a newsletter-esque way. What I’m currently thinking is to use that account for covering a variety of interesting things like the prediction market stuff, used car sale trends, other interesting topics, etc. My goal is to use data+prediction markets to give differentiated high quality insights instead of just being another blog rattling off unverifiable opinions or the same old news flow. If that sounds interesting, feel free to subscribe to that account here.

Clarity Markets Newsletter

Dive into the intersection of prediction markets and deep public equity research

This blog will likely continue to be my own personal one covering my investments/what I’m up to/etc. In recent months that’s largely been working on the Clarity website which now has CVNA/KMX/CDLX alt data available with both paid and free versions. We couldn’t quite get our prediction market beta ready for this Q1 earnings season, but are trying to get a beta up in the next couple weeks for Q2.

What we’ll likely do is a smaller rewards pool as we work through the initial rollout and technical complexities needed to support live trading. Once we have things ready I’ll throw out links here and on Twitter which you can follow if you’d like to stay up to date.

As many of you followed me for my investing commentary, I regret to inform you that I’ve not dug a ton into investment opportunities. Thankfully the primary holding in my portfolio is CVNA, which albeit off from highs, has led to my portfolio somehow being positive YTD! I continue to strongly believe in that business and the momentum they’ve had is phenomenal. As will soon be available on our site for free (and is currently available to paid users), growth has been quite consistent in Q2 despite macro uncertainty.

I did end up divesting BLND to add to CVNA in the $180-$190 range, as well as the fact I simply keep up with CVNA far better since we cover it.

CDLX has not had much of note occur in the last 6 months or so. This morning BofA did notify them they do not wish to renew their contract, but also stated they’d like to continue coverage into 2026. This isn’t very material to the CDLX results as BofA had functionally no offers for awhile. I did believe CDLX had some positive optionality in BofA revamping the program and efforts have been made from the CDLX side, so it’s unfortunate to see the lack of renewal, but there is also nearly a year to potentially sort out some kind of solution. In hindsight selling CDLX at $20 on the AmEx news would have been a great decision. That said, given the stock price has fallen so much, I don’t mind keeping a smaller position around on the chance they can figure out some growth. While correct about Card Linked Offers being a bigger trend in the future, CDLX has not seen much success from the trend. There’s likely a lesson in there that can be pulled apart after we see everything unfold.

Anyways, below is the article I’ve mentioned, please enjoy and let me know if you’d like to see more or less of this type of content going forward.


Intro:

The Alternative Alpha brand has now rolled into the Clarity Markets brand. This blog used to be solely dedicated to covering our CVNA alternative data. We still cover that (and KMX/CDLX) via our website, but aim to significantly expand our coverage in the future. As a result, this blog will just be about general Clarity Markets things, which includes in-house alternative data, interesting public data trends, and of course prediction markets.

As an idea, our future blogs will cover trends we’ve seen in Make/Model of the used car industry, Google trends and social media stats for retailers, consensus expectations for inflation, shifts in geopolitical expectations such as Taiwan invasion possibility, etc. The goal is to provide readers with a variety of data points on current topics, hopefully providing some “clarity” to their worldview. If that sounds interesting, feel free to subscribe now

Intro (To the article this time):

Clarity Markets currently is home only to alternative data coverage of a few companies. The goal is to expand that dramatically over time with the help of prediction markets.

If you are unfamiliar, prediction markets are essentially platforms where one can trade opinions. Users can buy “Yes” or “No” tokens for questions like “Will Jerome Powell say “inflation” during his next press conference?” These Yes and No tokens trade around an implied odds from 0%-100%. If I believe he absolutely will say inflation, I can buy Yes tokens at the market rate of 90% and get my snazzy ~10% return next month. If he doesn’t, my Yes tokens are worth 0 and the No tokens 10x from 10% to 100%.

Prior to 2024 prediction markets were for the most part a niche hobby for those with an extreme interest in politics that wanted to trade election results. After a variety of legal battles around regulation, we’ve started to see them enter the mainstream with numerous billions of dollars of volume on things such as Trump’s election odds. They’ve even started eeking their way into the finance world with some small rollouts at Robinhood and IBKR if you’d like to trade weather events and sports from your brokerage.

There are a variety of companies in the space today such as Kalshi/Polymarket/Manifold/PredictIt/etc. Some utilize real money for their markets, others utilize fake currency. At the core, all that really matters for the purpose of research is how well they can answer questions, and the evidence thus far points to pretty well.

That idea is why we plan to roll-out some tests with a prediction market on our platform in the coming weeks (more on that in later blogs). For now, let’s dive in to the article.

Navigating Uncertainty:

I believe it’s safe to say we are currently experiencing uncertain times. Irrespective of politics, the current administration has created a significant amount of business and geopolitical uncertainty in recent months with the VIX index hitting levels not seen since 2020 Covid.

From an investing perspective, the best we can hope to do is try and get an understanding of both what’s actually happening, as well as what consensus believes is happening. That is where I believe prediction markets and data can be of great help.

Below I’m going to go through a variety of topics that have existing markets and cover where consensus is at in those markets. This is a blog format I haven’t done before, but am considering making recurring coverage. If you love it or hate it, feel free to let me know!

As a reminder, the way these markets work is questions either resolve “Yes” or “No” to 100%, but trade at an implied odds in the meantime. They have order books and market makers. Due to option value and time value of money, odds near extremes such as 0-5% or 95-100% may err farther from 0/100 than the market actually expects. Afterall why lock-up money for 1 year to buy a 99% Yes where you make 1% if you’re correct?

Taiwan Invasion Risk:

With geopolitical tensions rising, particularly with China, the Taiwan question has popped up a bit more lately. Below I’ve compiled predictions from Polymarket (real money) and Manifold (fake money). The consensus on these actually hasn’t changed much despite recent tariff shenanigans. These markets do believe it’s roughly a coinflip that China and Taiwan have a military clash sometime in the next decade. That timeframe is long enough to be of questionable accuracy given option value/time value of currency, let alone the decade between there and now, but I thought it worth highlighting where consensus is currently pegged. Plugging said expectations into a model for TSM can be an interesting exercise.

Polymarket:

Manifold:

Tariffs:

Tariffs are front of mind for many given the broad impact to companies both large and small. In aggregate, the prediction market consensus is that a deal with China is likely to happen, but only to lower tariffs, not remove them. The markets also believe the reciprocal tariffs in general will be reduced/removed before the deadline. The general read seems to be that the tariffs have a predominantly Chinese focus without much sticking power elsewhere besides Mexico/Canada.

Polymarket:

Kalshi (real money):

Manifold:

May Fed Decision:

Macro economics can be fun to spectate given the many moving parts. We can clearly see rate cut expectations moving significantly upwards in early April when tariff chaos was getting started. As things have slightly stabilized, expectations have moved back towards consensus of no cuts in May. These rates markets have tens of millions of dollars in volume which is substantially higher than many other markets on the platforms. The Manifold website tends to skew more politics and has less volume on these macro questions.

Kalshi:

Polymarket:

Manifold:

2025 Rate Cuts:

Overall rate consensus for 2025 is all over the place, with a relatively even distribution from 0 to 4 cuts, skewed towards 2/3. Expectations have come down from peak uncertainty in early April, but are still higher from the start of the year likely related to recession risk.

Kalshi:

Polymarket:

Manifold:

Recession in 2025:

A result of trade being all over the place is of course recession risk. The consensus here has exploded up from ~20% at the start of the year during initial Trump election euphoria to ~60% currently. This is despite expectations for Q1 GDP to be positive real growth, implying high odds for negative comps for Q2-Q4. The odds here have only slightly come down from highs despite beliefs that tariffs will come down meaningfully later in the year.

Kalshi:

Polymarket:

Manifold:

Kalshi GDP:

Polymarket GDP:

TTWO and GTA6:

As a sample of what can be possible for single stock KPI’s, markets like these trying to predict the release date of GTA6 can be interesting.

Kalshi:

Conclusion:

Our future vision for Clarity Markets is to have a variety of single stock and market wide KPI’s that answer key questions for whatever you may want to know. This blog (and our website data subscriptions) can then be used to disseminate high signal consensus on what matters most.

We’ll also explore a variety of other topics, with an upcoming blog exploring trends in used car sales. For now, all blog posts here will be free as we establish core coverage and cadence.