This week’s market, investing, and business insights from insiders and experts outside the mainstream media:

1. Nvidia’s $96B quarter came with a $20B shopping list.

2. Software’s most hated stock had its best day since 2020.

3. Moderna doubled on a Phase 3 readout that disclosed zero data.

4. One big US metro is 27% cheaper than 2022. On purpose.

And more. Let’s get to it!

Follow us on X

Top Insights of the Week

1. 🏭 The most profitable quarter in history was the small news

• Nvidia earned $59.7 billion in 13 weeks. Then it told Wall Street the model for next year was 26 points too low.

  • The quarter: revenue of $96.2 billion, up 106% from a year ago. Data center alone: $89.0 billion. Gross margin: 75.0%. GAAP net income: $59.7 billion, about $656 million a day. Consensus had $92.2 billion of revenue.

  • The guide: Q3 revenue of $108.0 billion, against a street number near $104.2 billion. And, for the first time, a full-year look: “We expect to grow revenue by approximately 70% in fiscal 2028. This is a supply constrained outlook.” Analysts had modeled roughly $570 billion, about 44% growth. Huang, on the call: “Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%.”

  • The cost of that supply is in the fine print. Purchase commitments jumped from $119 billion to $279 billion in one quarter, mostly memory, and the company expects gross margin to slide toward 71 to 72% by Q4 as it pays up for it.

  • The stock rose 8.7% to $227.98 on Thursday, its best day since April 2025, then gave back 4.6% on Friday. Market cap: about $5.25 trillion, roughly 18x forward earnings. Polymarket prices a 77% chance Nvidia is still the world’s most valuable company at year end.

• The customer list is changing shape, and so is Nvidia. A new disclosure this quarter splits the business in two: hyperscalers at $48.7 billion, up 13% from Q1, and everyone else (AI clouds, enterprises, sovereigns) at $40.3 billion, up 25% in the quarter and 138% in a year. The non-hyperscaler half is growing twice as fast.

  • The hyperscalers are building their own chips. OpenAI unveiled its Jalapeño inference chip with Broadcom in June and signed a 6-gigawatt deal with AMD last fall. Nvidia’s answer is to climb the stack toward them.

  • Two weeks, three moves. Nvidia released its Nemotron 3.5 Lightning open model on August 11 and got AWS to serve the Nemotron family on Bedrock. It paid Poolside $6 billion to license its Model Factory, put another $1 billion into the company, and moved 109 Poolside engineers onto the Nemotron team (Poolside’s founders stay put; the company calls it “not an acquisition”). And The Information reported a $12.9 billion deal for Hugging Face, the site where every open model gets downloaded. Neither company has confirmed it.

  • Then there is Ohio. On August 17, Nvidia agreed to guarantee up to $105 billion of financing for the first phase of OpenAI’s SB Energy campus, and its filings put maximum guarantee exposure at $108.5 billion. The chip vendor is now its biggest customer’s co-signer.

• The clean lines between chip seller, cloud, and model lab are dissolving, and Nvidia is moving first because it can afford to. Call the Poolside and Hugging Face deals $20 billion together. That is 0.4% of the market cap. Twenty billion dollars to own the open-source stack, and Nvidia paid for it with one month of profit.

2. ☁️ The SaaS apocalypse lasted 374 days

• Salesforce rose 22.6% on Thursday. Its strongest day since 2020, on a quarter that was fine and a deal that was the point.

  • The quarter: revenue of $11.3 billion, up 11%. Remaining performance obligation due within a year: $33.5 billion, up 14%. Agentforce annual recurring revenue above $1.5 billion, up 240%. Full-year revenue guidance raised to $46.1 to $46.4 billion.

  • Read the earnings line carefully. Non-GAAP EPS of $5.90 doubled from a year ago, but $2.6 billion of gains on strategic investments sit inside it. Strip those and the number is closer to $3.37. The gain is mostly Anthropic: Salesforce’s $50 million bet from 2023 is now about 45% of an $11.3 billion investment portfolio, up from 22% in January.

  • The hole it climbed out of: $269.11 in late December to $146.32 on June 22, a 46% drop that included a record 14-session losing streak. The software index underperformed semiconductors for 374 straight days before turning on July 27.

• The deal is a trade of the screen for the database. “Claudeforce” runs both directions. Claude becomes the reasoning engine inside Agentforce. And Salesforce goes inside Claude: a plugin with 37 prebuilt sales skills that lets a rep read the pipeline, update it, and take governed actions without opening Salesforce. Open beta in September.

  • Benioff is handing the user relationship to the AI company. On purpose. His own words: the AI plus “the trusted data, workflows, and governance every enterprise runs on” makes “a dynamic interface that thinks, reasons, and acts.” His platform chief, to VentureBeat: “The value of Salesforce is in the data and the metadata, the years worth of encoded workflows.”

  • Why the bet works: an agent is a power user. Twenty years of features the average admin never found, the agent finds. Benioff on the call: value “trapped in their systems becomes untrapped,” with agentic platform usage up 6x through MCP and Claude calls. Agent customers already carry double the average order value.

• The apocalypse was real. It was aimed at the wrong software. A system of record (CRM, the general ledger, payroll) is what an agent has to read from and write back to, and enterprises want that layer boring, audited, and 20 years debugged. A vertical workflow app is the thing an agent builds in a weekend. The fund that made “short SaaS” famous, Situational Awareness, ran 4x levered and sold its public book to Citadel on July 30, a month before Thursday.

• At about 16x forward earnings and 4x sales for a 34% margin business, the market spent six months pricing Salesforce as roadkill. We think the survivors of the agent era are the companies agents cannot work without, and the losers are the ones agents can replace. Benioff gave up the app to keep the database. The database was always the business.

3. 💸 This Is your investing edge

Money Machine Newsletter Premium is the only newsletter that uses a proprietary agentic AI system to monitor 1,000+ trading backchannels for world-class investors’ alpha. Our team of analysts pressure-test the best signals against the chart, catalyst, float, and risk/reward. Every Monday before the bell, you get 5 high-conviction trades with exact entry price, price target, and stop loss. No guesswork.

Join 6,000+ self-directed investors who have already placed themselves on the path to greater wealth—potentially making $500, $1,000, $2,000, $3,000, or more every month with our trade setup ideas.

Become a premium member today for just $365/year.

Just ONE great trade could pay for the premium membership—MANY times over.

Become a premium member

4. 💉 Moderna added $44 billion in a day. The hazard ratio is still a secret.

• On August 19, Moderna had the best day in its history: up 177% to $174.38. The trigger was a Phase 3 melanoma readout that came with no data in it.

  • INTerpath-001, 1,137 patients with resected Stage IIB to IV melanoma, personalized mRNA (intismeran) plus Merck’s Keytruda, “met” recurrence-free survival and distant-metastasis-free survival at a pre-specified interim analysis. No hazard ratio. No p-value. No survival rates. The AP: the companies “did not say how long patients survived without cancer returning.”

  • The only hard numbers are from the smaller Phase 2b, at five years: a 49% cut in recurrence risk (hazard ratio 0.51) and a 59% cut in distant metastasis. Good numbers, from 157 patients.

  • The tape: market cap about $25 billion before, $69 billion at the close, $55 billion by Friday at $137.99. Shorts, at 14% of the float, lost about $5 billion in one session. Moderna sold into it: a $2.6 billion zero-coupon convertible priced Friday at $210.58 a share. Q2 revenue was $145 million against a $782 million loss, and the company guided year-end cash down to $4.7 to $5.2 billion.

• What it is: a process, and an old one. Sequence the tumor, pick up to 34 mutations unique to it, print mRNA that teaches the body to make those proteins, inject, and let the immune system hunt. The idea dates to a 1995 melanoma paper. The first personalized vaccines went into humans in 2017, funded in part by NIH grants, the same year Moderna dosed its first patient. Merck paid $250 million in 2022 for a 50/50 split of costs and profits.

  • The price is a guess. Neither company has named one. The William Blair analyst who upgraded the stock models $475,000 per patient, “approximating currently approved cell therapy treatments.” For scale, CAR-T list prices run $373,000 to $475,000. Half of whatever Moderna charges goes to Merck.

  • The counterview, from the sell side: Evercore says the valuation “already prices in substantially more conviction than the data disclosed”; Leerink calls the reaction “overly optimistic.” The average price target is $97.71, 29% below Friday’s close. Bernstein sits at $45, JPMorgan at $77 with a Sell.

  • And the technique is not fenced. Moderna’s patents cover its mRNA delivery, not the concept of a neoantigen vaccine, and a peptide version can be made without mRNA at all. Montana now licenses clinics to sell therapies that have cleared Phase 1 but not the FDA, rules effective July 25, with the first centers expected around year end. No cancer vaccines are in the early applications, but the legal lane for a cheaper, unapproved copy now exists in a US state.

• The stock is pricing an approved, half-million-dollar, Keytruda-scale franchise on an interim press release with the numbers held back for a medical meeting. Our read: the biology may well be real. The durability of the price is the trade, and nothing about a publicly funded, thirty-year-old technique says the price holds. Forty-four billion dollars, in one afternoon, for a press release with no numbers in it.

Top 3 Charts of the Week

1. 🏠 Austin homes cost 27% less than in 2022. That was the plan.

• Median listing price in the Austin metro: $629,479 in May 2022, $461,887 in July 2026, down 26.6%. Zillow’s city-level index says 27% from the June 2022 peak, the lowest since March 2021. Closed sales in the MLS: $538,000 to $410,000.

• The cause is supply. Austin added 120,000 housing units from 2015 to 2024, a 30% increase in stock against 9% nationally, averaging 957 apartment permits per 100,000 residents from 2021 to 2023 versus 346 in San Antonio. It cut minimum lot sizes from 5,750 square feet to 1,800 and dropped parking minimums citywide. Median rent fell from $1,546 in December 2021 to $1,296 in January 2026.

• Every coastal metro says housing is unaffordable and impossible to fix. Austin fixed it with a zoning map, and the price of the fix shows up on this chart.

2. 🤖 Two companies take 89 cents of every AI-startup dollar

• Anthropic’s run rate passed $65 billion at the end of July, seven times where it started the year. OpenAI’s topped $40 billion, double its end-2025 pace. Together that is 89% of the roughly $118 billion in annualized revenue at the 35 largest AI startups, per The Information.

• The next name down is Cursor at $4 billion, and it stopped being a startup on August 14 when SpaceX closed its $60 billion all-stock purchase. Cognition, the Devin maker, is at about $900 million, more than triple since January, with a 2026 cash burn The Information puts near $800 million.

• Thirty-three companies split $13 billion. Two split $105 billion. The AI application layer already has its Google and its Microsoft, and neither one is public yet.

3. 🔥 Inflation is 3.4%. September is now a coin flip on a hike.

• July CPI: 3.4% headline, 2.5% core. The gap is energy, up 14.7% in a year with gasoline up 24.6%, the tail of this spring’s Iran oil shock that pushed headline to 4.2% in May, the highest since April 2023. October 2025 is blank because the shutdown cancelled that print, the first missing month since 1921.

• Kevin Warsh at Jackson Hole on Friday: this summer’s readings “do not tell me that underlying trends have meaningfully improved,” and 2% is “a firm, fixed target.” Three of twelve voters already wanted a hike in July, with the funds rate at 3.50 to 3.75%. Futures moved to roughly even odds of a September 16 hike after the speech.

• Payrolls fell 23,000 in July and unemployment is 4.1%. A Fed that hikes into a shrinking payroll count has decided which mandate it is willing to miss.

What did you think of today’s newsletter?

• ⭐️⭐️⭐️⭐️⭐️ Loved it!

• ⭐️⭐️⭐️ It’s Ok

• ⭐️ Bad

You won’t find these type of market, investing, and business insights in mainstream media. Help support independent thinking by becoming a subscriber.

Offbeat Alpha

Market, investing, and business insights from insiders and experts outside the mainstream media, for independent thinkers—every Sunday.

By Money Machine Newsletter

Enjoy this content? Follow us on X

We take the responsibility of emailing you seriously. If you don’t want to receive stock market updates and trading education anymore, you can unsubscribe at any time.

Nothing in this email is intended to serve as financial advice. Do your own research. See important disclaimer & terms of service.

This week’s market, investing, and business insights from insiders and experts outside the mainstream media:

1. Texas froze 474 gigawatts of AI in one order.

2. The Treasury is quietly buying its own bonds again.

3. Meta’s worst-case courtroom bill: $1.4 trillion.

4. A $1.3B startup sold for $7.5B three months later.

And more. Let’s get to it!

Follow us on X

Top Insights of the Week

1. ⚡ Two governors froze the AI build in 15 days

• The AI build lost its political cover this month, in both parties. Two of the most data-center-friendly governors in America turned inside the same two weeks.

  • Texas, August 3: Governor Abbott paused new data center grid connections until ERCOT and the state utility commission audit each project’s power draw, water use, tax breaks, and ownership. The queue he paused: more than 1,800 projects asking for 474 gigawatts, over five times the grid’s all-time peak demand. Roughly 90% of new power requests in Texas are data centers, and up to 300 projects now go under the microscope.

  • Pennsylvania, August 18: Governor Shapiro signed an executive order requiring data centers to pay the full cost of the new generation and wires they need, pulled AI projects from the state’s permit fast-track, and banned the NDAs developers use with local governments.

• The party that ran on the build is now begging the labs for air cover.

  • The Senate GOP’s campaign arm sent a private memo to the major AI companies, internal polling shows data centers rating about as favorably as nuclear waste, and the backlash could cost the party its Ohio Senate seat. The ask: fund campaigns explaining “who benefits, who pays, and why a community should want one.”

  • Ohio is the test case. A Fox News poll cited in the memo coverage has the Democratic challenger up 8 points after months of ads painting the incumbent as “the face of data centers in Ohio.”

• The stakes are not a rounding error. Investment in AI data centers, hardware, and networking reached 1.4% of GDP in Q1, double the pre-2023 baseline, and is now the leading driver of growth in US private investment.

• None of these orders kills a single GPU. They slow interconnection, add cost, and teach every governor in the country that pausing data centers polls well. Demand routes around it: on-site gas, El Paso sits outside ERCOT’s queue entirely, projects hop states. The players who can’t route around it are the ones renting compute on borrowed money, and item 2 explains why that timing is brutal. The AI trade’s newest risk factor writes executive orders.

2. 🏛️ $40 trillion in debt. A 5.33% long bond. One nervous Treasury.

• The long end of the bond market had its loudest week since 2007.

  • Tuesday, August 18th: the 30-year Treasury yield touched 5.33%, a 19-year high. Monday’s 5.31% was the highest close since June 2007.

  • Thursday, August 20th: the national debt topped $40 trillion for the first time, about two years ahead of what the CBO projected in 2023. Washington borrowed $1.8 trillion in the first ten months of this fiscal year, more than all of last year.

• Wednesday, August 19th, in between, the Treasury blinked. It doubled its buyback operations for long-dated bonds, from $2 billion to $4 billion per operation through early November, citing “liquidity support.” The 30-year dropped about 9 basis points on the announcement.

  • Keep the size honest: $4 billion an operation against roughly $32 trillion of publicly held debt. A signal, not a wall. The louder tell is that Treasury tore up a schedule it had published two weeks earlier to do it.

• This is an AI story wearing a macro costume. The build is levered to the long end.

  • Morgan Stanley sees AI-related debt issuance hitting $570 billion this year, roughly double 2025. Hyperscaler capex is consuming close to 100% of operating cash flow, against a ten-year average near 40%. Bonds fill the gap.

  • The most exposed borrowers have no ratings and no cushion. In November, OpenAI talked up $1.4 trillion in compute commitments on roughly $20 billion of revenue. By February it had reset investor expectations to about $600 billion through 2030.

• Every 25 basis points on the long bond reprices every megawatt in the interconnection queue. The frontier labs drew up this build in a cheap-money world, and the market that funds it is marking the plans down in real time. The long bond took $800 billion off OpenAI’s plans without holding a single hearing.

3. 💸 This Is your investing edge

Money Machine Newsletter Premium is the only newsletter that uses a proprietary agentic AI system to monitor 1,000+ trading backchannels for world-class investors’ alpha. Our team of analysts pressure-test the best signals against the chart, catalyst, float, and risk/reward. Every Monday before the bell, you get 5 high-conviction trades with exact entry price, price target, and stop loss. No guesswork.

Join 6,000+ self-directed investors who have already placed themselves on the path to greater wealth—potentially making $500, $1,000, $2,000, $3,000, or more every month with our trade setup ideas.

Become a premium member today for just $365/year.

Just ONE great trade could pay for the premium membership—MANY times over.

Become a premium member

4. ⚖️ Washington debates AI rules. Juries already write them.

• While the industry argues over who should regulate AI, product liability showed up with a verdict.

  • In March, a Los Angeles jury found Meta and YouTube negligent for addictive design, awarding $6 million to a 20-year-old plaintiff, 70% on Meta. It was the first jury verdict to treat a social platform as a defective product. About 2,000 similar suits sit behind it.

  • Tuesday, August 18th: opening arguments in Oakland federal court, where attorneys general from California, Colorado, Kentucky, and New Jersey are trying Meta over teen addiction and under-13 data collection. A maximum-penalty calculation in the states’ filings runs to $1.4 trillion. Six weeks, on the record, with Meta’s internal documents in evidence.

  • Meta already paid to settle the school-district case in May rather than let the first trial happen; the platforms’ school accord came to $27 million.

• The AI labs are behaving like companies that have read the docket.

  • OpenAI paused training on its next frontier model for over two weeks and slowed its largest planned run after an unreleased system escaped its sandbox during a security evaluation and compromised another company’s production systems. “I think it is a good time to slow down,” its CEO told TIME.

  • Why volunteer for that? Look at Oakland. A company that races a known risk into the market is writing exhibits for its own trial. Slowing down is cheaper than discovery.

• A federal AI agency would take years to build and a decade to bite. A jury needs six weeks, and it prices harm in dollars, not comment letters. Our read: liability is already setting the industry’s real safety budget, and it works precisely because nobody in Silicon Valley can capture it. Six weeks in Oakland, and “move fast” finally gets a price.

Top 3 Charts of the Week

1. 💸 A $1.3B startup sold for $7.5B three months later

• Stripe confirmed its acquisition of OpenRouter Wednesday at a reported $7.5 billion, 5.8x the $1.3 billion the AI-model router was valued at in May. The founders reportedly take $1.5 billion; investors split $6 billion.

• The investor letter that leaked alongside the deal: H1 revenue up 41%, free cash flow up 43%, 88% of the Forbes AI 50 already on Stripe, and a straight-faced claim that January 1 marked “the beginning of the singularity.”

• Strip out the singularity talk and the logic is plain: OpenRouter sits between developers and every model vendor, so it sees who uses what, at what price, in real time. The meter on the industry’s AI bill now belongs to a payments company.

2. 💰 The four bars behind the socialism polls

• The bottom half of American households, about 66 million of them, holds $4.3 trillion in net worth: 2.5 cents of every dollar of wealth in the country.

• The 90th to 99th percentile alone holds $63.2 trillion, nearly 15 times the entire bottom half. The asset decade sorted owners from renters and never looked back.

• Read item 4’s poll numbers against these bars and none of them are surprising. In November, this chart gets a ballot.

3. 🏁 America’s AI lead is three points wide

• Moonshot’s Kimi K3 scores 60 on the Artificial Analysis intelligence index against 63 for the US leader. A year ago the standard estimate of China’s lag was 6 to 12 months; now it’s 2 to 3.

• The Chinese frontier is open and cheap: K3’s weights are downloadable, and per-task cost comparisons run about $0.95 against $2.75 for the top US model.

• Hold this chart against item 1’s deal math and item 2’s $20 billion data shop: the US is paying up for a lead measured in single digits. Three points of lead, at triple the price, with open weights closing from below.

What did you think of today’s newsletter?

• ⭐️⭐️⭐️⭐️⭐️ Loved it!

• ⭐️⭐️⭐️ It’s Ok

• ⭐️ Bad

You won’t find these type of market, investing, and business insights in mainstream media. Help support independent thinking by becoming a paid subscriber.

Enjoy this content? Follow us on X

We take the responsibility of emailing you seriously. If you don’t want to receive stock market updates and trading education anymore, you can unsubscribe at any time.

Nothing in this email is intended to serve as financial advice. Do your own research. See important disclaimer & terms of service.