Recent valuations imply confidence, with the cyclically adjusted price-to-earnings ratio near levels seen only during previous market extremes. Earnings forecasts are equally demanding, with investors expecting another year of unusually strong profit growth. Price action on the other hand looks less comfortable. The S&P 500 has struggled to make sustained progress while semiconductors and other parts of the AI trade have become increasingly volatile. Investors are not abandoning the theme, but they are becoming less willing to accept every data-centre project or increase in capital expenditure at face value. The difficulty is that the fundamental case for AI remains strong. Demand for computing capacity continues to rise and the largest technology groups can finance much of their spending internally.

The portfolio problem is how much of the eventual economic value will accrue to today’s listed companies and how much has already been reflected in their valuations. Expensive markets can become considerably more expensive before they correct. Moving into cash too early can be as damaging as remaining fully invested for too long, particularly while the companies leading the market continue to report strong earnings. The more realistic response is diversification.

Protection has an immediate cost

A portfolio designed to reduce the damage from a future correction will usually lag while the current leaders continue rising.

The parts of the market creating the greatest concern are also producing the strongest earnings growth and attracting the most capital. Reducing exposure to them means accepting lower returns if the rally continues.

An investor can conclude that semiconductor valuations are excessive, move into cheaper sectors and then watch chip stocks rise another 40 per cent. The concern may remain valid, but the portfolio will appear wrong for as long as the market continues rewarding that theme. A defensive portfolio that underperforms for several quarters is often abandoned before the protection becomes useful to the investor. A complete exit from AI-linked equities requires considerable conviction, while a smaller reduction in concentration is easier to maintain. Part of a capitalization-weighted index allocation could be redirected towards equal-weighted funds, cheaper sectors or international markets.

The next correction may travel through bonds

The current market does not resemble the housing bubble in every respect.

Household balance sheets are stronger than they were before 2008 and large companies generally carry less financial leverage than banks and property-related businesses did during that period. However, the vulnerabilities have moved elsewhere.

• The federal government is running large deficits while the economy remains relatively strong

• Treasury issuance is heavy and debt-servicing costs are rising

• Data centres, power generation and AI infrastructure are also creating additional demand for capital

Together, these pressures place greater strain on long-term interest rates.

Higher bond yields affect equities in two ways:

1. They raise the discount rate applied to future earnings

2. Provide investors with a more credible alternative to expensive stocks.

Both effects become more important when valuations are already high. So far, earnings growth has offset the increase in rates. The position becomes less stable if earnings revisions slow while borrowing costs remain elevated. More financial activity has also moved into hedge funds, private credit and other non-bank structures. These markets are less transparent than traditional banking and often rely on leverage, derivatives or short-term funding.

The current structure may be less vulnerable to a banking collapse, but more exposed to a bond-market repricing.

Treasuries are no longer a complete hedge

For much of the period after the financial crisis, the case for a traditional stock-and-bond portfolio was straightforward: weak growth pushed equities lower and treasury prices higher. Bonds generated income during normal periods and appreciated when the economy deteriorated. Recently however, inflation has weakened that relationship.

When the shock comes from higher energy prices, tariffs or supply constraints, stocks and bonds can decline together. Equities fall because margins and growth expectations weaken, while bonds fall because investors expect inflation to remain elevated and central banks to keep rates high. Renewed disruption in the Strait of Hormuz illustrates the problem. Higher oil prices are arriving while inflation remains above target. In that environment, long-dated government bonds may provide less protection than they would during a conventional recession.

Of course they still have a role: a sharp slowdown accompanied by falling inflation would probably send Treasury yields lower and long-duration bonds remain useful protection against that outcome. The problem is that they perform poorly against several other risks facing the market. On the other hand short-term government debt offers income without substantial duration risk. Inflation-linked bonds provide a real yield for investors able to hold them to maturity. Cash creates capacity to rebalance after a correction. Commodities can respond directly to supply shocks that damage both nominal bonds and equities.

None is a perfect hedge, but together they cover a wider range of outcomes than long-dated Treasuries alone.

The Role of Cash

Money-market funds and short-term government securities still offer meaningful yields. Cash will of course not match equity returns, but it no longer earns nothing while investors wait. Nonetheless, liquidity remains important. An investor with no liquidity during a correction may have to sell one asset to buy another, while an investor holding cash can add exposure without disturbing the rest of the portfolio.

Cash allocation therefore needs rules, given that capital can and should be (partially) redeployed when valuations fall by a defined amount, when expected returns improve or through scheduled rebalancing.

The Role of Commodities

Commodity exposure has a particular role in the current cycle because AI infrastructure is itself resource-intensive.

Data centers require electricity, natural gas, uranium, copper and specialized materials, while grid expansion requires transformers, cables and new generation capacity. This investment cycle supporting technology shares is also creating demand across energy and industrial supply chains. Commodities can therefore protect against inflationary supply shocks while also benefiting from the physical infrastructure required by AI. That overlap explains why energy producers, utilities and metals businesses have become part of the wider technology trade.

Still, commodity prices are volatile and sensitive to supply responses. High prices encourage new production, substitution and weaker demand. Commodity producers add another layer of risk. They may benefit from rising prices, but they remain businesses with operating costs, debt and management teams. Direct commodity exposure provides the cleaner inflation hedge. Producers may offer better long-term compounding, but they should be analyzed as companies rather than treated as substitutes for the commodity itself.

Defensive equities still carry risk

Reducing exposure to expensive technology stocks does not automatically create a defensive portfolio.

For example, healthcare is less dependent on data center spending and may benefit as AI applications move from infrastructure towards diagnostics, drug development and administration. The sector also has durable demand and generally lower sensitivity to the economic cycle. It does however carry its own risks, as drugmakers depend on patent protection and clinical outcomes, insurers face political intervention and medical devices companies can trade at high valuations despite being labelled defensive.

Japan offers another alternative. Corporate governance has improved, shareholder returns have become more important and valuations remain less demanding than in the US. Smaller companies may benefit as management teams use excess cash more efficiently. The market however remains exposed to global manufacturing, currency movements and the semiconductor cycle. The same applies to Europe, given that its indices contain more banks, healthcare companies, industrial groups and energy producers. They are less concentrated in US technology, but they also face weaker structural growth and greater sensitivity to energy costs.

Diversification works because these markets respond differently to the same economic environment and does not require them to be safer in every respect.

Conclusion

Investors have to choose between remaining fully exposed to an expensive and concentrated market or accepting some underperformance in exchange for a broader set of return drivers.

The second option is unsatisfying because it does not produce a dramatic recommendation. It means reducing concentration gradually, separating the bond allocation by purpose, holding some liquidity and adding assets that respond to inflation rather than falling alongside it.

• The AI investment cycle may have much further to run

• Productivity gains could justify spending that currently appears excessive

• Earnings may continue rising fast enough to support valuations

Nonetheless, a strong technology can still produce weak investment returns when purchased at today’s valuations.

Subscribe now

Unlock Member Access

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For further information read the full disclaimer here.

These transactions do not necessarily begin with a new view of fundamental value. They may instead arise from portfolio constraints, benchmark requirements, contractual obligations or risk-management rules. Even so, the resulting flows can move prices materially, particularly when they arrive in a market with limited liquidity.

For retail investors, the objective is not to identify every institutional order, as that information is rarely available in real time and many institutional trades are deliberately fragmented across venues and execution windows. The more realistic task is to estimate where mechanical pressure may be developing, how much liquidity is available to absorb it and whether the movement is confirmed by related markets.

This requires a different way of looking at short-term price action, as volume alone is not enough and neither is a large options print, a dark-pool transaction or a change in ETF ownership.

Institutional Trading and Mechanical Demand

Institutional activity is often discussed as though every large order represents a high-conviction forecast. In practice however, many trades are driven by implementation.

For example:

• An index fund buys a newly included security because its mandate requires it.

• A dealer buys futures because an options book has become too short delta.

• An authorised participant purchases an ETF basket because an arbitrage opportunity has opened between the fund and its net asset value.

• A benchmarked manager accelerates an order near the close because performance will be judged against the official closing price.

In each case, the institution may be relatively insensitive to the precise execution price. The transaction must be completed for mechanical reasons, which can make the associated flow more persistent than an ordinary discretionary order.

This distinction is noteworthy because markets react differently to information-driven trading and constraint-driven trading. A discretionary investor can stop buying if the price moves too far. Meanwhile, a dealer managing a risk limit or a fund facing a redemption may not have the same flexibility. The resulting order can continue even after the initial price move, creating short-term trends that appear disconnected from news or valuation.

The strongest institutional-flow signals therefore tend to answer four questions:

• What type of flow is likely entering the market?

• Is the trade discretionary or mechanically required?

• How much liquidity is available to absorb it?

• Do other assets or markets confirm the same interpretation?

A large order linked to an ETF rebalance, options expiry or policy shock is considerably more informative.

ETF Creations and Redemptions Can Reach the Underlying Securities

Most ETF trading occurs between buyers and sellers on an exchange. Those transactions do not automatically require the fund to purchase or sell its underlying holdings. The connection becomes more direct when authorised participants create or redeem ETF shares.

When an ETF trades above the value of its underlying basket, an authorised participant may buy the constituent securities, deliver them to the fund sponsor and receive newly created ETF shares. When the ETF trades below the basket, the process can work in reverse, with ETF shares exchanged for the underlying securities. This mechanism is designed to keep the ETF’s market price close to its net asset value. It also creates a channel through which demand for the ETF can become demand for the securities held inside it. A creation worth several hundred million dollars may have little visible impact on highly liquid megacap stocks. However, it may have a larger impact for smaller constituents, securities with limited free float or assets trading during periods of market stress.

Moreover, if a stock represents a large weight across several popular funds, simultaneous inflows can create overlapping basket demand. A stock may be included in a basket that institutional investors are being forced to purchase. The academic debate focuses less on whether this channel exists than on its broader effect. Some research finds that ETF arbitrage can transmit volatility to underlying securities, particularly when arbitrage activity is intense or the underlying assets are relatively illiquid. Other work argues that ETFs improve price discovery, facilitate risk transfer and help information move across related markets more efficiently.

Those conclusions are not necessarily contradictory, given that ETF trading can improve market-level efficiency while still producing temporary pressure in individual securities. The same arbitrage process that keeps prices close to fair value can generate short-run volatility as the basket is assembled or unwound.

For practical analysis, ETF turnover and ETF creations should not be treated as the same variable. Heavy exchange volume may simply reflect investors trading existing ETF shares with one another. Primary-market activity is more relevant because it is the point at which underlying assets are most likely to be purchased or delivered.

Daily changes in ETF shares outstanding are therefore a more useful proxy than ETF volume alone. They remain imperfect because they are delayed and provide no precise intraday execution time, but they offer a clearer indication of whether the fund has expanded or contracted.

A useful ETF-pressure estimate can combine:

• the change in ETF shares outstanding;

• the value of the fund’s net assets;

• the weight of each security in the basket;

• the normal dollar volume of each constituent;

• the concentration of ownership across multiple ETFs.

The same projected basket purchase can represent a negligible fraction of normal trading activity in one stock and several days of volume in another. Normalising the flow by liquidity is therefore essential.

Intraday estimates are more difficult, as ETF premiums, discounts and turnover can provide clues, but those signals are noisy. The calculated net asset value may rely on stale prices, particularly when the underlying assets trade in another time zone or in less liquid markets. A temporary premium may therefore reflect delayed pricing rather than imminent basket creation.

Order-Flow Imbalance and Market Depth

Trading volume does not reveal which side is applying pressure. A stock can trade heavily while buying and selling remain broadly balanced. By contrast, a relatively small amount of aggressive demand can move the price sharply when available liquidity is limited. For that reason, short-horizon analysis often focuses on order-flow imbalance rather than total volume.

Order-flow imbalance captures the net pressure placed on the order book. It includes trades executed against the bid or offer, together with changes in displayed liquidity. New limit orders, cancellations and quote revisions all alter the quantity available at the best prices. Prices can therefore move even without a large increase in executed volume.

Consider a stock where buyers repeatedly lift the offer, but sellers continue replacing the shares being purchased. Trading volume may be high, yet the price remains relatively stable because supply is continuously replenished. Now consider a smaller sequence of purchases after sell-side liquidity begins to disappear. Here the price can rise much further despite lower total volume because fewer shares are available at each successive price level.

Research by Cont, Kukanov and Stoikov found a strong short-term relationship between price changes and order-flow imbalance. The effect became stronger as displayed depth declined. The broader conclusion is that the same flow can produce very different price moves depending on the liquidity available when it reaches the market. A large imbalance in a deep market may be absorbed with little movement, while a moderate imbalance in a thin market can produce a much larger response.

The bid-ask spread provides another useful signal. A widening spread often indicates that liquidity providers are becoming less willing to take the opposite side of incoming orders.

Time of day also changes how these signals should be interpreted. Volume and liquidity are generally highest around the opening and closing auctions. Activity is usually lighter in the middle of the session, so an institutional order can have greater price impact at midday than near the close, even when the order itself is smaller.

The final minutes of trading deserve separate attention. Funds benchmarked to the official close often execute a meaningful portion of their orders during that window. Index rebalances, portfolio adjustments and end-of-day hedging can therefore create concentrated buying or selling pressure with little connection to new company information. Retail data offers only a partial view of this process. Public feeds usually show consolidated trades and top-of-book quotes rather than the full order book across every venue. Large institutional orders may also be divided into hundreds of smaller transactions, making the original parent order difficult to identify. Even with those limitations, order-flow imbalance remains useful when interpreted carefully.

Positive imbalance does not prove institutional accumulation. It indicates that aggressive demand is exceeding the visible liquidity available to absorb it. The signal becomes more credible when it persists, appears across several venues and is confirmed by ETF flows, options activity or broader market movements.

Benchmark Execution and Closing Flows

Institutional portfolios are frequently judged against benchmarks.

Some funds measure execution against VWAP and others use the opening auction, the closing price or a specific index level. These requirements can create concentrated demand during predictable periods of the trading session.

A portfolio manager may decide during the morning that a position should be increased. The execution desk, however, may spread the purchase across the day to minimise market impact and keep the average price close to VWAP. The order may therefore appear as a steady sequence of small trades rather than one visible block. This is one reason institutional accumulation can be difficult to identify from individual transactions.

As already mentioned, the closing auction is particularly important. Index funds, benchmarked managers and derivative desks often need their final exposure to match an official closing level. As the close approaches, the remaining portion of the order becomes less discretionary. The institution may have to complete it regardless of the final price. This can produce sharp price movements in the final minutes of the session, especially on index-rebalance dates or when closing-auction imbalances are unusually large.

Price Jumps Need Flow and Liquidity Confirmation

A sudden move does not reveal whether the next move is likely to continue or reverse.

Some jumps reflect new information and sustained institutional demand. Others are caused by temporary illiquidity, an execution error, one large order or a reaction that quickly proves excessive.

A price jump accompanied by persistent order-flow imbalance, wider spreads and declining depth is more consistent with sustained pressure. The market may still be trying to locate enough liquidity to complete a larger institutional order and fading the move immediately can therefore be costly. A jump without confirming order flow presents a different setup.If spreads remain stable, depth is normal and related securities fail to move, the jump may reflect a temporary imbalance rather than a broader repricing. Mean reversion becomes more plausible, although it is never guaranteed.

Cross-asset confirmation is also valuable.

A sharp move in a semiconductor stock is more credible when other chipmakers, the sector ETF and relevant options markets move in the same direction. A sudden decline isolated to one stock with no related news or flow may deserve more caution.

It helps answer whether the market is in:

• a flow-confirmed continuation regime;

• a temporary liquidity shock;

• a pre-breakout stress regime;

• an unconfirmed move that may be vulnerable to reversal.

What Retail Investors Can Observe

A serious retail framework does not require direct access to an institutional order-management system.

For example:

• Consolidated equity trades and quotes can support estimates of order-flow imbalance, spreads and displayed depth.

• Listed-options data can be used to evaluate strike concentrations, signed transactions and approximate delta exposure.

• ETF holdings and shares outstanding can identify likely primary-market activity.

• Treasury yields and inflation breakevens can help classify the macroeconomic regime.

These sources provide meaningful information, but their limitations remain visible.

Backtesting and Transaction Costs

Microstructure strategies are highly sensitive to data quality, timing and trading costs.

A backtest can appear profitable because it accidentally uses information before that information was publicly available. ETF shares outstanding, options open interest and institutional disclosures are frequent sources of this error. Intraday seasonality creates another problem. Volume, volatility and spreads follow predictable patterns during the session, causing a signal to appear effective simply because it trades near the open or close, when movements are naturally larger. Every intraday signal should therefore be compared with the normal conditions for the same time of day.

Transaction costs are equally important, as a strategy that earns several basis points before costs may lose money after accounting for bid-ask spreads, commissions, slippage and market impact.

Moreover, a stock may appear statistically low risk while offering very little depth at the price required. Position size should therefore account for normal dollar volume, current spread, displayed liquidity and the share of market activity the order is likely to represent.

Conclusion

Institutional flows become most informative when they arise from identifiable constraints.

ETF creations and redemptions, dealer hedging, benchmark execution and policy-driven repricing can all generate transactions that are separate from immediate changes in fundamental value. These flows can create short-term trends, reversals and liquidity events that standard price charts do not fully explain.

Retail investors cannot observe these channels perfectly. They can, however, estimate them using public trades, quotes, ETF disclosures and cross-asset price movements. The objective is to build a consistent interpretation of market pressure.

Used together, the inputs described above can improve the interpretation of short-term market moves. However, if used mechanically without publication timing controls, realistic execution costs or confirmation from related markets, they can create the appearance of information where none exists.

Subscribe now

Unlock Member Access

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For further information read the full disclaimer here.

The first part focused on the framework: how capital concentrates around themes, why 13F filings arrive too late and why options flow, dark pools and market microstructure can reveal pressure before it appears in ownership data. Here, the focus shifts from the framework to application: Vistra, MicroStrategy and the practical limits of trying to replicate institutional positioning as a retail investor.

Vistra Corp (VST)

Vistra Corp has emerged as a primary vehicle for institutional thematic concentration. Vistra’s strategic integration of its nuclear assets, compounded by its acquisition of Calpine to expand its natural gas and geothermal capacity, has driven strong fundamental growth. In Q1 2026, Vistra reported revenue of $5.64 billion (up 43.4% year-over-year) and EBITDA of $1.48 billion at a 26% margin, with normalized EPS rising 315% to $1.35.

The institutional investment thesis for Vistra is built on the pricing power of dispatchable, carbon-free baseload power required by hyperscaler data centers. In January 2026, VST finalized 20-year Power Purchase Agreements (PPAs) with Meta to support three of Vistra’s nuclear facilities in the PJM region, demonstrating its ability to secure long-term, high-margin revenue streams directly from technology hyperscalers. VST’s capital allocation model is highly supportive, with management projecting $10 billion in cash generation across 2026 and 2027, allocating $4 billion to growth capex and $3 billion to shareholder returns.

Vistra’s profile reveals highly active institutional interest. In the most recent quarter, 730 institutional investors added VST to their portfolios, while 736 decreased their holdings, signaling a high-velocity transfer of ownership.

The most notable microstructural validation came via congressional trading: on January 16, 2026, Representative Nancy Pelosi exercised 50 call options with a strike price of $50 (originally purchased on January 14, 2025) to acquire VST common stock valued between $100,001 and $250,000, aligning political capital with institutional flows.

MicroStrategy (MSTR)

MicroStrategy represents an alternative institutional play: a highly leveraged, regulated equity proxy for Bitcoin exposure. Following a 10-for-1 stock split in late 2024 to enhance liquidity, MSTR has traded at a persistent “MSTR Premium” relative to its underlying cryptocurrency reserves, which reached 818,334 BTC following a purchase of 3,273 BTC on April 27, 2026. This premium structure, however, exposes the stock to extreme volatility.

During Q2 2026, MSTR shares declined approximately 20% in response to a 14.1% correction in Bitcoin, which bottomed at $60,000 after peaking at $82,833 in early May. This correction was amplified by a severe Q1 2026 earnings miss, with the company reporting an EPS of -$38.25 against analyst expectations of -$0.86, alongside $25.7 million in corporate insider selling and zero corresponding purchases.

Despite these fundamental pressures, institutional options concentration on MSTR remains exceptionally high. As of July 2026, MSTR options carry an implied volatility (IV) of 96.70% (IV rank of 64.17%), with total open interest exceeding 2.78 million contracts.

A microstructural case study occurred on June 8, 2026, when unusual call option volume surged in contracts expiring August 21, 2026 (74 days to expiration). Over 51,700 call contracts were executed at the $127.50 strike (representing over 80 times the existing open interest) when MSTR was trading at $127.52. With a midpoint premium of $20.75, call buyers established a breakeven price of $148.25, requiring a 16.1% upward move in the underlying stock.

This massive concentration of call options allowed institutional buyers to build a highly leveraged, asymmetric position on a potential summer Bitcoin rally. Concurrently, institutional option sellers executed these trades to capture a 16.27% immediate premium yield ($20.75 / $127.50), translating to an attractive 6.5% monthly return over the 2.5-month holding period, complete with structural downside protection down to $106.64.

Tactical Execution and Structural Replication for Retail Portfolios

Short-Term Execution: Cumulative Volume Delta

To track aggressive institutional buying on public exchanges in real time, retail traders must use Cumulative Volume Delta (CVD).

CVD calculates the running total of net aggressive buying or selling pressure by isolating market orders executed at the ask from market orders executed at the bid. The mathematical model for CVD is defined as:

where:

• CVDt​ is the cumulative volume delta at the current interval

• CVDt−1​ is the cumulative delta from the previous interval

• Vask,t​ is the volume traded at the ask price (aggressive market buys)

• Vbid,t​ is the volume traded at the bid price (aggressive market sells).

When plotted on a tick-by-tick basis alongside price action, CVD reveals institutional accumulation and distribution patterns that are invisible on standard candlestick charts :

Read more

Within this macro environment, institutions have increasingly concentrated their capital around four primary strategic axes:

• AI/Technology Diffusion

• Future of Energy

• Multipolar World

• Societal Shifts

The magnitude of this concentration is historically unprecedented and has fundamentally altered market dynamics. While global equities have delivered robust returns, this capital concentration has pushed equity market concentration to record highs, compressing credit spreads to their tightest levels in years and introducing systemic vulnerabilities.

The risk of this thematic concentration became visible during the trading week ending July 6, 2026, which serves as an empirical case study of institutional crowding and sudden factor drawdowns.

Despite positive index returns (with the S&P 500 rising 1.76% and the Nasdaq Composite gaining 2.12% in response to a softer June payrolls report that lowered Federal Reserve interest rate hike expectations) a violent factor rotation occurred. Institutional capital rapidly exited crowded first-half AI infrastructure and semiconductor winners (the PHLX Semiconductor Index fell 4.4% on the week) to fund long positions in megacap software and technology names like Apple (+8.8%), Alphabet (+6.7%), Meta (+5.9%), and Microsoft (+4.7%).

This abrupt re-allocation inflicted severe performance drawdowns on long/short equity hedge funds. The Goldman Sachs Equity Fundamental Long/Short Performance Estimate fell 1.53% during this rotation, while Systematic Long/Short portfolios declined 2.09%, primarily driven by alpha losses on short-side hedges. This episode illustrates that institutional thematic concentration creates a highly volatile, momentum-driven market structure where crowded trades are prone to rapid liquidation.

To build a viable replication framework, retail investors must understand the technical and capital catalysts driving these institutional corridors.

Capital Expenditure and Technical Catalysts for AI-Energy-Cybersecurity

To appreciate the scale of institutional concentration, one must examine the capital expenditure (capex) commitments backing these themes. The capital cycle for artificial intelligence and data center development has moved beyond speculative enthusiasm into a multi-year, cash-flow-backed buildout.

Capital expenditures across major cloud hyperscalers are projected to exceed $360 billion, reflecting long-term commitments to physical compute infrastructure. Crucially, this capex is not being funded through destabilizing balance sheet leverage, rather, it is supported by robust operating cash flows and highly favorable debt pricing in corporate bond markets, signaling deep bondholder confidence.

This capital intensity is justified by a substantial gap between experimental adoption and full enterprise deployment. While 98% of Fortune 500 companies have experimented with generative AI, only 26% have deployed these systems at scale, with 85% of current usage driven by consumer-facing applications. This divergence implies a long, structural runway for enterprise software integration.

As enterprises transition from experimentation to full operational deployment, cybersecurity has emerged as a major beneficiary of institutional capital concentration. The expansion of AI-assisted coding, with companies such as Robinhood, Coinbase, Microsoft and Google now generating between 30% and over 50% of their new codebase using automated tool, has dramatically expanded the corporate attack surface. High-profile cyberattacks highlight the urgency of securing agentic AI models and sensitive data pipelines.

Consequently, institutions are heavily backing platform consolidation strategies. Palo Alto Networks (which is present in my Portfolio), for example, has completed approximately 1,400 platform consolidations, with an active target of 2,500 to 3,500 consolidations by fiscal 2030. This trend is further supported by strategic acquisitions, such as Palo Alto’s purchase of CyberArk to expand identity security, which remains a critical vulnerability as 89% of corporate security breaches involve credential theft.

This computational expansion has triggered a secondary surge in global energy demand. In the United States alone, electricity consumption driven by data centers is projected to rise by 10% per year over the next decade. This shift has expanded the “Future of Energy” theme from a pure supply-chain conversation into a complex political issue, as rising electricity bills become highly visible to retail consumers, directly impacting localized political dynamics.

To meet this demand, institutions are concentrating capital on low-cost, dispatchable carbon-free baseload power, carbon capture systems, grid evolution and nuclear energy. The global nuclear renaissance is drawing heavy capital as data center operators seek dedicated, non-intermittent power sources to sustain high-uptime compute grids.

Alternative Pathways to Portfolio Durability: Private Equity and Digital Assets

Beyond public equity themes, institutional allocators are modifying their alternative asset allocations to build portfolio durability against equity concentration and positive stock-bond correlations. In the private equity (PE) domain, core private equity has returned to operational fundamentals, utilizing corporate carve-outs, joint ventures and operational improvements to drive returns rather than relying on cheap financial leverage.

Geographically, institutional PE capital is exhibiting highly concentrated geographic preferences:

• Europe: While European public equity indices have lagged US benchmarks for more than a decade, the median European private equity buyout fund has consistently outperformed both its US peers and public benchmarks. This outperformance is driven by Europe’s fragmented markets, lower fund competition and an abundance of high-quality middle-market companies. Notably, technology and telecommunications account for approximately one-third of European private equity deal activity, compared to a mere one-tenth of the public MSCI Europe Index market capitalization, providing a play on digital transformation away from public markets.

• India: This region represents APAC’s fastest-growing private equity market, driven by secular economic growth, global supply chain diversification and accelerating domestic consumer demand.

• Japan: Institutional activity is surging around corporate carve-outs, where conglomerates spin off non-core divisions, and structural corporate governance reforms that unlock shareholder value.

Concurrently, institutional positioning in digital assets has undergone a profound regime shift.

According to the Coinbase and EY-Parthenon Institutional Investor Survey, 73% of surveyed institutional decision-makers plan to increase their allocations to digital assets, viewing them as long-term strategic assets rather than short-term trades. This institutionalization is characterized by a demand for regulated, transparent access points:

Regulated Exchange-Traded Funds (ETFs) and Exchange-Traded Products (ETPs) have become the default institutional on-ramp for digital asset exposure. Furthermore, stablecoins are being integrated, serving as standard settlement and treasury workflows rather than simple trading convenience vehicles. This institutional demand is supporting the expansion of regulated staking products, such as staked Ethereum trusts, which combine asset exposure with yield-generating consensus rewards.

Microstructure and Temporal Inefficiencies: The 13F Blind Spot vs. Real-Time Flow Detection

While retail investors frequently attempt to replicate institutional positioning by analyzing quarterly SEC Form 13F filings, this approach is fundamentally flawed due to severe temporal lags. Form 13F disclosures operate on a regulatory delay of up to 135 days from the actual date of trade execution. This temporal gap leads to rapid alpha decay.

Empirical microstructural research demonstrates that newly initiated institutional long positions generate an average of 36 basis points of positive incremental alpha during the first month following execution. However, this informational edge decays rapidly with a half-life of approximately four months. By the time a 13F filing becomes publicly available on Day 135, the predictive value of the data has evaporated, meaning that retail replication strategies relying solely on 13F disclosures are trading on outdated information.

To bypass this lag, real-time institutional flow detection models have emerged. By processing tick-by-tick order book data, transaction sizes and microstructural execution patterns, these models categorize order flow into retail, market-making and institutional categories. Advanced real-time flow classifiers demonstrate a 65.5% directional prediction accuracy for subsequent 13F filing changes, compared to a non-predictive 48.8% hit rate for retail order flow, providing a real-time proxy of institutional accumulation and distribution.

Unusual Options Activity (UOA) serves as a powerful microstructural leading indicator. Academic and market assessments indicate that equities experiencing pronounced UOA are five times more likely to undergo a significant price change within the subsequent trading sessions.

The most powerful institutional signals are identified by filtering for deep out-of-the-money (OTM) contracts with short-dated expirations, where the daily trading volume exceeds the existing open interest. This signature reflects highly leveraged, speculative positioning backed by a near-term informational advantage or a specific catalyst.

Crucially, monitoring the ask-side of options transactions provides a cleaner directional signal than bid-side flow. When institutional buyers are willing to pay the immediate asking price or higher, they display aggressive price-taking behavior. A surge in ask-side call options indicates aggressive bullish sentiment, while a corresponding surge in ask-side puts signals aggressive bearish sentiment.

Dark Pools as Market Quality Disruptors and Price Crash Catalysts

To fully understand institutional execution, retail investors must analyze alternative trading systems (ATS), commonly referred to as dark pools.

Dark pools are private trading venues where financial institutions execute large block trades anonymously, shielding their orders from public order books to prevent high-frequency trading (HFT) algorithms from front-running their positions. Historically, dark pools have captured a massive share of equity volume, reaching 47.2% of total U.S. equity trading volume in early 2021.

While dark pools successfully reduce execution costs for institutional block trades, their expansion has introduced severe market quality distortions and systemic structural risks. Microstructure research indicates that dark pools are highly active for large-capitalization equities, whereas over-the-counter (OTC) internalization is more common for small-capitalization stocks. This bifurcation is driven by payment for order flow (PFOF) arrangements, which are highly profitable for internalizing market makers when bid-ask spreads are wide.

Crucially, the entry of high-frequency traders and retail order imbalances into dark pools has degraded their execution quality. When HFTs penetrate a dark pool, they increase information leakage, turning institutional investors into sitting ducks. Because substitution across dark pools is remarkably low (with European Double Volume Cap (DVC) restrictions showing that suspending a stock on a single dominant dark pool causes a 50% drop in overall dark trading due to sticky, siloed platform relationships) institutions are frequently unable to escape predatory trading by migrating venues.

Furthermore, a critical University of Missouri study has established a direct link between dark pool trading activity, accounting manipulation and sudden stock price crashes. Uninformed, liquidity-driven retail traders are systematically drawn to dark pools by narrower bid-ask spreads and lower transaction costs. This migration drains uninformed volume away from public lit exchanges. As the public order book loses this stabilizing, liquidity-providing volume, the transaction costs for informed traders on lit exchanges rise.

Consequently, the incentive for independent research analysts and short-sellers to acquire and trade on private information is weakened. With this public disciplining force diminished, corporate management faces less scrutiny, enabling them to withhold or hoard negative material information for extended periods.

To sustain this bad-news hoarding, executives at firms with high dark pool concentration are statistically more likely to perform unusual accounting adjustments, artificially inflating earnings to postpone public disclosure. When these accounting manipulations eventually reach their structural limits, the accumulated negative information is forced into the open, triggering sudden, catastrophic stock price crashes. Dark pool concentration is thus a key microstructural indicator of delayed negative catalysts and elevated tail risk.

Subscribe now

Unlock Member Access

In the next article…

I will move from the framework to two live case studies: Vistra and MicroStrategy. Both sit at the center of these institutional themes, but for very different reasons. Vistra is tied to the AI power and nuclear infrastructure trade, while MicroStrategy is a leveraged equity proxy for Bitcoin exposure. The point will not be to copy institutional positioning blindly, but to show how flows, options activity, valuation, balance sheet risk and microstructure can change the way a retail investor interprets these trades.

Sources

1. BlackRock, “Thematic Outlook: Charting Trends for Investors” — blackrock.com

2. iShares, “2026 Thematic Outlook” — ishares.com

3. Invesco, “2026 Midyear Investment Outlook” — invesco.com

4. Morgan Stanley, “Investment Outlook 2026: Key Themes Shaping Global Markets” — morganstanley.com

5. J.P. Morgan Private Bank, “The New Frontier: 3 Themes Driving Alternatives in 2026” — privatebank.jpmorgan.com

6. Tickmill, “Institutional Insights: Goldman Sachs US Flow Themes” — tickmill.com

7. Exponential Tech, “The 13F Blind Spot: Why Institutional Flow Data Arrives 135 Days Late” — exponential-tech.ai

8. Hightower Signature, “Weekly Wisdom: 2026 Outlook and Investment Themes” — hightowersignature.com

9. The Motley Fool, “4 Nuclear Energy Stocks to Buy and Hold for the Next Two Decades” — fool.com

10. EY, “Volatility Drives Discipline, Not Retreat” — ey.com

11. InsiderFinance, “Live Options Flow and Unusual Options Activity” — insiderfinance.io

12. GIS Reports, “Dark Pools as an Evolution of Financial Markets” — gisreportsonline.com

13. American Economic Association, “Competing for Dark Trades” — aeaweb.org

14. Bocconi University, “Diving Into Dark Pools” — didattica.unibocconi.it

15. University of Missouri, “Mizzou Study Links Dark Pool Trading Activity to Stock Price Crashes and Accounting Manipulation” — showme.missouri.edu

16. TIKR, “Constellation Energy Stock vs Vistra Stock: Which AI Power Play Has More Upside?” — tikr.com

17. Quiver Quantitative, “Vistra Corp. Stock Opinions on Nuclear Energy Momentum” — quiverquant.com

18. TradingKey, “MSTR Stock Price Prediction 2026–2030: Is Strategy the Ultimate Bitcoin Leveraged Play?” — tradingkey.com

19. Trefis, “Strategy (MSTR)” — trefis.com

20. OptionCharts, “MSTR: Strategy Inc. Option Overview” — optioncharts.io

21. Barchart, “Investors Buying Unusually Heavy Volume of Strategy, Inc. Call Options” — barchart.com

22. Market Chameleon, “MSTR Option Breakdown: Strategy Class A” — marketchameleon.com

23. StockGro, “Cumulative Volume Delta (CVD): Indicators and Strategy” — stockgro.club

24. University of Iowa, “Hedge Fund Replication: A Model Combination Approach” — biz.uiowa.edu

25. The Ritz Herald, “Cumulative Volume Delta Made Simple: A Trader’s Guide to Market Intent” — ritzherald.com

26. Scribd / LiteFinance, “CVD Indicator: Cumulative Volume Delta Trading Guide” — scribd.com

27. GoCharting, “Orderflow Analysis: Footprint Charts and Delta” — gocharting.com

28. Bookmap, “Cumulative Volume Delta Trading Strategy” — bookmap.com

29. The Hedge Fund Journal, “Replication of Hedge Fund Strategies” — thehedgefundjournal.com

30. SEI Canada, “Hedge Fund Replication and Liquid Alternatives” — assetmanagementca.seic.com

31. Scholars’ Mine, “Hedge Fund Replication Using a Strategy-Specific Modeling Approach” — scholarsmine.mst.edu

32. TraderHQ, “FlowAlgo Review: Real-Time Options Flow Tracking for Active Traders” — traderhq.com

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For further information read the full disclaimer here.

Portfolio Lab Base is the model portfolio, the one I would actually run. Portfolio Lab + 5% MSFT LEAPS Overlay sits next to it for comparison and optionality.

Backtest results

Most backtests you see are closer to marketing than analysis, as survivorship bias flatters anything built on U.S. markets, cherry-picked windows flatter almost anything depending on where the chart starts and a strategy mined from history for what worked will always look great on the history it was mined from. This portfolio gets held to the same bar. The window runs straight through 2022 instead of stopping short of it and the sleeve structure came from household-debt data, AI capex figures and correlation behavior first, checked against a backtest only afterward, the reverse of how a mined strategy gets built. The specific limits of what this backtest does and doesn't prove are covered further down.

For context, over the same window:

Base beat both benchmarks on terminal value and did it with a smaller drawdown and a far less damaging 2022 period. The overlay beat Base on terminal value too, but it bought that extra return with volatility and drawdown that put it in a different risk category.

The Portfolio at a Quick Glance

Note that “Old aggressive” in the following charts refers to an experimental version of the Portfolio which I later abandoned. Still, I decided to keep it in the charts as it was a useful starting point to improve upon.

What was tested

The Base portfolio ran through an historical backtest using monthly returns on listed instruments. The question it answers is narrow on purpose:

What would this allocation have done over the tested window, using instruments an investor could actually buy?

The numbers are nominal, pre-tax, in USD. Adjusted prices capture dividends and fund-level expense ratios where those apply. They do not capture: taxes, platform fees, FX conversion, withholding tax, bid-ask spreads, or the everyday friction of running a real account.

However, none of that makes the comparison meaningless. Bid-ask spreads and commissions are small for a portfolio built from liquid, large-cap instruments that get reviewed monthly and rebalanced a few times a year rather than traded daily, closer to a few basis points of drag than a few percentage points. Taxes are the one item on that list that can genuinely matter, and the one that is investor-specific rather than portfolio-specific: a U.S. taxable account, a tax-advantaged account and a non-U.S. account will each keep a different share of the same pre-tax return. For this reason I deliberately decided to not model them in (although that can quickly be done if any of my subscribers desire). Base, the overlay, and the benchmarks all face similar tax treatment, so the relative ranking still holds up.

The LEAPS sleeve was a bit harder to backtest, as no free source has historical options-chain data going back to 2020, thus the overlay is modeled instead: MSFT’s real price history run through Black-Scholes, rolling a two-year, 20%-out-of-the-money call. That rolling history is there to test the mechanism, whether this kind of structure earns its place across a real stretch of calm years, a rate shock and an AI melt-up, not to price a specific contract.

The live position, when it goes on, gets struck off Microsoft's actual price and actual options pricing on the day it's opened, not off a six-year average.

You can read my take on this investment thesis on my dedicated article (Why Institutions Are Buying Long-Dated Microsoft Call Options).

What will be tracked live

The Synthetic LEAPS Impact

The LEAPS overlay improves the headline return at the expense of more possible drawdown.

None of that is an argument against MSFT LEAPS, but it is an argument for sizing them honestly. A 5% options sleeve does not behave like 5% of a portfolio, because the return distribution on a rolling call looks nothing like the return distribution on the underlying asset. For this reason it stays optional, for more aggressive investors.

Portfolio construction logic

A stock list starts with individual ideas:

• This company is cheap

• That one has a moat

• This one could compound for a decade

All of that can be true and still tell you nothing about what happens when the ideas sit next to each other in a portfolio. A portfolio has to answer a different question: not “is this a good business” but “what happens to the returns if rates rise, if the consumer weakens, if AI capex slows, if stocks and bonds fall together for once?”

A semiconductor name, a data-center infrastructure name, a software company and a quality compounder can look like four unrelated ideas. In times of crisis however (a liquidity shock for example) they can become much more correlated, thus unexpectedly increasing the risk profile of the portfolio.

Start with the consumer data

The headline numbers look fine:

• Unemployment sits near 4.3%

• Retail sales are at a record

• Household net worth is near a 30-year high

If we look closely however, that is not the correct conclusion to reach:

• The personal saving rate has fallen from roughly 6% two years ago to about 4% now.

• The share of credit card balances 90-plus days delinquency is at 13.12%, the highest in fifteen years.

• Auto loan delinquencies are at records.

• Total household debt is closing in on $19 trillion.

The marginal consumer is carrying more debt and more delinquency than the aggregate numbers admit and a portfolio that only owns “the economy” through broad index exposure has no way to express that gap.

Consumer-Stress Resilience exists because of the distance between the headline data and the household-level data (You can read more detailed information on this in my article: Fragile Consumers in an Asset-Driven Economy)

Owning AI infrastructure differently

Data-center capex is running near $725 billion for 2026 and the binding constraint on that spending has moved (Power availability). Utilities are re-contracting for nuclear capacity, transformer lead times are stretched years out and grid interconnection queues are the limit on how fast compute gets built.

Data-Center bottlenecks is thus its own sleeve instead of getting merged into “more AI stocks.” Owning the chip and owning the thing the chip cannot run without are different investments with different failure points and most portfolios built around an AI theme only own the first one.

Before any of this reached a backtest, I tested correlation

Equity positions that look unrelated on paper, a semiconductor name, a payments network, a cybersecurity vendor, tend to move together at 0.7 to 0.85 in ordinary markets and that correlation gets increases in a crisis.

Nominal Treasuries, the usual equity hedge, run close to zero correlation with stocks in calm periods and swing well above 0.5 in a rate shock, which is the specific mechanism that broke plain 60/40 portfolios in 2022. Gold is the one position in the book that stays close to zero correlated with equities across most regimes.

Consequently, I built the sleeve structure around the data before picking a single ticker, then checking the picks against how they behave together before anything gets backtested.

The sleeve structure

The exact holdings and weights are for paid subscribers. However, at a high level, the portfolio does five jobs.

Research equity is the main return engine and mainly regards: AI infrastructure, the physical bottlenecks constraining it, financial infrastructure, consumer-stress resilience and mission-critical software.

The broad anchor is not there to add return, rather it exists so the portfolio survives being wrong about its own thesis, which is a different failure mode than being wrong about a single stock.

Gold, Treasuries, TIPS, and managed futures can fail for different reasons and earn their spot in different regimes:

• Gold is not a universal hedge

• Bonds break when real yields rise

• Trend-following can go sideways for a year at a stretch.

The 2022 Case

2022 is the most useful stress period in this backtest and not because it was the worst year on record.

• SPY lost 18.2%

• VT lost 18.0%

• Base lost 9.1%.

What made 2022 different is that it was not a normal equity drawdown.

Rates, inflation and duration all repriced at once, which is exactly the environment where stock-bond diversification stops working.

A portfolio built purely on picking good businesses does not have an answer for a discount-rate shock, as good companies get marked down alongside bad ones when the rate used to value all of them moves. Moreover, a portfolio whose only real hedge is duration finds out the hard way that duration can take the same hit as the equity book, at the same time.

Base still lost money that year and that is no surprise, given that it is an equity-led portfolio, not an hedge fund. What matters is the size of the loss relative to the benchmarks and the fact that the full-period return held up anyway. That is the trade this construction makes: give up a little upside in the good years for a lot less pain in the bad ones.

Base’s worst drawdown across the entire window was -17.4%, a full 8.1 percentage points lower than VT’s -25.5% and 6.5 points lower than SPY’s -23.9%.

That gap opened during the exact same nine-month decline, December 2021 to September 2022, that hit every portfolio in the comparison. The difference in outcome came entirely from what was sitting inside the portfolio when the shock hit, not from timing or luck.

The more useful number for a reader is recovery speed. Base took six months from its trough to reclaim its old high, whereas SPY and VT both took fifteen. A 60/40 mix took eighteen. Base spent roughly nine fewer months “underwater” than a plain index position coming out of the identical drawdown. That’s the kind of number that never shows up if we just look at CAGR and matters a lot to anyone who actually has to sit through the drawdown.

8. What the backtest does and does not prove

The backtest shows the Base portfolio worked well over this specific six-year window. It does not cover every kind of recession and it does not include a full 2008-style credit event for every name currently in the book. This is evidence, not proof. It is however enough evidence to publish the portfolio and start tracking it live.

The full allocation percentages, deeper position-level rationale, implementation notes and live tracking process are found below

The upcoming section covers the exact holdings, the weights, why each position, what would have to change for a position to be wrong, how the optional LEAPS overlay was modeled, implementation issues for non-U.S. readers and what gets reported going forward.

Read more

Contextual Overview

A long-dated OTM call option is defined as a contract extending several months to multiple years into the future (typically structured as a LEAP when exceeding twelve months) with a strike price set significantly above the current trading spot price of the underlying asset. Since the strike price exceeds the current spot price, these options possess zero intrinsic value and are priced entirely on the time premium and implied volatility.

Unlike short-dated speculative options that decay rapidly due to accelerating time decay (theta), long-dated contracts offer a highly convex, asymmetric payoff structure. As the underlying stock appreciates toward the OTM strike price, the sensitivity of the option's price to the stock's price (delta) expands non-linearly. This sensitivity acceleration, known as positive gamma, drives rapid option value appreciation, allowing the investor to capture exponential upside.

What drove this accumulation of MSFT OTM call options ?

Institutional accumulation is driven by several overlapping catalysts:

• Consumption-Based Revenue Transition: Microsoft is shifting its enterprise software framework from flat per-seat licensing to a hybrid "seats plus consumption" architecture. Under this system, corporate clients pay a base fee plus variable overages linked to the actual usage of autonomous AI agents. This pricing shift represents a significant structural catalyst, as agentic models are projected to become the dominant enterprise workload.

• Asymmetric Compute Demand: Global usage of generative AI (measured in text tokens) has expanded dramatically, cementing a macro environment where compute demand consistently outstrips available semiconductor and infrastructure supply.

• Technical Drawdown Valuation: Following a deep technical correction, MSFT equity has experienced a drawdown of roughly 31% to 33% from its previous peak, hitting a 52-week low of $349.20. This sell-off has compressed Microsoft's forward price-to-earnings (P/E) multiple to approximately 20x to 21.5x, well below its three-to-five-year historical average in the high-20s to 30x range. Institutional investors are using long-dated options to establish a structural position at a perceived valuation floor.

Evidence of Options Flow and Activity

Derivative market data reveals a notable surge in MSFT options activity, dominated by large-block institutional trades and sweeps rather than retail speculation. On the active trading session of July 1, 2026, total options trading volume on MSFT reached 1.17 million contracts. Of this volume, call options accounted for 76.74%, while put options represented only 23.26%, demonstrating a structurally bullish bias. Total outstanding open interest climbed to 4.52 million contracts, representing 104.46% of the 30-day historical average, confirming that market participants are actively initiating new, unhedged directional positions. A volume-to-open interest (V/OI) ratio exceeding 10x is a strong technical signal of institutional trade entry. The market has recently observed multiple high-volume tranches targeting specific OTM strike horizons.

These structured option transactions highlight two distinct market behaviors. There is a tactical, near-term speculation play targeting strikes roughly 3% to 10% OTM with 30-to-200-day expirations, designed to capture immediate price reversals. Simultaneously, there is systemic accumulation of ultra-long-dated contracts (such as the December 2028 low-$700s LEAPs). This indicates a highly focused distribution of strikes, reflecting strong institutional conviction that the equity is structurally mispriced.

The institutional investment community is sharply divided on Microsoft's intermediate outlook, creating a structural debate between analysts focusing on multi-year compounding free cash flow and those warning of near-term margin erosion.

The Bullish Structural Thesis

Long-Term Compounding and Technical Floor On the bullish end of the spectrum, notable institutional allocators have backed their conviction with long-dated derivatives.

Read more

The specifics differ between industries, but the underlying logic has remained remarkably consistent. Investors estimate the size of a company’s addressable market, evaluate its competitive position, project future cash flows and then decide whether today’s price offers an attractive return relative to the risks involved. Whether the company sells enterprise software, semiconductors or cloud infrastructure, the valuation ultimately comes back to the same question:

Companies such as Anthropic, OpenAI and xAI are not simply software vendors, nor are they infrastructure providers in the traditional sense. They occupy a new category, one that combines elements of software, cloud computing, semiconductor demand, scientific research and national strategic infrastructure. Their products improve continuously, their cost structures evolve almost monthly and their competitive advantage depends heavily on access to computational power and data. That makes valuing them exceptionally difficult.

Until now, private capital has largely avoided confronting this uncertainty. Venture capital firms, sovereign wealth funds and strategic investors continue to finance frontier AI laboratories at increasingly ambitious valuations, often justifying those prices through long-term strategic optionality rather than conventional financial metrics. As long as these companies remained private, there was little pressure to reconcile these optimistic narratives with the quarterly reporting and the public ownership aspect.

If Anthropic proceeds with its expected Nasdaq listing later this year, investors will have the opportunity to assign a public market valuation to one of the world’s leading frontier AI developers. The valuation eventually assigned to Anthropic will almost certainly influence how investors price every major AI company that follows, from OpenAI to Mistral, Cohere and beyond.

The Numbers

Following its latest funding round, Anthropic reached a reported private valuation of approximately $965 billion, placing it within touching distance of becoming the first technology company to debut on public markets with a trillion-dollar valuation.

At the same time, the company has reported an annual recurring revenue run rate that has expanded at a pace rarely seen in enterprise software, driven primarily by corporate adoption of its Claude family of language models and a rapidly growing developer ecosystem.

Viewed in isolation, those figures paint the picture of a business experiencing almost a lot of commercial momentum, yet momentum alone has never fully justified a valuation. Revenue has accelerated rapidly across the sector, but so have infrastructure costs. Frontier models require enormous capital expenditure, sustained access to advanced semiconductors and cloud infrastructure that few companies can build independently. At the same time, the competitive landscape changes with unusual speed as new model releases compress product differentiation that, only months earlier, appeared significant. Unlike traditional software businesses, where gross margins often improve naturally as scale increases, frontier AI companies must continually invest billions simply to remain technologically competitive. Every new generation of models demands more computational power, larger training datasets and increasingly sophisticated inference infrastructure.

The result is an industry where revenue growth and capital intensity are rising simultaneously, making it far less obvious how today’s impressive top-line expansion will translate into free cash flow over the next decade.

The company undoubtedly possesses attributes that public investors tend to reward. Enterprise customers generate the overwhelming majority of its revenue, reducing its dependence on consumer subscriptions and advertising. Strategic partnerships with major cloud providers have given Anthropic access to distribution channels and computing resources that would have been prohibitively expensive to replicate independently. Claude has established itself as one of the leading foundation models for enterprise applications, particularly in software development, financial services and knowledge work, while the rapid adoption of Claude Code suggests the company has begun building products that extend beyond the underlying model itself. These strengths help explain why investors have been willing to finance Anthropic at increasingly ambitious valuations.

They do not, however, eliminate the questions that public markets will eventually ask:

• How durable is the company’s pricing power as frontier models become increasingly commoditised?

• Can revenue continue to grow faster than inference costs?

• What proportion of reported recurring revenue reflects recurring enterprise demand rather than unusually large commercial agreements signed during a competitive period for AI adoption?

• How much of today’s valuation depends on assumptions that extend many years into the future?

For investors, that is ultimately what makes this IPO so significant.

Before attempting to answer that question, however, it is worth understanding what Anthropic has actually built, how its business differs from many of its competitors and why those differences may prove to be important.

What Does Anthropic Actually Sell?

Anthropic is increasingly an enterprise infrastructure company.

Claude may be the product most users recognise, but from a financial perspective it functions more as an access point than as the business itself. The real commercial engine lies beneath: API usage, enterprise deployments, coding tools and long-term contracts that integrate Anthropic’s models directly into corporate workflows.

Consumer adoption can be spectacularly fast, but it is also notoriously difficult to monetise. Millions of users trying a chatbot does not necessarily translate into durable cash flows, particularly when serving those users requires expensive inference infrastructure. Enterprise customers, by contrast, purchase AI for productivity gains. They integrate models into software development pipelines, automate customer service, analyse internal documents or build entirely new applications on top of foundation models. Once those systems become embedded in daily operations, switching providers becomes considerably more costly.

The result is a revenue base that tends to be larger, more predictable and, at least in theory, more defensible. That strategic focus has shaped almost every major decision Anthropic has made over the past two years. While OpenAI became synonymous with consumer AI through ChatGPT, Anthropic concentrated on large organisations. Financial institutions, consulting firms, software developers and multinational corporations became its primary audience, mainly because they offered substantially higher lifetime value.

This approach also created a different competitive dynamic, as winning an enterprise contract is not about producing the model that scores highest on the latest benchmark. Procurement teams care about reliability, security, compliance, integration and cost just as much as raw model capability. A marginal improvement in reasoning performance is unlikely to justify replacing an AI system already embedded across thousands of employees unless the economic benefits clearly outweigh the migration costs. In other words, enterprise AI is a relationship business as much as it is a technology business. That is one reason Anthropic’s partnerships deserve far more attention than its model releases.

The company has spent considerable effort embedding Claude into the ecosystems where enterprises already operate rather than forcing customers to adopt entirely new workflows. Its relationships with major cloud providers have expanded distribution while reducing friction for corporate buyers that already rely on those platforms. Instead of convincing a Fortune 500 company to overhaul its technology stack, Anthropic can often meet customers where they already are. This distribution strategy resembles the evolution of enterprise software over the past two decades. The most successful software companies built products that integrated seamlessly with existing infrastructure, lowered implementation costs and became increasingly difficult to replace over time.

AI appears to be following a similar trajectory and that becomes particularly evident when looking at Claude Code. At first glance, coding assistants might appear to be simply another application built on top of a language model. Financially, however, they represent something much more attractive. Software engineers interact with these tools continuously throughout the workday and every generated function, code review, documentation request or debugging session increases model utilisation. Unlike occasional chatbot conversations, programming workflows produce sustained, high-frequency demand, creating a level of engagement that is considerably easier to monetise.

More importantly, successful coding assistants often become deeply integrated into development environments. As teams begin relying on AI-generated documentation, automated testing and code suggestions across large repositories, replacing one provider with another becomes increasingly disruptive. This is very interesting for investors, as this produces more resilient revenue than products dependent on occasional discretionary use.

However, coding assistants still remain computationally intensive, particularly when processing large repositories or performing complex reasoning tasks. Revenue can increase rapidly while infrastructure costs rise almost as quickly, particularly if customers expect ever more capable models without corresponding price increases. Revenue growth undoubtedly matters, as few companies have expanded their commercial footprint as rapidly as Anthropic reportedly has. Yet, investors ultimately care about operating leverage, thus the ability for profits to grow faster than revenue over time.

We therefore return to the computation intensity issue:

• Every interaction with a language model consumes computational resources.

• Larger models generally require more expensive hardware.

• More sophisticated reasoning often increases inference time.

• As customers demand better performance, infrastructure requirements tend to rise alongside usage rather than remaining largely fixed.

AI companies occupy an unusual position somewhere between software businesses and utilities: Like software companies, they generate value through intellectual property, algorithms and recurring subscriptions. Like utilities, however, they must continually invest enormous sums simply to deliver each additional unit of output.

That structure makes conventional valuation multiples considerably harder to interpret. A software company trading at twenty times recurring revenue might appear expensive but entirely reasonable if investors expect gross margins to approach 85% over time. The same multiple becomes much harder to justify if serving future demand requires tens of billions of dollars in additional infrastructure spending.

This is where Anthropic’s strategic partnerships become central to the investment case. Rather than attempting to build every component of its infrastructure independently, the company has relied heavily on alliances with hyperscale cloud providers. These partnerships provide access to computing capacity that would otherwise require years of construction and extraordinary capital expenditure.

From one perspective, this is an obvious competitive advantage, as Anthropic can scale considerably faster than would be possible on a standalone basis while preserving financial flexibility during an industry characterised by extraordinary infrastructure demand.

From another perspective, however, it raises a different question. How much of Anthropic’s competitive position ultimately belongs to Anthropic itself? If access to compute depends heavily on commercial agreements with a handful of technology giants, then we are also investing in the durability of relationships with some of the world’s largest and most influential infrastructure providers. That dependency is neither inherently positive nor inherently negative, but it does illustrate a broader reality that distinguishes frontier AI from previous generations of software businesses. The competitive moat is no longer defined solely by code or intellectual property. Increasingly, it is shaped by access: access to semiconductors, access to energy, access to data centres, access to cloud infrastructure and access to the capital required to finance all of the above.

The valuation argument becomes much clearer once we examine the numbers underpinning Anthropic’s valuation and whether a business approaching a trillion-dollar market capitalisation can realistically generate the financial returns such a valuation implies.

What Does a $965 Billion Valuation Actually Assume?

The latest Series H funding round valued the company at approximately $965 billion, making it the highest-valued privately funded AI company in the United States and placing it ahead of OpenAI’s most recent private valuation. More importantly, it established a reference point for investors. That is noteworthy as private market valuations and public market valuations answer different questions.

A private funding round reflects the price that a relatively small group of investors is willing to pay under negotiated terms, often with limited liquidity and information rights that public shareholders will never receive. An IPO removes those protections, as every institutional investor, mutual fund and retail shareholder evaluates the same financial statements under the same disclosure regime.

One of the more unusual aspects of Anthropic’s upcoming IPO is that there is remarkably little disagreement about the company’s growth trajectory. Enterprise adoption has been exceptional and revenue has expanded at a pace rarely seen in software, while Claude Code has quickly established itself as one of the dominant AI coding assistants. However, the debate begins once investors try to convert that commercial success into a market value.

Unlike a mature software business, Anthropic cannot be valued by simply extrapolating current earnings or applying a historical multiple to free cash flow. The company is still investing aggressively, the industry itself remains in its infancy and many of the variables that will ultimately determine long-term profitability, from inference costs to hardware efficiency, are still evolving. As a result, the market has largely converged on a different approach: valuing Anthropic based on the quality and durability of its future economics rather than its current financial statements. That distinction explains why the latest funding round attracted as much attention as it did.

The $965 billion valuation implies a forward revenue multiple of roughly 20x based on Anthropic’s reported annualised revenue run rate of approximately $47 billion. By itself, that figure says relatively little, as high-growth software companies have traded at similar multiples before. We should instead ask ourselves whether Anthropic eventually develops the characteristics that have historically justified those valuations. This is also where the discussion among professionals starts to diverge.

For Harrison Rolfes, Senior Late-Stage Company Research Analyst at PitchBook, the most important number in the entire IPO process is neither the valuation nor the revenue run rate, it is gross margin. His argument is that the S-1 will reveal, for the first time, whether Anthropic’s extraordinary commercial growth is beginning to translate into meaningful operating leverage. In his view, gross margin is the metric that will either validate or undermine the narrative private markets have been pricing for the past three years.

Other market participants are looking somewhere else entirely. Brian Mulberry, Chief Market Strategist at Zacks Investment Management, has argued that the central valuation question is not how quickly Anthropic can grow, but when that growth begins producing consistent cash flow and, eventually, earnings per share. Enterprise demand for Claude is already well established, but the uncertainty lies in the conversion of that demand into durable shareholder returns, particularly given the scale of infrastructure investment still required across the industry.

There is another aspect of the valuation that receives far less attention, yet could prove equally important over time. Reuters Breakingviews recently argued that this could become one of Anthropic’s greatest strategic advantages once it reaches public markets. If its equity continues to trade at a premium while specialised AI infrastructure providers command materially lower revenue multiples, Anthropic could use stock rather than cash to accelerate its access to compute through acquisitions. In an industry where securing GPUs, data centres and cloud capacity has become one of the principal competitive constraints, a richly valued equity can be as strategically valuable as cash on the balance sheet. How management deploys equity, whether through acquisitions, infrastructure investment or dilution, will become part of the investment case rather than simply a financing decision.

Ultimately, Anthropic’s IPO is unlikely to settle the valuation debate. It will, however, change its nature. Until now, discussions around Anthropic’s worth have been driven largely by private funding rounds and headline revenue figures. Once the company enters public markets, the conversation will shift towards audited financial statements, operating leverage, capital allocation and cash generation.

What to Look For in the S-1

An S-1 is designed to disclose the financial and operational risks that accompany becoming a public company. For Anthropic, the prospectus will almost certainly receive more scrutiny than the valuation itself.

Several areas deserve particular attention:

The first is revenue quality.

Anthropic’s reported annual recurring revenue has expanded at an extraordinary pace, yet ARR is not defined under GAAP and companies calculate it differently. The filing should provide a much clearer breakdown of recognised revenue, deferred revenue and the extent to which growth reflects recurring enterprise usage rather than unusually large commercial agreements signed during the recent surge in AI adoption.

The second area is gross margin.

As already mentioned, Harrison Rolfes recently described gross margin as the single most important figure likely to emerge from Anthropic’s IPO, arguing that it will provide the clearest indication of whether the company’s commercial momentum is beginning to translate into operating leverage. That is particularly relevant for frontier AI, where serving customers requires substantial ongoing expenditure on inference rather than simply recovering fixed development costs.

Cash flow deserves similar attention.

High-growth technology companies frequently prioritise growth over near-term profitability, and there are often sound economic reasons for doing so. Eventually, however, investors expect that growth to produce durable operating cash flow. How quickly Anthropic reaches that point, and whether it can sustain it as infrastructure spending continues to rise, may prove just as important as another update to annualised revenue.

Customer concentration and contractual obligations.

Anthropic’s commercial strategy has centred on large enterprise deployments and deep partnerships with hyperscale cloud providers. Those relationships have clearly accelerated adoption, although they also create dependencies that become increasingly important once quarterly reporting begins. The prospectus should provide much greater clarity regarding how diversified the company’s customer base has become, as well as the scale of its long-term infrastructure commitments.

None of these disclosures is likely to generate the same headlines as a trillion-dollar valuation. Collectively, however, they will determine whether investors continue valuing Anthropic primarily as a high-growth AI story or begin analysing it as a mature public company.

Conclusion

Anthropic’s IPO is about far more than one company’s valuation.

It will be the first real test of how public markets value such an important AI business once the excitement of private funding rounds gives way to quarterly earnings, audited financial statements and public ownership.

For years, investors have largely debated what artificial intelligence could become. Anthropic’s listing marks the point where the conversation begins shifting towards what these businesses can actually earn, how efficiently they can grow and whether today’s expectations can translate into durable shareholder returns.

Whether the IPO is ultimately viewed as a success will depend less on its opening valuation and more on what the first few years as a public company show us.

Subscribe now

Sources

Anthropic:“Series H Funding Announcement.”
https://www.anthropic.com/news/series-h

“Anthropic Confidentially Submits Draft Registration Statement for Proposed Initial Public Offering.”
https://www.anthropic.com/news/confidential-submission-draft-registration-statement

Reuters: “AI giant Anthropic confidentially files for U.S. IPO.” (June 1, 2026)
https://www.reuters.com/business/ai-giant-anthropic-confidentially-files-us-ipo-2026-06-01/

“Anthropic raises $65 billion, now valued at $965 billion.” (May 28, 2026)
https://www.reuters.com/business/anthropic-raises-65-billion-now-valued-965-billion-2026-05-28/

Reuters Breakingviews: “Anthropic IPO could train a large M&A model.” (June 2, 2026)
https://www.reuters.com/commentary/breakingviews/anthropic-ipo-could-train-large-ma-model-2026-06-02/

“VIEW: Anthropic IPO filing ratifies Wall Street’s AI obsession.” (June 1, 2026)
https://www.reuters.com/legal/transactional/view-anthropic-ipo-filing-ratifies-wall-streets-ai-obsession-2026-06-01/

PitchBook: Market commentary on Anthropic’s IPO (quoted by Reuters).

Zacks Investment Management: Market commentary on Anthropic’s IPO (quoted by Reuters).

It was indispensable for agriculture, manufacturing, electricity generation and human life itself, yet unlike oil, copper or natural gas, it rarely existed as an investable asset. Investors seeking exposure to the theme generally did so indirectly through utilities, treatment companies, irrigation equipment or infrastructure rather than through water itself.

However, nowadays water is not becoming a commodity in the traditional sense, nor is the United States selling its freshwater reserves to private investors as countless social media posts now claim. Nevertheless, water is increasingly being valued, financed and allocated through financial markets in ways that would have seemed unusual only a decade ago and AI has accelerated that process.

The rapid expansion of data centers has introduced one of the fastest growing industrial sources of water demand, particularly in regions already facing chronic water stress. At the same time, institutional investors have increased their exposure to water infrastructure, water rights have appreciated across several western states, governments are reconsidering how scarce water should be allocated and financial institutions have begun discussing the tokenisation of water rights as a potential solution to long-standing inefficiencies in existing markets.

Goldman Sachs recently described water as one of the world’s most underinvested forms of economic infrastructure, arguing that tokenization could eventually improve both water markets and infrastructure financing. Water is thus becoming an economic asset whose scarcity is beginning to influence investment decisions, infrastructure planning and capital allocation in ways that extend far beyond agriculture.

Water has never been free

Water is often described as a public resource, although from an economic perspective that statement has always been incomplete.

Supplying clean water requires reservoirs, pipelines, treatment facilities, pumping stations and distribution networks, all of which require substantial capital investment. Even where water itself is publicly owned, access to it depends on infrastructure whose construction and maintenance involve economic costs. The resource therefore has always possessed value and what changed over the past century was not the value of water itself but rather the assumption that developed economies would almost always have enough of it. That assumption is becoming increasingly difficult to maintain.

Population growth continues to increase municipal demand while industrial consumption has expanded alongside semiconductor manufacturing, battery production and artificial intelligence. Climate change has simultaneously reduced the reliability of water supplies across many regions, particularly throughout the American Southwest where prolonged drought has transformed water from an abundant input into a strategic constraint.

Water rights, not water itself

Investors are purchasing legal rights governing access to water and those rights differ substantially across jurisdictions.

Much of the eastern United States follows riparian principles, under which access depends largely on ownership of land adjacent to a water source. Much of the western United States instead follows prior appropriation, a doctrine developed during the nineteenth century whereby priority depends on who first put the water to beneficial use rather than who owns the surrounding land. This therefore determines who receives water during periods of scarcity.

Senior rights holders generally receive their allocation before junior users, meaning that identical parcels of land can differ substantially in value simply because one possesses older water rights. Consequently, farmland in parts of Arizona, Colorado and California frequently derives a significant proportion of its economic value from the security of its water allocation rather than from agricultural productivity alone.

This helps explain one of the most frequently quoted observations attributed to Michael Burry.

When asked how to invest in water, Burry reportedly concluded that the most effective approach was not purchasing water itself but buying farmland with reliable water rights. Scarcity increases the value of whichever asset controls access to the scarce resource. In western American agriculture, that asset is frequently the water right rather than the land itself.

Institutional investors, pension funds and agricultural investment vehicles have all expanded their exposure to farmland in water secure regions, partly because access to reliable irrigation has become progressively more valuable as drought conditions intensified.

Artificial intelligence introduces a new source of demand

For decades, discussions surrounding water scarcity focused primarily on agriculture, and while agriculture continues to account for roughly seventy percent of global freshwater withdrawals, we now have data centres. Training large language models requires extraordinary computing capacity, which generates heat that must be reduced through evaporative cooling systems that consume substantial volumes of freshwater every day. The exact amount varies according to climate, cooling technology and workload intensity, although the largest facilities can consume millions of gallons daily. More importantly, aggregate demand is increasing extraordinarily quickly as AI infrastructure expands.

Goldman Sachs estimates that global data centre water consumption could roughly double by 2030 while the International Energy Agency has similarly identified AI infrastructure as one of the fastest growing industrial sources of water demand.In addition, many of the regions attracting data centres development already experience persistent water stress. Cheap land, favourable tax incentives and available electricity have encouraged companies to concentrate investment throughout states such as Arizona, Texas and Utah, all of which face increasingly difficult water management challenges.

Recent academic work estimates that, if current trends continue, American data centres could require between roughly 700 million and 1.45 billion gallons of additional daily water capacity by 2030, comparable to the average daily water supply of New York City. Even under optimistic assumptions regarding efficiency improvements, demand remains substantial enough to create local capacity constraints across many host communities. Municipal infrastructure capable of supplying residential neighbourhoods may not possess sufficient spare capacity to support hyperscale computing facilities during periods of extreme summer heat. The consequence is that water itself begins determining where digital infrastructure can realistically expand. Freshwater therefore increasingly joins transmission capacity, semiconductor availability and permitting as a critical input into artificial intelligence.

That relationship has important implications extending well beyond technology companies. As water becomes more economically valuable, financial markets inevitably begin searching for mechanisms capable of pricing, financing and allocating it more efficiently.

USWR on Coinbase

United States Water Reserve (USWR) is listed on Coinbase.

At first glance, it sounded plausible that the United States had created some form of strategic water reserve or that water itself had begun to move onto the blockchain. Given everything discussed so far, namely growing water scarcity, rising AI demand and increasing institutional interest in water infrastructure, the idea did not seem entirely unreasonable.

However, despite its name, USWR is not backed by federal water reserves, does not represent ownership of legally recognised water rights and has no connection with the US government. Coinbase itself makes this distinction clear, as the asset appears on the platform with market data, although it is not listed as a standard Coinbase-traded asset and the pricing information is sourced from third parties.

Discussions around tokenisation are no longer confined to cryptocurrency projects. Goldman Sachs recently argued that blockchain technology could eventually improve water markets by digitising water rights, increasing market liquidity and creating new financing mechanisms for infrastructure. We, as already said, need to be careful about what is actually being tokenised: legal rights associated with using those resources and the financial claims attached to future infrastructure investment. Those are two very different things.

Owning a token called United States Water Reserve does not provide exposure to American freshwater in the same way that owning a gold-backed ETF provides exposure to gold. The token is simply a cryptocurrency whose value depends on whether other investors are willing to buy it. The underlying investment thesis, however, is considerably more difficult to dismiss with water becoming more and more valuable. That process was underway long before USWR appeared and it will almost certainly continue regardless of whether the token survives.

Conclusion

Water is unlikely to become an asset class in the same way as oil or gold. Legal frameworks, environmental considerations and the fact that access to clean water is widely regarded as a public good place natural limits on how far financialisation can go. Even so, it is becoming increasingly difficult to ignore the direction of travel.

Artificial intelligence has increased demand, climate change has constrained supply and decades of underinvestment have made water infrastructure a growing priority for both governments and private capital. As a result, investors are beginning to pay attention because its scarcity is becoming economically more visible.

USWR is therefore not evidence that the United States has tokenised its freshwater reserves. If anything, it demonstrates how quickly financial markets can attach themselves to a credible macroeconomic narrative. The token may ultimately prove insignificant, although the forces that made its story believable almost certainly will not.

Subscribe now

Unlock valuation Tools

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For further information read the full disclaimer here.

CrowdStrike was a good example of that. The company is outstanding, but the valuation was already asking for years of near-perfect execution and very little multiple compression. The business quality was clear, while the expected return was much less convincing. Arista Networks is a different kind of setup.

It is still expensive in the normal sense, as nobody looking at a stock trading around 46x forward earnings and 17x forward sales should pretend there is an obvious bargain here. However, the valuation looks more connected to the fundamentals than it did in the CrowdStrike case, because Arista has a cleaner cash-flow profile, a stronger balance sheet, very high margins and direct exposure to one of the most important physical bottlenecks in the AI buildout: networking.

AI does not only need GPUs, it also needs those GPUs to communicate with each other at enormous speed, with low latency and very little failure. If the network cannot keep up, the expensive chips sit idle for longer than they should. That is where Arista becomes interesting. The company is not the most hyped name in the AI stack, and that is part of the appeal. Arista Networks is basically selling part of the infrastructure that helps the chips work together.

All figures below are based on data available as of the June 26, 2026 market close.

The setup

Arista closed at $165.45 on June 26, giving the company a market value of roughly $208 billion and an enterprise value close to $196 billion. The balance sheet is one of the first reasons the case deserves attention: Arista holds about $12.35 billion of cash and marketable securities, with no debt. That is noteworthy because this is not a company funding its growth through leverage or constant capital raises. It is already very profitable, converts operating income into free cash flow, and has enough cash to keep investing through a cycle.

At a lower price, the base-case return looked more comfortably double-digit. After the latest run, the expected return is still interesting, but the margin of safety is thinner. This is not a “cheap AI stock” article, instead it is a case study on whether a high-quality AI infrastructure business can still offer a reasonable five-year return when the starting valuation is high, but not completely detached from cash flow.

What Arista actually does

Arista designs high-performance networking equipment and software for large data centers, cloud providers and high-performance computing environments.

That sounds technical, but the business is easier to understand if you start from the problem it solves. Large cloud and AI data centers need thousands, and sometimes tens of thousands, of servers and accelerators to communicate with each other reliably. As AI clusters get larger, networking becomes more important because poor data movement can reduce the efficiency of the entire system.

Arista’s core strength is not just the hardware. The company built its model around EOS, its Extensible Operating System, which gives customers a consistent software layer across its switches. That consistency is important for hyperscalers because it reduces operational complexity, improves automation and makes the network easier to manage at scale.

Here the moat starts to show up, as an a hyperscaler that builds part of its data-center fabric around Arista is not simply buying a replaceable box. Its engineers learn the software, its automation tools are built around the system and replacing that architecture can mean months of planning, testing and execution risk. That does not make Arista immune to competition, but it does make the relationship stickier than a standard hardware sale.

The AI angle has become more important because Ethernet is gaining attention as a networking standard for large AI clusters. Nvidia still has a powerful position through its own networking stack, especially where networking is bundled with GPUs, but many hyperscalers do not want to become completely dependent on one vendor. Arista’s open Ethernet approach gives them another path.

That is the investment case in simple terms: if AI infrastructure spending keeps growing, and if open Ethernet keeps taking share in large-scale AI networking, Arista has a credible way to remain one of the key beneficiaries.

The business quality is hard to ignore

The financial profile is unusually strong for a company exposed to hardware cycles.

In FY2025, Arista generated $9.0 billion of revenue, up 28.6% year over year and in Q1 2026, revenue grew 35.1% to $2.71 billion. Margins are also extremely high: GAAP operating margin was roughly 42.8% in FY2025 and 42.7% in Q1 2026, while non-GAAP operating margin was close to 48%.

Free cash flow is the other important aspect. Arista generated roughly $4.25 billion of free cash flow in FY2025 and about $1.64 billion in Q1 2026 alone. Stock-based compensation exists, but it is much less aggressive than what you often see in high-growth software. In FY2025, SBC was $439 million, compared with more than $4 billion of free cash flow. With a company like CrowdStrike for example, the reported free cash flow looked strong, but stock-based compensation absorbed a large part of the owner-earnings picture. With Arista, the cash flow is cleaner, as the gap between reported free cash flow and shareholder economics is much less severe.

The company is also asset-light. Normal capex is relatively low, although FY2026 should be modeled higher because of the Santa Clara building project. Even with that temporary capex increase, the free cash flow profile remains strong.

This is one reason the valuation can be discussed more constructively. A high multiple is much easier to justify when the company already produces real cash at scale.

The big risk is customer concentration

The main weakness is also clear.

Arista is highly dependent on a small number of very large customers. In FY2025, Microsoft represented roughly 26% of revenue and Meta represented roughly 16%. Together, those two customers accounted for about 42% of sales. That concentration is central to the risk profile. The positive interpretation is that Arista is deeply embedded with some of the most important AI and cloud infrastructure buyers in the world. If Microsoft and Meta keep spending aggressively on AI clusters and data-center upgrades, Arista benefits directly.

The negative interpretation is that any pause, digestion period, architecture shift or pricing pressure from those customers would matter a lot. Hyperscaler capex can move in waves and suppliers often feel those waves before the broader market notices.

There is also competitive pressure from Nvidia. According to the IDC Ethernet switch data, Nvidia’s Spectrum-X has recently gained meaningful share in the data-center Ethernet switch segment, narrowly ahead of Arista in that specific category. That does not break the Arista thesis, but it does make the competitive debate more serious. Nvidia has the advantage of bundling networking with GPUs. Arista’s counter is that hyperscalers often prefer open systems and do not want the entire AI infrastructure stack controlled by one vendor.

What the market is already pricing

The stock trades at roughly 46x forward earnings, about 17x forward sales and around 43x forward free cash flow. Those are rich multiples, especially for a company with hardware exposure. The reason the valuation is still worth discussing is that Arista’s growth, margins and free cash flow conversion are also far above average.

The reverse-DCF exercise helps make this more concrete: Using today’s enterprise value of roughly $196 billion and FY2025 free cash flow of about $4.25 billion, the market appears to be pricing in roughly 16–18% annual free cash flow growth over the next five years, assuming a 30x terminal free cash flow multiple and a 10–11% discount rate. That is demanding, but not absurd. It is above what a normal mature infrastructure company could deliver, but it is below Arista’s recent growth profile and not completely out of line with the company’s expected revenue trajectory. The issue is that the valuation becomes much less forgiving if the terminal multiple falls below 30x or if hyperscaler growth slows earlier than expected.

This is the difference when compared to the CrowdStrike case study: CrowdStrike’s valuation required a much richer terminal multiple to support modest forward returns. Arista still needs strong execution, but the bar is more reasonable because the starting free cash flow base is larger, cleaner and already very profitable.

The five-year return math

The table below is the core of the investment case. It starts from the current price and asks what kind of five-year return Arista could produce under different assumptions.

The base case is the most useful scenario because it does not require crazy assumptions, as it assumes Arista compounds revenue around 20% through 2030, maintains a 44% free cash flow margin and eventually trades at 30x free cash flow. That is still a premium multiple, but it is not an extreme one for a company with this margin structure, balance sheet and AI networking exposure.

Under that base case, the implied annualized return is about 9.7%. More importantly, it is supported by business performance rather than by assuming the market pays a hyper-growth multiple forever.

The bear case is also relatively important. If hyperscaler capex slows, Nvidia captures more of the networking stack, growth falls toward 12% and the terminal multiple compresses to 20x free cash flow, the five-year return turns slightly negative. That is not attractive, but it is not the same kind of downside profile as a company whose valuation only works under perfect assumptions.

The bull case is where the AI infrastructure upside shows up. If Ethernet wins more of the scale-out networking opportunity, Arista keeps growing above 20%, margins remain strong and buybacks reduce share count, the expected return can move much higher.

The stock has upside, but it is not free upside. It depends on Arista continuing to prove that open Ethernet networking will remain a critical part of the AI infrastructure stack.

Why the entry price should be noted

I would not describe Arista as obviously cheap today. The setup is better than many AI-linked names, but the stock has already moved enough that the margin of safety is not wide.

A pullback toward the $140 area would make the case materially more attractive. At that level, the base-case expected return would move closer to the 11–12% range, assuming the operating thesis remains intact. That would give investors more compensation for the customer concentration and competitive risks.

At the current price, Arista is a high-quality watchlist or Portfolio Lab candidate where the valuation is still reasonable enough to keep working on, but not low enough to ignore the risks.

What would make the thesis work

The first thing to watch is hyperscaler capex.

Arista’s customers are spending heavily because AI clusters require larger, faster and more complex networks. If Microsoft, Meta and other large cloud customers continue building aggressively, Arista has a strong demand backdrop. The company’s non-cancellable purchase commitments, which rose sharply in the latest filings, also suggest management is preparing for sustained demand rather than a short-lived spike.

The second thing to watch is Ethernet’s role in AI networking.

If open Ethernet standards keep gaining traction, Arista’s addressable market expands. If Nvidia’s vertically integrated networking bundle becomes the default choice for more AI clusters, Arista’s growth path becomes less certain.

The third factor is margins.

Arista has managed to operate with software-like profitability despite selling physical infrastructure. If gross margins stabilize in the low-to-mid 60s and operating margins remain in the mid-to-high 40s, the free cash flow case stays strong.

The fourth is customer diversification.

Microsoft and Meta are valuable customers, but the concentration risk is real. A third large hyperscaler, continued enterprise campus growth or stronger adoption outside the two biggest accounts would improve the quality of the growth.

Finally, capital allocation matters.

Arista has the balance sheet to buy back stock, invest internally and potentially pursue selective acquisitions. The model assumes modest share-count reduction over time and if free cash flow keeps compounding and buybacks are done at reasonable prices, per-share value can grow faster.

Portfolio Lab view

Arista is more interesting than a company like CrowdStrike from a five-year valuation perspective.

The business quality is high, the balance sheet is clean, free cash flow conversion is strong and the reverse-DCF bar is demanding but not unreasonable. Unlike CrowdStrike, the base case does not require the market to maintain an extremely rich multiple just to avoid disappointing returns.

That said, the stock is not cheap enough to be treated casually.

At $165.45, the base-case return is around 9.7% annualized, which is attractive but no longer comfortably above the double-digit line. The setup would become much more attractive closer to $140, where the same assumptions would provide a wider margin of safety.

It is not a deep-value opportunity and it still carries meaningful hyperscaler concentration risk. However, compared with many AI infrastructure names, the valuation is supported by actual cash flow, actual margins and a real five-year return framework.

Valuation Tools view

Arista is a good example of why valuation work should not stop at “AI beneficiary” or “expensive stock.” Both labels are true, but neither is very useful on its own.

The useful work is measuring what the current price already requires. In Arista’s case, the market is pricing in roughly 16–18% annual free cash flow growth over five years if the company exits at around 30x free cash flow. That is not a low bar, but it is much more reasonable than the valuation setup in many software and AI-adjacent stocks.

Final view

Arista is not cheap, but it is one of the cleaner valuation setups inside the AI infrastructure theme.

The company has a strong moat through EOS, deep hyperscaler relationships, very high operating margins, strong free cash flow conversion and a debt-free balance sheet with more than $12 billion of net cash. The main risk is concentration. Microsoft and Meta represent more than 40% of revenue, and Nvidia is becoming a more serious competitor in AI networking. The stock also needs a premium multiple to hold for the base-case return to work.

Still, the five-year math is reasonable. At the current price, the base case points to roughly a 9–10% annualized return. A pullback toward $140 would make the expected return much more compelling, while continued evidence of strong AI networking demand, stable margins and broader customer diversification would strengthen the case even at today’s price.

Subscribe now

Unlock Access to Valuation Tools

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For further information read the full disclaimer here.

The business is excellent:

• Revenue is still growing above 20%

• Free Cash Flow conversion is strong,

• Gross retention remains high

• Balance sheet carries more than $4.5 billion of cash with no debt.

The Falcon platform has also moved well beyond endpoint security, expanding into identity, cloud security, SIEM, browser security and AI-related workloads, giving investors a real business to underwrite.

The issue is the price

At roughly $701 per share before the 4-for-1 split, CrowdStrike has a market value of about $178 billion and an enterprise value close to $174 billion. That price implies that investors are paying for one of the best software businesses in the market to remain one of the best for a long time. That is noteworthy, as a great company can still become a difficult investment when too much future success is already reflected in the share price.

All figures below are pre-split. The 4-for-1 split changes the share count and share price mechanically, but it does not change the valuation.

Why this setup matters now

CrowdStrike has recovered from the July 2024 outage better than many investors expected. Customers have largely stayed, growth has continued and the regulatory risk around the Delta-related disruption has faded after the U.S. Department of Transportation closed its investigation without penalties. As a result, the market has moved back toward valuing CrowdStrike as a premium platform company rather than treating the outage as a permanent impairment to the business.

That is what makes the current setup interesting, making CrowdStrike a high-quality company where the market already believes a large part of the bull case.

CrowdStrike generated $4.81 billion of revenue in FY26, up 21.7% year over year. Q1 FY27 revenue grew 26%, while FY26 free cash flow reached $1.24 billion, equal to a 25.8% free cash flow margin. In Q1 FY27, free cash flow margin improved further to 33.8%. Those numbers are strong enough to justify a premium.

The valuation, however, is pricing something even more impressive.

A valuation like this does not leave much room for “pretty good.” It requires CrowdStrike to keep delivering numbers that are unusually strong even by software standards.

The bull case seems to be holding up

The reason CrowdStrike trades this way is not mysterious, given that the company has built one of the strongest platforms in cybersecurity and the operating data still supports the premium-business argument.

Falcon has become a broader security platform across endpoint protection, cloud security, identity, SIEM, threat intelligence and newer AI-security use cases. Since customers can keep adding modules inside the same platform, CrowdStrike has a natural path to grow revenue from existing accounts without having to win each dollar from scratch. That is already visible in the module adoption data, as the share of customers using six or more modules increased from 48% to 51%. Customers using seven or more modules moved from 32% to 35%, while customers using eight or more modules increased from 21% to 25%.

Those numbers are important because each additional module makes the platform more embedded. A customer using one or two modules can still treat CrowdStrike as a point solution. On the other hand, a customer using six, seven or eight modules is building more of its security architecture around Falcon.

Falcon Flex adds another layer to that argument. The model allows large customers to add, swap and expand modules with less procurement friction, which makes the platform easier to expand inside large enterprises. Falcon Flex accounts generated more than $1.9 billion of ending ARR in Q1 FY27, up more than 99% year over year, and renewals under the model have been associated with an average ARR uplift of roughly 26%.

This is the stronger version of the bull case: CrowdStrike is trying to become a larger share of the enterprise security budget and the data suggests that customers are, at least for now, moving in that direction.

A company with high retention, rising module adoption, strong free cash flow and no debt deserves to trade at a premium.

But….

What today’s price already assumes

As it usually happens in my case studies, a reverse DCF can become useful.

A reverse DCF starts with today’s price and works backwards, which is often more useful when a stock is already widely followed and the basic quality of the business is not in dispute.

For CrowdStrike, the reverse-DCF result is demanding. Using the current enterprise value of roughly $174 billion, the consensus free cash flow path, a 4% terminal growth rate and a normal required return, the valuation does not line up unless the market keeps assigning CrowdStrike a very high terminal multiple.

At a 10% discount rate, the implied enterprise value is far below the current market value. Even using 8% or 7%, the model still lands well below today’s price. The valuation only starts to reconcile with the current enterprise value at roughly a 6% discount rate.

This is telling us that, at today’s price, the market is effectively accepting a relatively low forward return unless CrowdStrike beats the expected path or keeps an unusually rich valuation multiple five years from now.

The other way to frame it is even clearer: at a more normal 10% required return, the current price needs something close to a 65x terminal free cash flow multiple. That is the assumption carrying most of the valuation.

Revenue growth obviously matters, margin expansion matters, AI-security adoption matters, Falcon Flex matters…

but the central question is whether investors will still pay a hyper-growth multiple for CrowdStrike once the company is larger and, in all likelihood, growing more slowly.

The five-year return math

The table below starts with FY26 revenue of $4.81 billion and looks five years out. This is not meant to produce a precise price target. The point is to make the assumptions visible and show what kind of outcome today’s price requires.

The base case is the most important part of the table, because it is not a bad operating outcome.

• An 18% revenue CAGR for five years would take CrowdStrike to around $11 billion of revenue.

• A 33% free cash flow margin would be excellent for a scaled software business.

• A 35x terminal free cash flow multiple would still be a premium valuation.

Even then, the implied value is around $493 per share, which is meaningfully below the current price. That is the problem with the starting valuation, as CrowdStrike can keep growing, keep expanding margins and keep taking share, yet shareholders can still earn poor returns if the multiple compresses faster than free cash flow grows.

The bull case is also worth paying attention to. Even if CrowdStrike compounds revenue at 24% annually for five years, reaches a 38% free cash flow margin and still trades at 45x free cash flow, the implied annualized return is only around 6%. That is not a disaster, but it is not much compensation for a scenario that already assumes exceptional execution.

This is why the stock is difficult at today’s price. The company needs to do extremely well while the market continues valuing those cash flows very generously.

The stock-based compensation issue

CrowdStrike’s free cash flow profile looks strong on the surface.

FY26 free cash flow was $1.24 billion and Q1 FY27 free cash flow was $468.5 million. For a company still growing above 20%, that is impressive. However, stock-based compensation makes it all less clean.

In FY26, stock-based compensation (SBC) and related payroll taxes were about $1.13 billion, equal to roughly 23.5% of revenue. Since SBC is added back in operating cash flow, reported free cash flow can make the business look more cash-generative than it is from a shareholder’s perspective. A strict owner-earnings adjustment would subtract the $1.13 billion of SBC from the $1.24 billion of reported free cash flow. On that basis, FY26 SBC-adjusted free cash flow would be only about $109 million.

The market is clearly not valuing CrowdStrike on that harsh measure today, and there is a reasonable argument for looking through some SBC while the company is still growing quickly. A high share price also reduces the dilution cost of equity compensation compared with issuing shares at a depressed valuation.

Still, this is not a small detail. If CrowdStrike is going to support a premium valuation over time, investors should want to see free cash flow remain strong while SBC falls as a percentage of revenue. That would make the cash flow quality improve, not just the headline number larger.

For me, this is one of the most important monitoring points.

What would make the investment thesis work

Revenue growth is the first test

If growth stays above 20% for longer than consensus expects, the current valuation becomes easier to defend, because that would suggest the market is still underestimating the depth of the cross-sell opportunity across identity, cloud, SIEM, browser security and AI workloads.

Falcon Flex is the second test.

If the model keeps driving strong renewal uplifts and deeper module adoption, then CrowdStrike’s growth duration becomes more credible. The platform story is much stronger if existing customers keep expanding rather than simply renewing.

The AI-security angle also deserves attention, although it needs to show up in numbers rather than marketing language. Products such as AIDR and Charlotte AI fit a real enterprise need, since companies adopting AI will need better controls around data access, identity, shadow AI usage and automated agents. If those products become meaningful recurring revenue streams, CrowdStrike gets another growth layer.

Lastly, the cash flow quality needs to improve.

A company growing above 20% with free cash flow margins moving toward the high-30s would deserve a rich valuation, but the case becomes much stronger if SBC intensity falls at the same time. That is the cleanest bull case: strong growth, expanding margins, lower dilution and a platform that keeps absorbing more of the enterprise security budget.

What could go wrong (The bear case)

Growth could simply slow faster than expected and consensus already points to deceleration over the next few years. Thus, if revenue growth moves lower sooner than expected, the market will probably stop paying the same multiple.

Competition could also pressure the model.

Microsoft has bundling advantages, Palo Alto Networks is pushing platformization aggressively, and SentinelOne competes at lower price points. CrowdStrike has a stronger platform than most direct competitors, but even modest pricing pressure could affect the multiple when the stock is priced this aggressively.

SBC remains another risk.

If stock-based compensation stays around the low-20s as a percentage of revenue, investors may eventually focus more on per-share economics than headline free cash flow.

The largest risk, however, is multiple compression.

CrowdStrike could keep growing and still disappoint shareholders if the market moves from paying a hyper-growth multiple to a more normal premium software multiple. That is the mechanism behind many frustrating software investments, as the company keeps executing, but the market stops paying the same price for each dollar of future cash flow.

Portfolio Lab view

CrowdStrike belongs on the watchlist.

The business is too strong to ignore, but the valuation is too demanding to force into Portfolio Lab at today’s price. The expected return profile depends too much on a rich terminal multiple surviving for another five years, and that does not leave enough margin of safety.

The setup would become more interesting in two cases.

1. The first is price. A move toward the $400–450 range would begin to create a more reasonable valuation cushion, assuming the operating numbers remain intact.

2. The second is fundamentals. If CrowdStrike re-accelerates revenue growth toward 25%+, expands free cash flow margins and shows clear SBC improvement, the intrinsic value estimate would move higher.

Until then, the classification is straightforward: Research Pipeline / Watchlist candidate: High conviction in the business, lower conviction in the stock at today’s valuation.

Final view

CrowdStrike remains one of the best software businesses in public markets. The platform is strong, the customer economics are attractive and the company is expanding into security categories that should matter for years.

None of that makes the stock automatically attractive at any price.

At $701 per share before the split, the market is already charging investors for much of that quality. The five-year math leaves little room for ordinary outcomes, since even the base case produces a negative annualized return if the multiple compresses. The bull case can work, but it requires CrowdStrike to keep growing at a very high rate, expand margins, improve SBC intensity and still trade at a premium multiple several years from now.

That is a lot to ask from one starting price.

CrowdStrike thus remains a watchlist name rather than a Portfolio Lab entry. The stock becomes more interesting either at a lower price or with clearer evidence that growth, margins and SBC improvement are moving faster than the market currently expects.

Subscribe now

Unlock Access to the Valuation Tools

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For further information, check the full disclaimer here.

1. The first is vibes: someone’s gut feeling about what’s going up, dressed up with confidence and a chart.

2. The second is a single backtest: “this allocation returned 9.2% a year,” built on one lucky stretch of history, with the taxes, the inflation and the years you’d have panic-sold all left out of the picture.

Neither survives contact with reality.

This series does something different. Before I ever show you an allocation, I want to give you the reasoning, the framework I run every Portfolio Lab portfolio through, so that when the numbers come in the paid editions, you understand exactly why they are what they are, and you could rebuild them yourself.

Let’s start with the thing almost everyone gets wrong.

The goal is not “maximum returns”

If the goal were simply the highest number, the answer would be boring: put everything in stocks, never sell, and wait. Over a long enough window that has usually won.

However, that’s not the real goal, because that’s not how humans actually behave with money. The real goal is to grow your purchasing power without big drawdowns.

We will then follow two ideas:

1. Real, not nominal. A portfolio that “doubles” over 25 years while prices also double has gone precisely nowhere. Every number in this series is real: after inflation, after taxes, after fees. The big nominal figure on a brokerage statement is the one that flatters you, but we should only care about the real figure.

2. The path matters, not just the destination. Two portfolios can end at the same place after 30 years and feel completely different along the way. One drifts up slowly over time, while the other plunges 45% in year three and takes a decade to recover. The biggest issue investors struggle with is investing in a good asset and not having the patience to deal with what comes with it.

Therefore, we will try to answer the question of:

The three questions that decide your portfolio

The uncomfortable truth is that there is no single best portfolio. There’s a best portfolio for a specific person with specific constraints.

Three questions set those constraints:

1. How long is the money invested? Time is the single most powerful variable. Over one year, stocks are a coin-flip. Over thirty, they’ve beaten inflation in every developed market on record. A longer horizon doesn’t just allow more risk, it changes which risks even count.

2. What would you actually do if it fell X%? Not what you’d like to think you’d do. What you’d actually do, watching the number drop month after month. If the honest answer is “I’d sell,” then a portfolio that can fall 45% is the wrong portfolio for you, no matter how good its long-run average looks on paper.

3. Where does the money live? Money inside a Roth IRA or 401(k) grows in a completely different tax universe than money in a regular brokerage account, and a brokerage account in California is taxed nothing like one in Florida. We’ll come back to this, because it’s where a lot of investment theses break.

Answer those three honestly and you’ve already narrowed the field more than any stock tip ever will.

Why I don’t trust a single backtest (and what I do instead)

When someone shows you “this portfolio returned X%,” ask one question: over which window? The answer is almost always cherry-picked, even when nobody’s lying. A few traps that wreck most analysis:

• Survivorship bias. The US has been the best-performing major market of the last century. Use it as your default and you’ll wildly overstate what’s normal.

• Cherry-picked periods. Start a bond chart in 2000 and they look amazing. Start it in 2022 and they look like a bad investment choice.

• Backtest overfitting. Any strategy “discovered” by searching history for what worked will look brilliant on that history and disappoint live. The past is easy to fit artificially with models and backtests.

• Marketing material. A firm earns far more from a 1.5%-fee product than from a 0.04% index fund. Material produced by the seller tells you about the seller’s incentives, not your expected return.

So instead of betting on one version of history, I simulate 100,000 possible paths, a Monte Carlo, with each path containing a different 30-year sequence of returns, inflation and the occasional crisis, drawn from distributions calibrated to a century-plus of global data. Then I read off the whole range of outcomes: the typical one, the lucky one and the one that actually matters, the unlucky one.

Fat tails and joint stock-bond crashes are built in, because pretending crashes are rare bell-curve events is exactly how people get blindsided. Moreover, every dollar is run through real US taxes and fees.

That’s the difference between “here’s a number” and “here’s how this behaves in actuality”.

The building blocks and what each one is actually for

You don’t build a portfolio by picking favorites, hence there are four blocks, each with a role:

• Global equities are the growth engine. Over decades they’re the only asset that has reliably outrun inflation by a wide margin. The catch is that they’re a terrible short-run inflation hedge and they crash hard. You hold them for the long run, knowing that they will test your discipline as an investor.

• Nominal Treasuries are the stabilizer and the deflation hedge, basically what tends to hold up when stocks fall in a recession. In the US they come with a benefit, as their interest is exempt from state income tax.

• Inflation-linked bonds (TIPS, I-Bonds) are purchasing-power insurance, as they pay a real yield plus whatever inflation turns out to be. Right now that real yield is the most attractive it’s been in years, which makes this block unusually compelling at the moment.

• The cash/short end acts as insurance, not a growth tool.

What matters now is the allocation percentages (which you can find in the different portfolio updates on Portfolio Lab).

The US advantage: where you hold it can matter as much as what you hold

Here’s where US investors have an edge most of them never use.

• Stocks are tax-favored: long-term gains and qualified dividends get preferential rates, and you don’t even owe them until you sell.

• Bond interest is tax-penalized: It’s taxed as ordinary income, every year, at rates that can exceed a third of the coupon for high earners.

• The federal code carves out two important exceptions: Treasury interest is exempt from state income tax and most municipal bond interest is exempt from federal income tax (and often state and local tax too, if you buy bonds issued by your own state).

Put those three facts together and the same dollar can be taxed wildly differently depending on what kind of account it sits in and which instruments fill each slot of the portfolio.

The real lever therefore is not about which state you live in, as the state taxes your investment income based on where you reside, not where the brokerage account is. For example, living in New York and opening an account with a Florida address doesn’t change your tax bill.

What does change your tax bill is the structure most investors never care to optimize:

• The account type: A taxable brokerage account pays tax every year on interest and dividends, plus capital gains whenever you sell. A 401(k), Traditional IRA, or Roth lets the same investments compound with zero tax friction until withdrawal. The ending balance will of course be very different.

• Asset location across your accounts: If you have both a 401(k) and a taxable brokerage, which assets you put in which account matters. The general rule, supported by decades of research: put the highest-tax-drag holdings (bonds, REITs, anything kicking off ordinary income) in the sheltered account, and put the most tax-efficient holdings (broad index funds you hold for years, qualified-dividend payers) in the taxable account. Most investors mirror the same allocation in both accounts and leave real money on the table.

• Instrument choice inside the taxable account: This is the closest legitimate version of moving your money to a state with lower taxes. For a New Yorker holding the bond sleeve in a taxable account, swapping corporate bonds for Treasuries removes the state tax bite. Swapping for New York municipal bonds can remove federal, state and city tax.

For a lot of people, fixing where and how their assets sit is a bigger upgrade than anything they’ll ever do to which assets they own.

Three types of Portfolios

Everything above converges into three archetypes. The precise percentages are detailed in the paid Portfolio Lab editions, here I’m focusing on the reasoning and general shape so you can see which archetype fits you.

• The Preservation portfolio is for capital you can’t afford to see fall much: a shorter horizon or simply a low tolerance for drawdowns. Its job is to beat inflation by a little, very reliably, with drawdowns small enough that you’d never be tempted to bail. Thanks to today’s real yields, this one can now do something it couldn’t a few years ago, which is protecting purchasing power and still growing it.

• The Moderate portfolio is the workhorse: for someone with a long runway who wants real growth but would still rather not watch their account halve. It accepts a meaningful dip in a bad year in exchange for materially more compounding. For most people with a decade-plus horizon, the answer lives somewhere in this zone.

• The Aggressive portfolio is for a long horizon and disciplined investors, someone who won’t flinch at a deep, multi-year drawdown in exchange for the higher upside. The honest catch, and one I’ll show you with the numbers, is that past a point, more stocks buy you a much bigger best-case scenario and a much worse bad-case while barely moving the typical outcome. Aggressive is therefore a bet on the right tail and your own discipline.

What’s coming next

In the paid Lab editions, I’ll open each of these up completely: the exact allocations, the reasoning behind every percentage, and for each slot two or three specific allocation options with the genuine pros and cons of each, so you can see the trade-offs and decide for yourself.

In the end, the goal was to find the approach most likely to grow what your money can actually buy while keeping the risk, the taxes, and the temptation to do something stupid all under control.

Get Access to all the Portfolio Updates

Subscribe now

This is educational content and thesis-building, not personalized investment advice. The simulations are estimates based on deliberately cautious assumptions, not predictions; markets can and will behave differently. Do your own research and consider speaking with a licensed, fee-only advisor about your specific situation. For further information check the full disclaimer here.

• Revenue grew 85% year over year

• Adjusted operating margin reached 60%

• Adjusted free cash flow margin reached 57%

• Management raised full-year revenue guidance to roughly $7.65–$7.66 billion

By most operating measures, it was the strongest quarter in Palantir’s history as a public company. This however did not stop the stock from falling.

After the earnings reaction, Palantir continued drifting lower and recently touched a fresh 52-week low near $116 (as of 24th of June 2026). Year to date, the stock is down roughly 27%, despite the company beating expectations and raising guidance.

That is the actual valuation setup, not whether Palantir is a good business, as the market has mostly moved past that debate. Palantir is growing fast, producing high margins, generating cash and expanding in areas investors care about: enterprise AI, U.S. commercial adoption, government software, defence data infrastructure and workflow automation.

What we should ask is:

Why has such strong operating performance not translated into a stronger share price?

The answer is that the numbers already need to be extraordinary for a long time. Palantir is being priced like an exceptional company that must remain exceptional.

There is also a second reason why this is a useful case study. Wall Street price targets on Palantir currently range from roughly $70 to $255. That is more than a 3x spread on a company with a market value near $300 billion and more than thirty analysts covering it. That kind of dispersion usually comes from disagreement about the long-term assumptions: how large the AI software market becomes, how much of that market Palantir captures, what margins survive competition and what multiple the market assigns once growth slows.

I will now break down the stock through a reverse-DCF framework. The goal is not to produce a false sense of precision, but rather to identify what today’s price already requires, compare that to consensus growth expectations and isolate the specific variables that matter most from here:

U.S. commercial growth, RPO, billings, margin durability, stock-based compensation, European sovereignty risk and the U.S. government budget cycle.

Record fundamentals, weaker stock

When a stock falls after strong results, investors usually reach for one of two simple explanations.

The bullish explanation is that the market is being irrational. The company is executing, investors are overreacting and the lower price creates a better setup.

The bearish explanation is that the market is seeing something that the data does not show. The company may still be strong, but expectations were too high, the valuation was too stretched or future growth is less certain than the latest quarter suggests.

With Palantir, both explanations contain part of the truth.

The operating business is clearly performing well. Q1 was not a weak quarter dressed up by adjustments, as revenue growth accelerated, U.S. commercial demand remained extremely strong, government revenue kept expanding, free cash flow was substantial and management raised full-year guidance.

Nonetheless, the stock was already carrying a valuation that required more than one great quarter. A stock trading at roughly 30–40x forward revenue and extremely high forward earnings multiples needs to keep proving that the long-term opportunity is larger, more durable and more profitable than consensus already assumes.

That is why this drawdown is interesting, as it is a case of the market asking whether even very strong results are enough.

That is the right way to frame the investment process: not “Is Palantir good or bad?” but “What level of future performance does the current valuation require, and is the evidence still moving in that direction?”

Three forces are hitting the stock at the same time

Palantir’s year-to-date decline should not be treated as one clean signal, since there are three separate forces moving together.

The first is sector-wide.

High-multiple software stocks have been repriced through 2026. The market has become less willing to pay extreme multiples for long-duration software cash flows, especially as investors debate whether AI will expand software economics or compress them. This means that Palantir is not falling in isolation, as when the market raises the discount rate applied to future software cash flows, expensive growth stocks are hit first, even when reported fundamentals remain strong.

The second is macro.

Higher rate expectations and pressure on long-duration equities directly affect valuation. A company like Palantir gets most of its implied value from cash flows far in the future. That makes the stock more sensitive to discount-rate changes than a slower-growth company trading on near-term earnings. Even if Palantir’s business performs exactly as expected, a higher required return can reduce the present value investors are willing to pay.

The third is company-specific.

Palantir is facing a set of risks that do not show up clearly in the headline revenue growth number:

• European data-sovereignty pushback

• Continued insider selling

• Dependence on large government programmes

• The challenge of sustaining unusually high margins as enterprise AI competition increases

Separating these layers is important. If the decline were only sector-wide, the case would be easier to dismiss as a temporary software selloff and if it were only company-specific, the operating thesis would be more clearly deteriorating. Instead, the current setup sits in the middle: Palantir remains a very strong business, but the market is becoming more demanding about the valuation.

Consensus is positive but not unified

The published analyst picture is broadly constructive. Palantir has a large number of covering analysts, with the average rating generally sitting around Moderate Buy or Buy depending on the data provider. Average price targets have recently sat somewhere around the high-$100s, while the target range stretches from roughly $70 on the low end to around $255 on the high end. A $70-to-$255 target spread on a mega-cap software company tells us that analysts are not just disagreeing about small modelling details.

One side is effectively saying Palantir becomes a dominant AI infrastructure layer with durable, very high margins. The other side is saying that even if Palantir remains a strong company, the current valuation already discounts too much success.

This is why we are using a reverse-DCF.

A traditional DCF asks: “What is the company worth under my assumptions?”

A reverse-DCF asks: “What assumptions are already embedded in the current price?”

For Palantir, the second question is more valuable. The stock is already widely followed and the basic facts are not hidden. The value comes from identifying which assumptions the market is implicitly underwriting and whether those assumptions are realistic, aggressive or fragile.

The valuation screen is misleading unless you ask what year you are valuing

Depending on the estimate source and the exact forward-year anchor, Palantir currently screens at roughly 30–40x forward revenue and extremely high forward earnings multiples. That sounds expensive and it is. However, a static multiple screen is not enough.

Palantir is growing so quickly that the multiple changes materially depending on whether you anchor to current revenue, next-year revenue, or a five-year revenue base. A company growing at 70% does not stay at the same revenue denominator for long.

That is the bullish response to the “too expensive” argument. If revenue compounds fast enough, today’s multiple can compress through growth rather than through a falling share price.

The problem is that this defence only works if growth remains very high for long enough. A one-year growth burst does not solve a valuation that depends on five years of exceptional compounding.

So… How much growth and margin durability are needed for the valuation to become reasonable without requiring the stock price to fall?

Reverse-DCF: what the current price requires

The model below starts with the current enterprise value and works backward.

The purpose here is to make the market’s implied expectations visible, and the model uses the following inputs:

Revenue compounds from the FY2026 base through 2031 and free cash flow is estimated as revenue multiplied by a normalised free cash flow margin. Each year’s cash flow is discounted at 10%, then a terminal value is calculated using a 3% terminal growth rate. The model then solves for the revenue growth rate required to justify today’s enterprise value.

Here is the output:

Even if Palantir sustains a 55% normalised free cash flow margin, the current price still requires revenue to compound at roughly 43% per year for five years, reaching about $46 billion by 2031.

At a 40% normalised free cash flow margin, which would still be excellent for a large software company, the required revenue CAGR rises to almost 53%, with implied 2031 revenue of nearly $64 billion.

Palantir thus needs to become much larger, very quickly, while preserving margins that most software companies never reach at maturity.

The key gap: consensus growth versus required growth

The reverse-DCF becomes more useful when compared with outside growth expectations.

Public third-party consensus data currently points to long-run revenue growth of roughly 30–31% annually. That is an extraordinary growth rate for a company already doing billions of dollars in revenue and would place Palantir well ahead of most mature software companies.

But… it is still below the growth rate required by the reverse-DCF.

That is the valuation gap: consensus can be very bullish in ordinary terms and still not bullish enough for the current price.

To make the point concrete, we can run the model forward using a 30% revenue CAGR from the FY2026 guidance base. Then we test different long-run free cash flow margins and estimate the equity value supported by those cash flows.

This is the central investment-process takeaway.

At consensus-level growth, even very generous margin assumptions do not support the current share price in this simplified DCF framework. That does not prove the stock is overvalued in some absolute sense, but it does mean that the market is not pricing consensus. It is pricing sustained outperformance versus consensus.

The stock can still work from here if Palantir keeps beating the numbers, but the required evidence is specific. The company needs to show that consensus is too conservative not only for one quarter, but for several years.

That is why the next several quarters matter so much. The key indicators are not only reported revenue growth. They are also U.S. commercial growth, RPO, billings, net dollar retention, margin quality and dilution.

Why discount rates matter more than most investors think

For a stock like Palantir, the discount rate is one of the main valuation drivers.

At a 40% normalised free cash flow margin, changing the discount rate produces a large change in the required growth rate:

If the market requires a higher return to own long-duration growth equities, then Palantir needs either faster growth, higher margins or a lower stock price to make the valuation work.

The software base rate: multiples compress before growth fully slows

Palantir’s current multiple should not be viewed in isolation.

High-quality software companies often trade at premium revenue multiples while growth is exceptional. The historical pattern however is fairly consistent: as growth decelerates from extraordinary to very good, the multiple usually compresses before the deceleration is complete.

This is the mechanism behind many “great company, disappointing stock” outcomes in software investing. The business continues growing, but the market stops paying the same multiple for that growth.

CrowdStrike is a useful reference point. It remains a high-quality, high-margin software company with strong security relevance, but once growth moved into a lower range, the revenue multiple compressed from its peak. ServiceNow is another useful reference: still an excellent enterprise software franchise, but at a more mature growth rate, the market assigns a much lower revenue multiple.

Palantir does deserve a premium while growth remains unusually high, but how long will that premium last?

The bull case: Palantir becomes the AI operating layer

The strongest bull case is not simply that AI spending grows, as that is too broad, many companies benefit from AI spending.

The stronger Palantir case is that large organisations need more than models. They need a controlled environment for deploying AI into real workflows, permissioning, auditability, governance, data integration, cost tracking, security controls and operational accountability. That is where Palantir’s AIP and ontology framework matter.

In this view, Palantir is trying to become the operating layer that allows complex institutions to use AI in production. If that works, the company’s addressable market is larger than a normal software replacement cycle.

The recent numbers support this case. U.S. commercial growth has been exceptional, government demand remains strong, net dollar retention is high, RPO and billings suggest visibility beyond the latest quarter. Palantir’s Rule of 40 score is far above what investors normally see in software at this scale and that is why a simple bearish view is too easy.

The bull case becomes credible if the next several quarters show that Palantir is not just benefiting from an early AI adoption wave, but becoming embedded infrastructure inside customer operations.

The risk that deserves more attention: European sovereignty pushback

The most under-discussed risk is not competition from another software company, but rather it is sovereign dependency.

Recent European developments suggest that some governments are becoming uncomfortable relying on a U.S. vendor for sensitive public-sector data infrastructure. France, Switzerland, the Netherlands, the United Kingdom and Germany have all produced headlines or policy signals pointing in the same direction: concern over data sovereignty, strategic dependency and the desire for domestic or European alternatives.

Near term, this probably does not break Palantir’s numbers. The company’s growth engine is currently U.S. commercial and U.S. government demand. European government revenue is not the core driver of FY2026 guidance.

Valuation however is not only about next year’s revenue. Palantir’s multiple reflects a long-duration assumption: that the company can become an essential software layer for large institutions, including allied governments. That is why European pushback can be a real pain for Palantir, since it raises questions about the duration and geographic breadth of the government opportunity.

The market may be willing to pay a very high multiple for Palantir as a default operating layer across Western institutions. It should pay a lower multiple if that opportunity becomes more concentrated in the United States and more politically contested abroad.

This is not a thesis-breaker yet.

Insider selling: not proof, but not irrelevant

Insider selling has also become part of the Palantir debate.

The fair interpretation is nuanced. Much of the selling appears to be connected to pre-scheduled 10b5-1 plans. That means it should not be treated as a clean signal that insiders suddenly believe the stock is overvalued today.

However, the absence of offsetting insider buying is still informative. If a company has fallen to a 52-week low after a major beat-and-raise quarter and insiders continue selling while no meaningful open-market buying appears, that does not prove anything by itself. It does however reduce the strength of any argument that insiders view the current drawdown as an obvious mispricing.

If the stock remains under pressure, do any insiders step in with open-market purchases?

That would be more informative than continued scheduled selling. Until then, insider activity should be treated as a cautionary data point, not a central thesis.

Government revenue and the U.S. budget calendar

Palantir’s government business is one of its strengths, but it also creates a specific risk that retail coverage often skips: the budget calendar.

A meaningful part of the company’s long-term case depends on large U.S. defence and government programmes scaling over time. If those programmes move from pilots into broader production, Palantir’s revenue visibility improves, and if funding is delayed by continuing resolutions, appropriations fights or election-year budget uncertainty, growth may be delayed even if the underlying demand remains intact. This is a timing risk.

Timing matters a lot for a stock priced this aggressively. If the market is expecting smooth acceleration and a government funding delay pushes revenue recognition out by several quarters, the valuation can become more vulnerable even though the long-term contract opportunity remains alive.

That is why we should monitor not only earnings releases but also the U.S. budget calendar, defence appropriations, continuing-resolution headlines and updates around major programmes such as Maven, TITAN and NGC2.

Scenario analysis

Putting the reverse-DCF, consensus-growth gap and qualitative risks together, the stock can be mapped into three practical scenarios.

The important point is that the base case is not a bad operating outcome. A company compounding revenue at 40% while producing 40%+ free cash flow margins would be one of the best software businesses in the market, but at Palantir’s current valuation, “one of the best” may not be enough. The stock needs evidence that Palantir is in the small group of companies capable of sustaining elite growth and elite margins together.

What would confirm the valuation

The valuation becomes easier to defend if several indicators move together.

U.S. commercial growth needs to remain far above ordinary software growth.

This is the highest-signal part of the thesis because it shows whether AIP is turning into a broad enterprise platform rather than a narrow government-heavy business.

RPO and billings need to grow faster than reported revenue.

That would suggest the company is building future visibility.

Net dollar retention needs to stay high.

This is critical because Palantir’s model depends not only on acquiring customers, but expanding deeply inside them. If existing customers keep increasing spend, the long-term revenue case becomes stronger.

Margins need to remain strong on a GAAP basis, not only on adjusted measures.

Adjusted free cash flow is useful, but we should not ignore stock-based compensation or dilution.

Finally, European sovereignty concerns need to remain contained.

If the issue stays mainly European and does not affect U.S. government demand or large commercial customers, it is manageable. If it spreads into broader procurement thinking among allied governments or regulated industries, it becomes more important.

What would weaken the valuation

The valuation weakens if the slope of the business changes.

The most important failure point would be U.S. commercial deceleration. If U.S. commercial growth starts moving rapidly toward consensus long-run assumptions, the market will likely reassess how much of the current AI adoption cycle is already priced in.

The second failure point is margin normalisation. Palantir can remain a great company with 35–40% free cash flow margins, but if the current valuation requires something closer to 50%+, that difference should be noted.

The third failure point is continued stock-based compensation pressure. For high-growth companies, free cash flow can look attractive while per-share value creation is diluted. The correct metric is cash generation per diluted share over time.

The fourth failure point is government timing risk. If major U.S. programmes face delays because of budget politics, the long-term thesis may remain intact but the valuation could still compress in the near term.

The valuation conclusion

Palantir is an exceptional operating business. The latest results show rare revenue growth, very high margins, strong cash generation, high customer expansion and a credible position in enterprise AI. The company belongs in any serious AI-software research pipeline. Nonetheless, the stock is not priced like a normal strong business and the reverse-DCF makes this clear.

Depending on the margin assumption, the current price requires roughly 43–62% annual revenue growth through 2031. That is far above the roughly 30% long-run growth rate reflected in third-party consensus expectations. At consensus growth, even generous margin assumptions produce values materially below the current stock price.

This is the real interpretation of the 27% drawdown. The stock has fallen because the gap between excellent fundamentals and even more demanding expectations is still wide.

The investment process should therefore focus on confirmation:

Palantir needs evidence that its growth duration, margin structure and competitive position can support the valuation already attached to the stock.

The next several quarters should be judged through five questions:

Until those answers become clearer, Palantir is best viewed as a high-quality but highly valuation-sensitive research candidate.

That is the core lesson of this Valuation Tools case study.

Running the Numbers monitoring framework

For future updates, I would track Palantir through the following framework:

A note on methodology: the reverse-DCF above is intentionally transparent. The goal is to show which assumptions matter most and allow members to adjust the model logic themselves. Founding Research Members receive the downloadable workbook and setup guide as part of the Valuation Tools library.

Disclaimer

This content is provided for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. For futher information check the disclaimer.

The goal from here is to build a proper research workspace for readers who want to understand markets and portfolio decisions with more structure.

A huge chunk of market commentary tells you what to think, but not how the conclusion was reached.

That can be interesting, but it is not enough for a serious investment process.

I believe the more useful question is what is really driving the investment case. Sometimes the answer is in the valuation. Other times it is in regulation, industry structure, consumer behaviour, debt dynamics or small data points that do not show up clearly in a basic model.

The goal is to connect those pieces and turn them into clear, useful commentary that makes financial markets easier to understand at a quick glance.

The new structure now includes four main areas: Content Hub, Research Pipeline, Valuation Tools and Portfolio Lab.

Valuation Tools

Valuation Tools is where the company analysis becomes practical.

This section is built for readers who want to move beyond headlines and understand what a company might actually be worth under different assumptions.It will include valuation case studies, selected model outputs, assumption guides and walkthroughs.

The main tool is the automated DCF/APV workbook, now available for Founding Research Members.

Pulling the data, organising the sheets, building the forecast and setting up the outputs would take lots of time. The workbook does all of this automatically and brings the valuation output into one place, with charts, sensitivity tables, cash flow forecasts and a summary dashboard designed to make an investment thesis easier to understand, compare and challenge.

Research Pipeline

The Research Pipeline is the forward-looking part of the research process.

This is where you can track what I’m currently watching across equities, rates, commodities, currencies, credit and macro themes. The point is to map the market debate in a more useful way.

For each asset, company or theme, the pipeline looks at the current consensus, what the market may already be pricing, the main tension, the next catalyst and the research priority. That makes it easier for the reader to separate noise from actual research opportunities.

A stock, sector or macro theme only becomes interesting when there is a real question behind it. Maybe the market is too confident, maybe the consensus is split or maybe the setup focuses too narrowly on a certain series of data, forgetting about other potential catalysts.

That is what the Research Pipeline is built to identify.

Portfolio Lab

Portfolio Lab is the next major section to be built out. This will be the place where selected research ideas are tracked over time.

Portfolio Lab is thus meant to show how investment theses develop after the first write-up:

• How is the idea monitored?

• What changes after earnings?

• What happens if the macro backdrop changes?

• Does the valuation still make sense?

• Has the thesis improved or weakened?

• Do we need to rebalance our portfolio?

  ---

Content Hub

The Content Hub is the archive of published research.

It is where I will organise market notes, company analysis, macro pieces, valuation work and public research so that the site becomes easier to navigate over time.

Instead of having older posts disappear into the feed, the Content Hub will act as a research library.

Why am I building this?

The aim is to make Running the Numbers more useful for people.

I want the site to become a place where readers can follow a research process rather than just read isolated opinions.

That means building tools, tracking assumptions, reviewing companies, watching macro variables and showing how different parts of the market connect.

I am aiming for less random commentary, more structure and more clarity around financial markets and the global economy.

The first pieces are live and more content will be added over the next few days, especially in Valuation Tools and Portfolio Lab.

You can go directly to:

• Content Hub

• Research Pipeline

• Valuation Tools

• Portfolio Lab

As always, everything published on Running the Numbers is for educational and informational purposes only. It does not constitute investment advice, financial advice, trading advice or a recommendation to buy, sell or hold any security, fund, commodity, currency or financial instrument. For further information check the disclaimer.

Become a Premium Member

As of right now, this is WINDOWS only. Mac compatibility will come soon.

Read more

It is built around one practical question:

What has to be true for this company to be worth more or less than the market price?

The goal is simple: take a company, organise the financial data, test the assumptions and understand what the valuation actually depends on. A lot of market content stops at the surface. A stock is described as cheap, expensive, high quality, overvalued or attractive, but the assumptions behind those claims are often left vague. Valuation Tools is designed to make the process more disciplined.

The section is designed to help readers move from vague opinions to a clearer valuation and investment process. Instead of saying a stock looks cheap, expensive or high quality, the aim is to test the assumptions behind that view.

What Valuation Tools includes

1. Understand what really impacts the valuation

The point of a valuation is not to produce a fancy spreadsheet. Much more important (and useful) is to understand what needs to happen for this company to be worth more than the market thinks?

Valuation Tools helps break that down, showing whether the investment case depends mainly on growth, margins, cash flow, debt, interest rates or something else entirely.

This makes your research process clearer and faster. Before spending time on a company, you can see what actually matters and what would need to go right.

2. Real company case studies

Case studies apply the framework to real companies, showing how the numbers change under different assumptions and what the market may already be pricing in.

The idea is simple: take a company people actually care about, run the numbers and show what the current price seems to imply.

• Is the market already expecting too much?

• Is the stock cheap for a reason?

• What would need to happen for the thesis to work?

That is what the case studies are built to answer.

3. Use templates and trackers instead of starting from zero

Premium Members will get access to structured templates and trackers for company research, valuation assumptions, thesis development and earnings updates.

These are designed to make the research process more repeatable.

Instead of starting from a blank page every time, members can work from a framework that separates inputs, assumptions, outputs and risks for their investments.

4. Automated DCF/APV workbook

The workbook is designed to help users move from ticker input to a structured valuation output.

Instead of spending hours pulling financial statements, setting up the model, linking the sheets and building sensitivity tables from scratch, the workbook brings the process into one workflow.

It automatically pulls the data via VBA, builds the forecast and brings the valuation output into one place, with charts, sensitivity tables, cash flow forecasts and a summary dashboard designed to make the result easier to understand, compare and challenge.

That means more time can be spent on the actual investment question: does the valuation make sense?

Moreover, each tool will be supported by walkthroughs and examples showing how to read the output and how the model can be used in practice.

How access works

• Free readers will see selected explanations, previews and examples.

• Running the Numbers Members will get access to valuation notes, walkthroughs, case studies, selected outputs, assumption guides and member-only valuation commentary.

• Founding Research Members will get access to the downloadable valuation workbook, full template files, model updates and setup guides.

The downloadable workbook will not include a shared API key. Users will need to insert their own Alpha Vantage API key, and a step-by-step setup guide will be provided.

This structure means that regular members can see how the valuation process works through case studies and model outputs, while founding members get the actual workbook and future file updates.

Unlock Valuation Tools

Why this matters for investors

Most investment content gives you an opinion.

Instead, Valuation Tools is built to show you the work behind the opinion.

It helps you understand what the market is assuming, what could make the investment case work and what could break it.

That is the whole point of this section: better stock research, clearer assumptions and a more disciplined way to evaluate companies.

In the United States, the Federal Reserve has accumulated a deferred asset of roughly $243 billion, representing the future net earnings it must retain before remittances to the Treasury can resume. The Bundesbank reported its first annual loss since 1979 for 2024, followed by a further €8.6 billion loss in 2025, while the Bank of Japan was carrying an unrealised valuation loss of roughly ¥45.4 trillion on its government bond portfolio at the end of the financial year in March 2026. In the United Kingdom, the Asset Purchase Facility is expected to leave the Treasury with a lifetime cash cost of roughly £134 billion.

These figures are easy to interpret as evidence that central banks expanded too aggressively, lost money when interest rates rose and now face the cost of repairing their own balance sheets. That reading, however, treats the central bank as if it were a private institution standing apart from the government. Economically, it is part of the state. Once the central bank’s accounts are consolidated with those of the treasury, claims held within the public sector cancel and what remains is the government’s combined exposure to the private sector. From that perspective, the recent losses are better understood as the fiscal consequence of decisions taken during the QE years, when the state exchanged part of its long dated fixed rate debt for reserve balances whose cost moved with the policy rate.

QE changed the maturity of government debt

The cleanest way to understand the losses is to treat quantitative easing as a change in the maturity structure of public debt.

When a central bank buys a long dated government bond, it pays by creating reserve balances for the banking system. On the consolidated public balance sheet, the government bond held by the central bank is both an asset of one public institution and a liability of another, so the two positions cancel. The reserve balances remain as a liability owed to the private sector. Those reserves are remunerated at a rate closely linked to the central bank’s policy rate and their cost therefore adjusts much faster than the coupons on the long dated bonds that were removed from private portfolios. To the extent that QE involved purchases of government bonds, it replaced part of the consolidated state’s fixed rate, long maturity debt with floating rate, short duration funding.

The scale of that maturity shift is visible in the US data:

The maturity transformation became very visible after 2008. In the United States, the floating rate component of the consolidated government balance sheet rose from a negligible level before the financial crisis to close to one quarter of GDP.

For most of the post crisis period, the maturity swap was highly profitable for the public sector. The central bank continued earning fixed income from securities purchased during the low rate period while reserve balances cost close to zero, leaving a positive carry that flowed back to the treasury. Between 2011 and 2021, the Federal Reserve remitted more than $920 billion to the US Treasury and cumulative remittances since the financial crisis rose above $1 trillion.

The Bank of England’s Asset Purchase Facility produced a similar pattern. It transferred £123.9 billion to the Treasury between 2009 and 2022, during the period when income from its gilt portfolio exceeded the cost of financing it. The assets continued paying coupons set when interest rates were low, while the cost of reserve balances adjusted rapidly with the policy rate.

In the United States, that funding cost eventually rose above 5%, turning a position that had generated positive carry for more than a decade into a source of losses.

There was nothing accidental about the reversal, with the Federal Reserve raising rates to bring inflation down, and the floating rate side of the consolidated public balance sheet repricing with it. The losses were the direct consequence of the interest rate exposure created by QE.

The fiscal exposure

The losses stem from the higher cost of the floating rate liabilities created during QE once short term interest rates rose. The transactions pass through the central bank’s income statement, although the ultimate effect belongs to the consolidated government budget.

By buying long dated bonds and paying with reserves, the state shifted part of the duration risk held by private investors onto the consolidated public balance sheet. That structure was cheap while short term rates remained low, but it became more expensive once policy rates rose because the reserve liability repriced far faster than the fixed coupon assets. The attraction of short maturity funding is that it often carries a lower expected cost, partly because investors value liquidity and money like claims, while the downside is that the government’s interest expense reprices much faster when policy rates rise. Recent central bank losses are that downside showing up in the fiscal accounts. They do not, on their own, show that QE was a bad decision, but they do show that the interest rate risk created by the policy has now been realised.

Accounting rules also impact how quickly that cost becomes visible. In Britain, the Treasury indemnifies the Asset Purchase Facility and makes direct cash transfers when the programme records losses. Active gilt sales also realise valuation losses sooner.

In the United States, the Federal Reserve records a deferred asset. Future net income will be retained until that balance is worked down, after which regular Treasury remittances can resume.

Within the Eurosystem, losses may be absorbed through provisions, reserves and retained future profits. The Bundesbank carried forward a loss of €19.2 billion from 2024, then added a further €8.6 billion in 2025.

The Bank of Japan holds government bonds at amortised cost. Changes in market value are disclosed but do not affect reported income unless the bonds are sold before maturity. Its ¥45.4 trillion unrealised loss is therefore economically relevant, particularly if sales become necessary, but it is not the same as a realised operating loss.

These arrangements produce very different headlines, although the underlying interest rate exposure is similar. They determine when the fiscal cost is recognised, how it enters the budget and whether current or future taxpayers absorb it.

Weighing the losses against the benefits

Given that these losses represent one side of a deliberate policy decision, they cannot be assessed in isolation. The relevant comparison includes the income earned before rates rose, the reduction in government borrowing costs and the macroeconomic purpose of the purchases.

The first offset is the long period of central bank profits. The Federal Reserve’s current deferred asset followed more than a decade of unusually large Treasury remittances. The Bank of England’s current indemnity payments followed £123.9 billion of earlier transfers in the other direction. Looking only at the years of loss turns one half of a complete interest rate cycle into the whole story.

The second offset is the effect of QE on government financing costs. Large scale purchases were intended to reduce long term yields, compress term premia and improve financing conditions across the economy. The Bank of England estimates that lower debt issuance costs associated with QE saved the Exchequer between £50 billion and £125 billion. That range offsets a substantial part of the £133.7 billion lifetime direct cost currently estimated by the OBR, with the upper estimate coming close to the full amount. Those calculations remain uncertain because they depend on a counterfactual yield curve, as no one can observe directly where government borrowing costs would have settled without QE. The same limitation applies to estimates for the United States and the euro area, although the direction of the effect is widely accepted.

The third offset is the macroeconomic objective itself. QE was introduced when policy rates were constrained and economies faced weak demand, financial instability or a serious risk of deflation. Its purpose was to lower borrowing costs, support asset prices, improve credit conditions and reduce the severity of recessions. Those benefits do not appear on a central bank income statement however and any serious evaluation still has to include them.

Cecchetti and Hilscher estimate that the maximum annual losses for the Federal Reserve, the Eurosystem and the Bank of England fall between approximately 0.3% and 1.5% of GDP. Moreover, the Federal Reserve’s $114.3 billion loss in 2023 was around 0.4% of US GDP. These are significant fiscal amounts, particularly when public finances are already under strain, although they remain modest relative to the output costs normally associated with a severe recession or deflationary shock. That comparison, however, does not establish how much economic damage QE actually prevented.

Whether QE was worth its cost ultimately depends on the counterfactual and that is a much harder question than whether the central bank reported an accounting loss.

The Swiss National Bank case

The Swiss National Bank stands apart from the other large central banks.

It lost CHF132.5 billion in 2022, an amount close to 17% of Swiss GDP. The scale of that loss was far beyond the experience of the Federal Reserve, the Eurosystem or the Bank of England.

The difference lies largely in the assets it owned, as the SNB accumulated foreign bonds and equities as part of its effort to limit appreciation of the Swiss franc. This created substantial exposure to foreign interest rates, equity prices and exchange rates. In 2022, falling bond and equity markets combined with an appreciating franc to produce the record loss.

A domestic government bond purchase rearranges claims within the same sovereign balance sheet. The central bank receives income from its own treasury, then returns any residual profit to that treasury. Changes in income affect taxpayers, bondholders and banks differently, although much of the flow remains within the domestic economy.

Foreign asset purchases on the other hand create a different distribution. Returns depend on foreign issuers and exchange rates, while any resulting loss is borne by the domestic public sector. The consolidated balance sheet approach therefore does more than reduce concern about central bank solvency, since it also shows where the economic gain or loss ultimately lands.

Negative equity does not operate as it would at a commercial bank

Commercial bank solvency depends on the value of assets remaining sufficient to meet liabilities. A central bank operates under a different constraint because its liabilities consist mainly of currency and reserve balances denominated in money that it issues. This does not mean that losses are irrelevant or that central banks possess unlimited economic resources. It means that ordinary corporate measures of equity do not determine whether monetary policy can continue.

A central bank can operate with negative accounting equity for long periods (Chile, the Czech Republic, Israel and Mexico have all done so without losing operational control of monetary policy). What matters is whether the institution retains the authority to issue currency, receives credible fiscal backing and can pursue its price stability mandate without being forced to create excessive money simply to finance itself. Hall and Reis show that a central bank can remain financially sound when positive net income is transferred to the treasury and losses can be covered by future earnings or fiscal transfers. In addition, Reis and Buiter make the broader point that central bank solvency depends on the present value of future seigniorage and fiscal support, rather than the accounting equity measure used for a private firm.

The Federal Reserve’s deferred asset fits within that framework (at the end of 2025 it stood at $243.481 billion). This is the amount of future net income that the Reserve Banks must retain before normal remittances to the Treasury resume. It records the cumulative earnings shortfall since remittances were suspended. The fiscal cost remains because the Treasury receives less income than it otherwise would, increasing its financing requirement relative to a counterfactual in which those remittances continued. The distinction is that the loss does not prevent the Federal Reserve from setting interest rates or meeting its obligations. Describing the deferred asset as evidence that the Fed has gone bankrupt therefore confuses a fiscal transfer with an operational failure.

The more serious risk

Central bank losses matter most when they begin affecting policy credibility.

Visible losses create an easy political argument. Legislators may ask why central banks are paying large amounts of interest to commercial banks while withholding remittances from the treasury and governments facing direct indemnity payments may question the cost of monetary independence. Public criticism can thus intensify even when the underlying accounting treatment has no effect on the central bank’s capacity to act.

A second concern is that central banks could become reluctant to raise rates when doing so would deepen their own losses. They might also cut earlier than required in an attempt to restore profitability.

Research by Gebauer, Pool and Schumacher examines precisely this conflict. Their model finds that higher rates can depress central bank profitability after a period of large asset purchases, but that profitability concerns do not undermine financial strength when future seigniorage is taken into account. The inflationary risk emerges if a central bank gives priority to repairing its income statement and allows that objective to influence the path of interest rates. For central banks with strong institutional credibility and reliable fiscal backing, there is little evidence that recent losses have materially altered monetary policy. The possibility still deserves attention because of political pressure.

How central banks account for the same underlying cost

The contrast between the Federal Reserve, the Bank of England, the Bundesbank and the Bank of Japan can make central bank finances look more complicated than the underlying economics.

• Britain pays the cost directly through the Treasury indemnity.

• The Fed records a deferred asset that must be worked down through future net income before remittances resume.

• The Bundesbank carries losses forward.

• The Bank of Japan reports market value changes without recognising them in income unless the bonds are sold.

However, none of these methods change the exposure.

That exposure was established when governments, viewed on a consolidated basis, replaced part of their long dated fixed rate debt with reserve balances whose cost followed the policy rate. During the low rate years, the position produced large profits, but once inflation forced central banks to raise rates, the same structure produced losses. The useful debate therefore concerns fiscal policy rather than central bank survival.

Did lower long term yields, easier financial conditions and protection against deeper recessions justify the interest rate risk transferred to the public balance sheet?

The answer will vary by country, programme and purchase period. The evidence assembled by Cecchetti and Hilscher suggests that the losses of the Federal Reserve, the Eurosystem and the Bank of England remain manageable relative to the size of their economies. Earlier remittances and lower financing costs also offset part of the direct fiscal bill. The Swiss case requires a different judgement because its foreign asset portfolio exposed the domestic public sector to much larger market and currency risks. Even there, the issue is the quality and distribution of the fiscal risk, rather than whether the central bank can continue issuing francs.

None of this makes the losses painless. The US Treasury will forgo remittances until the Fed works down its deferred asset, British taxpayers are financing current APF indemnity payments, the Bundesbank will use future profits to absorb accumulated losses and the Bank of Japan could realise part of its valuation loss if it sells bonds below cost.

These are all public costs and they should be scrutinised as such.

Sources

• Cecchetti, S. G. & Hilscher, J. (2024), Fiscal Consequences of Central Bank Losses, NBER Working Paper 32478 / CEPR DP 19088 — the consolidated-balance-sheet framing; loss magnitudes (0.3–1.5% of GDP; SNB ~17%); domestic vs foreign transfers.

• Greenwood, R., Hanson, S. G. & Stein, J. C. (2015/2016), A Comparative-Advantage Approach to Government Debt Maturity (J. Finance) and The Fed’s Balance Sheet as a Financial-Stability Tool (Jackson Hole) — reserves as floating-rate notes; consolidated interest-rate exposure.

• Hall, R. E. & Reis, R. (2015), Maintaining Central-Bank Solvency under New-Style Central Banking, NBER WP 21173.

• Reis, R. (2015), Different Types of Central Bank Insolvency and the Central Role of Seigniorage, NBER WP 21226; Buiter, W. (2008), Can Central Banks Go Broke?, CEPR Policy Insight 24.

• BIS — Bell et al. (2023), Bulletin 68, Why are central banks reporting losses? Does it matter?; Bell, Frost, Hofmann, Sandri & Shin (2024), Central bank capital and trust in money.

• Gebauer, S., Pool, S. & Schumacher, J. (2024), The inflationary consequences of prioritising central bank profits, ECB Working Paper 2985.

• Federal Reserve — November 2025 Balance Sheet Developments; CRS, The Federal Reserve’s Balance Sheet (deferred asset ≈ $243bn; 2023 loss $114.3bn, 2024 $77.6bn). FRED: WRESBAL, RRPONTSYD, WALCL; BEA GDP.

• Bank of England APF Quarterly Reports (2025 Q4); OBR Spring 2025 forecast (debt-servicing savings £50–125bn; lifetime cost ≈ £134bn).

• Deutsche Bundesbank annual accounts 2024 (−€19.2bn) and 2025 (−€8.6bn); Swiss National Bank 2022 result (−CHF132.5bn).

The number is easy to misread, as it is not a prediction that the United States will reach 210% debt to GDP, nor is it a claim that a crisis begins only once that threshold is crossed. It is better understood as a solvency boundary, as beyond that level, in PWBM’s model, no broad based labour income tax can generate enough revenue to finance the interest cost at market clearing rates. At that point, some form of default becomes almost unavoidable in real terms, whether through explicit Treasury default, a reduction in promised Social Security and Medicare benefits or the slow erosion of debt through inflation. Most commentary treats the number either as an alarmist countdown or as something too far away to matter. However, I think that both interpretations miss the point. The 210% ceiling does not define the exact moment the crisis begins, with the market risk starting well before then, when investors begin wondering whether the adjustment is still possible.

For scale, federal debt held by the public is around 101% of GDP in 2026. On the Congressional Budget Office’s current law path, it reaches roughly 120% by 2036. The ceiling is therefore still far above today’s level, but the path is no longer theoretical. PWBM’s median projection reaches the ceiling between 2045 and 2051, depending on how quickly healthcare costs grow relative to the economy. Under the higher cost growth assumption, which is closer to the historical experience of the United States, the model gives a 25% probability of reaching the ceiling within about 14 years, around 2040. Under that same scenario, the median path reaches the ceiling in 2045.

The risk is not hidden in an extreme scenario, it appears under the cost trend the United States has already lived with for decades.

The Tax Math Behind the 210% Limit

The phrase “no feasible tax rate” can sound abstract, so it is worth translating into a concrete adjustment. PWBM estimates that stabilising debt at the ceiling would require a permanent additional tax of about 15% on all uncapped labour income. That is larger than the combined employee and employer contributions to Social Security and Medicare Part A today.

A static calculation puts the required adjustment at around 5% of labour income, but that ignores the feedback loops, as higher debt pushes up interest rates, which crowds out private investment and shrinks the tax base. A large wage tax also suppresses work and saving, which shrinks it further. By the time those effects compound, the true required adjustment is roughly three times the static estimate and it also grows larger the longer the fix is delayed.

Delaying adjustment thus makes the “bill” larger, because the economy has to service a bigger debt stock at a higher required return while the tax base becomes less favourable. The ceiling is therefore the point where the fiscal adjustment becomes mathematically impossible under the model’s assumptions. The problem begins before then however, as investors start asking whether the political system can still deliver the adjustment needed to avoid it.

The Debt Path Comes Down to Interest and Growth Rates

At its core, the debt problem comes down to the relationship between the interest rate the government pays and the growth rate of the economy.

A country’s debt to GDP ratio rises when it runs primary deficits and when the interest rate on the debt exceeds nominal GDP growth. In simple terms, the ratio increases by the primary deficit plus the gap between the interest rate and growth, multiplied by the existing debt stock.

When nominal growth exceeds the interest rate, debt dynamics are forgiving. The economy can grow faster than the interest rate, which allows the debt ratio to stabilise or drift lower even with modest primary deficits. That was the case for most of the 2010s, when Treasury yields were low, inflation was subdued and nominal growth often exceeded the government’s borrowing cost. It was the world Olivier Blanchard had in mind in 2019, when he argued that public debt was less dangerous when interest rates remained persistently below growth rates.

That world has changed since then, as the 10 year Treasury yield now sits near 4.5%, nominal growth is no longer reliably above the interest rate and primary deficits remain large. When the interest rate creeps above growth while the government is still borrowing to fund current spending, the debt ratio starts compounding upward.

PWBM’s ceiling exists because the model allows interest rates to rise endogenously as debt increases. More Treasury supply has to be absorbed by investors and those investors require a higher return as the debt stock grows relative to the economy’s capital base.

Why the Ceiling Moved and Why that Is Not Entirely Comforting

PWBM first published its debt ceiling estimate in 2023. The 2026 update raised the sustainable limit slightly and the reason is important.

The value of capital held by US households outside Treasuries has grown faster than GDP. A larger private capital base gives the federal government more room to issue debt before it severely crowds out productive investment. In plain terms, if the private sector has more wealth and capital, it can absorb more government borrowing before the strain becomes binding. The complication is that this extra runway depends partly on asset prices being correctly valued today.

If risky assets were to reprice sharply, the private capital base woud shrink relative to the debt stock and the ceiling moves closer. That makes the topic more connected to markets than it first appears, as a large equity correction would not only damage portfolios, it would also reduce the capital base that helps make the debt path sustainable. PWBM even points to an AI driven equity correction as an example of the type of shock that could pull the ceiling back toward the present.

The fiscal runway therefore rests partly on a market environment where private asset values remain elevated. That is not necessarily wrong, but it means fiscal risk and equity valuation risk are not independent.

Debt’s Three Zones

The most useful part of the PWBM framework is not the 210% number itself, since that is a model dependent upper bound. Instead, the more useful idea comes from the Cole and Kehoe framework, where debt moves through three zones.

At low levels of debt, the government is safely solvent and lenders have no reason to doubt repayment, whereas at extremely high levels, default becomes rational regardless of lender behaviour. Between those two extremes sits the more dangerous zone, where the government is solvent if lenders keep refinancing it, but a shift in sentiment can become self fulfilling. Here, market risk begins, as the government may be able to service the debt under normal conditions only if investors continue rolling it over at reasonable rates. If confidence weakens, refinancing becomes more expensive, the debt path worsens and the market’s fear starts validating itself.

This is why the 2045 to 2051 closure window should not be read as the earliest point of danger. It instead is the latest point at which an orderly adjustment remains arithmetically possible under the model. As already mentioned, the market window can close earlier if investors begin demanding a higher term premium or if the buyer base becomes less reliable.

The UK gilt episode in 2022 was a small version of this dynamic. Britain was not insolvent in any mechanical sense, but the market quickly tested how much fiscal and institutional credibility it was willing to grant. Advanced economies can face liquidity driven stress long before they face insolvency.

How the Gap Can Actually Be Closed

There are only a few ways a debt path of this size can be resolved.

The most benign path is growth. If productivity accelerates enough to push nominal GDP growth back above the government’s borrowing cost, the debt ratio becomes easier to stabilise. This is the optimistic version of the AI investment cycle: higher productivity, stronger real growth, higher tax revenues and less pressure on debt dynamics. CBO and PWBM both allow for that possibility, although neither treats it as enough to reverse the fiscal path on its own.

The explicit fix is fiscal consolidation. That means the equivalent of a permanent large tax increase, major spending cuts or some combination of the two. In practice, the pressure would fall heavily on the programmes driving long run spending growth, especially Social Security, Medicare and the rest of the budget increasingly crowded out by interest expense. Politics would of course be a huge factor in the feasibility of this measure.

The more market relevant path is inflation and financial repression. The United States borrows in its own currency, so it does not face the classic emerging market problem of foreign currency debt. Its more likely adjustment mechanism is to reduce the real value of the debt over time.

The post-war example is the obvious template. Debt held by the public peaked near 106% of GDP after WW2 and a long period of negative real interest rates helped erode the burden while nominal growth did the rest.

The modern version of this would require real yields to remain low while inflation runs somewhat above target. This is especially important for bondholders, as a Treasury can pay on time in dollars and still deliver poor real returns if inflation erodes the purchasing power of those dollars.

The Case Against Panic

Japan is the obvious counterexample, having gross debt-to-GDP around 230% to 250% (well above the US ceiling) yet continuing to fund itself.

It is important to note however that on a net basis the ratio is considerably lower, as roughly 86% of the debt is domestically held and the Bank of Japan owns close to half the Japanese Government Bond (JGB) market, making much of the interest bill circular. Japan does not disprove the model, instead it shows that the ceiling sits higher when the buyer base is captive and the central bank dominates the market. The BOJ has gradually been reducing its bond purchases since 2024, with the 10-year JGB yield hitting 2.8% in May 2026 (a 29-year high) on a relatively modest reduction in monthly buying. That is exactly what the framework predicts, as debt capacity depends heavily on who owns the debt and it reprices quickly once that ownership shifts.

The dollar’s reserve currency status acts as the second counterargument. Treasuries remain the core safe asset of the global financial system, used as collateral, reserves and liquidity instruments across markets. The dollar’s share of allocated global reserves has fallen from the highs of the early 2000s, but it remains dominant and there is no rival at comparable scale. The euro is the closest alternative, while the renminbi remains too small and too constrained to replace the dollar in the global reserve system. That safe asset demand allows the United States to borrow more cheaply and at larger scale than a normal sovereign borrower, with no debt models fully capturing the political and institutional power of that privilege. Moreover, the 210% figure is an outer bound, not a forecast. It also assumes that markets will keep believing Congress will act, right up to the point where action becomes arithmetically impossible. In practice, that credibility erodes gradually, with the more likely path consisting of a slow repricing (higher term premia, less fiscal flexibility and a reassessment of how much real purchasing power a long-duration Treasury actually protects).

The more balanced synthesis is that reserve currency status, deep capital markets and the safe asset bid can defer the adjustment for a very long time. That is exactly why the risk is more likely to appear first as a gradual repricing rather than as a sudden event.

What This Looks Like in Markets

The fiscal year 2026 deficit is running near 5.8% of GDP, roughly $1.9 trillion on CBO’s estimate. Net interest has become one of the fastest growing lines in the budget. The United States paid about $970 billion in interest in 2025, roughly 3.2% of GDP and the highest share since 1991, while CBO projects net interest rising to $2.1 trillion by 2036 and moving toward 6.9% of GDP by the middle of the century.

Treasury is issuing large amounts of new debt while also refinancing older debt at higher coupons. Debt issued in the low rate years is gradually being rolled into a higher rate environment. Markets have responded in the way the framework would suggest at current debt levels (not with a run, but with a higher price for duration). The 10 year yield sits near 4.5% and CBO assumes yields will remain structurally higher than the 2010 to 2021 average.

Another important thing to note is that when Moody's downgraded the US' long-term issuer and senior unsecured ratings to Aa1 from Aaa in May 2025, Treasuries barely moved, as the market had already been pricing this trajectory for years. This muted reaction is also part of the signal, as the United States is not in a debt crisis, but the term premium is carrying more weight than it did a decade ago.

The buyer base also matters more, with PWBM assuming that foreign investors absorb about 40% of new Treasury issuance. If that falls to 20%, consistent with a more fragmented trade and capital flow regime, the closure year moves forward by two to four years and the real risk-free borrowing rate rises an additional 50 to 55 basis points over the horizon.

What Should Investors Take from This?

For investors, the point is not to sit around waiting for a debt crisis. They instead should focus more on what this debt path does to the assets people already own.

A government running deficits above 5% of GDP, while paying more each year in interest and issuing a heavy supply of Treasuries, makes it harder to believe that the long end of the curve simply drifts back to the pre-2022 world. The Fed can still control the front end, but the 10 year and the 30 year have to clear the market. If investors are being asked to absorb more duration, they are likely to demand a higher return for holding it.

That also changes how “risk-free” should be understood. For the US, the obvious risk is not that Treasuries suddenly stop paying coupons. The more realistic adjustment is that the debt is made easier to bear in real terms, through some mix of inflation, low real yields and financial repression. Nominal repayment and real purchasing power are not the same thing however and that distinction matters a lot more when the debt burden is large. The fragile point is therefore market confidence, not immediate solvency. The 210% ceiling may still be decades away under central assumptions, but markets do not need to wait for the endpoint before repricing the path. A weak auction, another fiscal standoff or a period of weaker foreign demand could be enough to push term premia higher well before the model says the arithmetic limit is reached.

There is also an uncomfortable link between fiscal risk and equity risk. Part of the debt runway depends on the size of the private capital base, which means elevated asset prices are doing some of the work. If equities reprice sharply, the government also loses part of the cushion that makes the debt stock easier to absorb. Treasuries may still rally in a classic flight to safety, but the hedge is less clean if the same shock also worsens the fiscal picture.

Therefore, the 210% figure is not a countdown to default, but rather a measure of how much fiscal space is left before the adjustment becomes impossible to do cleanly. Markets usually move before that point, only needing investors to demand more compensation for holding long term US debt in a world of persistent deficits, rising interest costs and less predictable demand.

Sources

• Penn Wharton Budget Model, “When Does Federal Debt Reach Unsustainable Levels? Spring 2026 – Onward,” June 4, 2026. https://budgetmodel.wharton.upenn.edu/p/2026-06-02-when-does-federal-debt-reach-unsustainable-levels/

• Congressional Budget Office, “The Budget and Economic Outlook: 2026 to 2036.” https://www.cbo.gov/publication/62105

• Congressional Budget Office, long-term outlook (summarized via CRFB, March 2, 2026). https://www.crfb.org/blogs/debt-rises-175-gdp-under-cbos-long-term-outlook

• Peter G. Peterson Foundation, “Interest Costs on the National Debt” (data through April 2026). https://www.pgpf.org/programs-and-projects/fiscal-policy/monthly-interest-tracker-national-debt/

• Cole, Harold L., and Timothy J. Kehoe, “Self-Fulfilling Debt Crises,” Review of Economic Studies 67 (1): 91–116, 2000.

• Blanchard, Olivier, “Public Debt and Low Interest Rates,” AEA Presidential Address, 2019.

• Moody’s Ratings, U.S. sovereign downgrade to Aa1, May 16, 2025. https://ratings.moodys.com/ratings-news/443154

• IMF, Currency Composition of Official Foreign Exchange Reserves (COFER), 2025Q3. https://data.imf.org/en/datasets/IMF.STA:COFER

• State Street Global Advisors, “Separating Fact from Fear in Japanese Investing” (Japan net vs. gross debt, ownership). https://www.ssga.com/us/en/institutional/insights/separating-fact-from-fear-in-japanese-investing

• U.S. 10-year Treasury yield and May 2026 payrolls, Trading Economics (as of June 6, 2026). https://tradingeconomics.com/united-states/government-bond-yield

That is the question Andrea Frazzini, David Kabiller and Lasse Pedersen aim to answer in their paper on Buffet’s Alpha1.

How much of Berkshire’s return came from stock selection, how much came from leverage and how much came from systematic exposure to factors that later became known as quality, value and betting against beta?

The answer proves to be more interesting than the myth.

The 19% return needs context

From 1976 to 2011, Berkshire generated an average excess return of 19.0% per year, compared with 6.1% for the broad US equity market.

Berkshire was however also volatile, with its annualised volatility close to 25%, compared with roughly 16% for the market. Thus, the relevant question should not just be how much return Buffett earned, but how much return he earned per unit of risk. On that basis, Berkshire still looks exceptional, as its Sharpe ratio was 0.76, almost double the market’s 0.39.

This does change the usual story, as Buffett is often described as a calm, low risk compounder, but the data show something more specific. Berkshire carried meaningful volatility, although it was rewarded unusually well for doing so. Its market beta was only around 0.68, meaning the company was less sensitive to broad market movements than the index itself, while still earning much higher excess returns.

The more precise measure of skill is the information ratio, which asks how much alpha a manager produces per unit of idiosyncratic risk. In regression terms, it takes the alpha from a market model and divides it by the standard deviation of the unexplained return (error term).

This is where Buffett’s record becomes harder to dismiss. Berkshire’s information ratio was 0.66. Frazzini, Kabiller and Pedersen compare it with every US stock and every actively managed equity mutual fund with at least thirty years of history. Berkshire ranks first on both Sharpe ratio and information ratio.

Over the longest horizon available and against surviving stocks and funds with similarly long records, no one produced a better risk adjusted performance.

The following chart plots the distribution of annualized information ratios across every common stock in the CRSP database with at least 30 years of return history. The arrow marks the information ratio of Berkshire Hathaway:

Public stocks, private holdings and the sources of return

Berkshire’s performance came mainly from two different sources: public equities and wholly owned private businesses.

The decomposition table makes the point clearer:

The public equity portfolio, visible through Berkshire’s 13F filings, delivered an excess return of 11.8% with a beta of 0.77 and an information ratio of 0.56. This was the more powerful return engine in risk adjusted terms and it reflected the listed stock selection that most people associate Buffett with (high quality companies bought at attractive prices and held for long periods).

The private business portfolio looked different. It generated a lower excess return of 9.6%, but with a much lower beta of 0.28. Its risk was mostly idiosyncratic rather than market driven, which makes sense for a collection of operating businesses whose cash flows were not mechanically tied to daily equity market movements.

Basically, the public side produced strong active returns, while the private side added a differentiated stream of cash flows with low market beta. The full 19.0% excess return, however, only appears once leverage is added on top.

The Leverage Question

Berkshire’s average leverage was about 1.6x, measured as assets relative to equity. That is important because leverage amplifies both return and risk. A manager who buys good assets with cheap borrowed money can improve equity returns substantially, provided the funding does not disappear at the wrong moment.

Still, leverage alone does not explain Berkshire’s record. Applying 1.6 times leverage to the broad equity market would have produced an excess return of roughly 10% per year. That is far better than the unlevered market, but still nowhere near Berkshire’s 19.0%.

The real question is why Berkshire could use leverage so cheaply, while most investors would face margin calls, refinancing risk or expensive funding costs.

The answer is insurance float.

Why Berkshire’s leverage was different

Insurance float is the part of Berkshire’s model that is hardest to replicate.

An insurance company receives premiums today and pays claims later. During the period between collection and payment, it can invest the money. Economically, the float behaves like borrowed capital. The cost of that borrowing depends on underwriting performance: if premiums exceed claims and expenses, the insurer is effectively paid to hold and invest policyholders’ money.

Berkshire’s underwriting discipline made it unusually cheap. Frazzini, Kabiller and Pedersen estimate that insurance float accounted for roughly one third of Berkshire’s liabilities, with an average cost of only 2.2% from 1976 to 2011. That was about three percentage points below the average Treasury bill rate.

The funding advantage is visible across the sample:

This is an extraordinary funding advantage, as Berkshire could run leverage without relying on fragile short term borrowing. Unlike margin debt, insurance float could not be pulled overnight by a broker and unlike normal corporate debt, it often cost less than the risk free rate. In some periods Berkshire was effectively being paid to borrow.

That funding structure is central to the Buffett record. Cheap leverage is useful, but cheap, long duration and stable leverage is much more powerful. It allowed Berkshire to hold positions through periods when other investors might have been forced to sell.

What was Buffett actually buying?

Lookin at a regression of Berkshire’s returns on six factors, the first four are familiar (market, size, book-to-value and momentum). The two additions, Betting-Against-Beta and Quality-Minus-Junk, do most of the work, and each is worth understanding on its own terms.

Berkshire loads positively on value, meaning it tends to own companies with attractive prices relative to fundamentals. It also loads positively on Betting Against Beta, meaning it favours safer, lower beta stocks. Same goes for Quality Minus Junk, meaning it favours profitable, stable and well managed businesses. At the same time, Berkshire has a negative loading on size, suggesting a tilt toward larger companies, and almost no meaningful exposure to momentum.

In plain terms, the regression says that Buffett bought large, cheap, safe and high quality companies, then applied cheap leverage to raise the overall return profile.

That sounds simple, but it is very different from how many investors try to beat the market. A constrained investor who wants higher expected return often buys higher beta stocks because they cannot use leverage. Meanwhile, Buffett could do the opposite, buying safer businesses and using Berkshire’s cheap funding to lever the portfolio.

That is exactly the intuition behind the Betting Against Beta factor. Many investors face leverage constraints, so they overpay for high beta stocks to increase expected returns. Lower beta stocks become relatively underpriced, and a portfolio that buys low beta assets and applies leverage can earn attractive risk adjusted returns.

Quality Minus Junk captures another part of the same philosophy. High quality companies are profitable, stable, less levered and more likely to return cash to shareholders. Markets however do not fully price those characteristics, as quality firms trade at a premium, but not as large a premium as their fundamentals would justify, so they can still generate superior risk adjusted returns.

This is the explanation behind Buffett’s famous shift away from buying cheap “cigar butt” stocks toward buying wonderful businesses at fair prices.

The alpha shrinks after controlling for factors

The most interesting result in the paper is what happens to alpha as more factors are added.

Under a standard four factor model, Berkshire’s alpha is large and statistically significant. Once Betting Against Beta and Quality Minus Junk are included, Berkshire’s alpha declines to 6.3% with a t-statistic of 1.58, which means it is no longer statistically distinguishable from zero.

If alpha shrinks once you control for factors, was Buffett simply harvesting risk premia that the literature has since identified?

That interpretation is too harsh, on two counts. First, these are largely risk and constrained-arbitrage premia, not a free anomaly (BAB pays precisely because most investors can't or won't lever low-beta stocks, so naming the factor doesn't relax the constraint or erode the return). Second, the factors were formalised long after Buffett had been applying the strategy, so the regression explains the source of his returns only in hindsight (it identifies them without diminishing the achievement).

Reading the regression table

Run the full regression and the philosophy is right there in the loadings:

Positive on value (HML), positive on safety (BAB), positive on quality (QMJ), negative on size (SMB) and essentially zero on momentum.

As Buffett said: “It is far better to buy a wonderful company at a fair price than a fair company at a wonderful price”.

As already said, under the standard four-factor model the alpha is large and clearly significant. Add Betting-Against-Beta and Quality-Minus-Junk and it becomes no longer distinguishable from zero. In the controlled, statistical sense of the word, Buffett’s alpha disappears.

Can it be replicated?

Because the style reduces to a set of factor exposures, we can reverse-engineer a “Buffett-style” portfolio. Strip out his passive market exposure, rebuild the active part from his factor tilts, then rescale it to match his idiosyncratic volatility before adding the market and the risk-free rate back on top:

The result is a diversified portfolio matched to Berkshire on beta, idiosyncratic risk, total volatility and relative factor loadings.

On paper it tracks the real thing well, but the catch is everything the regression can’t capture. Almost no one has access to negative-cost insurance float and even fewer have the temperament to hold a cheap-safe-quality book through fifty years of cycles without flinching.

So… what does the evidence actually say?

Buffett’s returns were extraordinary, but they were not mysterious. They came from buying cheap, safe, high quality businesses, using unusually cheap and stable leverage, and holding the strategy for long enough to let the premia compound. Once those exposures are controlled for, the unexplained alpha becomes much smaller, although the implementation remains exceptional.

In that sense, Buffett’s record shows that markets may be competitive, but they are not frictionless. Leverage constraints, behavioural limits, institutional pressures and funding costs create opportunities for investors who can operate differently from the crowd.

Leave a comment

Subscribe now

At first glance, this looks like the long awaited return of UK equities. The more interesting explanation is different, as the FTSE 100 is being rewarded for the same traits that investors ignored for years.

It has very little technology exposure and is heavily tilted toward banks, oil, miners, defence, pharmaceuticals and consumer staples. In a world of higher rates, energy disruption, European rearmament, commodity volatility and stretched US valuations, it suddenly looks useful.

The index is finally performing because the market environment has changed in favour of its old economy structure. The question is whether that change is durable, or whether this is just a rotation that fades once oil cools, geopolitical risk eases and US growth stocks recover their leadership.

The FTSE 100 is less British than people think

The first mistake is treating the FTSE 100 as a clean expression of the UK economy.

It is listed in London, quoted in sterling and described as Britain’s benchmark index, although its revenue base is mostly global. Around three quarters or more of FTSE 100 revenues are earned outside the UK, which makes the index much closer to a portfolio of global multinationals than a direct bet on the British economy.

UK Growth expectations for 2026 remain modest, consumer confidence has weakened and inflation is still awkward enough to complicate the Bank of England’s job. A strong FTSE 100 therefore does not automatically signal a strong UK economy. In some ways, the index can benefit from conditions that hurt UK households, with weak sterling boosting overseas earnings when translated back into pounds and higher oil prices squeezing consumers while supporting Shell and BP.

Moreover, rising defence spending reflects a more dangerous geopolitical environment, yet it also supports companies such as BAE Systems, Rolls Royce and Babcock.

Thus, the FTSE 100 can do well precisely because it is not a pure UK economy index, as it is a global, value heavy, commodity exposed and income oriented index that happens to sit in London.

The old economy suddenly matters again

The strongest part of the FTSE story as of right now is sector composition.

Defence is the clearest example. with European rearmament turning what used to be a quiet industrial theme into a structural investment case. Rolls-Royce, BAE Systems and Babcock are no longer peripheral names in the story of UK equities. They are part of the reason the index has started to look relevant again.

Energy has played a similar role. Shell and BP are not growth stocks in the Silicon Valley sense, but they generate cash when oil prices are high and energy security moves back onto the policy agenda. In a world where geopolitical risk keeps supply concerns alive, the cash flow profile of the oil majors becomes harder to ignore. Mining also adds another layer, given the FTSE meaningful exposure to companies such as Rio Tinto, Glencore, Anglo American, Antofagasta and Fresnillo.

Banks also fit the current regime better than they did during the low rate years. Higher rates improve net interest margins, especially for large international banks such as HSBC. The same rate environment that pressures households can support bank profitability, at least while credit quality holds up.

Cheap, although not without reason

The FTSE 100 remains much cheaper than the S&P 500, with a forward price earnings ratio around 13x, compared with roughly 21 to 22x for the S&P 500, while its dividend yield sits around 3.3% to 3.4%, against roughly 1.1% for the US benchmark.

That gap is large, although it has been large for years. Cheap markets can remain cheap when investors doubt their growth prospects, liquidity or earnings quality, and the UK has dealt with all three concerns.

Still, the discount becomes more interesting when the macro regime shifts. If investors want technology led growth at almost any price, the FTSE is the wrong index. If however they want income, cash generation, commodities, banks and defence exposure at lower multiples, the index becomes much more attractive. AJ Bell’s 2026 Dividend Dashboard forecasts a record £88 billion in ordinary dividends from FTSE 100 companies, above the previous 2018 peak, alongside £29.4 billion in announced share buybacks.

Nonetheless, we need to pay close attention to where these dividends come from, with a large share of dividends coming from a small group of companies, especially banks, energy and consumer staples. If commodity prices fall, banks face credit pressure or large payers reduce distributions, the income case becomes less comfortable.

That doesnt’t destroy the argument, but it does keep it grounded, as the FTSE is indeed cheap and income rich, although the cheapness reflects real structural questions.

The missing tech problem has not disappeared

The FTSE’s strongest recent feature is also its biggest long term weakness: it has almost no technology exposure.

That helped during a rotation toward value, commodities, defence and financials. It however risks becoming a problem if US technology earnings continue to dominate global equity returns. Over the past decade, the S&P 500 has dramatically outperformed the FTSE 100, helped by mega cap technology, deeper capital markets and stronger earnings momentum.

We also need to remember that the recent rally started from a weak base. A strong year can mark the beginning of a regime shift, but it can also be the rebound of a market that had already been priced at lower levels, meaning that the FTSE can perform well for a period without solving its deeper problem. It still lacks the high growth technology complex that drove global equity returns for much of the last decade. Banks, oil, miners and defence can generate strong cash flows, but they do not necessarily provide the same long term compounding profile as the best US growth companies.

That is why the FTSE rally should be taken seriously without being overstated.

London’s listing problem is the real counterargument

The strongest counterargument to the FTSE bull case is market structure, as London has been losing companies.

EY reported that 88 companies delisted or moved their primary listing away from the main market in 2024, while only 18 new IPOs came to London. Some companies cited weaker liquidity and lower valuations compared with other markets, especially the US, where capital pools are deeper and trading volumes are higher. If this trend continues, the FTSE 100 risks becoming even more concentrated in mature global incumbents. That can still work for income investors, but it weakens the idea that the UK market is becoming a broad growth opportunity.

The departures of companies such as CRH, Flutter and Ashtead point to the same problem. If companies believe they can secure higher valuations and better liquidity in New York, London may keep losing the very businesses that could improve its growth profile.

Who should actually own the FTSE 100?

As already stated, the FTSE 100 makes most sense for investors who want income, value exposure and global diversification away from US growth concentration.

It suits a view that commodities remain strategically important, defence spending has a longer runway, banks can still earn acceptable returns in a higher rate environment and sterling weakness may continue to support reported earnings. It also suits investors who believe US equity valuations leave little margin of safety and want exposure to cash generative companies at lower multiples.

On the contrary, it is less attractive for investors looking for pure growth, innovation exposure or a clean domestic UK recovery trade. The FTSE 100 is not really a bet on the British economy and it is not built around the sectors that drove global equity returns for most of the past decade.

That is why comparing it mechanically with the S&P 500 can be misleading. The S&P 500 is a growth heavy, technology driven global benchmark with premium valuations. The FTSE 100 on the other hand is a value and income oriented index with global revenues, commodity exposure and very limited tech. As such, these two answer different investor needs.

Regime change or rotation?

The FTSE 100’s rally makes sense in the current macro environment. Defence spending is no longer a short term political reaction, energy security has moved back to the centre of policy and commodity markets remain sensitive to supply disruption, infrastructure demand and geopolitical risk. Banks also look different in a world where rates are no longer pinned near zero, while investors are becoming more willing to question whether the valuation premium attached to US equities can keep expanding indefinitely.

The FTSE is full of companies that benefit from a world of higher nominal growth, harder assets, defence spending, dividends and cash flow discipline. If that world lasts, the index’s old weaknesses start to look less damaging. The alternative view is that this is still a rotation. Oil prices can fall, geopolitical risk can ease and central banks can still deliver a cleaner disinflation path than markets currently expect. In that environment, the same sector mix that helped the FTSE could become less attractive again, especially if earnings momentum slows or companies continue to seek deeper liquidity and higher valuations outside London.

The most reasonable view sits between those two outcomes. The FTSE 100 has become relevant again because the current environment rewards what it actually owns: cash flows, dividends, banks, defence, energy and real assets. At the same time, the index still carries the same structural weaknesses that made it cheap in the first place, from limited technology exposure to weak domestic growth and a listing market that keeps losing important companies.

Concluding, if the coming years continue to reward income, commodities, defence, valuation discipline and real assets, the FTSE may have more room to run. If leadership returns to cheaper energy, falling rates and US technology dominance, the rally may look more like a rotation than a lasting reset.

Either way, the FTSE 100 has become somewhat interesting again, which is already a meaningful change for an index that spent years being ignored.

Source list

Morningstar UK, FTSE 100 reaches 10,000 for the first time, January 2026.

The Times, FTSE 100 breaks 10,000 and records best annual gain since 2009, January 2026.

AJ Bell Dividend Dashboard, 2026 FTSE 100 dividend and buyback forecasts, March 2026.

EY, London stock market 2024 IPO and delisting review, January 2025.

RTE Business, London Stock Exchange delisting data, January 2025.

The Guardian, FTSE 100 10,000 milestone and 2025 performance drivers, January 2026.

Morningstar UK, FTSE 100 outlook and potential path to 11,000, February 2026.

Morningstar UK, UK market opportunities and valuation commentary, May 2026.

Averages, however, can be misleading. They smooth over the differences that often matter most at turning points. An economy can show positive growth, low unemployment, rising household net worth and resilient consumption while a large part of the population is becoming more financially fragile. This is one of the defining features of a highly financialized economy: wealth, asset ownership, income and debt are not distributed evenly, so the households that move asset prices are often not the same households that determine the durability of the real economy.

That distinction matters, as financial markets are increasingly shaped by the balance sheets of wealthy households and institutional capital, while the business cycle is still exposed to the condition of the marginal consumer, the household closest to the constraint, the one most sensitive to rent, food, energy, insurance, interest costs, credit availability and employment conditions. Once these consumers weaken the economy does not break immediately, on the contrary, spending can continue for a while because households still need to cover non-discretionary expenses. They draw down savings, use credit cards, delay larger purchases, refinance where possible and keep consuming because many costs cannot simply be avoided.

Stress can thus appear in lower savings rates, rising revolving credit, elevated auto-loan stress, student-loan defaults and pressure in the earnings of consumer-facing companies. By the time the weakness is obvious in headline data, markets may already be late.

The “average consumer” does not really exist

A large part of traditional macroeconomic analysis still relies, explicitly or implicitly, on the idea of a representative consumer. That can be useful inside a model, but it becomes much less helpful once we move into the real economy.

One household may own stocks, real estate, treasury bills and money-market funds, while another may be dealing mainly with rent, a car loan, credit-card balances and higher grocery bills. They are both part of the same economy, but they are not exposed to it in the same way. As a consequence, if equity markets rise, the gains flow mostly to households that already own financial assets.

When interest rates rise, the pressure falls more heavily on households that depend on borrowing. When inflation hits essentials such as rent, food, energy or insurance, the burden is much harder to absorb for households with lower disposable income. The same macro event can therefore strengthen one balance sheet while weakening another. That is why asset prices should not be treated as a neutral mirror of the whole economy, as they are shaped disproportionately by the people and institutions that own such assets.

The Federal Reserve's Financial Accounts reported that household and nonprofit net worth reached $184.1 trillion in Q4 2025 (the households-only figure, excluding nonprofits, was approximately $175 trillion). On paper, that is an enormous figure, but aggregate wealth says little about how resilient the average household actually is. The Federal Reserve's Distributional Financial Accounts make the same point clearly: the bottom half of households owns only a very small share of corporate equities and mutual fund assets, while the top wealth groups hold a dominant share of financial assets. Data from World Inequality Report 2026 shows a similar trend with respect to wealth ownership.

A bull market can therefore create a powerful wealth effect for the top percentiles while doing very little for households whose main economic asset is labor income. For asset owners, rising markets increase wealth, collateral and investment income, whereas for lower- and middle-income households, rising prices in rent, food, energy and borrowing costs reduce the cash left after necessities. This is the structural divergence at the center of the argument. The stock market can remain optimistic because the owners of stocks are still in good shape, even as the broader consumption base becomes more stretched.

Wealth Concentration Matters for Markets

Wealth concentration is often discussed as a social issue, but for investors it also matters as a market mechanism.

When financial assets are concentrated near the top of the income and wealth distribution, markets become more sensitive to the liquidity, confidence and allocation decisions of that group. Wealthier households generally spend a smaller share of each additional dollar they receive and have more capacity to invest it. Dividends, capital gains, interest income, tax cuts and asset appreciation are therefore more likely to be recycled into financial markets rather than spent immediately on goods and services.

This creates a reinforcing loop where rising asset prices increase the wealth of asset owners, their portfolios generate more income and part of that income can be reinvested into equities, bonds, private assets, real estate or money-market funds. Capital can keep circulating near the top of the system even while the ordinary consumer is under pressure. This helps explain why markets can sometimes feel detached from everyday economic reality. It is not always irrationality. Often, it is simply a reflection of who owns the assets. The S&P 500, for example, is not a broad survey of household financial health. It is a capitalization-weighted index of large public companies, owned largely through institutions, pension funds, mutual funds, ETFs, foreign investors and wealthier households. It can rise because margins are strong, buybacks continue, AI capex accelerates, discount-rate expectations improve or global capital flows into U.S. assets. None of that requires the median household to feel financially comfortable.

Housing works in a similar way. Rising home prices increase the wealth of existing owners, but they also worsen affordability for renters and first-time buyers. What appears as wealth creation on one side of the balance sheet appears as a higher entry cost on the other. That is the distributional problem inside modern macro.

Resilience or Leverage?

Over the last few years, one of the dominant market narratives has been the resilience of the consumer and there is some truth to that, as spending has not collapsed and nominal consumption has continued to grow.

The question is what kind of resilience we are looking at. Resilience funded by rising real incomes is very different from resilience funded by thinner savings and more expensive debt. The BEA reported that the U.S. personal saving rate fell to 3.6% in March 2026, near the lower end of its post-COVID range and matching levels last seen in mid-2008 (though it had fallen as low as 2.2% in June 2022). What matters is the direction, with the partial recovery in household savings built between 2023 and early 2024 is now being unwound.

However, the conditions driving it differ from that period: spending in March appears to have been partly front-loaded ahead of tariff-driven price increases, rather than reflecting recession-era contraction. It is important to note that the level of the buffer is the same and that matters.

The same March data also showed a clear gap between nominal and real spending, with the current-dollar PCE rising by 0.9%, while real PCE increased only 0.2%. Part of the "strong spending" story was therefore price-driven rather than volume-driven, and the monthly PCE price index rose 0.7%, its fastest pace since 2021. That matters for both companies and households. A business may report higher nominal revenues even if consumers are not buying much more in real terms, while the marginal consumer can appear resilient simply because essentials cost more. Inflation in rent, food, energy or insurance can keep nominal spending elevated while real purchasing power deteriorates. Moreover, credit-card balances have risen roughly $482 billion since their pandemic low in Q1 2021, reaching a record $1.277 trillion in Q4 2025 before dipping seasonally to $1.252 trillion in Q1 2026 (the quarterly decline is a normal pattern, not a reversal). In addition, total household debt reached $18.8 trillion in Q1 2026.

So the "strong consumer" story becomes harder to read. If households are spending because real incomes are rising, savings are healthy and debt burdens are manageable, demand is on firmer ground. If, on the other hand, spending is being supported by lower savings and higher credit balances, the economy can look stronger than the underlying balance sheet really is. Debt-financed consumption can keep growth alive in the short run, but it also leaves the system more fragile later.

The Timing Problem

It is also worth addressing the timing problem directly before going further. A fragile consumer does not automatically mean an imminent recession, as markets can remain optimistic for longer than the data seems to justify.

There is also a real counterargument in the recent credit data, with stress not deteriorating uniformly across all categories. Credit-card and auto-loan delinquency transitions stabilized at elevated levels in Q1 2026 rather than worsening sharply. The New York Fed's own researchers describe delinquency transition rates as "mostly steady" and note that most households remain on relatively stable footing, while student loans are the clearest pressure point, but the broader credit system has not broken. That makes the argument more about vulnerability than timing a crash.

Distributional fragility increases the risk that markets are overestimating the durability of growth, especially when valuations are high and credit spreads are compressed.

Student Loans and Signs of Strain

The clearest recent example of stress showing up outside the usual headline data is the student-loan market.

Roughly 1 million federal student loan borrowers entered default in Q4 2025, the first quarter in which new defaults began appearing on credit reports, given that missing 270 days of payments is required for a federal loan to enter default status. An additional 2.6 million borrowers defaulted in Q1 2026, bringing the two-quarter total to approximately 3.6 million. The 90-plus-day delinquency rate on student loans rose to 10.3%, up from 9.6% in Q4 2025.

Federal collection efforts (wage garnishment, tax refund seizures, Social Security offsets) remain suspended for now with no clear timeline for resumption. The New York Fed researchers also flag a potential second wave, as roughly 7 million borrowers in the now-defunct SAVE repayment plan remain in forbearance, and as they reach the nine-month repayment threshold, a further round of defaults becomes plausible.

The New York Fed notes that borrowers who have defaulted tend to have elevated delinquency rates across other credit product and that their financial struggles likely extend beyond student loans. At the same time, the same researchers conclude that spillover to broader credit markets is "likely to be limited," partly because student-loan borrowers represent a relatively modest share of overall credit usage. That counterargument deserves equal weight.

Still, consumer stress rarely appears everywhere at once. It usually starts in pockets, hits specific cohorts first and shows up in less-followed segments before it becomes obvious in aggregate data. Student-loan defaults show how fragile balance sheets can begin to surface after a long delay, and when collection efforts eventually resume, the pressure on disposable income will become visible in the data.

Debt redistribution and Cash Flows

Household debt is often described as a way to pull future consumption into the present. That is true, but it misses the other side of the balance sheet: every liability for the borrower is an asset for someone else. The interest paid on credit cards, auto loans, mortgages and consumer finance products becomes income for lenders, banks, bondholders and investors.

That creates a distributional transfer inside the economy. Income moves away from households that are more likely to spend each additional dollar and toward institutions or asset owners that are more likely to save, invest or recycle it back into financial markets. The same debt burden that weakens the marginal consumer can therefore support financial-sector profits and investment income.

The borrower's constraint becomes the lender's income, while the consumer's reduced future spending power becomes current cash flow for the financial system. This is one of the more overlooked mechanics of modern macro cycles, as debt does not simply disappear into the economy. There is this reallocation of income across balance sheets, and while that can support asset prices in the short term, it can also weaken the consumption base that corporate revenues ultimately depend on.

Higher rates, uneven effects

Interest rates work through the same distributional channel. In a simplified macro model, higher rates reduce demand by making borrowing more expensive and saving more attractive. In practice however, the effect depends heavily on where a household sits on the balance-sheet spectrum.

Lower-income households are more likely to rent, borrow and rely on expensive forms of credit. They are more exposed to auto loans, credit-card balances and the need to refinance or borrow at higher rates. For them, tighter monetary policy shows up directly in monthly cash flow. Wealthier households are in a different position, as higher rates still tighten financial conditions overall, but the income effect is not the same across households.

The current environment makes this more complicated because inflation is not purely demand-driven. March 2026 data showed headline PCE inflation at 3.5% year-over-year and core PCE at 3.2%. Note however that the Dallas Fed estimates that tariff effects added around 0.80 percentage points to core PCE in March, and absent those effects, core inflation would have been closer to 2.3% (much nearer the Fed's 2% target).

Federal Reserve Board staff research reaches a similar conclusion. This means a significant portion of the current inflation reading reflects a policy-driven supply shock rather than excess consumer demand, which changes the character of the monetary policy trade-off substantially.

Thus, growth does not need to collapse for households to feel squeezed. Cutting rates too quickly risks loosening financial conditions while inflation remains above target, while keeping rates elevated prolongs the squeeze on households already carrying expensive debt.

Are markets reacting late?

Markets do not price current conditions as much as expectations about future earnings, margins, default risk, liquidity, inflation, policy rates and discount rates.

Mispricing appears when those expectations become too optimistic relative to the underlying structure of the economy. Consumer stress is especially easy to misread because it usually builds gradually, with households first absorbing pressure by reducing savings, then by relying more on credit, and eventually by rotating obligations (using one form of debt to manage another). Discretionary spending tends to weaken later, and defaults usually rise only when the labor market softens or credit availability tightens. That early phase can look deceptively stable, as spending continues, GDP holds up, credit spreads stay tight and equity markets keep moving higher. The adjustment becomes faster only when the stress begins to affect corporate earnings, credit performance or refinancing conditions.

High-yield spreads often remain compressed late in the cycle because recent default data still looks manageable and investors continue reaching for yield. In late May 2026, the ICE BofA U.S. High Yield Option-Adjusted Spread was around 2.75–2.80%.

That does not mean credit has to reprice immediately, rather it means that investors are receiving limited compensation for the possibility that consumer weakness eventually shows up in margins, revenues, defaults or refinancing risk. The real danger is that markets may be pricing a consumer that looks stronger in aggregate than the marginal consumer actually is.

Post 2008 and the Consumer Balance Sheet

The years after the 2008 financial crisis are a useful example of this dynamic. By 2010 and 2011, many headline indicators had started to improve. GDP was recovering, financial markets had stabilized and unemployment would eventually move lower.

From a distance, the economy looked as if it was slowly returning to normal. Household balance sheets told a slower story however, with many consumers still dealing with mortgage losses, lost home equity, weak wage growth and the need to deleverage after the previous credit boom. The private sector was still repairing its balance sheet, which made a normal rate cycle much harder to sustain than headline data alone suggested.

That helps explain why interest rates stayed lower for longer than many expected. The economy could not easily absorb rapid monetary normalization while households were still rebuilding after the crisis. Investors who understood that were better positioned for the post-crisis regime, as long-duration government bonds benefited from persistently low rates, growth equities benefited from lower discount rates and gold found support from concerns about monetary expansion, currency debasement and financial-system fragility. The broader point is that the balance sheet mattered more than the headline recovery.

Post-COVID

The post-COVID period showed a different version of the same divergence. The policy response to the pandemic injected enormous liquidity into the economy.

At first, fiscal transfers, unemployment benefits and monetary stimulus supported households broadly. Over time, though, the asset-price response benefited households that already owned financial assets much more than those that did not. Wealthier households also built savings because their spending opportunities were temporarily limited. Travel, restaurants and services were restricted, while portfolios benefited from rising markets. For lower-income households, the picture was less comfortable. The initial fiscal support helped, but the later inflation shock hit essentials such as rent, groceries, gasoline, utilities, insurance. That is how the post-COVID economy could produce strong equity markets and visible consumer pressure at the same time.

The rallies in equities and gold were not as contradictory as they may have looked. Equities reflected liquidity, earnings resilience, mega-cap concentration and expectations of productivity gains. Gold reflected inflation uncertainty, low or negative real rates at different points, geopolitical risk and concerns about the purchasing power of money.

What investors should actually watch

If the marginal consumer matters, investors need to look beyond the usual headline indicators. GDP growth and unemployment still matter, but they are not enough. The more useful signals are found in household balance sheets and credit conditions (personal saving rate, credit-card balances, student-loan defaults, auto-loan delinquencies, real wage growth, rent burden, consumer loan charge-offs, lending standards and high-yield spreads). Consumer-facing earnings are also important because stress often appears there before it is visible in aggregate data. Retailers, restaurants, travel companies, payment networks, banks, auto lenders and consumer-credit firms can show early signs of pressure through weaker volumes, softer guidance, higher provisions or changes in customer behaviour.

The key is to separate spending from spending capacity. The practical market question is not just whether the consumer is still spending, but whether that spending is being funded by income, savings or debt.

Conclusion

Modern macro analysis needs to take distribution more seriously. In a high-inequality, asset-heavy economy, markets can remain strong while the consumption base weakens gradually. Wealthy households and institutional investors can support asset prices, while the marginal consumer keeps spending by reducing savings and taking on more debt. The New York Fed itself has used the phrase "K-shaped economy" to describe what is visible in the current data.

The better macro question is whether the marginal consumer still has enough income, savings and credit capacity to keep absorbing shocks.

Sources

Federal Reserve, Financial Accounts of the United States, Z.1 Release, Q4 2025.

Federal Reserve, Distributional Financial Accounts.

Bureau of Economic Analysis, Personal Income and Outlays, March 2026.

Federal Reserve Bank of New York, Quarterly Report on Household Debt and Credit, Q1 2026.

FRED / ICE BofA, U.S. High Yield Option-Adjusted Spread.

Dallas Fed, research on tariff effects on PCE inflation.

Bank for International Settlements, research on household debt and macroeconomic effects.

Federal Reserve Bank of Minneapolis, research on the K-shaped consumer.

European Central Bank, research on household saving dynamics and monetary-policy transmission.

BlackRock, 2026 investment outlook and AI-related capital expenditure commentary.

J.P. Morgan Asset Management, 2026 Year-Ahead Investment Outlook.