The Diversification Paradox
Traditional portfolio hedging is breaking down. The culprit isn't a macroeconomic shock—it's the sheer scale of AI capital expenditure.
Gabriela Santos, J.P. Morgan Asset Management's Chief Market Strategist for the Americas, made a startling admission in a recent CNBC appearance: finding genuine diversification away from the AI trade is becoming nearly impossible. Her team built an "AI factor basket" to test this hypothesis. The results were sobering—most assets now move in sync with the broader AI trade, whether their holders realize it or not.
Summer's momentum unwind hit AI-related equities hardest in July, with effects rippling into August. Yet the deeper structural story isn't the pullback. It's the architecture of correlation that the AI buildout has created—one that renders traditional portfolio construction increasingly obsolete.
The bytecode of modern markets is rewriting itself. AI capital spending has become a systemic factor, and most investors are still trading like it's a sector theme.
Beyond the Sector Thesis
Let me be precise about what's happening here. AI capital expenditure isn't a single-industry phenomenon anymore. It's a macro factor—comparable in scale and scope to China's industrialization in the 2000s or the shale oil revolution of the 2010s. When an investment theme grows large enough to influence essentially every asset class, from public equities to fixed income and private markets, it stops being a trade and becomes an environmental condition.
Santos's key insight cuts against the prevailing narrative. She isn't bearish on AI. Her point is more subtle and more challenging: you can be very, very bullish on AI and still need to think very, very carefully about portfolio construction. That distinction matters. It separates investors who understand the nature of correlation risk from those who will discover it the hard way.
The problem is structural. During my years auditing Layer 2 protocols and DeFi systems, I learned that when a single variable dominates a system's behavior, diversification becomes illusory. You can hold twenty different tokens, but if they're all exposed to the same underlying vulnerability—say, a shared dependency on Ethereum's base layer—you haven't diversified at all. You've just multiplied your exposure to a single point of failure.
The AI trade has reached that point in traditional markets. The old industry groupings—hyperscalers, chipmakers, software companies—no longer move as cohesive units. The internal differentiation is accelerating. Some companies within these sectors are executing on AI monetization effectively; others are burning capital with no clear path to returns. Basket-buying an AI index is no longer a strategy. It's a gamble on beta that may not exist.
The AI Factor: A New Correlation Regime
J.P. Morgan's AI factor basket testing reveals something uncomfortable for traditional asset allocators. Most assets now exhibit synchronized movement with the broader AI trade. The diversification that once came from holding different sectors, different geographies, different asset classes—much of it has evaporated.
What remains genuinely non-correlated? The list is short: Treasuries, gold, core real estate, and European equities. That's it. A handful of asset classes that sit far enough from the AI capital expenditure chain to retain some independent pricing dynamics.
This isn't a judgment about AI's long-term value. It's a statement about market structure. When a single investment theme becomes large enough, it creates its own gravitational field. Assets that seemed unrelated begin to orbit the same center of mass. The AI factor becomes a hidden variable in every portfolio, whether the investor recognizes it or not.
The fixed income dimension makes this more complex. The stock-bond correlation has broken down in ways that challenge the traditional 60/40 portfolio logic. As inflation and interest rates fluctuate, capital competition returns—AI capital expenditure competes with everything else for the same pool of investable funds. This isn't a temporary technical condition. It's a structural shift in how risk propagates through the financial system.
We didn't see the correlation regime shift coming. Neither did most institutional investors. The question now is what the new architecture demands.
The Hidden Exposure Problem
Here's what the standard analysis misses. The AI factor isn't just affecting assets that are obviously AI-related. It's penetrating everything—through energy costs, through data center demand for electricity, through cloud computing price structures, through the cost of capital itself. Infrastructure demand ripples outward in ways that are difficult to trace.
Consider the transmission channels:
- Hyperscalers are committing massive capital to data center construction, affecting construction costs, power infrastructure, and regional economic development.
- Chipmakers face order visibility that's increasingly divorced from end-user demand, creating inventory cycles that echo through the supply chain.
- Software companies are being forced to integrate AI capabilities, changing their cost structures and competitive positions.
- Every other industry is absorbing AI's effects through productivity changes, competitive pressure, and shifting customer expectations.
The result is that "AI exposure" is no longer a property of specific holdings. It's a property of the entire portfolio. And most investors don't have the tools to measure it.
During my research into DeFi protocols in 2020, I saw a similar pattern. The yield farming mania created a web of correlated positions across protocols that looked diversified on the surface—different platforms, different tokens, different mechanisms. But when the liquidity crunch hit, everything fell together. The correlation that mattered wasn't between individual positions. It was between the entire ecosystem and its shared dependencies.
Volatility is noise. Architecture is the signal. The architecture of AI capital expenditure creates a specific kind of systemic risk: a simultaneous exposure that diversification cannot mitigate because it isn't a sector risk—it's a factor risk.
The False Comfort of Indexing
The traditional response to uncertainty is broad diversification. Own everything, the logic goes, and you'll capture the winners without betting on any single loser. The AI factor breaks this logic.
When J.P. Morgan's testing shows that most assets move in sync with the AI trade, the implication is clear: owning a broad index doesn't diversify away from AI exposure. It concentrates it. The index itself is saturated with the factor.
This creates a genuine dilemma for institutional investors. The assets that would provide genuine diversification—Treasuries, gold, core real estate, European equities—aren't necessarily attractive on their own merits. They're attractive because they're non-correlated. That's a different investment thesis from buying them for their intrinsic value. It's a portfolio construction thesis, and it requires a different kind of conviction.
The hidden implication is that these "safe haven" assets may be repriced as investors increasingly recognize their scarcity value. Gold, for instance, is no longer just an inflation hedge. It's a hedge against AI correlation risk. That's a different demand driver, and it may persist even if inflation moderates.
Rethinking Risk Budgets
The challenge for investors isn't deciding whether AI is a good investment. It's deciding how much AI exposure the portfolio can tolerate, and then finding assets that genuinely reduce that exposure.
Santos's framework—control position sizing, manage leverage, maintain diversification—sounds simple. But in a regime where most assets carry hidden AI exposure, implementation is far from straightforward. The first step is recognizing that the problem exists. The second is building the tools to measure and manage it.
A few signals worth tracking:
- Hyperscaler capital expenditure guidance — the single most important indicator of whether the AI buildout continues or decelerates
- AI factor basket correlation data — updated quarterly, revealing whether diversification is becoming harder or easier
- Real interest rates and gold price dynamics — validating whether non-correlated assets maintain their scarcity premium
- Intra-sector dispersion — rising dispersion within AI-related industries signals that stock selection is replacing beta as the alpha source
- Stock-bond correlation — a return to negative correlation would partially restore traditional portfolio construction logic
The architecture of the AI trade will determine whether the next market downturn is contained or systemic. Capital expenditure guidance is the load-bearing wall.
The Regulatory Dimension
One aspect that deserves more attention than it typically receives: the regulatory implications of factor concentration. If AI capital expenditure has indeed created a systemic factor, then regulators and policymakers need to understand this new architecture.
The summer momentum unwind offered a preview. When AI-related equities sold off, the effect rippled through the broader market. A larger, more concentrated unwind could have consequences that extend well beyond equity markets. Institutional investors heavily exposed to the AI factor through multiple asset classes could face simultaneous margin calls and forced selling across their portfolios.
This isn't a prediction of crisis. It's a statement about the changed nature of risk. The 2008 financial crisis was fundamentally about correlated exposure to a single factor—real estate—hiding in multiple asset classes. The possibility that AI capital expenditure creates a similar dynamic cannot be dismissed out of hand.
My experience auditing smart contracts taught me that every protocol has assumptions embedded in its design. The same is true of financial markets. The assumption that diversification works is embedded in portfolio construction logic. When that assumption breaks, the consequences propagate through the system.
Positioning for a New Correlation Regime
The most important takeaway from Santos's analysis is that AI capital expenditure has transformed the fundamental architecture of market risk. Traditional diversification logic assumes that different assets respond to different drivers. That assumption no longer holds when a single factor—AI investment—drives outcomes across most asset classes.
This isn't a bearish view of AI. It's a recognition that the AI trade has grown so large that it now functions as a macro factor. And macro factors require macro responses: portfolio-level risk management, not sector-level position adjustments.
The investors who navigate this regime successfully will be those who measure their AI exposure at the portfolio level, not the holding level. They'll allocate to non-correlated assets based on their correlation properties, not just their expected returns. And they'll recognize that diversification is a risk management tool, not a return enhancement strategy.