Goldman Sachs’ latest report proposes a key risk framework for AI investments, distinguishing between 'aggregate shocks' that threaten the total value of AI and 'distributional shocks' that alter the allocation of winners and losers. Five market events over the past 18 months illustrate that aggregate shocks triggered broad market declines, spikes in the VIX, and widening credit spreads; in contrast, during distributional shocks—such as Google’s equity offering and Meta’s cloud business developments—the S&P 500 remained unchanged while individual stocks experienced significant turnover. Currently, AI stock volatility sits at the 99th percentile of its 15-year range, while implied correlation has fallen to a historic low, indicating that distributional divergence is dominating the market. Broad-based exposures can hedge aggregate risks using macro tools, whereas AI-specific exposures face challenges in hedging against distributional shocks.
Over the past month, semiconductor stocks have come under collective pressure, and hedging AI-themed investments has once again become a focal point. However, a more fundamental question is often overlooked in these hedging discussions: what kind of AI risk is the market actually concerned about?
According to Chasing the Wind Trading Desk, Goldman Sachs’ Global Markets Commentary published on July 23 provided an analytical framework: AI investments face two fundamentally distinct types of risk. One threatens the size of the pie—the total value created by AI shrinking; the other alters how the pie is sliced—total value remains unchanged, but winners and losers are reshuffled.
These two types of risk transmit through cross-asset markets via distinctly different channels and require entirely different hedging instruments. Although strong earnings reports may temporarily outweigh valuation concerns and the AI investment frenzy is expected to continue, Goldman Sachs still advises investors to closely monitor potential challenges—the key lies in identifying the nature of those challenges.
USD 26 trillion vs. USD 9 trillion
This distinction is urgent because AI valuations are becoming increasingly reliant on optimistic assumptions.
Since November 2022, the market capitalization of AI-related equities (including private companies) has grown by approximately USD 26 trillion, or roughly USD 23 trillion after adjusting for baseline returns. Under Goldman Sachs’ baseline assumptions, the present discounted value (PDV) of capital income that U.S. corporations could capture from AI-driven productivity gains is only about USD 9 trillion. Even under the most optimistic scenario—combining higher productivity growth, faster adoption rates, and a larger capital share—this figure rises only to approximately USD 28 trillion.
In other words, the current market capitalization increase in AI stocks is already approaching the upper bound that would be justified only if all optimistic assumptions were perfectly realized.
In Goldman Sachs’ framework, five variables drive the total value of AI: productivity growth, adoption speed, capital share (corporate monetization capability), international share, and discount rate. A downward revision in expectations for any one of these variables would shrink the pie. Macro shocks—such as monetary tightening, rising oil prices, or weakening employment—can also transmit through profit expectations and discount rates. Given the high valuations, concentrated positioning, and substantial financing needs of AI stocks, such macro shocks could disproportionately impact the AI sector.
An easily overlooked boundary: here, the “pie” refers specifically to the AI-derived value accruing to U.S. corporate sectors. Even if the total global economic value of AI remains unchanged, any reallocation of value away from U.S. corporations toward consumers or non-U.S. producers would effectively shrink the pie from the perspective of U.S. equities.
The allocation landscape is already shifting. To date, AI-generated value has primarily flowed to U.S. AI firms and select key Asian enterprises. Over the past 6–9 months, the market has particularly rewarded the supply side—infrastructure providers such as semiconductor and memory chip manufacturers. Since late 2025, memory chips have surged in price due to accelerated shortages driven by AI demand, representing essentially a terms-of-trade shock: chip producers benefit while consumers suffer. As the pie grows, the way it is sliced is also changing rapidly.
The same event, two entirely different shocks
How can one determine whether an AI-related risk pertains to the 'size of the pie' or the 'division of the pie'?
Some judgments are intuitive. Slower adoption and limited use cases—classic signs of a shrinking pie. Declining market willingness to finance AI projects, driving up discount rates—also indicative of a smaller pie.
But many situations are far less clear-cut.
Is it difficult for AI products to monetize, or is model competition compressing innovation costs? If returns shift from firms to consumers, the slice held by U.S. corporations shrinks—but consumers benefit, and lower costs may even accelerate adoption. What about expanded semiconductor capacity or improved chip efficiency? Producers may suffer while consumers gain; the total size of the pie might remain unchanged, but its distribution shifts. Could AI disrupt traditional industries? At its core, this is about redistribution—but if the winners aren’t listed in U.S. equity markets, the overall pie may still contract.
Goldman Sachs offers a key example: hyperscale cloud providers cutting capital expenditures.
The same action—if driven by pessimism about AI investment returns or tighter financing conditions—signals a shrinking pie; if driven by the realization that existing infrastructure is sufficient or by discovering more efficient utilization methods—it leaves the pie unchanged, merely reallocating a slice from suppliers to the cloud providers themselves.
Different origins lead to entirely different market outcomes.
The market has already delivered its verdict
Five market events over the past 18 months have clearly illustrated this distinction.
First, consider three aggregate shocks—the DeepSeek event (January 2025), broad-based AI concerns (February 2026, centered on doubts about the sustainability of high capital expenditure), and a non-farm-payrolls-driven surge in interest rates (June 2026). Market reactions were highly consistent: the S&P 500 fell by 1.5%, 1.7%, and 2.6% respectively; the VIX spiked by 20.5%, 20.9%, and 39.7%; AI-related stocks and semiconductors significantly underperformed the broader market; credit spreads widened; defensive equities rose against the trend; and the 10-year U.S. Treasury yield declined by 9 basis points in the first two episodes. In the third episode, interest rates themselves were the source of the shock, resulting in an opposite directional move. The first two episodes aligned with downward revisions to growth expectations, while the third reflected a hawkish policy shock.
Next, examine two allocation shocks—Google’s announcement of a share issuance to finance AI-related capital expenditures (June 2) and Meta’s announcement of launching a cloud business to sell excess computing capacity (July 1).
The picture was strikingly different. The S&P 500 barely moved, U.S. Treasury yields were unchanged, and the VIX showed no reaction.
Yet beneath the surface, turbulence was intense: on the day of Google’s announcement, semiconductors rose 5.8% while hyperscale cloud providers fell 2.4%; in the case of Meta’s announcement, the move was entirely reversed—hyperscale cloud providers gained 2.5% while semiconductors dropped 6.4%. Winners and losers switched dramatically, offsetting each other at the index level and leaving macro assets nearly unaffected.
The classification of the DeepSeek event warrants special clarification. Breakthroughs of this kind, which lower the cost of innovation, may reallocate value to consumers (including corporate consumers) but simultaneously reduce the overall share of AI-related profits accruing to the U.S. corporate sector—globally, it changes how the pie is sliced; for U.S. equities, it shrinks the pie itself. Hence, it is categorized as an aggregate shock.
Volatility data tells the same story. The average implied volatility of S&P 500 constituents currently sits at the 99th percentile over the past 15 years, yet implied correlation has fallen to its lowest level in 15 years—indicating sharp dispersion among individual stocks, even as consensus on the overall valuation of AI remains intact. Over the past six to nine months, aside from macro events such as the Iran conflict, only aggregate shocks have pushed up both implied correlation and index-level volatility.
The more dominant allocation shocks become, the less pressure they exert on index-level volatility.
Who can hedge—and who cannot
Investors holding broad-based U.S. equity exposure face aggregate shocks as their primary threat. Fortunately, aggregate shocks follow clear macro transmission channels—the equity index and interest rates react most sensitively, while signals from FX and commodities are weaker. Except when interest rates themselves are the risk source, AI-related aggregate concerns typically push U.S. Treasury yields lower. Macro hedging instruments are available, and cross-country and cross-sector diversification can absorb volatility arising at the allocation level.
Investors with AI-specific exposures or overweight positions face a more challenging situation. While aggregate risk can still be hedged macro-wise, allocation shocks also impact their portfolios but cannot be reliably protected against using macro assets—since allocation shocks cancel out at the macro level, and correlations between other assets and core AI holdings are unreliable.
If you only hold broad market positions, the size of the pie is your concern—not how it’s sliced. But if you’re heavily invested in AI, you’ll be exposed to both risks—and the latter is harder to hedge against.
Edited by Jeffy