A leading trader from Goldman Sachs has recently cautioned that the simplistic trading logic previously surrounding the AI theme is gradually unraveling, with market fragility on the rise.
In a recent report, the firm's trader Lee Coppersmith pointed out that investors are beginning to tap the brakes on AI trades, primarily for three reasons: first, a need for market adjustment following significant prior gains; second, heightened concerns over accumulated leverage and excessively concentrated positions; and third, the market is starting to reassess the potential value of cyclical opportunities outside of AI, especially against a backdrop of easing geopolitical risks, declining oil prices, and reduced interest rate pressures.
Coppersmith simultaneously emphasized that this does not equate to a blanket bearish view on AI. Semiconductors remain the most direct beneficiaries within AI infrastructure build-out, demand from hyperscale cloud providers remains robust, and overall AI earnings revisions are still resilient. However, the market is shifting from a crude "buy all AI" strategy to a more discerning, selective phase, which will have a substantive impact on asset allocation thinking.
Overall, the broader AI narrative remains largely intact, but internal divergence is intensifying, significantly elevating the importance of sector selection. Goldman Sachs is closely monitoring whether this divergence will widen further and if capital will rotate to areas outside of AI. For investors, the execution risk of a broad-brush thematic strategy is rising, with the key to future alpha lying in the meticulous selection based on position in the supply chain, earnings delivery capability, and valuation.
Intensifying Return Dispersion Signals a Shift to Selective AI Plays
Coppersmith noted that the most significant recent change is the severe divergence in returns within the AI theme. Data shows the standard deviation of returns for a basket of AI-related stocks surged to 53 percentage points in the second quarter, the highest level since the launch of ChatGPT.
Structurally, the divergence pattern is clear: infrastructure-related names generally recorded positive returns, while application-layer companies showed mixed performance. This indicates that the previous strategy of indiscriminately buying AI concept stocks is no longer viable, requiring investors to make more nuanced choices between different segments of the value chain.
Concurrently, positioning and leverage within the AI sector remain highly concentrated. Data from Goldman Sachs' prime brokerage shows hedge funds have been net buyers of global stocks for a fourth consecutive week, 0.8 standard deviations above the one-year historical average; US stocks saw their fastest pace of net buying since November 2025, 2.1 standard deviations above average.
Notably, this buying activity was primarily driven by short covering rather than new long positioning. Even during the recent tech sector pullback, net selling in the Information Technology sector was relatively modest (-0.7 standard deviations), and semiconductor positioning remained largely unchanged. Currently, both gross and net exposure to the tech sector remain at historically elevated levels—at the 92nd/98th and 95th/99th percentiles within one-year and five-year lookback periods, respectively. Overall, positioning remains tilted long but has not yet seen panic selling.
Soaring AI Capital Expenditure Puts Cloud Providers in "Prove the Returns" Phase
Coppersmith highlighted that the market narrative for hyperscale cloud providers is undergoing a fundamental shift. Over the past decade, the market awarded them a high premium for their high margins, asset-light model, and consistent buybacks, but the massive investment in AI infrastructure is rewriting this logic.
Consensus expectations indicate that capital expenditure from cloud providers will reach approximately $770 billion this year, nearly equivalent to their total operating cash flow. Simultaneously, buybacks have contracted sharply, net leverage has increased by about $170 billion since early 2025, and share counts have also begun to rise.
Coppersmith believes a critical inflection point occurred last week: investors are no longer buying the "build story" for AI infrastructure and are instead demanding to see tangible returns on capital. The market is transitioning from an "all-out build" phase to a "prove the returns" stage.
Comments