Goldman Sachs has issued a warning that the artificial intelligence (AI) capital expenditure cycle may be approaching an inflection point, with two of three key risk signals already flashing. In a report published on October 12th, 2026, Goldman Sachs senior macro trader Bobby Molavi estimated that the total asset scale of the AI ecosystem – encompassing major technology companies, cloud service providers, AI laboratories, chips, and computing infrastructure – has reached 25.00 trillion to 30.00 trillion U.S. dollars. This valuation relies heavily on continued expansion of infrastructure investment, and historical experience suggests that stock prices typically peak before capital expenditure does.
The three signals identified by Molavi are: stock prices diverging from actual earnings, credit markets repricing AI borrower risk, and capital expenditure growth actually decelerating. The first two signals have already materialized. On valuation, Molavi noted that forward price-to-earnings ratios may understate true valuation levels. For instance, in March 2000, theStandard & Poor’s (S&P) 500’s forward Price-to-Earnings Ratio (P/E) was 25.20 times, but based on realized earnings it reached 39.30 times. In 2007, the forward P/E was only 15.10 times, yet realized earnings implied 31.60 times – suggesting that forward estimates can mask valuation risk if corporate earnings fail to materialize. On the financing side, more than 500.00 billion U.S. dollars in debt issuance has been used to support AI infrastructure this year. Yields on several prominent AI financings have continued to rise, and investment-grade and high-yield credit spreads have begun widening, indicating that creditors are demanding higher risk compensation.
The third signal – capital expenditure growth deceleration – remains to be confirmed but is expected to emerge soon. Goldman Sachs forecasts that hyperscaler capital expenditure growth will peak at 116.00 percent year-over-year in the third quarter of 2026, then decline quarter by quarter to around 20.00 percent by the end of 2027. Growth peaking does not mean total spending will immediately fall, but if markets have already priced in continued acceleration, a slowdown in growth could still alter investor expectations for future returns.
Market structure amplifies the vulnerability. The ten core AI-related stocks now account for approximately 40.00 percent of the S&P 500’s total market capitalization – exceeding the concentration at the peak of the 2000 internet bubble. The best-performing AI-related stocks have gained an average of 58.00 percent this year, far outpacing the S&P 500’s 13.00 percent and the Nasdaq’s 22.00 percent, while the median S&P 500 constituent has actually declined 15.00 percent. Molavi characterized the current market as “running desperately but standing still,” warning that once AI investment expectations shift, the adjustment could be rapidly amplified by the highly concentrated market structure.