The AI capex super-cycle: A math problem for the S&P 500

The scale of investment in artificial intelligence by S&P 500 companies—specifically the "hyperscalers" like Microsoft, Alphabet, and Meta—has reached levels that are historically unprecedented. (See figure below. Data source: Goldman Sachs.) For the 2026 fiscal year, collective capital expenditures for these firms are now forecasted to hit a staggering $674 billion, up from roughly $400 billion in 2025. To put that in perspective, this spending represents about 2.2% of U.S. GDP, a figure that is more than four times the investment level seen as recently as 2023. We are witnessing an infrastructure "arms race" that dwarfs the historical buildouts of the interstate highway system and the moon landing combined.

To justify this massive $600 billion-plus annual burn, the earnings of these S&P 500 giants will need to grow at an extraordinary clip over the next few years. For the current level of AI investment to generate a positive return on invested capital (ROIC), AI-driven revenues would need to scale to approximately $2 trillion annually by 2030. Currently, actual AI-related revenue for these firms is estimated to be in the neighborhood of only $20 billion to $40 billion. This implies that the market is banking on a 100-fold increase in top-line AI contribution within just four to five years to keep pace with the depreciation and operating costs of the hardware being installed today.

From a structural and historical standpoint, this "hockey stick" growth requirement is bordering on the mathematically impossible. While the S&P 500 is currently projected to grow earnings per share (EPS) by roughly 12% to 14% in 2026, those gains are largely driven by cost-cutting and existing software margins, not yet by a massive influx of AI-native profits. When you factor in the short three-to-five-year useful life of AI chips compared to traditional industrial infrastructure, the "earn-back" hurdle becomes so high that even a "goldilocks" economy may not be enough to prevent a significant valuation correction.