Goldman sees hyperscalers slashing buybacks as AI capex surges to record levels

Goldman Sachs is warning that the AI boom is becoming a capital-allocation story, not just a demand story. The bank said the five largest hyperscalers, Amazon, Alphabet, Meta, Microsoft and Oracle, are set to spend a combined $755 billion on capital expenditures in 2026, an 83% jump from a year earlier, while that spending could consume 100% of their cash flows from operations and leave little room for shareholder returns without more debt.

That matters because buybacks have long been one of Big Tech’s most reliable support beams for investors and employee wealth creation tied to equity awards. Goldman said repurchases could fall to 20% of spending from 34%, a change that would reduce a long-standing source of total return for hyperscaler shareholders and force the market to separate the names that can keep compounding from the ones that are simply funding the race.

The scale of the spending reset is already showing up in Wall Street estimates. Goldman said consensus 2026 capital spending for AI hyperscalers has risen to $527 billion from $465 billion at the start of the third-quarter earnings season. At the same time, Goldman forecast S&P 500 capital expenditures to grow 33% in 2026 and 22% in 2027, underscoring how quickly the AI buildout is spreading beyond the mega-cap platforms.

Goldman’s view is that the payoff question is no longer just whether AI adoption will justify the spend. The bigger issue is the size and timing of the infrastructure buildout itself, from chips and servers to data centers and the power grid. Goldman said rising hyperscaler reinvestment is already affecting power demand forecasts and data-center infrastructure planning, and it now expects data-center power demand to grow 220% by 2030 versus 2023 levels.

The market consequences are starting to look less uniform than in the first phase of the AI trade. Goldman said large public AI hyperscaler stocks have become less correlated to one another, a sign investors are becoming more selective about which companies can translate capex into durable earnings power. That selectivity could shift more of the upside toward semiconductor suppliers, datacenter builders, power infrastructure companies and other capex beneficiaries rather than the platforms themselves.

Goldman also put the current cycle in historical context. It said AI hyperscaler capex would need to reach $700 billion in 2026 to match the spending intensity of the late-1990s telecom boom, when tech, media and telecom stocks spent more than 100% of operating cash flow on capex and R&D. By comparison, Goldman said the current AI cycle was running at about 72% at that point, a reminder that the buildout is already massive, but still not at the most feverish level of the dot-com era.