Goldman Sachs lifts AI capex outlook to $7.6 trillion through 2031

Goldman Sachs is telling clients that the AI boom is no longer just a software story. It is becoming a coverage problem for bankers, analysts and product specialists who now have to track semiconductors, data centers, power capacity, cooling, cabling and the financing behind all of it.

The bank’s updated June 2026 analysis lifts expected combined capex for Meta, Microsoft, Amazon and Alphabet to $5.3 trillion from fiscal 2025 through fiscal 2030, up from a prior $4.5 trillion estimate. Goldman’s baseline aggregate capex outlook now stands at $7.6 trillion between 2026 and 2031 across compute, data centers and power. The broader hyperscaler set also includes Oracle, which Goldman says plans to add roughly $2 trillion in AI-related assets to balance sheets by 2030.

For Goldman employees, the shift matters because the spend is moving the firm’s internal center of gravity. Coverage teams that once treated AI as a software monetization story now have to understand where the physical bottlenecks sit and who gets paid along the way. The bank said in a May 1 report that the build-out is highly sensitive to assumptions about how long AI chips last, what new data centers cost, the mix of chips and architectures, and bottlenecks in power, labor and equipment. Small changes in those assumptions can swing cumulative spending by hundreds of billions.

That means more work for people who can speak the language of data-center economics, grid constraints and chip supply, not just cloud revenue growth. Goldman’s November 2025 research projected data-center power demand would surge 175% by 2030 versus 2023 levels, equal to adding another top-10 power-consuming country. The implication for the bank is straightforward: more client demand around utilities, equipment, infrastructure financing and project execution, and more pressure on coverage bankers to connect those themes back to capital markets and advisory mandates.

The spending surge is already showing up in market expectations. Goldman said consensus 2026 capex for hyperscalers had climbed to $527 billion from $465 billion at the start of third-quarter earnings season. Separate estimates have put combined 2026 spending by Alphabet, Microsoft, Meta and Amazon near $700 billion, with investors growing more selective and favoring companies that can show a clearer link between capex and future revenue growth.

For Goldman, that selectivity changes the job. The people who can move between semis, power infrastructure, data-center build-outs and AI finance are likely to matter more in client pitches, staffing decisions and the next round of promotion conversations than those still treating the AI trade as a pure software race.