The five largest US technology companies plan to spend $650–690 billion on AI infrastructure in 2026, nearly doubling their combined outlay from the previous year, according to a Bridgewater Associates estimate1.
The capital flows into data centres, GPU procurement, and power infrastructure. Meta's capex guidance and Oracle's planned workforce-to-infrastructure conversion are two expressions of a sector-wide pattern: labour budgets becoming infrastructure budgets at accelerating rates2.
The scale creates its own momentum. Data centres take two to four years to plan, permit, and build. They consume electricity at densities far exceeding traditional computing, adding grid constraints to the capital lock-in. Once tens of billions are sunk into physical infrastructure, the economic incentive to automate enough work to justify the investment intensifies. The capital demands utilisation, which means finding more tasks to transfer from workers to machines. The current wave of layoffs is the front end of a capital cycle that will generate sustained pressure on labour costs through the rest of the decade.
