AI Capex Supercycle Pushes Hyperscaler Spending Past $400 Billion

The four largest U.S. hyperscale cloud providers will collectively spend more than $400 billion on capital expenditure this calendar year, according to guidance issued alongside third-quarter results. That figure is more than double 2024 spending and roughly five times the level of 2022.

The bulk of the money is going into accelerated computing infrastructure — graphics processors, custom silicon, high-bandwidth memory, and the networking gear needed to run them at scale. Data center construction has become the single largest non-residential building category in the U.S., surpassing both warehouses and offices.

"This is not a cyclical investment, it is a re-platforming," said Wei Zhang, technology analyst at Cascade Equity Research. "Every workload assumption that enterprises made in the cloud era is being rewritten for inference economics."

The spending has created acute bottlenecks elsewhere in the technology supply chain. Advanced packaging capacity at leading foundries is booked through 2027. High-bandwidth memory shipments are constrained by test equipment lead times. Optical transceivers at 800 gigabits and beyond are in short supply.

Power has emerged as the binding constraint. Utilities in Virginia, Texas, and Arizona report interconnection queues of five to seven years for large data center projects. Several hyperscalers have signed long-term power purchase agreements with nuclear generators, and at least two are exploring small modular reactors.

Investors have so far rewarded the spending. Combined shareholder returns at the four largest hyperscalers are up 38% year-to-date, well ahead of the broader market. But scrutiny is rising: several large asset managers have publicly questioned the return on invested capital trajectory.