Hyperscaler AI capital spending is raised to $750 billion for 2026, on track to cross $1 trillion in 2027
The capital numbers keep re-rating upward. Hyperscaler AI capex has been raised to $750 billion for 2026, up from $670 billion, and is set to cross $1 trillion in 2027, while US data-center electricity demand climbs from 23 GW in 2023 to 42 GW in 2026. The spending is real, and it is increasingly bounded by what the grid can deliver.
A $750 billion annual outlay, revised up mid-year, is the clearest signal that the buildout is accelerating rather than plateauing. The upward revision from $670 billion is the part that matters: forecasts are chasing demand upward, not settling, and a trillion-dollar 2027 is now the base case rather than a bullish outlier.
The electricity figure is the constraint hiding inside the capex figure. US data-center demand nearly doubling to 42 GW in three years is the load that has to be found, delivered, and interconnected, and it is growing faster than the grid was built to accommodate. The dollars can be committed faster than the megawatts can be energised.
The consequence is that the scarce input has shifted from silicon to power. A lab can commit capital and order chips on a timeline of quarters; it cannot conjure a grid connection on the same schedule. That mismatch is why the compute story of 2026 is a story about substations and power-purchase agreements as much as about GPUs.
Help Net Security — Power shortages could slow AI data center expansion → · Utility Dive — AI data center growth could force US utilities to rethink generation plans, BofA says →