The US data centre power shortage is widening as AI companies build infrastructure far faster than the electricity grid can expand to serve it, with one industry expert predicting that around 50 to 60% of data centre projects will be delayed beyond the one-to-two-year windows developers are targeting.
Kathryn Burke, who leads US specialty energy and power growth at insurance firm Marsh, told Fortune that access to power is ‘probably the number one, if not top five bottlenecks for data centre development in the US right now’.
She said companies are requesting power at a rate the grid cannot handle, and that utilities are demanding upfront financial commitments from developers so they are not left exposed if projects fall through.
Grid demand racing ahead of supply
Data centres are projected to consume nearly 12% of all US electricity by 2030, almost six times their pre-AI share in 2018, according to the Lawrence Berkeley National Lab.
The grid watchdog NERC‘s 2025 long-term reliability assessment projects summer peak electricity demand in North America will grow by more than 224 GW over the next ten years, 69% above the growth it projected a year earlier, driven largely by new AI data centres.
On the winter side, NERC projects demand will grow by 245 GW over the same period, a 65% increase from last year’s projection of 149 GW growth, underlining how sharply the forecasts have shifted in a single year.
Within the Western US grid region, planned data centres already account for an average of 10% of demand forecasts, rising as high as 40% in some areas.
US data centre power shortage reflects years of under-investment
Rob Gramlich, president of consulting firm Grid Strategies, said the US electricity industry had ‘got out of practice on building new infrastructure’ after 25 years of weak demand growth.
Large grid expansions in the 1980s and 1990s left utilities with spare capacity, and the period from 2000 to 2023 did not replicate the surges seen when domestic appliances and air conditioning spread widely. The system was not positioned for what followed.
‘But now we have electric vehicles, we have electric space heating, we have new manufacturing, much of which is using electricity more than traditional manufacturing, and then we have data centres,’ Gramlich said. ‘The growth is about half data centres and half those other new things, but those new quickly expanding uses of electricity are happening all at once.’
He said tech firms are ‘famous for moving fast’ while utilities ‘notoriously move very slowly’ because maintaining an interconnected grid requires ‘deliberate study and planning’ that can take years. ‘There is a real disconnect there,’ he said.
What delays look like in practice
A gigawatt of electricity, enough to power 750,000 US homes, can take far longer to bring online than the two years developers typically request. Power projects that came online in 2025 spent a median of five years from requesting a grid connection to operating commercially, according to separate Berkeley Lab research.
Gramlich said utilities will not generally connect new customers they cannot reliably serve. Rather than blackouts, the likely outcome is that data centres face multi-year waits for full service, or accept provisional connections under which supply can be interrupted when the grid is under strain.
‘There is scarcity on the grid,’ he said. ‘Not everybody’s going to get the full level of service that they want, at least until the system can catch up to these new demands.’
Burke said the scale of uncertainty remains high. ‘It’s hard to predict how many of these data centres are actually going to get built at the end of the day,’ she said, pointing to financing and multiple supply constraints as unresolved questions. NERC’s 2025 assessment is due to be reviewed by reliability planners ahead of next year’s update.
