Big AI models are running into a surprisingly old-fashioned bottleneck: juice. Leaders from Microsoft and OpenAI warned that raw computing power is only part of the race; reliable electricity and finished data centers matter just as much. Firms are pausing projects, signing new power deals, and bracing for market shifts if massive cheap energy arrives. The struggle over racks, cooling, and grid access is turning power into a strategic weapon in the AI era.
Microsoft’s CEO made the issue blunt and memorable when he said, “The biggest issue we are now having is not a compute glut, but it’s power,” and added, “It’s not a supply issue of chips. It’s actually the fact that I don’t have warm shells to plug into.” That line captures the scene: thousands of AI accelerators sitting idle because data centers are incomplete or lack sufficient energy and cooling. It’s a reminder that silicon without infrastructure is just expensive hardware in a warehouse.
The podcast where the comment landed has drawn attention from investors and engineers alike, and the source clip is available for those who want the full context.
Microsoft has responded by slowing or halting some early-stage data center initiatives while it sorts capacity and power contracts. In an April post, Microsoft Cloud Operations and Innovation leadership said the company is “slowing or pausing some early-stage projects” as part of the largest infrastructure buildout in its history. Pausing projects is painful but it avoids filling facilities that can’t be reliably powered and cooled once the gear arrives.
Other cloud giants are juggling similar trade-offs. Google has been moving workloads away from peak grid stress periods and cutting back temporarily when utilities ask, and Amazon has been pushing efficiency projects and grid-support programs. These operational fixes help, but they don’t change the underlying math: more powerful models and lower per-cycle costs mean demand can explode faster than new generation and transmission get built.
OpenAI’s CEO offered a related warning focused on contracts and market disruption, saying, “If a very cheap form of energy comes online soon at mass scale, a lot of people are going to be extremely burned with existing contracts they’ve signed,” which highlights a different risk. Long-term power deals can look great until a disruptive energy source changes the economics overnight, leaving firms locked into high-cost supply. That kind of shock could ripple through AI providers, cloud operators, and corporate customers alike.
Energy companies and tech firms are responding by cutting strategic deals to lock in capacity and cleaner supply, including long-term renewable power arrangements meant to guarantee steady flows. Large developers and utilities have inked major contracts that promise dedicated generation or 24/7 carbon-free energy to key campuses and cloud nodes. Those arrangements add complexity to project economics but give cloud operators the predictability they need to bring new data centers online.
Forecasts back up the urgency. Analysts expect data center electricity use to climb sharply over the decade, with projections that global consumption could roughly double and approach many hundreds of terawatt-hours by 2030. That increase would make data centers a much bigger slice of total power demand and put pressure on regional grids, especially where growth is concentrated near tech hubs.
The upshot is that AI’s infrastructure story now reads like a power market one. Idle AI chips, paused construction, and complex energy contracts are reshaping how companies plan capacity and costs. For anyone building or buying AI services, the delicate dance between compute capability and reliable, affordable electricity has become the defining constraint to watch.
