AI is becoming a power-grid problem
Building AI used to be a software problem; now it is increasingly an electricity problem. Training and running frontier models requires data centres at industrial scale, and the bottleneck has shifted from chips to the gigawatts of power those data centres consume. Local communities are pushing back, utilities are racing to build, and the question of where AI gets built is starting to be answered by where the electricity is.
Timeline
- September 20, 2024
Constellation Energy signs a 20-year agreement with Microsoft to restart Three Mile Island's Unit 1 reactor to power Microsoft's data centers. Restarting a closed nuclear plant for AI sets a new bar for what AI infrastructure now demands.
- December 4, 2024
Meta announces a $10 billion AI data center in northeast Louisiana, its largest in the world. The site selection signals that AI infrastructure is moving where the power and land are, not just where the talent is.
- January 21, 2025
OpenAI, Oracle, and SoftBank announce the Stargate Project, a venture planning to invest $500 billion over four years building new AI infrastructure in the United States. The plan reframes AI competition as an industrial-policy and power-supply contest.
- July 1, 2025
Dominion Energy reports it has nearly doubled data center capacity under contract to 40GW, up from 21GW a year earlier. Virginia, home to more data centers than any state, faces an estimated 183% increase in energy demand by 2040 driven by AI data centers.
- March 1, 2026
Data center moratorium bills begin spreading across state legislatures, with similar legislation introduced in Georgia, Maryland, Vermont, Virginia, Oklahoma, and New York. Public opposition to data center developments explodes over the past year.
- May 19, 2026
Denver City Council unanimously approves a one-year moratorium on new data center construction after community testimony about power and water demands. Denver becomes the first major US city to enact a formal data center construction pause.
Where things stand right now
Power and land have become the binding constraints on AI build-out, not chips. The frontier labs and hyperscalers are racing to lock down generation capacity through nuclear restarts, behind-the-meter generation, and long-term utility contracts — while local communities are pushing back on the industrial footprint AI requires. Where AI gets built in 2027 will be decided as much by zoning boards and utility commissioners as by engineering teams.