The AI boom just turned a relic of America's nuclear past into the largest single data center bet ever placed.
The Summary
- NextEra Energy and Brookfield are converting a Cold War-era uranium enrichment plant in Kentucky into a $100+ billion data center campus, the largest infrastructure project of its kind in U.S. history
- The partnership pairs America's biggest power utility with institutional capital to solve AI's energy problem at unprecedented scale
- Turning weapons-grade infrastructure into compute infrastructure: a literal conversion of 20th century power into 21st century intelligence
The Signal
NextEra Energy, the largest U.S. power utility by market cap, is partnering with Brookfield Corp. to transform a shuttered uranium enrichment facility into what will be the world's largest data center campus. The $100 billion price tag isn't a typo. It's bigger than the GDP of most countries, and it's all going into Kentucky dirt that once enriched uranium for nuclear weapons.
The site choice is not arbitrary. Cold War enrichment facilities came with three things data centers desperately need: massive electrical infrastructure already in place, transmission line connections built to industrial scale, and government-cleared land with security protocols baked in. The bones are good. Now they're getting a new body.
"The AI training runs that define the next decade will happen in places that solved the hardest infrastructure problems of the last one."
What makes this signal rather than noise is the partnership structure. NextEra brings grid expertise and regulatory relationships. Brookfield brings $850 billion in assets under management and a track record of converting distressed industrial sites into cash-flowing infrastructure. They're not building a data center. They're building a power plant that happens to have servers attached.
Key differences from standard hyperscale buildouts:
- Integrated generation on-site, not grid-dependent
- Reuse of existing federal infrastructure and security protocols
- Scale designed for training runs, not inference workloads
- Capital structure mixing utility debt with infrastructure equity
The timeline matters too. Construction at this scale takes years. The partners are betting that by the time this comes online, AI training demand will have outstripped every existing facility. They're building for models we haven't imagined yet, trained on datasets that don't exist, by companies that might still be in Series A.
The Implication
Watch where power meets capital at infrastructure scale. The AI buildout isn't happening in garages or even in typical data centers anymore. It's happening where century-old electrical infrastructure can be repurposed, where land is cheap and power is abundant, and where patient capital can wait years for returns measured in decades.
If you're building agents that need serious compute, your deployment options just shifted. The hyperscalers will keep building, but the next tier of infrastructure is coming from utilities and infrastructure funds that think in transmission lines and bond yields, not cloud regions and spot pricing.