Paducah's $100B AI Campus Has 4.6GW of Power and No Named Tenant
The U.S. Department of Energy has picked Brookfield Asset Management and NextEra Energy for one of the largest AI infrastructure proposals yet: a privately funded campus at the former Paducah Gaseous Diffusion Plant in western Kentucky, with a headline value above $100 billion.
The numbers are huge even by 2026 data-center standards. The plan calls for up to 1.2GW of compute capacity, roughly 1.8GW of utility capacity, and 4.6GW of dedicated generation built around the project. The power stack is expected to include about 2GW of gas generation and 2.6GW of battery storage.
The missing number is the tenant. No anchor customer has been publicly named.
The Deal Shape
| Component | Paducah proposal |
|---|---|
| Total private investment | $100B+ |
| Compute capacity | 1.2GW |
| Utility capacity | 1.8GW |
| Dedicated generation | 4.6GW |
| Gas generation | 2GW |
| Battery storage | 2.6GW |
| Construction jobs | ~8,000 |
| Permanent jobs | ~600 |
Brookfield is the site developer and operator. NextEra is responsible for the power infrastructure. Big Rivers Electric, Jackson Purchase Energy Cooperative, and Paducah Power System sit in the regional utility layer. DOE controls the underlying site: a Cold War-era uranium-enrichment facility that has been in cleanup mode since operations ended.
The project is framed as private investment rather than federal capex. That distinction matters because the political promise attached to AI infrastructure is that surrounding ratepayers will not subsidise hyperscaler power demand. The plan still needs to survive execution details: definitive agreements, power-service approval, utility interconnection, site remediation constraints, and tenant commitments.
Power First, Tenant Later
Paducah is the latest example of the AI buildout moving from “find cheap land near fibre” to “secure power before the customer is public.” The site has three advantages that ordinary greenfield campuses do not: federal ownership, industrial history, and a power strategy built into the announcement rather than bolted on afterward.
That is also what makes the missing tenant conspicuous. A 1.2GW compute campus is not a speculative office park. It implies billions of dollars in GPUs, networking gear, memory, backup power, and long-term operating commitments. Without a named hyperscaler, frontier lab, or sovereign customer, the announcement is a site-control and power-development story more than a fully contracted AI factory.
This pattern is becoming common. Capital partners announce gigawatt-scale campuses, power partners attach generation, and customer names arrive later if they arrive at all. The industry has learned that power is the hardest gating asset, so the race starts there.
Why Paducah Is Different
The location gives the project a political and symbolic edge. AI infrastructure is being placed on land built for the nuclear-weapons supply chain, now repurposed for compute and energy infrastructure. That fits Washington’s current industrial-policy framing: federal sites, private capital, domestic power, and AI capacity treated as strategic infrastructure.
It also creates a sharper permitting and public-health context than a normal data center. The former gaseous diffusion plant is not empty farmland. Environmental cleanup, community trust, and industrial redevelopment all sit inside the project perimeter. The promised ratepayer protection does not answer questions about gas emissions, water use, local grid effects, or long-term site liability.
The Real Milestone
The next meaningful signal is not another headline investment number. It is a binding customer announcement or a filed power-service agreement with enough detail to show who pays for what.
Until then, Paducah is a giant option on AI demand. Brookfield and NextEra are trying to pre-package land, power, transmission, and capital before the compute buyer steps into view. If a frontier lab or hyperscaler signs, the site becomes one of the largest dedicated AI campuses in the U.S. If not, the $100 billion figure remains an intent marker in a market already full of phantom gigawatts.
That distinction is the whole story. AI infrastructure is no longer short of announcements. It is short of contracted power, named tenants, and execution paths that survive contact with regulators and local communities.