DOE Turns a Kentucky Uranium Site Into a Test of America’s AI Infrastructure Strategy

The Department of Energy has selected Brookfield to develop a $100 billion AI complex at a contaminated former uranium plant in Kentucky. The proposal pairs a 1.8-gigawatt data center with two gigawatts of natural gas generation and 2.6 gigawatts of battery storage.

A Google News headline can make the plan sound like a simple property conversion. It is much more consequential. Washington is trying to transform federal land, private capital, and dedicated power generation into a faster route for expanding American AI capacity.

The real contest is between construction speed and public accountability. Brookfield, NextEra Energy, federal agencies, utilities, regulators, and prospective computing customers must turn an industrial concept into an approved operating system. They must do that while cleanup continues beneath the site.

The plan also establishes a national comparison. A much larger federal project was announced in March at Ohio’s former Portsmouth uranium-enrichment site. Together, the projects show how the federal government wants to reuse Cold War properties for the AI era.

What DOE Actually Selected at the Paducah Site

The federal decision identifies a developer and an infrastructure model, but it does not deliver a finished data center or fully approved power plant.

The Department of Energy selected Brookfield Asset Management to develop and operate the proposed campus at the Paducah Gaseous Diffusion Plant. The federal government owns the property, which lies west of Paducah in McCracken County.

NextEra Energy would build and own the energy components. According to the reported plan, those components include two gigawatts of gas generation, transmission improvements, and 2.6 gigawatts of battery storage.

The infrastructure would support a 1.8-gigawatt AI data center campus. At that scale, the computing load would resemble a large industrial power system rather than an ordinary commercial development.

The project’s reported investment value reaches $100 billion. That figure covers a proposed complex whose construction is expected to continue through 2031, according to the project details.

Brookfield had not publicly identified a final group of technology tenants when the project was announced. A company spokesperson said discussions with prospective commercial partners were continuing.

That distinction matters. Brookfield can prepare land, financing, utilities, and buildings, but customers determine how much computing capacity becomes economically useful. Large campuses often develop in phases because chip supply, customer commitments, and power delivery do not arrive simultaneously.

The batteries would not generate energy. They would store electricity, respond to short demand swings, and help balance computing equipment that needs consistent power. The gas plant would provide the principal new generation described in the announcement.

Excess electricity could enter the regional grid, according to the reported framework. However, the commercial arrangements governing that power still require regulatory scrutiny.

The power service agreement must receive approval from Kentucky utility regulators. Air permits, water arrangements, construction reviews, and environmental controls will also shape the final design and schedule.

This means DOE’s selection changes the project from a federal solicitation into a developer-led proposal. It does not remove the engineering, regulatory, environmental, or commercial gates ahead.

That nuance can disappear when readers encounter the event through Google News or another headline feed. The federal government has chosen a path and a development team. It has not guaranteed every proposed component.

Why Federal Land Has Become an AI Policy Tool

Washington is treating available power and buildable land as strategic AI resources, not merely local development assets.

DOE first identified Paducah as a preferred location for AI infrastructure in July 2025. The department later sought private proposals for data centers and accompanying energy projects at the property.

The solicitation aligned the development with federal policies covering AI leadership, domestic energy production, and faster permitting. It invited companies to combine computing facilities with the power infrastructure needed to operate them.

This structure addresses one of the AI sector’s hardest constraints. A company can order servers faster than a utility can always add generation, substations, and high-capacity transmission connections.

Placing new generation beside a data center can shorten part of that process. It also moves financing and development responsibilities toward the companies benefiting from the added capacity.

Federal property offers another advantage. The government can coordinate land access, existing infrastructure, security considerations, and redevelopment goals through one site-level process.

Paducah contains roughly 3,550 acres and has a long industrial history. Its roads, utility corridors, workforce, and access to water make it an unusual location for another large industrial use.

The site’s history also creates its greatest complication. The plant began producing enriched uranium in the 1950s, first for national security programs and later for commercial nuclear fuel.

Commercial enrichment stopped in 2013. DOE then assumed responsibility for deactivation, decommissioning, surveillance, waste management, and environmental remediation.

DOE’s own site history describes a property still moving through cleanup and reuse. The AI development would therefore coexist with an active federal environmental mission.

That combination explains why Paducah is attractive now. It has industrial land and infrastructure, while the government has a financial interest in finding productive uses for portions of the property.

A successful lease could also help support regional employment after the enrichment era. Local leaders have pursued new uranium, energy, and technology investments to keep the site economically relevant.

The federal strategy is broader than Kentucky. DOE has evaluated multiple properties for data centers and energy generation, effectively turning parts of its real estate portfolio into AI infrastructure policy.

This is the point that matters beyond a single Google News result. The government is not simply approving privately selected sites. It is packaging federal land as an input to the national computing race.

That choice pressures traditional data center markets. Regions with expensive land, crowded transmission queues, or local construction resistance now compete with government-controlled industrial properties.

It also pressures utilities and regulators. They must decide whether dedicated generation truly isolates costs or merely moves new risks elsewhere in the power system.

The Core Tradeoff Is Speed Versus Accountability

Bundling land, computing, gas generation, storage, and transmission can accelerate development, but it also concentrates several unresolved risks in one project.

The Paducah concept treats power as part of the data center rather than a utility service added later. That approach can give developers more control over scheduling and capacity.

NextEra would own the generation assets, while Brookfield would develop and operate the computing campus. Local utilities and transmission partners would connect the project to surrounding infrastructure.

The model can reduce dependence on waiting for enough existing grid capacity. Yet a two-gigawatt gas plant remains a major power project with fuel, pipeline, emissions, maintenance, and permitting requirements.

Battery storage adds flexibility. It can absorb electricity during lower-demand periods and release it during peaks or equipment transitions.

However, a battery system cannot replace a continuous energy supply indefinitely. Its practical contribution depends on its energy duration, cycling strategy, grid connections, and operating rules.

Those details had not been fully disclosed when the development team was announced. The public capacity number describes maximum power, not how many hours the batteries can sustain that output.

The computing side contains similar uncertainty. A 1.8-gigawatt campus represents potential load, but actual consumption depends on construction phases and installed equipment.

AI hardware changes quickly. New chips can increase computing output per unit of electricity, but rising demand can consume those efficiency gains.

The project must also secure tenants willing to make long-term commitments. Brookfield’s continuing discussions suggest that customer selection remained an active commercial task.

Those customers will influence the campus design. Training large AI models, running cloud services, and serving real-time inference workloads create different network, storage, and reliability demands.

The uncertainty does not make the proposal empty. Large infrastructure projects routinely begin before every building and customer has been finalized.

It does mean readers should separate announced capacity from operating capacity. A multiyear development can reach only part of its headline scale or alter its technology mix.

The same distinction applies to investment. A $100 billion development value can cover staged spending over years rather than committed capital delivered immediately.

The government’s chosen model makes the bet easier to understand. Federal land addresses siting, Brookfield organizes development capital, and NextEra provides an energy pathway.

The tradeoff is equally clear. Combining these elements can speed decisions, but failures become interconnected.

A delayed pipeline can hold back gas generation. A delayed power agreement can slow the campus. A missing tenant can weaken financing for later phases.

Environmental restrictions can alter where foundations, pipes, wells, and utility corridors are placed. Each change can affect the economics of surrounding components.

The Paducah plan is therefore an infrastructure orchestration test. Its success depends less on any single building than on coordinating several regulated systems.

Google News Headlines Cannot Show the Cleanup Problem

The development sits on a federally managed contamination site where construction, groundwater protection, and remediation must remain coordinated.

The former enrichment plant is listed under the federal Superfund program. Investigators have identified contamination in groundwater, soil, surface water, and sediment.

Contaminants associated with the site include trichloroethylene, technetium-99, polychlorinated biphenyls, and uranium. Trichloroethylene is an industrial solvent historically used for metal degreasing.

EPA reports that human exposure is currently controlled. However, its Superfund profile says available evidence remains insufficient to determine whether contaminated groundwater migration is stabilized.

That difference is important. Controlled exposure means protective measures are addressing current contact pathways. It does not mean every contaminated area has been excavated, treated, or cleared for unrestricted development.

DOE continues to investigate groundwater conditions and evaluate long-term remedies. Its Decision 2029 process aims to consolidate cleanup decisions for buildings, soil, groundwater, waste areas, and other environmental units.

The development team must know which parcels can support construction and which require continued controls. It must also avoid interfering with monitoring wells, treatment systems, and future remedial work.

Water deserves particular attention because data centers reject large amounts of heat. Cooling designs can use water directly, rely more heavily on air, or combine several methods.

The project’s final annual water demand had not been publicly established when the development selection was reported. That missing number should not be replaced with estimates from unrelated campuses.

Byron Gary, a senior attorney with the Kentucky Resources Council, warned that redevelopment and groundwater withdrawals require careful execution. His concern reflects the site’s documented conditions rather than general opposition to computing.

The question is not whether development must wait until every cleanup activity ends. Federal contaminated sites can support carefully controlled reuse while remediation continues.

The question is whether construction plans, water sourcing, and environmental safeguards remain transparent enough for regulators and nearby communities to evaluate them.

Gas generation adds a second layer of review. A large combustion plant can emit carbon dioxide, nitrogen oxides, and other regulated pollutants.

Actual emissions will depend on turbine technology, operating hours, fuel consumption, pollution controls, and permit limits. None should be inferred solely from the plant’s maximum capacity.

The campus could deliver excess power to the grid, which supporters may present as a regional benefit. Regulators must still examine when that energy would be available and who pays for associated infrastructure.

They must also determine how costs are assigned if the computing campus develops more slowly than expected. A protective rate structure should prevent unrelated customers from inheriting stranded investments.

DOE’s cleanup strategy shows why the environmental issue cannot sit in a footnote. Redevelopment and remediation are happening within the same physical system.

This is the strongest skeptical case against treating the project as settled. The developer selection is real, but environmental compatibility remains an engineering and regulatory claim requiring continuing evidence.

Ohio Shows the Scale of the Federal Strategy

Paducah is not an isolated reuse project, but a smaller version of a federal model already advancing at another former uranium site.

In March 2026, federal officials announced a related development at the Portsmouth Gaseous Diffusion Plant near Piketon, Ohio. That proposal was substantially larger than the Kentucky project.

The Ohio plan described a data center reaching 10 gigawatts. Its power portfolio included as much as 10 gigawatts of new generation, with 9.2 gigawatts coming from natural gas.

By comparison, Paducah’s planned 1.8-gigawatt data center and two-gigawatt gas plant look more contained. They are still among the largest industrial additions proposed for Kentucky.

The Portsmouth precedent confirms that DOE is following a repeatable strategy. It is pairing former uranium infrastructure with private data center capital and dedicated energy construction.

The two sites share important characteristics. Both occupy large federal properties, supported uranium enrichment, and entered lengthy cleanup programs after operations ended.

They also contain communities seeking durable employment after federal industrial missions declined. AI developers arrive with large investment projections, construction demand, and long operating horizons.

Yet the Ohio comparison also reveals the risk of headline capacity becoming the story. Ten gigawatts is a planning ceiling, not evidence that ten gigawatts of servers are already installed.

The Ohio project faces its own questions about gas supply, customers, permitting, financing, and phased deployment. Paducah will encounter the same categories at a different scale.

Other private developers have pursued former power plants because those properties already connect to transmission networks. The federal model adds direct government control over land and site redevelopment.

That control can simplify coordination, but it can also blur institutional roles. DOE is simultaneously a landlord, cleanup authority, national energy agency, and promoter of AI infrastructure.

Independent state and federal regulators therefore remain important. Their reviews provide checks that a development-oriented landlord cannot replace.

The competition is not simply Kentucky versus Ohio. It is federal-site development versus conventional data center siting across Virginia, Texas, Georgia, and other established markets.

Conventional markets offer customer clusters, fiber networks, and experienced contractors. They increasingly face congested grids, local opposition, water constraints, and longer connection schedules.

Federal sites offer land and political sponsorship. Their disadvantages include contamination, smaller regional technology ecosystems, and the difficulty of coordinating new power plants with computing demand.

There is also a technology competition inside the energy strategy. Gas turbines can be built faster than many nuclear plants, while batteries support short-term balancing.

Nuclear generation offers low operational carbon emissions and continuous output, but new reactors often require longer development and licensing periods. Renewables can reduce fuel exposure but require storage, transmission, or firm backup.

Paducah’s proposal resolves that near-term choice in favor of gas and batteries. It does not settle the longer debate over how AI infrastructure should obtain reliable energy.

Ironically, a former uranium-enrichment property will initially host an AI power plan centered on natural gas. That reversal captures the federal priority: secure near-term electricity first, then manage the associated tradeoffs.

Three Signals Will Show Whether the Plan Is Real

Regulatory filings, tenant commitments, and site-specific environmental disclosures will reveal more than additional announcements.

The first signal is the power service agreement submitted for state approval. Its terms should explain cost allocation, grid exports, infrastructure ownership, and protections for other electricity customers.

This review will test a central promise of dedicated generation. If project developers bear the costs and risks tied to their load, the federal model gains credibility.

If utilities seek broad cost recovery from existing customers, the model becomes harder to defend. The filing should also clarify how batteries and excess generation interact with the regional system.

The second signal is a binding tenant or capacity commitment. Brookfield has said it is speaking with commercial partners, but discussions do not establish long-term demand.

A named customer with a defined development phase would strengthen the $100 billion proposal. It would also provide clues about chip density, construction timing, fiber requirements, and reliability standards.

The absence of a major tenant would not immediately end the project. It would weaken the case for building every announced component on the current schedule.

The third signal is a detailed environmental and water package. Readers should watch for air permit applications, cooling specifications, water sources, withdrawal volumes, and parcel-level construction plans.

Those documents should explain how new infrastructure avoids contaminated groundwater and preserves access for cleanup. They should also identify the monitoring requirements that continue during construction and operation.

Clear disclosures would strengthen the argument that industrial reuse and remediation can coexist. Delays, missing consumption estimates, or unclear construction boundaries would increase execution risk.

Construction milestones matter only after these three signals begin aligning. Site preparation without commercial demand or regulatory clearance can create activity without proving the full campus.

The reported 2031 completion target provides a useful deadline, but it should not be treated as a single opening date. Projects of this scale usually bring generation and computing capacity online in stages.

For developers and enterprise AI buyers, Paducah offers another possible source of computing capacity. More capacity can ease shortages, diversify locations, and support workloads that require substantial processing.

For knowledge workers, the immediate effect is less direct. The project illustrates how every AI assistant, search feature, and generated document depends on physical infrastructure with local consequences.

People following the story through Google News should look beyond investment totals and gigawatt figures. The decisive evidence will appear in permits, contracts, utility proceedings, and cleanup documents.

The Paducah proposal has already cleared one important gate: DOE selected a development team. It still needs to show that customers, regulators, energy systems, and environmental obligations can move together.

Watch those three records over the coming months. If they converge, Paducah becomes evidence that federal industrial sites can accelerate AI construction. If they diverge, the project will expose the limits of solving an infrastructure shortage through land access alone.