DOE Selects Brookfield for $100 Billion Kentucky AI Data Center and Power Complex

The U.S. Department of Energy has selected Brookfield to transform a former Kentucky uranium site into a $100 billion AI infrastructure complex. The plan pairs a 1.8-gigawatt data center campus with new gas generation, batteries, and transmission upgrades.

That combination creates the central conflict behind this Google News story. A federal cleanup site built for Cold War uranium enrichment is becoming a foundation for another strategic competition, the race for AI computing capacity.

The announcement is also more complicated than a conventional data center deal. Brookfield still needs customers, NextEra Energy needs regulatory approvals, and environmental work at the Paducah property remains unfinished. The project is a development framework, not a completed campus.

What DOE Approved at the Paducah Site

DOE is offering federal land and coordinated infrastructure, while private investors assume the cost and execution risk.

The department selected Brookfield Asset Management to develop and operate an AI data center campus at the government-owned Paducah Gaseous Diffusion Plant. DOE announced the selection on July 29, 2026, according to the project report.

NextEra Energy would build and own the associated power assets. The proposed system includes 2 gigawatts of natural gas generation and 2.6 gigawatts of battery storage. Transmission upgrades would connect these resources with the campus and surrounding grid.

The planned computing campus would require up to 1.8 gigawatts. That load approaches the electricity demand of a large metropolitan area, depending on utilization and operating conditions.

The gas plant would become Kentucky’s largest if completed at the announced capacity. Batteries would support reliability and load management, but they would not create primary energy. Their performance would depend on how often they cycle, their duration, and the resources charging them.

DOE expects construction to reach completion in 2031. That schedule leaves five years for commercial agreements, engineering, permitting, equipment procurement, utility approvals, construction, and commissioning.

The department began seeking offers for the site in October 2025. Its federal solicitation asked companies to propose both computing facilities and supporting energy projects.

This structure reflects a shift in federal industrial policy. DOE is not simply leasing an empty parcel. It is using federal land, existing infrastructure, and coordinated permitting to shorten an otherwise difficult development process.

Brookfield brings experience across real estate, energy, and digital infrastructure. The firm reports more than $1 trillion in assets under management, although its total portfolio does not represent committed Paducah capital.

A Brookfield spokesperson estimated that infrastructure construction would represent about 30% of the announced investment. Servers, networking equipment, and semiconductor systems would account for most of the remainder.

That distinction matters because the $100 billion figure describes a potential full buildout. It does not mean the entire amount has already been financed, contracted, or spent.

Brookfield also said discussions with prospective computing customers were continuing. No hyperscaler, AI laboratory, or cloud provider had been publicly confirmed when the project was announced.

The proposal therefore joins three linked businesses. Brookfield coordinates capital and campus development, NextEra supplies energy infrastructure, and future tenants purchase computing space or capacity.

Each part depends on the others. A customer needs credible power delivery before committing. A power developer needs dependable demand before building. Investors need both sides to support years of construction spending.

The Paducah AI project tries to resolve that coordination problem within one development package. That integrated structure is the announcement’s most consequential feature.

Why a Former Uranium Plant Fits the AI Buildout

Paducah offers industrial land and an energy history, but its value comes from coordination rather than an instantly usable facility.

The Paducah Gaseous Diffusion Plant began producing enriched uranium in 1952. It initially supported national defense and later supplied material for commercial nuclear fuel.

Gaseous diffusion separated uranium isotopes by repeatedly moving uranium hexafluoride gas through porous barriers. The process consumed enormous amounts of electricity and has since been replaced by more efficient centrifuge technology.

Commercial enrichment operations ended in 2013. DOE then assumed responsibility for deactivation, decommissioning, waste management, and long-term environmental work.

The location covers roughly 3,500 acres near Paducah in western McCracken County. The former production area occupies only part of that federal property.

Large industrial sites can suit data center development because they offer space, controlled access, utility connections, and communities familiar with energy-intensive operations. Federal ownership can also reduce fragmented land negotiations.

Those qualities do not make Paducah construction-ready. Existing transmission capacity, environmental restrictions, water systems, road access, and buildable parcels still require detailed engineering.

Yet the site provides something increasingly scarce: a place where computing demand and new generation can be planned together. Developers can design the power complex around a known campus instead of waiting for conventional grid expansion.

DOE identified Paducah among 16 federal sites that might host AI infrastructure. The selection followed a broader policy to use federal properties for data centers and dedicated energy projects.

Paducah is not the government’s only test. DOE selected a related development at the former Portsmouth Gaseous Diffusion Plant in Ohio. That project also connects an old uranium complex with large-scale computing and power.

The department later chose Amentum to negotiate a lease at South Carolina’s Savannah River Site. That proposal involves a separate data center and power development on another former nuclear production property.

Together, these projects establish a repeatable federal model. The government supplies land and administrative coordination, while private companies finance, build, and operate the commercial infrastructure.

The historical reversal is striking. Gaseous diffusion once converted electricity into strategically important nuclear material. The new campus would convert electricity into strategically important computing capacity.

However, the physical assets are not simply changing functions. Old enrichment buildings will not become server halls through a straightforward renovation. New facilities require modern foundations, cooling, networking, substations, security, and environmental controls.

That makes “repurposing” a policy description more than a literal construction plan. The project reuses the site’s industrial position and federal ownership, not necessarily its original production equipment.

The location also carries a workforce argument. Western Kentucky has experience with industrial construction, energy operations, environmental remediation, and regulated facilities.

Those capabilities can support development, but data centers require specialized electrical, mechanical, and network skills. Local hiring outcomes will depend on training, contractor decisions, and the mix of permanent operations roles.

Construction could produce thousands of temporary jobs, according to federal officials. The lasting employment effect remains less certain because hyperscale campuses operate differently from manufacturing plants.

The Kentucky AI data center nevertheless gives DOE a visible economic-development narrative. A property associated with industrial decline becomes a platform for private investment and national technology policy.

That narrative now faces a practical test. The site must host new development without allowing commercial urgency to weaken cleanup obligations or community protections.

Google News Focuses on the Real Bottleneck: Power

The project treats electricity as part of the AI product, not as a utility service ordered after the servers arrive.

AI infrastructure demand has exposed a difficult constraint. Companies can order processors faster than utilities can build generation, substations, and high-voltage transmission.

A gigawatt-scale data center cannot rely on an ordinary commercial connection. It requires continuous power, redundant systems, careful voltage management, cooling capacity, and backup arrangements.

The Paducah proposal responds by bundling the computing campus with its energy supply. NextEra would develop the generation, battery, and transmission components alongside Brookfield’s data center construction.

This model reduces one source of uncertainty. A tenant can evaluate a package containing land, power, and development capital rather than negotiating each element separately.

It also moves the power question into the project’s financial core. Electricity availability determines how quickly customers can install accelerators and begin generating revenue from them.

The national demand forecast explains why developers are taking this approach. The Energy Information Administration expects rising computing demand to drive sustained U.S. electricity growth through 2027.

EIA projected electricity consumption growth of 1% in 2026 and 3% in 2027. That would produce the strongest four-year demand expansion since 2000.

The agency separately found that U.S. electricity demand grew about 1.7% annually from 2020 through 2025. The comparable rate from 2005 through 2019 was only 0.1%.

Utilities built many planning assumptions during that earlier period of nearly flat demand. AI campuses now challenge those assumptions with concentrated loads that can arrive faster than traditional industrial projects.

Natural gas offers developers several advantages. Combined-cycle plants can supply dependable output, gas infrastructure is familiar, and turbine projects generally mature faster than new nuclear reactors.

Those advantages explain Paducah’s proposed 2-gigawatt gas plant. They also reveal the tradeoff at the center of the project.

Gas can provide firm power for AI workloads, but it creates long-lived fuel dependence and carbon emissions. A campus scheduled for 2031 would operate through decades of changing energy policy and technology.

Battery storage helps manage rapid changes in demand and temporary disruptions. It can shift electricity across hours and support grid stability, depending on its technical configuration.

However, the announcement provided battery power capacity rather than energy duration. A 2.6-gigawatt battery can behave very differently if it stores two hours of energy instead of eight.

That missing detail limits conclusions about how much the battery system can reduce gas use. It also prevents a complete assessment of the campus’s resilience during extended outages.

The planned gas capacity exceeds the stated data center load by 200 megawatts. DOE said excess electricity could flow to the regional grid.

That possibility gives the project a public-grid dimension. The plant might support regional supply when campus demand falls, but utility rules will determine dispatch, cost recovery, and customer exposure.

The model also pressures competing data center regions. Northern Virginia, Texas, Ohio, and other established markets must add power while managing transmission queues and public opposition.

Paducah offers a different proposition: build on federal land beside new generation, then coordinate the approvals as one strategic infrastructure program.

That proposition becomes compelling only if it saves time. A 2031 target is ambitious enough to matter, yet distant enough for competitors to expand their own powered-land pipelines.

The project’s main opponent is therefore not another developer. It is the conventional sequence in which data centers request power after site selection and wait for utilities to catch up.

Paducah reverses that order. It begins with an integrated energy and computing plan, then seeks tenants capable of consuming the resulting capacity.

The $100 Billion Promise Still Has Regulatory Gaps

Paducah has a selected developer and an announced design, but it does not yet have every approval, customer, or construction commitment.

The largest immediate hurdle is Kentucky utility regulation. A power service agreement must receive approval from state regulators before the proposed supply arrangement can proceed.

That review will address questions beyond the campus boundary. Regulators must evaluate costs, reliability, contractual protections, and the treatment of electricity delivered to the broader grid.

Ratepayer risk is especially sensitive. Residents and small businesses do not want to finance generation built mainly for a private data center customer.

Developers can reduce that concern through long-term contracts, minimum payment commitments, collateral, and exit protections. The announcement did not disclose a complete commercial structure.

The absence of a named tenant creates another gap. Brookfield confirmed that discussions were underway, but prospective occupants had not been identified publicly.

A 1.8-gigawatt campus could serve one major company or several customers. The choice would affect construction phases, equipment density, network design, and financial concentration.

Customer identity also affects emissions accounting. Major technology companies have different climate commitments, clean-energy purchasing strategies, and standards for matching electricity consumption.

The project’s headline value needs similar caution. The $100 billion estimate includes computing equipment that future tenants would install across the campus.

Servers and chips can be purchased in phases, replaced during operations, or omitted if demand changes. The number is therefore a full-development scenario, not a guaranteed opening-day investment.

The timeline faces equipment constraints too. Gas turbines, transformers, switchgear, generators, and high-voltage components have experienced long procurement cycles amid rising power demand.

A large developer can reserve equipment and coordinate suppliers early. It cannot eliminate factory capacity limits or every permitting delay.

Transmission is another dependency. Even a campus with on-site generation needs grid connections for balancing, backup supply, exports, maintenance periods, and market participation.

The project must therefore satisfy both behind-the-meter and regional requirements. Behind-the-meter generation sits near a customer, but it does not automatically operate independently from the public grid.

Fuel delivery also requires scrutiny. A 2-gigawatt gas facility needs dependable pipeline capacity under peak conditions, when homes and other power plants may compete for supply.

Neither battery storage nor excess generation resolves a pipeline constraint. Developers must secure firm transportation and demonstrate how the plant operates during extreme weather.

Water use remains unclear as well. Data centers need cooling, although consumption varies widely by cooling design, climate, server density, and operating strategy.

The site’s 2026 advisory materials referenced an existing industrial water system and anticipated gas-related demand. The final design must explain withdrawal, recycling, discharge, and drought management.

The critical issue is not whether one technology always uses more water. It is whether the complete campus fits local water capacity without shifting costs or environmental burdens.

Commercial success presents a final uncertainty. AI spending is strong, but a project finishing in 2031 must compete against future chips, more efficient models, and other large campuses.

Higher processor efficiency can lower the energy needed for each computation. It can also increase total demand by making AI services cheaper and more widely used.

No forecast can settle that tension today. Long-term contracts and phased construction will reveal how much demand customers are willing to guarantee.

The federal selection lowers development risk, but it does not remove market risk. Paducah becomes real as contracts, permits, equipment orders, and completed buildings replace forecasts.

Cleanup Is Part of the Infrastructure Story

The former uranium site is not an empty industrial park, and ongoing contamination controls must shape where construction occurs.

The Environmental Protection Agency lists the Paducah Gaseous Diffusion Plant as a Superfund site. Uranium operations produced radioactive, hazardous, mixed, and conventional wastes over more than five decades.

Investigations identified contamination in groundwater, soils, surface water, and sediment. Contaminants include trichloroethylene, polychlorinated biphenyls, technetium-99, uranium, and other radioactive elements.

EPA added the property to the National Priorities List in 1994. DOE, EPA, and Kentucky regulators have operated under a federal cleanup agreement since 1998.

The current Superfund profile says human exposure is under control. That finding means identified exposure pathways are being managed under current conditions.

It does not mean cleanup is complete. EPA separately reports that groundwater migration is not under control across the entire site.

DOE continues groundwater treatment, monitoring, waste disposition, building deactivation, and demolition. The federal cleanup program anticipates additional work extending through 2065.

That schedule overlaps the data center’s proposed operation by several decades. Construction and remediation will need compatible boundaries, access controls, monitoring systems, and emergency plans.

The site’s size makes coexistence possible. New development can occupy selected parcels while cleanup continues elsewhere, provided regulators verify that each parcel supports its proposed use.

Precedent already exists at Paducah. DOE previously leased a parcel to General Matter for a proposed commercial uranium enrichment facility.

That arrangement demonstrated that reindustrialization can advance before every cleanup action finishes. It did not establish that every parcel presents the same conditions or constraints.

A data center introduces different requirements. Heavy electrical infrastructure, underground utilities, water lines, foundations, and roads can disturb soil or alter drainage.

Developers must understand subsurface conditions before finalizing those systems. Construction plans also need procedures for unexpected contaminated material.

The environmental issue is therefore more specific than a choice between development and cleanup. The harder question is whether reuse supports remediation without creating new exposure pathways or administrative conflicts.

DOE argues that leasing land can lower long-term federal costs and return portions of the property to productive use. Private investment may also improve roads, utilities, and emergency capabilities.

Critics can reasonably ask whether commercial deadlines will pressure regulators to accept incomplete characterization. They can also question who pays if construction encounters additional contamination.

Those concerns require enforceable agreements, not broad assurances. Lease boundaries, cleanup responsibilities, monitoring access, liability allocation, and stop-work procedures must be clear.

Public reporting matters because the project combines several oversight systems. DOE manages federal land, EPA oversees Superfund obligations, Kentucky reviews utility arrangements, and local agencies address community impacts.

Fragmented authority can provide multiple safeguards. It can also obscure accountability when each institution controls only part of the decision.

The Paducah development will need a public record that connects these reviews. Residents should be able to see how environmental findings affect construction and how commercial changes affect cleanup.

Trust will depend on measurable conditions. Groundwater data, approved parcel boundaries, construction monitoring, water plans, and incident reporting offer stronger evidence than investment announcements.

The project’s Cold War history gives it symbolic force. That same history imposes obligations that a greenfield campus would not carry.

Three Signals Will Decide Whether the Kentucky AI Data Center Works

Regulatory approval, binding customer commitments, and visible construction milestones will determine whether Paducah becomes a model or remains an ambitious plan.

The first signal is the power service agreement submitted to Kentucky regulators. Its terms should show who pays for generation, transmission, battery capacity, and unexpected cost increases.

A strong agreement would place project-specific risks on developers and large customers. It would also protect existing utility customers if campus demand arrives late or falls below forecasts.

Regulators should explain how excess electricity enters the grid and how associated revenue is allocated. They should also address plant retirement obligations and long-term fuel exposure.

Approval with transparent protections would strengthen DOE’s integrated-development model. A disputed proceeding, major redesign, or rejection would weaken the claimed schedule advantage.

The second signal is a binding anchor-customer announcement. Brookfield needs tenants willing to commit to substantial capacity before the largest construction phases make economic sense.

A credible announcement should identify capacity, timing, and contractual duration without exposing sensitive commercial details. It should also explain the customer’s energy and emissions strategy.

A memorandum of understanding would provide less certainty than a contracted power and occupancy commitment. Investors and local officials should distinguish between the two.

A named hyperscaler would validate the campus’s scale. Multiple customers could reduce concentration risk, although coordinating their delivery schedules would add complexity.

Continued silence would not prove failure, since negotiations can remain confidential. However, prolonged uncertainty would make the $100 billion estimate increasingly speculative.

The third signal is synchronized progress across permits, equipment, and cleanup boundaries. Announcements alone do not reveal whether those tracks are moving together.

Useful milestones include approved parcel plans, major turbine reservations, transmission studies, environmental reviews, and the first data center construction package.

Progress in only one category would be insufficient. A reserved turbine does not produce an operating campus without fuel, transmission, tenants, and a cleared construction site.

The 2031 target becomes credible when these milestones form a sequence. Land preparation should align with utility approvals, while equipment delivery should align with contracted customer demand.

This sequence will also test whether federal land truly shortens development. If Paducah moves faster than conventional campuses, DOE can apply the model at other properties.

If coordination stalls, the project will expose the limits of federal site selection. Land control cannot manufacture turbines, guarantee customers, or replace state utility oversight.

The broader industry should watch the financing structure closely. AI infrastructure increasingly combines real estate, energy, computing hardware, and long-term customer contracts.

That structure distributes risk among more parties. It can also make projects harder to evaluate because the announced total blends distinct spending categories and schedules.

Developers evaluating similar campuses should preserve every decision, permit, technical study, and contract assumption in a searchable project record. A structured knowledge base can help teams compare changing claims with source documents.

For readers following Google News coverage, the next headline matters less than the next filing. Watch the Kentucky regulatory docket, Brookfield’s customer commitments, and DOE’s parcel-level construction approvals.

Those records will answer the central question: can federal land, private capital, and dedicated gas generation deliver AI capacity faster without transferring hidden costs?

Paducah is now a test of that model. Track the three signals, compare each milestone with the 2031 schedule, and treat the $100 billion figure as conditional until contracts become operating infrastructure.