⚡ Key Takeaways

Brookfield and NextEra Energy announced a $100 billion AI data center campus on the DOE’s former Paducah Gaseous Diffusion Plant site in Kentucky, paired with up to 4.6 gigawatts of dedicated natural gas and battery storage generation. No anchor tenant has been named, but the power infrastructure was secured first — a sequencing that inverts most 2026 hyperscaler deals.

Bottom Line: AI infrastructure developers should treat co-located, dedicated power generation as an emerging requirement for winning both regulatory approval and hyperscaler tenancy, not just a nice-to-have differentiator.

Read Full Analysis ↓

🧭 Decision Radar

Relevance for Algeria
Medium

Algeria is building out its own data center and cloud ambitions under Sonatrach and Algérie Télécom-linked digital infrastructure plans, and the Paducah deal is a live example of how power procurement — not chip supply — is becoming the binding constraint for large-scale AI compute anywhere, including in energy-rich markets like Algeria.
Infrastructure Ready?
Partial

Algeria has the primary energy resource (gas) that underpins Paducah’s model, but lacks the dedicated co-located generation-plus-storage engineering, permitting speed, and private-capital project-finance structures that made this deal move from RFO to $100B announcement in under nine months.
Skills Available?
Limited

Structuring a multi-party power-plus-compute megaproject requires specialized project finance, utility-interconnection, and data center engineering talent that is still scarce in Algeria’s energy and telecom sectors.
Action Timeline
12-24 months

Algerian energy and digital economy planners should treat this as a model to study now, since any comparable domestic project (gas-backed data center campus) would take a similar multi-year path from site identification to financial close.
Key Stakeholders
Ministry of Energy, Sonatrach, Algérie Télécom, data center investors
Decision Type
Educational

This article illustrates an emerging global template — power-first, tenant-agnostic AI infrastructure development — rather than requiring immediate action from Algerian stakeholders.

Quick Take: Algerian energy and digital infrastructure planners should study Paducah’s power-first structure as a template for using the country’s gas and renewable capacity to attract AI data center investment, rather than waiting for hyperscaler demand to arrive first. The key lesson is that guaranteed, dedicated power — not compute commitments — is now the scarce asset that draws AI infrastructure capital.

Advertisement

From Cold War Uranium Enrichment to AI Compute

The Paducah Gaseous Diffusion Plant ran continuously from 1952 to 2013, enriching uranium first for nuclear weapons and later for commercial reactor fuel before it became the last government-owned enrichment facility in the United States to shut down. The Department of Energy took the roughly 3,556-acre site back into full federal control on October 21, 2014, and it has spent the years since in environmental cleanup status — a Cold War industrial relic waiting for a second use.

That second use arrived on July 29, 2026, when Brookfield and NextEra Energy announced plans to build a $100 billion AI data center campus on the site, funded entirely by private capital. The DOE had issued a request for offers on November 4, 2025, with a response deadline of January 30, 2026 — Brookfield’s winning bid moved from solicitation to public announcement in under nine months.

The Power Stack: 4.6 Gigawatts Before a Single GPU Ships

What distinguishes the Paducah deal from most AI campus announcements is that the power comes first and is contractually inseparable from the compute. Per PowerMag’s reporting on the deal structure, Brookfield will lease the land from the DOE and develop and operate the data center campus itself, while NextEra Energy separately builds, owns, and operates up to 2 gigawatts of new natural gas-fired generation and up to 2.6 gigawatts of battery energy storage — a combined 4.6 GW of dedicated capacity.

That figure is deliberately larger than what the campus needs to run. The project targets up to 1.8 GW of utility-supplied capacity and more than 1.2 GW of actual compute load, meaning the 4.6 GW of generation and storage exceeds committed demand — PowerMag notes the surplus is designed so excess electricity can flow onto the regional grid rather than competing with it. Construction is targeted for completion by 2031, with full buildout following in 2032.

Three regional utilities round out the coalition: Big Rivers Electric Power Corporation, Jackson Purchase Energy Cooperative, and Paducah Power System. Big Rivers CEO Don Gulley framed the arrangement as additive rather than extractive, saying the project “will provide significant benefits to our members without compromising affordability or reliability,” according to PowerMag.

Why the DOE Wanted This Specific Site

The Paducah announcement is part of a broader federal push to reuse decommissioned DOE land for AI infrastructure rather than leave it in cleanup limbo indefinitely. Tim Walsh, the DOE’s Assistant Secretary for Environmental Management, tied the deal directly to that strategy: “We are intently focused on finding ways to put federal land back to use for American taxpayers. That commitment is at the heart of our office’s American Energy Hubs initiative,” he said, as reported by PowerMag.

For NextEra, the framing is about self-sufficiency rather than grid strain — a direct answer to the interconnection-queue bottlenecks now surfacing in states like Texas. NextEra CEO John Ketchum said “the data center will bring its own power, pay for its own power infrastructure and create good-paying jobs for local workers,” per PowerMag’s coverage. Brookfield CEO Bruce Flatt described the campus as the opening move in a larger commitment, calling it the seed of the firm’s plan to invest $100 billion in AI infrastructure with the goal that “innovation and affordability go hand in hand.”

No anchor tenant — the hyperscaler or AI lab that will actually rent the compute — has been publicly named. TechTimes reported the project moving forward with power infrastructure secured before any customer commitment was disclosed, a build-it-first sequencing that is itself notable given how tightly compute deals and power deals have been coupled in most 2026 hyperscaler announcements.

That sequencing inverts the pattern set by most of the year’s marquee AI infrastructure deals, where a hyperscaler or AI lab signs a multi-gigawatt compute commitment first and power procurement follows as a downstream negotiation with utilities and generation partners. Paducah instead locks in the harder-to-secure asset — dedicated, permitted, under-construction generation on federally controlled land — before a tenant is named, betting that guaranteed power will itself be the scarce resource that draws hyperscaler interest once the campus nears energization.

Advertisement

Coalition Economics: Who Carries the Risk

The four-party structure splits risk in a way that differs from a single-developer campus. Brookfield carries the real estate and data center construction risk on land it leases rather than owns outright, keeping the capital-intensive building shell separate from the balance sheet holding the underlying property. NextEra, as the sole owner-operator of the 4.6 GW generation and storage stack, carries the commodity and interconnection risk on the power side, a business closer to its core utility operations than to data center leasing. The three regional utilities — Big Rivers, Jackson Purchase, and Paducah Power System — absorb neither construction nor generation risk directly, but their cooperation is what turns a private mega-project into one the surrounding community and state regulators are willing to approve without the kind of pushback now playing out in Texas and New York.

What This Means for AI Infrastructure Developers

1. Treat co-located generation as the new baseline, not a differentiator

Paducah’s 4.6 GW of dedicated gas and battery capacity is a direct response to the interconnection queues now stalling projects in ERCOT and other grid territories. Developers still planning to rely primarily on grid draw should model the risk that co-located generation becomes a de facto requirement for winning both regulatory approval and hyperscaler tenancy commitments.

2. Federal land reuse is now a live sourcing channel

The DOE’s American Energy Hubs initiative turned a Cold War industrial site into a nine-month path from RFO to $100 billion announcement. Developers should track other decommissioned federal sites — former weapons, energy, or industrial facilities — as a sourcing pipeline that can move faster than greenfield permitting in contested markets.

3. Build power-to-grid surplus into project economics, not just campus load

By sizing generation above campus demand, Brookfield and NextEra positioned the project to sell into local utility relationships rather than simply extract from them. That surplus-to-grid design is a template for winning cooperation from regional utilities like Big Rivers and Jackson Purchase, whose support is often a prerequisite for local and state approval.

4. Expect tenant-agnostic power deals to become more common

Sequencing power infrastructure ahead of a named compute tenant reduces financing risk tied to any single hyperscaler’s capex cycle. Investors evaluating similar deals should weigh whether power-first structuring, rather than tenant-first, better insulates a project from the kind of capex pullbacks that have periodically rattled AI infrastructure valuations in 2026.

Where This Fits in 2026’s Power-Constrained Buildout

Paducah lands at a moment when power, not chips, is the binding constraint on AI infrastructure growth. Texas paused new data center grid interconnections on August 3, 2026 after its queue hit 474 gigawatts of requests it could not physically accommodate, and New York enacted its own approval pause in July 2026. Against that backdrop, a $100 billion project that brings 4.6 GW of dedicated generation on federally reused land — rather than requesting it from an already-strained regional grid — reads less like an isolated mega-deal and more like the industry’s answer to the exact bottleneck regulators in Texas and New York are now trying to contain. Whether Paducah’s power-first model becomes the standard playbook, or remains a one-off enabled by unusually favorable federal land access, will shape how much of the next wave of AI capacity gets built in already-congested grid territories versus reused industrial sites like this one.

Follow AlgeriaTech on LinkedIn for professional tech analysis Follow on LinkedIn
Follow @AlgeriaTechNews on X for daily tech insights Follow on X

Advertisement

Frequently Asked Questions

What is the Paducah AI data center campus and how much will it cost?

Brookfield and NextEra Energy announced on July 29, 2026 a plan to build a $100 billion AI data center campus on the Department of Energy’s former Paducah Gaseous Diffusion Plant site in western Kentucky. The project pairs the data center campus with up to 4.6 gigawatts of dedicated natural gas and battery storage generation, funded entirely by private capital.

Why did Brookfield and NextEra choose a former nuclear enrichment site?

The Paducah Gaseous Diffusion Plant operated as a uranium enrichment facility from 1952 to 2013 before the Department of Energy took the roughly 3,556-acre site back into full federal control in 2014. The DOE has been pushing to reuse decommissioned federal land for AI infrastructure under its American Energy Hubs initiative, and issued a request for offers for the Paducah site in November 2025 that Brookfield won within months.

Has a hyperscaler or AI company signed on to use the Paducah campus?

No anchor tenant has been publicly named as of the announcement. Brookfield and NextEra secured the power infrastructure — 4.6 GW of dedicated generation and storage — before any compute customer commitment was disclosed, a sequencing that inverts the pattern set by most 2026 hyperscaler deals, where compute commitments typically come before power procurement.

Sources & Further Reading