Nuclear power purchase agreements for AI refers to the 2024–2026 pattern of AI-scale data-center operators signing long-term Power Purchase Agreements (PPAs) with nuclear generators to meet gigawatt-scale electricity demand. NERC 2025 Long-Term Reliability Assessment (LTRA) warns that this demand is overloading the grid. The defining transactions involved Microsoft and Three Mile Island, Meta and Constellation, and Amazon and Talen Energy.
Rationale
Operators and analysts cite four properties of nuclear generation that fit AI data-center load. Nuclear provides firm, 24/7 baseload power, which matches the flat load profile of AI data centers, as opposed to peaking loads, more closely than intermittent renewables. Individual nuclear units are gigawatt-scale, so a single reactor can power a frontier training run. Nuclear electricity is carbon-free, allowing technology companies to count it against ESG commitments. Long-dated PPAs, typically 15–20-year terms, lock in prices and reduce developer risk for nuclear operators.
Deals
Microsoft signed a 20-year PPA with Constellation in September 2024 to restart the Three Mile Island Unit 1 nuclear reactor, which had been retired in 2019, to be renamed the "Crane Clean Energy Center." It was the first deal to physically restart a retired reactor specifically for AI.
Meta signed a 20-year PPA with Constellation Energy in June 2025 to operate the Clinton Clean Energy Center in Illinois. The Guardian reported the deal on June 3, 2025 under the headline "Meta signs deal with nuclear plant to power AI and datacenters for 20 years."
Amazon agreed in 2024 to purchase power from Talen Energy's Cumulus nuclear complex in Pennsylvania. FERC reviewed the deal and in 2024 limited its grid-interconnect structure.
The pattern extended to reactor-building startups: Valar Atomics, which builds small nuclear reactors for data centers, was reported on July 16, 2026 to be in talks to raise $1 billion at about a $6 billion post-money valuation, in a round Sequoia Capital had discussed leading (Source: New Developments Log/2026-07-16-2205-ai-developments.md).
Counterparties
Constellation Energy, the largest US nuclear operator, was the principal counterparty for both Microsoft and Meta. Talen Energy, based in Pennsylvania, was Amazon's counterparty. Vistra in Texas and Dominion in Virginia, whose service territory overlaps "Data Center Alley," have also drawn data-center interest.
Regulatory frictions
NERC (NERC 2025 Long-Term Reliability Assessment (LTRA)) warns of 224 GW of summer peak demand growth over ten years, driven by data centers, with high-risk regions including MISO, PJM, ERCOT, WECC-Northwest, WECC-Basin, and SERC-Central. FERC interconnect reviews have subjected data-center-nuclear "behind-the-meter" arrangements to scrutiny. Small Modular Reactors (SMRs) remain experimental, drawing technology-company interest but facing slow licensing.
Nuclear PPAs address electricity supply but not other resource constraints. A January 27, 2026 New York Times report, "Microsoft Pledged to Save Water. In the A.I. Era, It Expects Water Use to Soar," noted that nuclear power does not resolve data-center water use. The Information reported on December 16, 2025 that data centers were falling behind schedule ("Why Data Centers Are Falling Behind Schedule"). A NERC report dated January 29, 2026 described data centers increasing power use and warned of shortages.
Relationships
- supports: AI Environmental Impact, Data Center Siting / AI Power Politics, Stargate Project.
- related: NERC 2025 Long-Term Reliability Assessment (LTRA), IEA — Energy and AI, Bain & Company — 6th Annual Global Technology Report (2025), RAND — AI's Power Requirements Under Exponential Growth (2025), Epoch AI — How Much Power Will Frontier AI Training Demand in 2030?, (Source: Raw Sources/SemiAnalysis - Stargate Data Center Layer Cake.md), GAO-25-107172 — Generative AI's Environmental and Human Effects.
- related: AI Bubble Debate — nuclear PPAs are a component of the $500B annual capex figure at the center of the sustainability debate
- related: Circular Financing in AI — long-term PPAs are one mechanism by which lab spending creates obligations that compound the circular-financing pattern