Grid Squeeze: Big Tech Bets $9.8B on Nuclear for AI Boom

Big Tech has committed over 9.8 GW in nuclear deals by 2026, including Three Mile Island restart, as AI data center demand pushes grids to limit with 267% wholesale price surges. Learn how nuclear is powering the AI boom.

big-tech-nuclear-ai-2026
Facebook X LinkedIn Bluesky WhatsApp
en flag nl flag

By 2026, the collision of surging AI energy demand with grid capacity limits has triggered an unprecedented wave of nuclear-backed power purchase agreements, making this the most consequential energy-technology inflection point of the year. Hyperscale AI data centers have pushed electricity grids to their limits, with wholesale power costs rising 267% near major hubs and interconnection delays exceeding three years. In response, Microsoft, Google, Amazon, and Meta have collectively committed over 9.8 GW in nuclear power deals — including the historic restart of Three Mile Island — marking a strategic pivot from renewable-only strategies to baseload nuclear as the cornerstone of AI infrastructure.

Context: The Great Grid Squeeze

Global data center electricity consumption is projected to reach 945 TWh by 2030, up from 415 TWh in 2024, according to the International Energy Agency. The United States alone accounted for 183 TWh in 2024 — 4% of total U.S. electricity — and that figure could hit 580 TWh by 2028. AI-optimized racks draw 60+ kW per rack versus 5-10 kW for standard racks, because AI training clusters run at near-100% utilization for weeks. An NVIDIA H100 server draws ~10 kW compared to 300-500W for a standard server.

This demand surge has overwhelmed grid infrastructure. Interconnection queues in the PJM market, which serves 65 million people across 13 states, currently hold 1,006 active requests representing 118.81 GW of capacity. Processing times for generation interconnection agreements now stretch one to two years even under expedited processes. The AI data center energy crisis has become a defining challenge for grid operators nationwide.

Wholesale electricity prices near major data center clusters have surged 267% over the past five years, according to Bloomberg analysis of Grid Status data. Prices in cities like San Francisco ($35/MWh), Baltimore ($38/MWh), and Phoenix ($21/MWh) have more than doubled since 2020. Data centers now account for 39% of Virginia's electricity consumption, 33% of Oregon's, and 14% of Nebraska's. Utilities filed $29 billion in rate hike requests in just the first half of 2025, up 142% from a year earlier.

Big Tech's Nuclear Pivot: A $9.8 GW Commitment

Faced with these constraints, the four largest tech companies have collectively committed over 9.8 GW of nuclear capacity through power purchase agreements, direct investments, and project development deals. This represents more nuclear energy capital commitment than any prior decade in U.S. history.

Microsoft: Restarting Three Mile Island

Microsoft secured a $16 billion, 20-year power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1 in Pennsylvania — now rebranded as the Crane Clean Energy Center. The 835-megawatt reactor, which ceased operations in 2019 due to economic pressures, is expected to return to service in late 2027. The project is backed by a $1 billion loan from the U.S. Department of Energy and is expected to create roughly 3,400 jobs and generate $16 billion in GDP for Pennsylvania. Microsoft's deal makes it the first hyperscaler to receive nuclear-powered AI data center electrons.

Amazon: 1.92 GW from Susquehanna Plus SMR Investment

Amazon Web Services signed a 17-year PPA with Talen Energy for 1.92 GW of power from the 2.5 GW Susquehanna nuclear plant in Pennsylvania, running until 2042 with an option to extend. The agreement shifted from a behind-the-meter arrangement (rejected by FERC over cost and reliability concerns) to a front-of-the-meter model where power flows to the PJM grid. Amazon also led a $500 million financing round for SMR developer X-energy, with plans for up to 12 Xe-100 small modular reactors. The company has committed $20 billion to Pennsylvania, bringing 1,250 jobs.

Google: 500 MW from Kairos Power SMRs

Google signed the world's first corporate agreement to purchase nuclear energy from multiple small modular reactors, partnering with Kairos Power for a fleet totaling 500 MW by 2035. The first deployment is targeted by 2030. Kairos Power uses fluoride salt-cooled, high-temperature reactor (KP-FHR) technology and received a construction permit from the U.S. Nuclear Regulatory Commission in 2023 for its Hermes demonstration reactor in Tennessee. Google has signed over 115 clean energy agreements totaling 14 GW since 2010.

Meta: Up to 6.6 GW Across Three Partnerships

Meta announced landmark agreements with Vistra, TerraPower, and Oklo to unlock up to 6.6 GW of new and existing nuclear energy by 2035, making it the largest corporate nuclear purchaser in American history. The Vistra deal supports operating plants (Perry, Davis-Besse, and Beaver Valley) in Ohio and Pennsylvania, extending their lifespan and increasing output. TerraPower, co-founded by Bill Gates, will develop up to eight Natrium units generating 2.8 GW of baseload power plus 1.2 GW of storage, with delivery as early as 2032. Oklo will build an advanced nuclear campus in Pike County, Ohio, adding up to 1.2 GW by 2030. The Meta nuclear energy strategy reflects a broader trend where 17% of corporate clean energy deals in 2025 came from around-the-clock sources like nuclear, geothermal, and hydro.

Why Nuclear? The Baseload Advantage

Nuclear power's key advantage for AI data centers is its ability to provide clean electricity 24/7 — when solar cannot — at high capacity factors exceeding 90%. Unlike wind and solar, which require massive battery storage to firm intermittent output, nuclear plants deliver consistent baseload power that matches the relentless energy demands of AI training clusters. The nuclear energy vs renewables for data centers debate has shifted decisively as tech companies confront the physical limits of grid interconnection.

Existing reactor restarts deliver power fastest: Microsoft's Three Mile Island restart targets 2027, while new-build SMRs offer greater scale but longer timelines — most won't deliver power before the mid-2030s. This leaves a gap filled by natural gas and renewables in the interim. The cost dynamics also favor existing nuclear: SMR costs range from $50-$130 per megawatt-hour for later plants, while existing nuclear is among the cheapest sources of clean power.

Implications for Energy Markets and Regulation

This nuclear pivot carries profound implications. Residential ratepayers face $31 billion in rate hikes as utilities pass on grid upgrade costs, according to an analysis by Informed Clearly. The utility regulation and data center costs debate is intensifying, with proposals for data center-specific tariffs and high-load tariffs emerging in multiple states.

Tech capital expenditure has surpassed oil and gas investment in 2026. The four largest tech companies — Alphabet, Amazon, Meta, and Microsoft — are preparing to pour a combined $650 billion or more into AI infrastructure in 2026 alone, representing the largest single-year corporate investment cycle in history. This dwarfs the oil and gas sector, where capital reinvestment rates are projected at ~50% with Brent crude averaging under $60 per barrel.

The PJM grid, which serves the Mid-Atlantic and parts of the Midwest, is at the epicenter of this transformation. With 118.81 GW of active interconnection requests and data center construction pipeline adding 140 GW of new load nationally, the grid faces unprecedented strain. The Nuclear Regulatory Commission has approved construction permits for advanced reactors, but regulatory frameworks for colocation and behind-the-meter arrangements remain contested.

Expert Perspectives

"The scale of AI energy demand is unlike anything we've seen in the power sector," said a senior energy analyst at the IEA. "Without significant transmission infrastructure investments, up to 20% of planned data center projects could face delays."

"Nuclear power is the only carbon-free source that can deliver the 24/7 reliability that AI infrastructure requires," noted a Constellation Energy executive. "The Three Mile Island restart is a symbol of how far the industry has come since 1979."

"These deals represent more nuclear energy capital commitment than any prior decade in US history," said a nuclear industry analyst. "But most projects won't deliver power before the mid-2030s, leaving a critical gap."

FAQ

Why are tech companies investing in nuclear power?

AI data centers require 24/7 reliable electricity at massive scale. Nuclear power provides carbon-free baseload energy with over 90% capacity factor, unlike intermittent renewables that require storage. Grid interconnection delays exceeding three years and wholesale price surges of 267% have made on-site or contracted nuclear power an attractive solution.

How much nuclear capacity have Big Tech companies committed to?

As of May 2026, Microsoft, Google, Amazon, and Meta have collectively committed over 9.8 GW of nuclear capacity through power purchase agreements and development deals. Meta leads with up to 6.6 GW, followed by Amazon (1.92 GW), Microsoft (835 MW), and Google (500 MW).

When will Three Mile Island restart?

Three Mile Island Unit 1, now called the Crane Clean Energy Center, is expected to return to service in late 2027. Microsoft has a 20-year PPA for its 835 MW output. The $1.6 billion revival is backed by a $1 billion federal loan from the Department of Energy.

How does this affect residential electricity rates?

Residential ratepayers face an estimated $31 billion in rate hikes as utilities invest in grid upgrades to serve data center demand. States like Connecticut (up 18.4%) and Maine (up 36.3%) have seen steep increases. Debates are emerging over whether tech giants or consumers should foot the bill.

What are small modular reactors (SMRs)?

SMRs are advanced nuclear reactors with capacity typically under 300 MW per unit, designed for factory fabrication and modular deployment. Companies like Kairos Power, TerraPower, Oklo, and X-energy are developing SMRs that tech companies plan to deploy by the early 2030s to power data centers.

Conclusion: A New Energy Paradigm

The nuclear pivot by Big Tech represents a fundamental shift in energy strategy. As AI infrastructure demands continue to grow — with data center capacity projected to nearly double from 80 GW in 2025 to 150 GW by 2028 — the need for reliable, carbon-free baseload power will only intensify. The success of these nuclear deals will depend on regulatory approvals, construction timelines, and public acceptance. But one thing is clear: the AI boom has made nuclear power indispensable to the future of digital infrastructure.

Sources

Related

nuclear-data-centers-smr-2026
Energy

Nuclear-Powered Data Centers: Big Tech's $400B SMR Bet in 2026

Global data center electricity consumption exceeds 1,000 TWh in 2026 as Big Tech commits $400B+ to nuclear-powered...

ai-nuclear-energy-markets-2026
Ai

AI-Nuclear Nexus: How Hyperscalers Reshape Energy Markets in 2026

AI data centers will consume 1,100 TWh in 2026, driving Microsoft, Google, Amazon, and Meta to sign 9.8 GW of...

smrs-nuclear-ai-data-centers
Ai

SMRs: Tech Giants Bet on Nuclear Power for AI Boom

Tech giants Google, Amazon, Microsoft, and Meta are investing billions in small modular reactors (SMRs) to power AI...

nuclear-power-ai-data-centers-2026
Energy

Nuclear Bargain: Big Tech Bets on Atomic Power for AI Boom in 2026

AI data centers will consume 1,000 TWh by 2026, driving a 49 GW US power shortfall. Microsoft, Google, and Amazon...

ai-nuclear-data-centers-2026
Ai

AI-Nuclear Nexus: Data Center Power Demand Reshapes Energy in 2026

AI data centers will consume 1,000+ TWh in 2026, driving a nuclear energy renaissance. Tech giants sign 47 GW in...

ai-data-center-power-grid-2026
Energy

AI Power Paradox: Data Centers Strain Global Grid in 2026

AI data centers will consume nearly 1,000 TWh by 2026, driving a 165% surge in power demand. Grid delays force...