Hyperscalers Lock in Nuclear Power for AI Campuses

Major tech firms sign multi-decade nuclear power purchase agreements, driving generation asset valuations as data center electricity needs intensify.

Tradesnaut Quant Research Desk · September 21, 2026 · 6 min read · AI Data Centers

Hyperscalers Lock in Nuclear Power for AI Campuses

Key takeaways

What changed

A new chapter is unfolding in how hyperscalers power their digital infrastructure. Rather than solely pursuing intermittent renewable energy, major technology firms are now directly underwriting multi-decade nuclear power agreements to fuel their rapidly expanding AI data centers. This strategic pivot became acutely clear over the past year. Google, for instance, in a significant move reported by Reuters on September 9, 2026, announced an investment of at least €13 billion (

5.1 billion) in AI infrastructure in Finland, including three new data centers, backed by a 22-year deal to purchase up to 50% of the output from Finland's Loviisa nuclear power plant through Fortum. This marks Google's first nuclear energy deal outside the United States.

Meanwhile, Microsoft committed to a 20-year,

6 billion power purchase agreement with Constellation Energy (CEG) to restart the 835-megawatt (MW) Three Mile Island Unit 1 reactor, rebranded as the Crane Clean Energy Center, with commercial operation targeted for late 2027 or 2028. Meta Platforms has also been active, securing over 2.1 gigawatts (GW) from Vistra's (VST) existing nuclear reactors in Ohio and Pennsylvania through 20-year PPAs, alongside 433 MW of capacity upgrades set to come online in the early 2030s. Meta also has a 1.1 GW PPA with Constellation Energy for its Clinton Clean Energy Center in Illinois, effective June 2027. Amazon Web Services similarly signed a 20-year PPA with Vistra in September 2025 for 1,200 MW from its Comanche Peak nuclear facility in Texas. These deals underpin a significant shift in energy procurement for the technology sector, extending far beyond traditional grid purchases or standard renewable energy certificates.

The mechanism

This aggressive pursuit of nuclear power stems from the physics of artificial intelligence. Training and running advanced AI models demands continuous, firm power at a scale unprecedented in corporate energy consumption. A single next-generation AI campus can require 500 MW to over a gigawatt continuously, comparable to a mid-sized city. Traditional renewable sources like solar and wind, while carbon-free, are intermittent and cannot provide the 24/7 baseload power necessary for these operations without extensive and costly battery storage, which itself has an energy footprint. Goldman Sachs Research projected in May 2026 that US data center power demand will more than double from 31 GW in 2025 to 66 GW in 2027. TrendForce estimated global data center power demand capacity will reach 161 GW in 2026, an approximately 31% year-over-year increase, with AI servers driving 33.4% of that total.

The existing electricity grid infrastructure, designed for distributed consumption, struggles to accommodate such concentrated, rapidly growing demand. Interconnection queues in AI hotspots now span four to seven years, and transformer orders can take up to five. This grid bottleneck is forcing hyperscalers to become their own power developers, seeking direct-to-site or dedicated baseload generation. Nuclear power, with its high capacity factor (often above 90% for existing US reactors) and energy density, offers the reliability and scale that AI workloads require. Furthermore, co-locating data centers behind the meter with nuclear facilities can circumvent transmission constraints and ease grid-connection queues. The willingness of tech companies to sign long-term, pre-commercial contracts and even invest in SMR development reflects a deep-seated need for energy security and price stability over decades, rather than short-term cost arbitrage.

Who is exposed

The companies directly benefiting from this shift are the established nuclear utility operators and emerging Small Modular Reactor (SMR) developers. Constellation Energy (CEG), a leader in nuclear generation, is a clear beneficiary. The company's stock last traded at 255.65, up 0.94 (0.37%), with a volume of 6,547,116 shares. Its ability to restart mothballed plants, as with Three Mile Island Unit 1 for Microsoft, or extend the life of existing facilities like Clinton for Meta, demonstrates the value of its installed nuclear fleet. Vistra (VST), another key player, has leveraged its diversified generation mix, including nuclear assets, to secure significant PPAs with Meta and Amazon Web Services. Vistra's stock last stood at 141.35, marking a 0.68 (0.48%) increase, on a volume of 5,268,950 shares. The company reported Q2 2026 Ongoing Operations Adjusted EBITDA of

.77 billion, up more than 30% from Q2 2025, reaffirming its 2026 Adjusted EBITDA guidance of $6.8 billion to $7.6 billion. Vistra's commitment of up to
billion to Helix Digital Infrastructure alongside NVIDIA and KKR further positions it as a preferred power provider for AI data centers.

SMR developers, while still in earlier stages of commercialization, are seeing a surge in interest and investment. NuScale Power (SMR), the only company with an SMR design certified by the U.S. Nuclear Regulatory Commission, is at the forefront. The SMR ticker last traded at 8.32, up 0.05 (0.60%) with a volume of 64,383,612 shares. NuScale's April 2026 agreement with Standard Power to develop two SMR-powered facilities for data centers in Ohio signals a move from theoretical application to concrete site and technology selection. Other SMR companies like TerraPower, Oklo, Kairos Power, and X-energy are also involved in hyperscaler-backed projects. These agreements offer SMR developers critical, upfront capital, accelerating technology development years before reactors generate power.

Quantitative Outlook

The market data for Constellation Energy (CEG) and Vistra (VST) reflects positive sentiment, with both stocks showing modest gains at the close. CEG's last price of 255.65, up 0.37%, and VST's last price of 141.35, up 0.48%, suggest that investors are reacting to the long-term revenue visibility these nuclear PPAs provide. Vistra's strong Q2 2026 Adjusted EBITDA of

.77 billion and reaffirmed 2026 guidance reinforce its financial positioning amid the AI power boom. The NuScale Power (SMR) stock, trading at 8.32, with a 0.60% increase, indicates growing investor attention as the SMR industry moves closer to commercial deployment, despite the absence of fully operational commercial SMRs in the US as of mid-2026.

The outlook for continued growth in hyperscaler nuclear power demand remains robust. AI server power consumption is projected to surpass that of conventional servers by 2027. What would change this picture are significant delays in SMR commercialization, unexpected regulatory hurdles for existing nuclear plant extensions or restarts, or a material slowdown in AI infrastructure buildouts. Investors should closely watch for further definitive deployment agreements involving SMRs, especially those with clear timelines and funding, and monitor utility earnings calls for updates on PPA execution and new contracts. The success of these nuclear ventures will dictate the long-term energy security for the AI revolution and the financial performance of the utilities powering it.

Tags: Constellation Energy, Vistra, NuScale Power, AI Data Centers, Nuclear Energy