The insatiable power appetite of AI is forcing hyperscalers to think beyond solar farms and wind turbines. With data centers projected to consume up to 12% of U.S. electricity by 2030—up from 4% in 2024—Microsoft, Amazon, and Google are betting billions on nuclear power to secure reliable, carbon-free baseload energy for their AI infrastructure.
What's New
The past 18 months have witnessed an unprecedented wave of nuclear deals from Big Tech. Microsoft signed a landmark 20-year agreement with Constellation Energy to restart the 835MW Three Mile Island Unit 1 reactor—now renamed Crane Clean Energy Center—with commercial operation targeted for 2028. Amazon acquired Talen Energy's 960MW Cumulus Data nuclear-powered campus for $650 million, securing direct access to the Susquehanna nuclear plant's output. Meanwhile, Google committed to 500MW of advanced nuclear power from Kairos Power by 2035, betting on next-generation small modular reactors (SMRs).
But the innovation doesn't stop at legacy reactors. Startups like Deep Fission are developing 15MW microreactors designed to operate one mile underground, while Prodigy Clean Energy is engineering transportable nuclear plants scaling from 1MW to 1,000MW—offering modular solutions for edge data centers and remote deployments.
Technical Deep Dive: Why Nuclear for AI?
The physics of AI training creates a unique energy profile that nuclear power is ideally suited to address. Large language models and generative AI systems require sustained, high-density power delivery—a single hyperscale AI data center can demand 1-5 gigawatts of continuous power. Unlike solar or wind, nuclear provides 24/7 baseload generation with capacity factors exceeding 90%.
According to Deloitte analysis, power demand from AI data centers in the United States could grow more than thirtyfold by 2035, reaching 123 gigawatts. Nuclear energy could meet up to 10% of data center electricity demand by 2035, providing critical grid stability as intermittent renewables scale.
The technical specifications of current deals reveal the scale of commitment:
- Microsoft/Constellation: 835MW pressurized water reactor, 20-year PPA, targeting 2028 restart
- Amazon/Talen: 960MW campus with direct nuclear feed, $650M acquisition
- Google/Kairos: 500MW from 6-7 KP-FHR (fluoride salt-cooled high-temperature) SMRs, first unit online by 2030
- Deep Fission: 15MW underground microreactors at 1-mile depth, $30M funding secured
- Prodigy Clean Energy: Transportable units from 1MW-1,000MW for flexible deployment
Kairos Power's approach is particularly notable. Their Hermes demonstration reactor at Oak Ridge, Tennessee started nuclear construction in May 2025, representing the first non-water-cooled reactor to receive NRC construction approval in over 50 years. The August 2025 agreement with TVA to supply 50MW from Hermes 2 to Google data centers marks the first U.S. utility PPA for an advanced Generation IV SMR.
Market Impact
The nuclear renaissance is reshaping energy infrastructure investment. S&P Global projects data center grid-power demand will rise 22% in 2025 to 61.8 GW, nearly tripling by 2030. This demand surge is creating a seller's market for nuclear capacity, with utilities commanding premium pricing for guaranteed baseload access.
The competitive landscape is intensifying. Meta issued an RFP for up to 4 gigawatts of new nuclear power generation, signaling that the hyperscaler nuclear race extends beyond the current leaders. Amazon, Google, and Meta have joined pledges to triple global nuclear capacity, institutionalizing Big Tech's role as nuclear's most important new customer segment.
For reactor developers, this represents a transformational market opportunity. The U.S. SMR pipeline now exceeds 23.5 GW—half of the global 47 GW total—with hyperscaler demand providing the long-term offtake agreements necessary to secure project financing.
What It Means for Engineers and Businesses
For infrastructure architects: Nuclear-powered data centers will require new approaches to facility design. The guaranteed baseload enables higher power density deployments and reduces the need for on-site backup generation, but requires proximity to reactor sites and long-term site commitment.
For capacity planners: The 2028-2035 timeline for major nuclear projects means current AI infrastructure buildouts must bridge with existing power sources. Hybrid approaches combining grid power, renewables, and battery storage remain essential for near-term deployments.
For enterprise IT leaders: Colocation providers near nuclear facilities will command premium pricing for carbon-free compute. Organizations with aggressive sustainability targets should evaluate nuclear-adjacent hosting options as they become available.
For the broader industry: The nuclear pivot signals that AI's energy demands are fundamentally reshaping utility-scale infrastructure investment. Companies planning decade-scale AI initiatives should factor nuclear availability into geographic strategy.