Microsoft just signed a deal to restart Three Mile Island’s nuclear reactor. Yes, the same plant that had America’s worst nuclear accident in 1979. They’re paying billions to bring it back online for one reason: their AI data centers need power. Lots of it. More than the existing grid can reliably provide.
Google is investing in small modular reactor projects. OpenAI is in talks with nuclear energy companies about dedicated reactors. Amazon is buying nuclear power through long-term contracts. This isn’t a trend, it’s a fundamental reckoning with AI’s energy consumption. The AI boom is creating a nuclear power renaissance, and the implications are massive.
The Energy Problem Nobody Saw Coming
When ChatGPT launched in November 2022, everyone talked about capabilities and ethics. Almost nobody mentioned energy consumption. That was a mistake.
Training GPT-4 consumed roughly the annual electricity of 5,000 American homes. But training is just the beginning. Running inference (answering queries) consumes far more over time. The explosive growth of AI coding tools alone represents millions of daily queries, each consuming significant energy. A single ChatGPT query uses approximately 10 times the electricity of a Google search. Multiply that by billions of daily queries, and the numbers get astronomical.
Google’s electricity consumption increased 48% from 2021 to 2023, almost entirely due to AI. These companies are adding load to the grid faster than new generation capacity is being built. The obvious question: why not just build more solar and wind? They are, but renewables have a fundamental mismatch. AI data centers need consistent, 24/7 power. Solar fades at night, wind is intermittent, and battery storage at the necessary scale doesn’t exist yet.
Nuclear is the only proven carbon-free technology that provides the massive, consistent baseload power these data centers require.
The Nuclear Solution Takes Shape
Microsoft’s deal to restart Three Mile Island’s Unit 1 reactor is both symbolic and substantive. The reactor, shut down in 2019 for economic reasons, will provide 835 megawatts of power directly to Microsoft’s data centers for 20 years. It’s a win for both the company and the operator, providing dedicated power and guaranteed revenue.
But traditional plants are slow and expensive to build, which is why the industry is pivoting toward Small Modular Reactors (SMRs). These factory-built reactors are smaller (50-300 megawatts) and theoretically faster to deploy. Google has signed agreements to purchase over 500 megawatts from SMR developers like Kairos Power.
The vision is to have SMRs co-located with data centers, giving AI companies energy independence from the grid. This insulates them from price volatility and grid instability, a strategic necessity for maintaining AI leadership. For tech companies used to controlling their infrastructure, owning their power source is the logical next step.
What It Means for the Industry
This nuclear bet will reshape AI development if it pays off. AI clusters will likely concentrate geographically near nuclear plants, creating new tech hubs. Energy access will become a primary competitive moat, where companies with dedicated power have a massive advantage over those relying on the public grid.
The global race is already underway. China is building nuclear capacity faster than anyone, with 22 reactors under construction explicitly to support AI and semiconductor industries. France and South Korea are positioning themselves as AI hubs with abundant nuclear power. In the US, regulatory hurdles have historically made nuclear slow and expensive, but the urgency of AI energy needs is shifting the landscape.
The Environmental Debate
The marriage of AI and nuclear power creates a philosophical dilemma. AI companies present it as a clean energy solution, allowing them to scale without increasing carbon emissions. Pro-nuclear environmentalists agree, arguing that nuclear is essential for decarbonization regardless of AI.
However, anti-nuclear groups argue we’re building massive infrastructure to power technology that may not even be necessary, creating new risks of accidents and waste. The US still has no permanent disposal solution for radioactive waste. Adding thousands of tons of waste annually from new SMRs exacerbates this issue.
Furthermore, the immense power draw of AI data centers raises electricity prices for everyone else and competes for clean energy resources needed to electrify transportation and heating. If SMRs fail to deliver on cost or timeline promises, or if a nuclear accident turns public opinion, the strategy could collapse, leaving AI companies with stranded assets or forcing a return to fossil fuels.
The Bottom Line
AI companies going nuclear is one of the strangest and most consequential business stories of 2025. It reveals that AI’s energy consumption is sustainable only with massive infrastructure changes. It’s driving a nuclear renaissance that seemed impossible five years ago.
Whether this is brilliant strategy or reckless expansion depends on whether you believe the value of AI justifies the costs and risks of nuclear expansion. What is certain is that the computational power behind your chatbot has to come from somewhere. By 2030, there’s a good chance it will be coming from a nuclear reactor built specifically for that purpose.
The industry is betting billions that nuclear technology can scale as fast as their algorithms. We’re about to find out if they’re right.
Sources: Energy industry announcements, technology company disclosures, nuclear regulatory filings.
