Microsoft's Nuclear Bet: Why AI's Power Demands Just Brought Three Mile Island Back to Life

In a deal that surprised much of the energy world, Constellation Energy agreed in September 2024 to spend $1.6 billion restarting a reactor at Pennsylvania's Three Mile Island plant. Microsoft committed to buying every megawatt the facility produces for 20 years. The site, long associated with the 1979 partial meltdown, now carries a new name: the Crane Clean Energy Center. Constellation targets a 2028 restart, with hopes of accelerating to 2027 if regulators move quickly.

The transaction reveals a hard truth for the tech industry. Artificial intelligence consumes electricity at a scale that caught even hyperscalers off guard. Training a single large model can burn through as much power as thousands of homes use in a year. Running those models in production multiplies the load. Data centers now draw power like small cities. Tech companies face a choice: slow AI expansion or secure always-on generation.

Renewables alone can't deliver the constant output these facilities require. Natural gas fills gaps but brings emissions and price swings. Nuclear plants run 24/7 without carbon emissions and achieve capacity factors above 90 percent. That reliability commands a premium. Microsoft reportedly pays over $100 per megawatt-hour for the Crane output, well above typical renewable contracts.

Microsoft isn't alone in this strategy. Google signed agreements with Kairos Power for small modular reactors. Amazon invested in the Susquehanna nuclear site and is exploring multi-gigawatt deals. Meta issued requests for 1 to 4 gigawatts of new nuclear capacity. Collectively, big tech has committed to more than 10 gigawatts of nuclear-related projects in the past year, according to industry data.

Restarting an existing reactor is faster than building new ones, but it's still complex. Constellation needs Nuclear Regulatory Commission approvals, equipment upgrades, fresh fuel, and grid interconnection fixes. Recent transmission delays threatened to push full operations past 2030. The Federal Energy Regulatory Commission granted waivers, and the current target is 2027.

Constellation CEO Joe Dominguez framed the stakes clearly: "Powering industries critical to our nation's global economic and technological competitiveness requires an abundance of energy that is carbon-free and reliable every hour of every day. Nuclear plants are the only energy sources that can consistently deliver on that promise."

Critics note the symbolic weight of reviving a plant linked to America's worst commercial nuclear accident. The Unit 1 reactor being restarted operated safely until its 2019 economic shutdown. Local communities express both hope for jobs and lingering unease. Pennsylvania officials support the restart as economic development and a signal of readiness for high-tech industry.

The broader picture is stark. U.S. data center electricity use could double or triple by decade's end, according to forecasts. Wholesale power prices in some regions have surged as much as 267 percent over five years. Consumers feel the impact through higher utility bills. Some states are exploring rules requiring data centers to pay for full grid upgrades.

Small modular reactors promise faster deployment and factory-built standardization, but no commercial fleet operates in the U.S. yet. First-of-a-kind risks, supply chain bottlenecks, and lengthy regulatory reviews remain hurdles. Existing nuclear plants offer quicker wins. Constellation is exploring uprates across its fleet. Vistra signed a 20-year agreement to supply Meta with over 2.6 gigawatts from upgraded reactors.

Analysts at BloombergNEF project U.S. nuclear capacity could grow 63 percent by 2050 under optimistic scenarios, adding more than 50 gigawatts after 2035. That outcome depends on policy continuity, financing, and supply chain development. The industry shrank for decades; reversing that trend requires sustained commitment.

Data center developers now scout locations based on power availability first. Land, water, and fiber come second. Some build their own generation assets. Others pursue co-location with nuclear facilities, effectively creating private grids for AI workloads. A single large training cluster can draw 1,000 megawatts. Forecasts suggest U.S. data center demand could reach 74 gigawatts by 2028, with potential shortfalls of 49 gigawatts.

Microsoft's move on Three Mile Island set a precedent. It forced the industry to confront the true cost of scaling artificial intelligence. Energy now ranks as a core strategic input for tech companies, equal to chips or talent. The Crane Clean Energy Center, when operational, will power the equivalent of 800,000 homes - all flowing to Microsoft's data centers, bypassing much of the traditional grid.

Nuclear power, long stuck in political and economic limbo, finds fresh purpose in the age of AI. Three Mile Island's revival stands as the most visible symbol of that transformation. Its success or struggles will shape decisions for years ahead.

Source: Webpronews