Three Mile Island, the site of the worst nuclear disaster in U.S. history, is planning a comeback to sell power to one single customer: Microsoft.
Constellation Energy signed a 20-year deal with the tech giant to restart the remaining undamaged unit, idled since 2019, and sell the output to Microsoft to help satiate the surging appetite for the juice that fuels artificial intelligence.
And Microsoft is not alone. Google and Amazon are also aggressively sourcing nuclear energy to power their own AI initiatives. These developments mark a fascinating twist in the long and often disappointing history of atomic energy, enabled by private capital and design innovations that may pave the way for a larger role in the transition to zero carbon.
Growth in U.S. electrical consumption has remained relatively flat for the past two decades. Since 2004, the economy measured by GDP has expanded by 50% after inflation, while demand for electricity has risen by just 8% or less than 0.5% per year. Thanks largely to the deployment of artificial intelligence, that is changing dramatically.
Big tech companies are quaffing increasingly copious amounts of electrical power, far beyond the capability of the existing utilities to provide. These firms have made commitments to carbon neutrality by 2040, pledges that cannot be honored by leaning on traditional gas and coal-fired plants. Renewables like solar and wind will contribute, and data center operators have tied up much of the new generation scheduled to come online, but the potential supply from these sources is simply insufficient. Furthermore, their inherent intermittency is problematic for massive computing arrays that run 24/7. This looming demand-supply imbalance has prompted artificial intelligence companies to deploy the nuclear option to fill the gap.
Large language AI models are power hogs. According to Goldman Sachs, a single ChatGPT query consumes 10 times the electricity of a Google search, enough to light an LED bulb for nearly 20 minutes. McKinsey and Company estimates that adoption of AI will require an additional 50 gigawatts (50 billion watts) of generating capacity by 2030, sufficient to power 35 million homes. Data center demand will grow from around 3% of total electrical generation today to 12% by 2030.
While Microsoft and Constellation are restarting a large legacy reactor, Amazon and Google are turning to a newer technology: small modular reactors. These are scaled-down versions promising lower construction and operating costs and increased scalability to meet growing demand. Small modular reactors generate 300 megawatts (300 million watts) or less, compared with behemoths like Three Mile Island with capacities of 1,000 megawatts or more. These reactors can be prefabricated in manufacturing plans and transported to the generation site for assembly, offering substantial cost efficiencies over larger bespoke facilities. Massive legacy nuclear plants have also been plagued with delays and cost overruns, which combined with a trio of high-profile accidents have sapped public support for atomic energy.
Several small modular designs currently in development have attracted interest from technology companies, utilities, private investors and the Department of Energy, while the explosion of interest in artificial intelligence may provide the impetus to move forward with broader deployment.
Google recently announced an agreement to purchase up to 500 megawatts from Kairos Power beginning in 2030. Kairos will build and operate a fleet of six to seven small modular reactors devoted to furnishing clean, continuous power to Google’s data centers. Kairos has also received a construction permit for a demonstration project near Oak Ridge, Tennessee, in collaboration with Oak Ridge National Laboratory and the Tennessee Valley Authority to test its innovative cooling technology.
Google released a sustainability report in October that noted its greenhouse gas emissions have increased by 48% since 2019 due to reliance on fossil fuels to meet surging data center demand, emphasizing the need for additional clean power.
Following Google’s announcement, Amazon unveiled a $500 million initial investment in X-Energy Reactor Co. to fund the completion of its own reactor design and a fuel refining facility, also in Oak Ridge. X-Energy will construct a four-unit 320 megawatt modular complex in central Washington state, with the option to bolt on eight more reactors.
Amazon also agreed to purchase a Pennsylvania data center campus from Talen Energy for $650 million adjacent to an existing Talen nuclear plant. Amazon plans to build out the facility for its Amazon Web Services cloud division and purchase power from Talen.
AI is not the only driver of surging electrical demand. Widespread adoption of electric vehicles will strain the grid, and cryptocurrency mining consumes huge amounts of power. And clean hydrogen, a promising near-zero-emission replacement for petroleum in transportation and industrial applications like steelmaking, must likely also rely on nuclear power to be profitable.
Of course, a renaissance in nuclear power is not without challenges. Storage of nuclear waste remains an unsolved problem, and operating costs are still not competitive with cheap natural gas-fired generation. However, the U.S. has amassed substantial experience with safe atomic energy in smaller packages over 70 years in Naval propulsion. The U.S. Navy has operated 273 reactors since the Nautilus nuclear sub was launched in 1958 and currently maintains 83 nuclear-powered vessels on active duty.
Moving away from fossil fuels is an imperative, and passive clean technologies like solar and wind are simply not sufficient to meet the dramatic increase in forecasted demand. Smaller, factory-built modular reactors have attracted attention and capital, and are likely to play a major role.
Christopher A. Hopkins, CFA, is a co-founder of Apogee Wealth Partners in Chattanooga.
Christopher Hopkins