TerraPower Leverages Nuclear Reactor Innovation to Power AI Data Centers

TerraPower, a nuclear energy startup co-founded by Bill Gates, is set to announce its first data center project this year, according to Bloomberg. The company, which already has a power plant under construction in Wyoming, plans to break ground on its second facility in 2027. While the identity of the customer remains undisclosed, TerraPower previously announced that Meta had agreed to purchase eight of its Natrium power plants. Unlike traditional nuclear reactors, TerraPower’s 345-megawatt molten salt-cooled reactor is designed to complement intermittent energy sources like wind and solar. This is achieved through an innovative energy storage system that uses molten sodium to store excess heat generated by the reactor. When demand spikes, the stored heat can be converted into steam to power turbines, allowing the reactor to respond quickly to fluctuations in energy demand. Nuclear reactors typically operate at full capacity, with a capacity factor of 92.5% in the U.S. However, they are slow to adjust output, only capable of changing by about 5% per minute. In contrast, small modular reactors (SMRs) can adjust more rapidly, by about 10% per minute. Despite this, operating at reduced capacity is economically unviable for nuclear plants due to their high capital expenditures. TerraPower’s approach aims to mitigate this by pairing its reactor with energy storage, enabling it to operate efficiently even when demand is low. Data centers, particularly those involved in AI training, face significant challenges due to their fluctuating energy demands. These demands can cause stress on traditional power sources like natural gas turbines. To address this, TerraPower’s system allows for flexible energy generation, making it well-suited for integration with renewable energy grids or data center operations. This innovation positions TerraPower as a potential leader in the race to provide reliable, scalable power for AI infrastructure.