TerraPower, a nuclear startup founded by Bill Gates, is positioning its innovative nuclear reactors as a prime energy source for AI data centers, addressing their critical demand for reliable power. The company plans to announce its first data center project this year, with construction expected to begin in 2027. This will be TerraPower’s second Natrium power plant, following the one currently under construction in Wyoming. Notably, TerraPower has already secured a deal with Meta, which agreed to purchase eight Natrium plants, underlining the growing interest in nuclear power to support massive computing infrastructures.

A major challenge for nuclear reactors in powering data centers is their limited ability to quickly adjust output, which is essential to handle the variable and spiky electricity demands of AI workloads. Traditional reactors generally operate best at full capacity and can only ramp output by about 5% per minute. While newer small modular reactors (SMRs) improve this to 10%, running reactors below full power is economically inefficient due to their high capital costs. Data centers, especially those using on-site power supplies, face difficulty managing rapid load fluctuations, a problem that conventional energy plants and natural gas turbines struggle with, sometimes causing equipment stress or failure.

To tackle this, TerraPower has developed a 345-megawatt molten salt-cooled reactor that uniquely integrates thermal energy storage. Instead of modulating the reactor’s power output, the reactor runs continuously at full power, and excess heat is stored in a large reservoir of molten sodium. When electricity demand spikes, the stored heat is used to produce additional steam and electricity on demand. This system allows the nuclear plant to maintain a high capacity factor akin to renewable energy sources, blending steady nuclear power with the flexible response capabilities traditionally associated only with batteries or solar and wind.

This energy storage approach positions TerraPower's reactors as highly adaptable partners for renewable energy grids and AI data centers alike, potentially giving them a decisive advantage in the competitive landscape for powering next-generation technology facilities. By smoothing out the supply-demand curve without turning down the reactor, TerraPower can amortize its large investment over more hours of operation and offer power with the flexibility that AI workloads demand. This combination could accelerate the adoption of nuclear energy as a core component in supporting the future of AI computation.