Nuclear energy startups are positioning themselves as the ideal solution for the insatiable, always-on power requirements of AI data centers. Bill Gates’s TerraPower is the latest to join this trend, with Bloomberg reporting that the company plans to unveil its first data center project this year.
While TerraPower has not yet disclosed the customer, it had announced in January that Meta had agreed to purchase eight of its Natrium power plants. The new data center project, expected to commence construction in 2027, will mark the company’s second power plant, as its initial facility is already under construction in Wyoming.
Not all nuclear reactors are suitable for data center operations, but TerraPower holds a crucial advantage: integrated energy storage. This feature, surprisingly enhanced by advancements in renewable power like wind and solar, promises to give it a competitive edge.
Nuclear reactors, including TerraPower’s designs, perform optimally when operating at maximum output. Among all power plant types, nuclear reactors boast the highest capacity factor, generating at peak power 92.5% of the time in the U.S. This continuous operation is often a necessity, as existing reactors are slow to adjust power levels, capable of increasing or decreasing only about 5% of their total rated output per minute, according to the National Laboratory of the Rockies (NLR).
Newer small modular reactors (SMRs), which many startups are pursuing, can respond more swiftly, adjusting by about 10% of their rated output per minute, as per NLR. However, operating below full capacity is not economically ideal, as profitability is directly tied to consistent power generation.
While this is true for any power plant, it is especially pronounced for nuclear power, which incurs the highest capital expenditures among all generating technologies. Startups hope that widespread manufacturing of SMRs will lower these initial costs, but this remains to be demonstrated. Even if successful, it could take a decade or more to realize the full benefits. Given that all startups acknowledge their initial power plants will be costly, maximizing peak capacity operation is a logical strategy.
For data centers, particularly those relying on behind-the-meter power, this presents a considerable challenge. Their power demands, especially during AI training or prompt processing, can fluctuate dramatically as GPUs respond to tasks. These extreme power swings have been so taxing that natural gas turbines have been failing under the stress. To smooth these fluctuations, data centers often require extensive battery banks, which further escalate costs.
TerraPower engineered its 345-megawatt molten salt-cooled reactor to overcome these difficulties. A primary design objective was to ensure the reactor could quickly modulate its output to complement intermittent electricity sources like wind and solar. While TerraPower initially focused on renewable integration, not data centers, when developing its plans, both scenarios involve similarly fluctuating, albeit inverse, power demands.
To achieve rapid adjustments, TerraPower avoids increasing or decreasing the reactor’s power output directly. Instead, atom splitting continues constantly, and any excess heat is stored in a massive reservoir of molten sodium. When electricity demand surges, the power plant can tap this stored energy to generate additional steam, thereby spinning the turbines faster. This approach allows expensive equipment to operate continuously even during low demand periods, maximizing investment amortization over more operational hours.
This strategy combines the inherent advantages of nuclear power—a high capacity factor—with an energy storage technology that enables TerraPower to integrate effectively into grids with significant renewable penetration or when connected to a data center. This flexible methodology could provide the startup a substantial competitive advantage in the race to power the expanding demands of AI.