TerraPower’s nuclear reactor emerges as key solution for powering AI data centers

Nuclear power startups are positioning themselves as the solution to AI data centers’ most pressing challenge: reliable, always-on energy. TerraPower, founded by Bill Gates, is the latest entrant, with Bloomberg reporting that the company plans to unveil its first data center project this year.

While TerraPower has not disclosed the specific customer, it previously announced in January that Meta agreed to purchase eight of its Natrium power plants. The upcoming data center project, slated for groundbreaking in 2027, will be the company’s second facility, following its first plant currently under construction in Wyoming.

Not all nuclear reactors are suitable for data center operations, but TerraPower holds a distinct advantage through integrated energy storage, giving it a competitive edge. This capability is derived from its integration with renewable sources like wind and solar.

Nuclear reactors, including TerraPower’s, operate most efficiently at full capacity. Nuclear plants boast the highest capacity factor of all power types, generating at maximum output 92.5% of the time in the U.S. However, this efficiency comes with a trade-off: existing reactors are slow to adjust, capable of changing output by only about 5% per minute, according to the National Laboratory of the Rockies.

New small modular reactors (SMRs), a focus for many startups, can adjust output faster—approximately 10% of rated capacity per minute, per NRL. Yet, operating below full capacity is financially inefficient, as revenue depends on generating electricity.

This issue applies to all power plants but is particularly acute for nuclear energy, which carries the highest capital expenditures of any generation technology. Startups hope that mass manufacturing of SMRs will reduce these costs, though this remains unproven. Even if successful, realizing these benefits could take a decade or more. Since early plants are expensive, startups aim to operate them at peak capacity as often as possible.

For data centers, especially those using behind-the-meter power, this presents a significant challenge. Their energy loads fluctuate rapidly as GPUs handle AI training or user prompts. These swings are so severe that natural gas turbines have failed under the stress. To stabilize operations, centers must invest in large battery banks, further increasing costs.

TerraPower designed its 345-megawatt molten salt-cooled reactor to address these issues. A key design goal was ensuring the reactor could complement intermittent renewable sources like wind and solar, requiring the ability to ramp up and down quickly. Although TerraPower originally envisioned renewables rather than data centers, both share similar intermittency challenges, just from opposite perspectives.

To achieve rapid response, TerraPower does not adjust its reactor’s core power output. Instead, it maintains continuous fission, storing excess heat in a large vat of molten sodium. When demand spikes, the plant draws from this reservoir to produce more steam and drive turbines. This allows expensive equipment to operate continuously, spreading investment costs over more operational hours.

This approach combines nuclear power’s high capacity factor with energy storage technology, enabling TerraPower to integrate seamlessly into renewable-heavy grids or connect directly to data centers. This flexibility may give the startup a competitive advantage in powering AI infrastructure.