Nuclear startups are touting an antidote to what plagues AI data centers: always-available power. TerraPower, founded by Bill Gates, is one of the latest companies to jump into the ring, with Bloomberg reporting that the startup plans to announce its first data center project this year.
TerraPower has not disclosed who its customers will be, but in January it announced that Meta had agreed to buy eight sodium power plants. The data center project, scheduled to break ground in 2027, will be the company’s second power plant, with the first already under construction in Wyoming.
While not all nuclear reactors are suitable for data center duties, TerraPower has one key advantage that promises to give it an edge over its competitors: energy storage. And it’s all thanks to renewable energy sources like wind and solar.
Nuclear reactors, including TerraPower, work best when operating at full tilt. Among various types of power plants, nuclear reactors have the highest capacity factor. In the US it produces maximum power 92.5% of the time. But in a way, it has to be. According to the National Laboratory of the Rockies (NLR), existing nuclear reactors ramp up and down slowly, increasing or decreasing their total rated power by only about 5% per minute.
New small modular reactors (SMRs), which are being pursued by many startups, can react faster, at about 10% of the rated power per minute per NLR. However, operating at reduced capacity is not ideal. If you don’t generate electrons, it’s difficult to make money.
This is true for any power plant, but it is especially true for nuclear power, which has the highest capital investment of all power generation technologies. Startups hope that mass production of SMRs will reduce capital expenditures, but that remains unproven. And even if it works, it could take a decade or more to reap the benefits. Every startup recognizes that early power plants are expensive, so it makes sense to run them at peak capacity as often as possible.
This poses a challenge for data centers, especially those that rely on behind-the-meter power. Especially when training AI or responding to prompts, the load can drop and rise quickly as the GPU responds to the task. The shaking is so severe that natural gas turbines can break under the stress. Smoothing the curve requires the use of large amounts of batteries, further increasing costs.
TerraPower designed a 345-megawatt molten salt-cooled reactor to avoid these challenges. One important consideration was to ensure that the reactor could supplement intermittent power sources such as wind and solar. Power plants had to be ramped up and down quickly. When TerraPower drew up its plans, it had renewable power in mind, not data centers, but the two are just different sides of the equation and similarly intermittent.
For rapid ramp-up, TerraPower does not increase or decrease the reactor output. Rather, the atoms continue to split and the excess heat is stored in a huge container of molten sodium. When electricity demand spikes, power plants can tap into their reservoirs to generate more steam to turn turbines. Expensive equipment continues to operate even when demand is low, allowing the company to amortize its investment over a longer operating time.
This approach takes full advantage of nuclear power’s high capacity utilization benefits and combines it with energy storage technology, allowing TerraPower to operate well on renewable-heavy power grids or when connected to data centers. This is a flexible approach that could give startups an edge in the race to power AI.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
