TerraPower’s nuclear reactor has a secret weapon for powering AI data centers
TerraPower, the nuclear energy startup founded by Bill Gates, is preparing to announce its first data centre power project this year, according to Bloomberg. The move is part of a broader push by nuclear startups to court AI companies, which need constant, reliable electricity to run power-hungry data centres, and it matters because it could position nuclear as a more stable alternative to the batteries and gas turbines currently struggling to cope with volatile AI computing loads.
The project, expected to break ground in 2027, would be TerraPower's second power plant after one already under construction in Wyoming, though the customer has not been named; in January the firm said Meta had agreed to buy eight of its Natrium reactors. TerraPower's edge lies in its 345-megawatt molten salt-cooled reactor design, which stores surplus heat in a tank of molten sodium rather than adjusting the reactor's output. This lets it keep splitting atoms at a constant, efficient rate while still ramping electricity generation up or down quickly to match sudden swings in demand, whether from intermittent renewables like wind and solar or from surging AI workloads, and without the added cost of large battery banks.
- TerraPower plans to announce its first AI data centre project this year.
- Its reactor stores heat in molten sodium to handle demand swings cheaply.
- Second plant expected in 2027; customer undisclosed, Meta already buys reactors.
New here? Start with this
AI data centres need large amounts of electricity around the clock, and their demand can rise and fall quickly as computing work changes. This has led technology companies and energy suppliers to look for power sources that can provide a steady supply while responding to those swings.
TerraPower is a US nuclear energy company founded by Bill Gates. Its Natrium design uses a reactor cooled by liquid sodium and stores surplus heat so electricity production can be adjusted without frequently changing the reactor’s output. The company is also building a plant in Wyoming, and has said Meta agreed to buy eight Natrium reactors.
The approach matters because nuclear power could offer data centres a dependable source of electricity while accommodating changing demand. Whether it can do so at the scale and cost needed will depend on how the projects are built and operated.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Nuclear energy is essential for meeting global climate goals whilst satisfying growing electricity demands, and AI data centres represent a pragmatic opportunity to deploy new nuclear capacity. TerraPower's molten salt reactor specifically addresses a real technical challenge—matching variable loads without expensive battery systems—making nuclear economically competitive in this sector. This demonstrates that advanced reactor technologies can meaningfully displace fossil fuel generation and play a vital role in decarbonisation alongside modern infrastructure needs.
The case against
Nuclear plants face persistent cost overruns and lengthy construction delays that reduce their economic competitiveness compared to renewables and storage, and directing scarce new nuclear capacity toward powering AI data centres may misallocate resources when climate urgency demands rapid decarbonisation of essential services like transport and heating. The decision to expand nuclear primarily to facilitate AI consumption, rather than replacing existing fossil fuel capacity in vital sectors, raises legitimate questions about whether this represents sound energy policy or serves commercial interests at the expense of broader climate priorities.