TerraPower's Secret Weapon for AI Data Centers: Molten Salt Energy Storage

Nuclear startups are pitching themselves as the antidote to what ails AI data centers: power that is always available. Bill Gates-founded TerraPower is the latest to enter the race, with Bloomberg reporting that it plans to announce its first data center project this year.

TerraPower has not said who the customer will be, though in January it announced that Meta agreed to buy eight of its Natrium power plants. The data center project, expected to break ground in 2027, would be the company's second power plant; its first is already under construction in Wyoming.

Not every reactor is suited to data center duty. Nuclear plants perform best at full tilt — U.S. reactors run at maximum output 92.5% of the time — but existing reactors ramp slowly, adjusting only about 5% of rated output per minute, while small modular reactors (SMRs) manage around 10%.

Data center loads, however, swing rapidly as GPUs respond to training and inference tasks — swings so demanding that natural gas turbines have been breaking under the stress. Smoothing the curve typically requires large banks of batteries, which add cost.

TerraPower's answer is energy storage. Its 345-megawatt molten salt-cooled reactor keeps splitting atoms instead of throttling output, storing excess heat in a giant vat of molten sodium; when power demand spikes, the plant taps that reservoir to generate more steam and spin the turbines.

The design was originally conceived to complement intermittent renewables like wind and solar — loads that are similar in shape to data center demand, just on the other side of the equation. Expensive equipment keeps working even when demand is low, letting the company amortize its investment over more operational hours.

That combination — nuclear's high capacity factor paired with storage flexibility — could give TerraPower a genuine edge in the race to power AI.

What to watch next: the identity of the first data center customer, the pace of Meta's eight-plant order, and whether SMR mass manufacturing can actually bring capital costs down.