TerraPower pitches its Natrium reactor at AI data centers - the timing may finally be right

TerraPower is arguing its sodium-cooled Natrium design has a structural advantage over rival small reactors when the customer is a hyperscaler in a hurry—and the customer is now.

In plain terms. TerraPower - Bill Gates' nuclear startup - is telling data center customers that its Natrium reactor design gives it an edge over rival small modular reactors when it comes to timing, cost predictability, and grid integration. In an AI power-wall context where every hyperscaler is being told to wait years for grid upgrades, that pitch is landing differently than it did two years ago.

Context

The last twelve months have industrialized nuclear-for-AI as a category. Sequoia led a $1B round into fusion-adjacent Valar Atomics (#1785); Amazon is building a gas plant in Texas to bridge the wait (#1873); New York became the first US state to ban new data centers outright (#1793); the largest US grid, PJM, is now warning about curtailment (#1983). Every hyperscaler is short of firm, dispatchable, low-carbon power on a five-year horizon. Nuclear is the obvious answer—and every SMR vendor knows it.

What the article says

TechCrunch reports that TerraPower "possesses a strategic advantage over competitors, especially when chasing after data center deals." The article frames this as a "secret weapon" but does not spell out what specifically constitutes that advantage in its headline pitch. The rest of this section is our editorial reading of the plausible components, drawn from TerraPower's public documentation (linked in sources).

Our reading of the advantage

Two structural features of TerraPower's public Natrium design likely underpin the pitch: (1) integrated thermal storage that decouples reactor output from data-center load profile—attractive for spiky AI training runs; (2) the lead project site is a retired coal plant, meaning an existing high-voltage grid interconnect rather than a new queue slot. Neither of these is unique to TerraPower in principle, but together they constitute a "site + storage" package rather than a bare reactor module. That's a different sales motion from NuScale or X-Energy.

Scenarios

• Base case (55%): TerraPower signs a first hyperscaler PPA in 2027–2028, but grid revenue-share fights delay commissioning to 2032+.
• Fast case (25%): Federal loan guarantee accelerates a second unit at a co-located DC campus by 2030.
• Skeptical case (20%): Cost overruns push customers toward NuScale/X-Energy modular kit.

Risks

Sodium-cooled fast reactors have a difficult history internationally (Superphénix, Monju). Regulatory approval for a first-of-a-kind commercial fast reactor in the US remains the binary variable. Backend fuel handling for sodium-cooled designs is not yet solved commercially.

So what

If you are a CFO signing multi-year DC power contracts, the question is no longer "nuclear vs. gas"—it's "which nuclear vendor can actually put steel in the ground before your training cluster ships." Brownfield sites with existing interconnects are the scarce asset. Watch first PPA announcements, not press releases.