TerraPower will announce its second Natrium reactor this year, sized for a data-center customer and using molten-salt storage to ramp from 345 MW baseload to 500 MW on demand — specs and announcement are solid Tier 2 reporting (Bloomberg, TechCrunch).
The open question is not the engineering; it’s the calendar. The first Natrium plant took roughly six years from non-nuclear groundbreaking to a 2030 target completion, and plant two is only expected to break ground in 2027 — set against that precedent, “on demand” power is more likely a 2032–33 story than a 2027–28 one (assessed with moderate confidence).
Confidence: high on the specs and the announcement; moderate on the licensing-timeline forecast, which rests on one precedent case, not a confirmed schedule for the second plant.
The Natrium specs and the customer demand case
TerraPower’s Natrium design is a 345-megawatt sodium-cooled fast reactor paired with a molten-salt thermal storage system: excess heat from steady fission is banked in the salt and released to push output to 500 MW for more than five hours when demand spikes. [Established — Bloomberg, 18 August 2026; TechCrunch, 19 August 2026.] CEO Chris Levesque said the company will “unveil its next project — designed for a data center — before the end of the year,” though he declined to name the customer. [Tier 2 — reported via Bloomberg, 18 August 2026.] The plant is expected to break ground in 2027. [Tier 2 — TechCrunch; Briefs.co coverage of the same reporting.]
The demand case is genuinely strong, and it deserves to be stated before any skepticism about timing. AI training and inference loads swing sharply — GPU clusters can “sink and soar” in power draw over minutes, not hours, and TechCrunch reports that natural-gas turbines pressed into this role “have been breaking under the stress” of those swings. A reactor that holds steady output while banking a five-hour buffer of dispatchable power solves a real engineering mismatch that batteries and gas peakers handle poorly at this scale. The commercial signal is also real: Meta separately agreed in January 2026 to buy up to eight Natrium units [Established — company and press reporting.], and TerraPower has a construction partnership with Hyundai Engineering & Construction aimed at standardizing the build across up to eight reactors to shorten each subsequent project. [Tier 2.] A data-center operator willing to sign a long-dated power-purchase agreement for a design like this is buying certainty that gas and grid interconnection increasingly cannot offer on a predictable timeline. That is the case for Natrium, and it is not a weak one.
The NRC timeline: precedent, not promise
The regulatory picture is where the announcement’s implicit optimism runs into an institution TerraPower does not control. Kemmerer 1, the first Natrium plant, broke ground on its non-nuclear support facilities in June 2024. The NRC did not issue a construction permit for the nuclear portions of the site until 4 March 2026 — the first construction permit for a commercial non-light-water reactor in more than forty years, and the first commercial reactor construction permit of any kind in nearly a decade. [Established — NRC, Department of Energy, American Nuclear Society reporting.] Nuclear-related construction began the following month. TerraPower’s own target for completion is 2030. [Tier 2 — American Nuclear Society, PowerMag.] Groundbreaking to target completion: roughly six years, on a project the NRC itself moved through eight months ahead of its own schedule and, by the agency’s account, under budget.
That acceleration is worth crediting — it suggests the NRC has genuinely learned to review a novel design faster than its own internal estimate. What it does not suggest is that a six-year pipeline compresses to two. We found no public evidence that TerraPower’s second plant will use a pre-approved standard design certification that would meaningfully shorten site-specific NRC review; Kemmerer proceeded under the traditional Part 50 construction-permit pathway, which is site-specific by design. [Assessed with moderate confidence — absence of evidence is not proof of absence here, and TerraPower’s Hyundai partnership is explicitly built around replicable construction, which could shorten the build phase even if it doesn’t shorten NRC review.] A second plant breaking ground in 2027 and following anything close to the Kemmerer sequence would plausibly reach completion in the early 2030s, not 2027–28.
The supply-gap context
The urgency behind this announcement is real and immediate. Estimates of the AI data-center power shortfall vary by source and methodology — one industry analysis puts it at roughly 7 GW of near-term capacity that is announced but not yet under construction; Goldman Sachs projects a structural shortfall widening from 9.3 GW in 2026 to 45 GW by 2028. [Tier 2 — figures diverge across analysts; the exact number is assessed with low confidence while the existence and direction of the gap is established.] What’s consistent across estimates is the timing: the shortage is a 2026–2028 problem. Natrium’s second plant, on the timeline its own precedent suggests, is not.
The institutional question
This is the frame worth holding onto: TerraPower’s molten-salt storage is a genuine answer to a genuine engineering problem, and the NRC has shown it can move faster than its own default pace when a design earns confidence through review. Neither of those facts changes what a construction-permit process for a novel reactor design actually costs in years. The Natrium announcement is a demand-side signal — proof that data-center operators will sign for nuclear on a multi-year horizon — more than it is a supply-side fix for the shortage straining the grid today. Readers should treat “second Natrium plant” as evidence the market believes in the 2030s, not as evidence the 2027 gap gets filled by fission. [Assessed.]