Bill Gates' Nuclear Reactor Just Broke Ground. At $11,594 Per Kilowatt, Here's What 'Affordable' Actually Means.

TerraPower began construction on the first non-light-water commercial reactor permitted in the United States in over 40 years. At $4 billion for 345 megawatts, the cost per kilowatt looks 28% cheaper than Vogtle. But the fuel it needs barely exists on American soil, and the nth-of-a-kind target requires a cost reduction steeper than any nuclear program has ever achieved this early.

Four billion dollars. That is what TerraPower says it will cost to build a 345-megawatt sodium-cooled fast reactor outside Kemmerer, Wyoming, on the site of a retiring coal plant. Bechtel mobilized crews for full construction on April 23, 2026, after six years and seven months had elapsed since the Department of Energy selected Natrium as one of two flagship advanced reactor demonstrations, a timeline that makes the project's ambition feel less like a sprint and more like a slow grind through regulatory reviews, fuel shortages, and the institutional inertia that has defined American nuclear construction for half a century. Pause on what that $4 billion actually buys, because the number sounds enormous in isolation but tells a different story when measured against recent American nuclear construction.

Divide the $4 billion by 345 MW of nameplate capacity and you get $11,594 per kilowatt, a number that is not cheap by any reasonable standard but is cheaper than every alternative that has actually been attempted in the American nuclear sector in the past two decades. Georgia Power's Vogtle Units 3 and 4 cost ratepayers $36 billion for 2,234 MW combined, which works out to $16,115 per kilowatt. NuScale's cancelled UAMPS project in Utah had escalated to $9.3 billion for 462 MW before subscribers pulled out in November 2023, meaning $20,130 per kilowatt for a reactor that was never built.

Natrium's first-of-a-kind cost is 28% lower per kilowatt than Vogtle, 42% lower than NuScale's final estimate, and roughly half the DOE's reference figure for new large light-water reactors. Progress. Real progress. But context matters enormously, because no first-of-a-kind reactor in history has come in at its stated budget, not one, and Vogtle, originally estimated at $14 billion, finished at $36 billion in a 157% overrun that drove its contractor, Westinghouse, straight into bankruptcy.

Where the Money Comes From

TerraPower is not building alone. Up to $2 billion of the $4 billion total comes from the Department of Energy's Advanced Reactor Demonstration Program under a 50-50 cost-sharing agreement. That means American taxpayers are covering roughly half the bill for a reactor designed and majority-owned by a company Bill Gates founded in 2008. The DOE framing is straightforward: the government absorbs first-of-a-kind risk so the private sector can prove the design works, then commercial orders follow at lower cost. It is the same logic behind NASA's Commercial Crew program, which paid SpaceX and Boeing to develop capsules the companies then sell seats on. Whether Natrium follows that trajectory depends entirely on what the second and third units cost.

The Nth-of-a-Kind Problem

TerraPower has stated an nth-of-a-kind target of approximately $3,600 per kilowatt. Getting there from $11,594 requires a 69% cost reduction, which is brutal math even by the standards of an industry that has never once met its own cost forecasts in this country.

Nuclear construction learning rates have historically ranged from 5% to 20% cost reduction per doubling of cumulative capacity, with most U.S. experience closer to the low end. France's fleet-build program in the 1970s and 1980s achieved roughly 15% learning per doubling across 58 reactors, but that was a single standardized design (the Pressurized Water Reactor) built by a nationalized utility under a government mandate, conditions unlikely to be replicated in the American market.

If Natrium achieves a 20% learning rate per doubling, which is optimistic for nuclear but standard in manufacturing-intensive industries like solar and automotive, the math requires roughly three doublings of cumulative capacity to reach the NOAK target. Three doublings from one unit means eight cumulative units must be built and operated before the ninth approaches $3,600/kW. Eight reactors. At $4 billion per unit declining over the series, the total capital deployed before reaching the target would be on the order of $18 to $22 billion, and that assumes no schedule overruns, consistent regulatory treatment, and a fuel supply chain that does not exist yet.

Bechtel, TerraPower's construction partner, has framed its approach as "a delivery model designed for repeatability." Smart framing. It is a deliberate contrast with Vogtle, where Bechtel was brought in to rescue a project already deep in crisis, inheriting a bespoke Westinghouse AP1000 design with thousands of on-site engineering changes and a workforce that turned over multiple times during the fifteen-year build. Repeatability is the right goal. But eight consecutive builds of a sodium-cooled fast reactor? That is a commitment the American nuclear industry has not made since the Eisenhower administration.

The Fuel That Does Not Exist

Natrium runs on high-assay low-enriched uranium, or HALEU, enriched to between 5% and 19.75% uranium-235. Standard reactor fuel? About 5%. HALEU is more energy-dense, which means smaller reactor cores, longer refueling intervals, and better neutron economy. It is also, in practical terms, almost entirely unavailable from Western suppliers.

Only Russia and China currently operate commercial-scale HALEU enrichment infrastructure, and that is it, nobody else on the planet. The United States banned Russian uranium imports in May 2024, removing the only existing supply option. Centrus Energy, the sole American company with an operating HALEU demonstration cascade at Piketon, Ohio, had produced and delivered approximately 920 kilograms to the DOE by mid-2025. Natrium's first core requires on the order of six tonnes.

That is a factor-of-six gap between total American production and one reactor's initial fuel load, a disparity so large that it should be treated as the central risk to the entire project rather than a footnote in press coverage that focuses on the reactor's design innovations. TerraPower already delayed its schedule by over a year in late 2022, citing precisely this bottleneck. Since then, Urenco received NRC authorization to enrich up to 10% at its New Mexico facility, and Orano began developing Project IKE at Oak Ridge, Tennessee, with DOE funding. A new Urenco HALEU facility at Capenhurst in the UK is targeting production by 2031. None of these will reach the scale Natrium needs before 2028 at the earliest, which means the reactor's construction timeline and its fuel timeline are on a collision course that nobody in Washington is publicly discussing with the urgency the numbers demand.

TerraPower's 2030 operational target depends on a fuel supply chain that, as of April 2026, has produced less than one-sixth of a single core load across all Western facilities combined. If HALEU production scales on schedule, Natrium will be the proof that advanced reactors can work in the American grid. If it does not, Natrium becomes the most expensive construction site in Wyoming.

Why Coal-to-Nuclear Matters

Kemmerer sits next to PacifiCorp's Naughton Power Station, a coal plant with units either retired or slated for conversion, and coal-to-nuclear siting is not incidental to TerraPower's economics but central to them in ways that most coverage of the project has failed to explain. Retired coal plants bring existing grid interconnection points, transmission lines rated for hundreds of megawatts, cooling water infrastructure, road access, and workforces with industrial construction experience. A 2022 DOE study identified over 300 retired and retiring coal sites across the country suitable for nuclear conversion, representing a combined capacity opportunity exceeding 200 GW.

Reusing grid interconnections is the single largest cost advantage, because new transmission lines cost $1 to $3 million per mile and interconnection studies for greenfield sites routinely take three to five years just to navigate the queue. Naughton's existing 345 kV transmission infrastructure could shave hundreds of millions of dollars and years of permitting from the project total, costs that do not appear in the reactor's $4 billion price tag but would otherwise kill the economics of a standalone site.

The Molten Salt Trick

Natrium's design includes an integrated molten-salt thermal energy storage system that allows the plant to boost electrical output from 345 MW to 500 MW for more than five hours, which is the single cleverest engineering decision in the entire project. Conventional nuclear plants run at constant output because ramping reactor power up and down is slow, mechanically stressful, and economically wasteful. Natrium's reactor runs at constant thermal power while the storage system absorbs excess heat during low-demand periods and releases it during peaks, effectively turning a baseload nuclear plant into a dispatchable one that can chase price signals across the day.

This matters because the electricity markets Natrium is entering are increasingly shaped by variable renewable generation. Wind and solar. In PacifiCorp's service territory, which spans six Western states, wind and solar penetration reached 31% in 2025. During midday solar peaks, wholesale prices frequently hit zero or go negative. A nuclear plant that can only sell power at $35/MWh around the clock will struggle against a plant that sells at $15/MWh during gluts and $80/MWh during evening peaks. Natrium's storage system does not change the reactor, but it changes the revenue profile. That could make the difference between a project that earns its capital cost back and one that operates at a loss for decades, which is exactly what happened to several French EPR projects that lacked dispatchability in markets flooded with cheap renewable power.

The Strongest Counterargument

Every advanced reactor company has a first-of-a-kind story, and the good ones really are good. NuScale had the only SMR design ever certified by the NRC, and it still collapsed under its own cost growth. Kairos Power's Hermes test reactor is under construction in Tennessee, but at 35 MW thermal it is a technology demonstrator, not a commercial plant. X-energy, the other ARDP awardee, has not yet submitted a construction permit application for its Xe-100 pebble-bed reactor.

Sodium-cooled fast reactors have an uncomfortable track record. Japan's Monju prototype operated for a cumulative 250 days across 16 years before permanent shutdown, derailed by a 1995 sodium leak and fire. France's Superphénix? Seven percent. That was its lifetime capacity factor before closure in 1997, a number so catastrophically low it would be comical if French taxpayers had not paid for it. The United States' own Experimental Breeder Reactor-II ran successfully for 30 years at Idaho National Laboratory, but it was a research facility with a captive government funder, not a commercial plant competing in wholesale electricity markets against gas turbines and wind farms and batteries. Sodium is a highly effective coolant with excellent thermal properties, and it operates at atmospheric pressure, eliminating the massive pressure vessels required for light-water reactors. It also catches fire. Sodium leaks are not maintenance problems. They are fires. TerraPower's design incorporates an intermediate sodium loop that physically separates the radioactive primary sodium from the steam generators, a lesson learned directly from Monju and Superphénix, and whether that engineering solution performs as designed over a 40-year plant life is the bet Kemmerer represents.

What This Analysis Does Not Prove

Natrium's $4 billion estimate has not been independently verified through a detailed cost review. ENR and Power Engineering report the figure as stated by TerraPower. No public construction cost breakdown by system has been released. Reactor island, balance of plant, storage, owner's costs: undisclosed. All of it. Without that granularity, it is impossible to assess where the FOAK premium concentrates and how plausible the learning-rate assumptions are for each subsystem, which is exactly the kind of information that would distinguish genuine confidence from hope dressed up as a business plan.

Additionally, this comparison uses nameplate capacity, not levelized cost of energy, which would incorporate capacity factor, fuel costs, operating expenses, decommissioning, and financing structure over the plant's lifetime. A full LCOE comparison would be more informative but requires data that TerraPower has not disclosed and that will not be available until the plant operates commercially.

Finally, the HALEU production timeline data comes from public DOE reporting and World Nuclear Association databases. Enrichment capacity projections are manufacturer-stated targets, not independently audited schedules. Centrus, Urenco, and Orano each have incentives to project optimistic timelines to secure government funding.

The Bottom Line

TerraPower breaking ground at Kemmerer is the most significant moment in American nuclear construction since Vogtle Unit 3 reached criticality in 2023, not because the reactor is guaranteed to work but because it tests a fundamentally different proposition: can a smaller, non-light-water reactor built on a standardized delivery model with integrated energy storage compete in a deregulated electricity market where gas is cheap, wind is free, and patience for nuclear cost overruns evaporated a decade ago?

If you work in energy policy, the single number to watch is not the $4 billion FOAK cost but whatever TerraPower quotes for Kemmerer Unit 2, because that is where the learning curve either validates itself or collapses into another nuclear fantasy. A second unit at $3.2 billion or below would validate the trajectory. One at $3.8 billion would not. No second unit within three years of the first? Natrium joins the long list of nuclear prototypes that worked once and never scaled.

If you are an investor evaluating nuclear as a data-center power source, pressure TerraPower, Kairos, and X-energy to publish detailed cost breakdowns by subsystem, because first-of-a-kind total costs tell you nothing about where the NOAK savings will come from, and you should demand the same transparency from nuclear companies that the market demands from solar and wind developers, who publish dollars per watt by module, inverter, racking, and labor.

And if you live in Kemmerer, a town of 2,651 people that built its economy on coal, this is the most consequential economic development in a generation. PacifiCorp's Naughton plant employed roughly 200 people. Bechtel estimates 1,600 construction jobs at peak build and 250 permanent operations positions. Whether those jobs materialize depends on whether a reactor that runs on fuel America cannot yet produce reaches completion on time and on budget, which is something no American nuclear project has ever achieved.