On April 23, 2026, in Lincoln County, Wyoming, crews moved earth on what will, if the schedule holds, become the first commercial-scale advanced reactor to deliver electrons to the U.S. grid since the 1990s. The site sits across a service road from the retiring Naughton coal plant. The reactor going in is TerraPower’s Natrium — an 840 MWt pool-type sodium-cooled fast reactor mated to a nitrate molten-salt thermal storage island sized to push the plant from its 345 MWe baseload up to 500 MWe for roughly five and a half hours of peak demand. The Nuclear Regulatory Commission granted the construction permit on March 4, 2026, the first such permit ever issued for a non-light-water commercial reactor in the United States.
Two things deserve to be said before the analysis. First, Kemmerer is a demonstration unit in name only — it is grid-scale, utility-owned (PacifiCorp will offtake), and large enough to matter on Wyoming’s interconnection. Second, the project is no longer a lab story. Meta signed a multi-gigawatt offtake with TerraPower in January 2026 for up to eight Natrium units totaling 2.8 GW of firm baseload, all of it earmarked for AI data centers. That contract is what flipped Natrium from a single demonstration to a fleet program.
Why Sodium, Why Now
Liquid sodium has been the favored coolant of fast-spectrum reactor designers since EBR-II ran at Argonne in the 1960s. The reason is thermodynamic, not nostalgic. Sodium boils at 883 °C at atmospheric pressure, which means a sodium reactor can run at 500–550 °C outlet temperature without pressurizing the primary loop. Compare that to a light-water reactor like the AP1000, which has to hold roughly 15.5 MPa of primary pressure to keep water liquid at 320 °C. Drop the pressure and you drop the wall thickness, the pump head, the loss-of-coolant accident envelope, and the containment volume in tandem. The Natrium primary tank operates near atmospheric pressure. The economic consequence is that the nuclear island is roughly half the steel of an equivalent pressurized water reactor.
The fuel form matters too. Natrium uses sodium-bonded metallic uranium-zirconium fuel clad in HT9 ferritic-martensitic steel — the same fuel chemistry validated by tens of thousands of pin-hours at EBR-II and FFTF. Metallic fuel has a thermal conductivity roughly twenty times that of UO2, which means the centerline temperature at full power is a few hundred degrees instead of a few thousand. That cooler fuel has a very large prompt negative reactivity feedback on power excursions — the physics-based safety case that lets the design dispense with the high-pressure emergency core cooling architecture that dominates light-water plant cost.
The 300 MWh Built-In Battery
The reason Natrium is being noticed outside the nuclear community is the salt loop. The reactor delivers its 840 MWt of heat into an intermediate sodium loop, which exchanges into a nitrate molten-salt loop — the same hot-salt and cold-salt tank arrangement used at concentrated solar plants like Crescent Dunes. The salt tanks decouple the reactor from the turbine. At baseload the reactor charges the hot tank at the rate the turbine drains it. When the grid needs more electrons, the turbine pulls extra salt out of the hot tank and runs harder — up to 500 MWe gross — while the reactor continues steady at 840 MWt. When the grid needs less, the turbine throttles back and the reactor charges the tank without changing power level. The reactor never has to chase load.
The numbers behind that flexibility are interesting. The hot tank holds enough thermal energy to push the turbine to 500 MWe for roughly 5.5 hours — call it 850 MWh-thermal of storage, or about 300 MWh-electric of dispatchable peaking on top of the 345 MWe baseload. In a market structure where intra-day prices spread by a factor of three or more — ERCOT, MISO, the western interconnect during evening ramps — that peaking capacity is the difference between a plant that earns at the average power price and a plant that earns at the peak. The salt loop is also the reason Natrium can sit next to a data center load and look, from the grid operator’s perspective, like a dispatchable resource rather than a must-take baseload generator.
The HALEU Knot
Every advanced reactor program in the U.S. shares one supply-chain dependency: high-assay low-enriched uranium, enriched between 5% and 20% U-235, versus the 3–5% used in legacy light-water plants. Until 2024, the only commercial source was Russia’s Tenex. The U.S. banned Russian uranium imports in May 2024, with a hard cutoff in 2028. The Department of Energy’s HALEU Availability Program is the answer, and Centrus Energy in Piketon, Ohio is the only American company currently producing HALEU at any scale. As of the most recent DOE accounting, Centrus had delivered 900 kg of HALEU under its DOE contract, which was extended to June 30, 2026 with options for up to eight additional years. DOE’s stated target is 21 metric tons of HALEU available by mid-2026.
For context, a single Natrium first-core load is roughly 7–9 tonnes of HALEU. The reload is smaller, but the supply-chain math is unforgiving: the U.S. needs to bring a second commercial enricher online and ramp Centrus to multi-tonne annual output before the late-decade reactor fleet hits its fuel demand wall. That is a procurement and policy problem more than a physics problem, and it is the single largest schedule risk for the entire Generation IV deployment in this country — including Kemmerer.
Why Meta Wrote the Check
Meta’s January 2026 nuclear announcement was the loudest signal yet that the AI buildout has moved past wind and solar PPAs. The deal stack is large: 1.92 GW from Talen / Susquehanna in Pennsylvania (Amazon), 500 MW with Kairos / Google in Oak Ridge, and Meta’s 6.6 GW across Vistra, Oklo, and TerraPower. The TerraPower piece is the interesting one because it is the only piece that is contracting for a reactor type the customer is also funding into existence. Meta is not buying nuclear electricity from Natrium today — it is buying schedule certainty on units two through eight, paying TerraPower to walk the fleet down the cost-learning curve faster than a single demo unit ever would.
The Prometheus campus in New Albany, Ohio is the immediate destination — one gigawatt of AI compute coming online in 2026. The Hyperion project in Louisiana, which Meta has indicated may scale to five gigawatts, is the secondary destination. A single hyperscaler is now in a position to absorb the entire planned Natrium fleet through 2035 before any of those electrons touch the public grid. That is a structural change in how nuclear projects get financed and built in the United States, and it has implications well beyond Wyoming.
The Competitive Field
Natrium is not the only advanced reactor with a path to first criticality this decade. The honest comparison matters because the AI offtake market is not yet locked.
X-energy Xe-100 is an 80 MWe high-temperature gas-cooled pebble-bed reactor that runs on TRISO fuel. The NRC is on schedule to complete the safety evaluation in November 2026. The Amazon / Energy Northwest / X-energy partnership in Washington state targets up to 12 modules. The Xe-100 wins on process-heat applications — its 750 °C helium outlet is hot enough to drive hydrogen production via high-temperature steam electrolysis or steam methane reforming.
Kairos Hermes is a 35 MWt fluoride-salt-cooled high-temperature reactor. The Google PPA covers up to 500 MWe by 2035; the DOE HALEU TRISO fuel contract is in place and the Hermes 1 demonstration is the first Generation IV reactor under construction in Tennessee. Kairos’s bet is on iteration speed — smaller cores, more builds, faster learning.
Oklo Aurora is a 15–75 MWe sodium-cooled microreactor with a heat-pipe primary. Meta’s deal includes up to 1.2 GW from Oklo units in Ohio starting as early as 2030. Oklo’s siting advantage — small footprint, simplified balance of plant — is real, but the NRC custom-licensing path for a microreactor is still less worn than the path TerraPower just finished.
Where Natrium leads, today, is two-fold. It has a construction permit in hand, on a real site, with a real coal-plant replacement story and a real grid interconnection. And it is the only design in this list that integrates dispatchable storage at scale on the steam side of the plant. For a hyperscaler thinking about a 1–5 GW data center campus with daily load swings, that storage matters. For a coal-replacement project that has to honor the original plant’s interconnection rights and capacity factor expectations, that storage matters more.
What Sodium Fast Reactors Mean for Industrial Heat
The fission story usually stops at the busbar. Most analyses ask only how much electricity comes out of the plant and at what cost. That framing misses the second-largest single use of energy in the U.S. economy: industrial process heat. Refineries, ammonia plants, ethylene crackers, cement kilns, and steam-methane reformers consume roughly a quarter of U.S. primary energy and emit roughly a fifth of U.S. CO2, and almost none of that demand is electrifiable with current technology because it requires temperatures above 400 °C and steady, dense heat that wind and solar cannot deliver.
A 500 °C sodium-fast plant is not yet at refinery hydrocracker temperatures, but it is well above the 250–300 °C steam most light-water plants deliver. It can supply process steam to a crude distillation unit’s reboilers, to a delayed coker’s heater preheat, or to an ammonia plant’s syngas process. The economics turn on whether the nuclear plant is sited adjacent to the demand. A refinery taking process heat directly from a co-located reactor at $5/MMBtu is a fundamentally different industrial unit than one buying natural gas at $4 plus a $40/tonne CO2 charge. That is the unlock that Natrium-class reactors create — and it is largely orthogonal to the data center story Meta is buying.
At Porritt Inc., the work we do in process safety and refinery economics points to a near-term opportunity: pairing modular advanced reactors with new-build modular refineries to displace natural gas as the primary process-heat input in distillate-focused units. That is not a hypothetical decade-out vision. The reactor side of the equation now has a construction permit. The refinery side — modular, skid-mounted distillate-first units in the 5,000–15,000 BPD range — is where our equipment vendor pipeline and process engineering work sit.
The Next 18 Months
The schedule between today and Kemmerer first criticality is dense. TerraPower has to complete the non-nuclear balance of plant — the turbine island, the salt tanks, the cooling infrastructure — while the NRC operating license application proceeds in parallel. The salt loop itself is novel at this scale for a nuclear application, and the integration tests between the intermediate sodium loop, the salt loop, and the steam cycle will determine whether the 500 MWe peaking specification holds in practice. The HALEU fuel manufacturing pipeline has to deliver, on schedule, a core load that does not yet exist at the requisite enrichment in commercial quantity inside the United States.
If TerraPower hits its 2030 commercial operation target, Kemmerer Unit 1 will be the first non-light-water commercial reactor to generate electricity in the United States since the closure of Fermi 1 in 1972. The schedule has slipped before — the original Natrium plan targeted 2028. But the work happening today on a Wyoming construction site is the most concrete proof point yet that the U.S. has, for the first time in three decades, a functional pipeline for delivering new nuclear capacity at utility scale. That matters for the AI buildout. It matters more for the industrial decarbonization problem nobody yet has a credible answer to.
If you’re a refinery operator, an EPC, a hyperscaler infrastructure team, or a DOE program manager thinking about how to connect new nuclear capacity to industrial demand — we should talk. Porritt Inc. builds the process safety and engineering compliance tooling, the modular refinery designs, and the dual-use industrial-heat economics that make the next decade of advanced nuclear deployments deliver more than just baseload electrons. Reach out through porrittinc.com/contact.
Timothy Porritt is founder of Porritt Inc., building AI-powered tools for process safety, engineering compliance, and industrial operations. Based in Salt Lake City, Utah.