If you have followed how Big Tech plans to feed its AI habit, you already know the pitch: small modular reactors, factory-built, quick to deploy, powering data centers without a fifteen-year construction saga. Most of them are still slideware. Almost none of the designs pulling in billions of headlines can legally sell a single watt in the US yet.
So it is worth slowing down on what Oklo just did in Idaho, because it is the opposite of a slide.
On June 11, the Department of Energy’s Idaho Operations Office approved the preliminary safety case for the company’s first Aurora powerhouse, a compact sodium-cooled fast reactor under construction at Idaho National Laboratory. It was not a ribbon-cutting, and nobody set off fireworks. It was a stack of paperwork clearing a federal review, which in nuclear is the part that usually takes years and quietly kills projects.
And the machine on the other side of that paperwork is a strange one. According to the Nuclear Regulatory Commission, the Aurora is a liquid metal-cooled, metal-fueled fast reactor rated at a maximum of 75 megawatts of electricity. It will run on uranium recovered from a reactor that shut down in 1994. And it is deliberately not racing to the finish line everyone else in its program is sprinting toward.
Oklo is not chasing the Fourth of July
There is a reason “July 4” keeps coming up around new reactors right now. Back in May 2025, an executive order created the DOE’s Reactor Pilot Program with a stated goal of getting at least three experimental reactors to criticality by July 4, 2026, using DOE authorization instead of the usual NRC licensing route. In a field where new builds are measured in decades, the government gave companies about a year.
Some of them are pulling it off. A startup called Antares hit the first criticality under the program in early June at that same Idaho site, and Valar Atomics followed by taking a test reactor critical in the Utah desert. Two brand-new reactors are already running. We covered the reactors racing to switch on by the Fourth and the safety rulebook that got a lot thinner to make the deadline.
Oklo is in that program too. Its flagship Aurora, though, is not one of the sprinters. The June 11 approval was the preliminary documented safety analysis, or PDSA, which the American Nuclear Society describes as the third of four steps toward DOE authorization to build. The fourth and final step, the full documented safety analysis, is the one that actually lets you operate. No company in the program has cleared it yet.
So Aurora sits a couple of steps back from criticality, on purpose. Oklo’s own target for first power at Idaho is late 2027 or early 2028, and its June 11 announcement did not mention July 4 at all. This is the reactor built to sell electricity, not to light a candle on the country’s 250th birthday.
The reactor runs on fuel older than half the company building it
Oklo was founded in 2013, and for a while its best-known link was Sam Altman, who chaired its board until he stepped down in April 2025 to avoid a conflict of interest as OpenAI started talking to nuclear suppliers. The company name comes from Oklo, a spot in Gabon where uranium deposits ran a natural chain reaction on their own nearly two billion years ago. Which is a fitting name for what the reactor actually is.
The Aurora is a fast reactor. That means it splits atoms with fast-moving neutrons and skips the water that conventional plants use to slow them down. Instead of water, it uses liquid sodium to pull heat off the core.
Sodium carries heat far better than water and does not need to sit under crushing pressure. It also catches fire if it touches air or water, so it stays sealed inside its own loop, away from anything it can react with. That is the trade every sodium reactor makes.
The design is not new so much as revived. It is built on the Experimental Breeder Reactor II, or EBR-II, a sodium fast reactor that ran at Idaho from 1964 to 1994 and, in a landmark 1986 test, shut itself down safely with its cooling pumps switched off and no operators intervening. Oklo leans on that heritage for its safety pitch: the physics pull reactor power down when it overheats, without an operator or an active safety system stepping in.
Then there is the fuel, which is the part that sounds made up. Oklo won a competitive DOE award in 2019 for roughly 5 metric tons of metallic uranium recovered from EBR-II itself. So the first cores of a reactor that broke ground in 2025 will run on fuel from a reactor switched off before a good chunk of Oklo’s engineers were born. It is high-assay low-enriched uranium, or HALEU, the richer fuel most of these advanced designs need, and Aurora can also run on used fuel pulled from other reactors.
It builds its own fuel plant next door
Most reactor companies buy their fuel from someone else. Oklo is building the fuel plant next to the reactor.
The Aurora Fuel Fabrication Facility, which Oklo shortens to A3F, is going up at the same Idaho site. It is meant to turn that recovered EBR-II material into finished metallic fuel assemblies for the reactor. It cleared its own DOE safety agreement in November 2025 and its preliminary safety analysis that December, which made it the first facility approved under a separate DOE program aimed at fast-tracking advanced fuel manufacturing.
Oklo is now finishing the last safety step for the fuel plant and fitting out the building with its non-nuclear gear. The company expects A3F to be turning out its first metallic fuel rods in that same late-2027 to early-2028 window, lined up with the reactor next door.
That on-site setup is the whole strategy. Control the fuel, and you are far less exposed to the HALEU supply crunch that has snagged nearly every other advanced reactor, including Bill Gates-backed TerraPower’s sodium reactor in Wyoming, which lost roughly two years when it cut ties with its Russian fuel supplier.
Oklo is not the only one betting on sodium
Sodium fast reactors are not Oklo’s private idea. The same physics is showing up all over American nuclear right now, in wildly different sizes.
The biggest example is Natrium, the 345-megawatt sodium reactor TerraPower is building in Wyoming with a molten-salt tank bolted on to store heat and shove its output up to 500 megawatts when the grid spikes. That is a full utility-scale power plant. Aurora is the compact version, a 75-megawatt block you truck to a site, aimed at data centers, remote industrial plants, and military bases rather than the regional grid.
Then there are the reactors that ditch pumps entirely. Westinghouse’s eVinci heat-pipe reactor, headed to the same Idaho test bed, moves heat through sealed tubes with nothing pumping inside them, the way the little pipe in your laptop does. Same goal as Aurora, take moving parts out so there is less to break, different route to it.
And for contrast, there is the boring option: a 300-megawatt water reactor from GE Vernova Hitachi that the Tennessee Valley Authority wants to build, using the same basic technology that has run on the US grid since the 1970s. Boring is a genuine selling point in nuclear. Oklo’s bet is that a small, waste-eating, self-fueling machine can be new and safe enough to sell at the same time, which is a much harder pitch than boring.
Who is actually going to buy the power
The reason any of this gets billions in attention is demand. AI data centers are pulling power off the grid faster than utilities can build it, and reactor companies are signing up buyers before a single one of their plants exists.
Oklo’s anchor customer is Meta, which signed a binding agreement in January for 1.2 gigawatts of power from an Aurora campus Oklo plans to build in southern Ohio, kicking in upfront cash and power prepayments to help Oklo get there. On paper, that one campus could eventually hold more than a dozen Aurora reactors.
Then came the fuel piece. In June, Oklo and Centrus Energy signed a letter of intent for Centrus to supply enough HALEU to power up to five Auroras at that same 1.2-gigawatt Ohio site, with deliveries starting in 2029. The fine print matters, though. The Centrus deal is a non-binding letter of intent, not a signed contract, and the campus itself is not expected to reach full output until the mid-2030s. Oklo also expanded a construction-planning agreement with Kiewit, its lead builder, for the same Ohio project.
Beyond that, the numbers get big and soft fast. Oklo says it holds more than 14 gigawatts of demand under non-binding letters of intent, including a 12-gigawatt partnership with data center firm Switch. And in late May, the DOE picked Oklo for advanced talks on a program to turn surplus weapons-grade plutonium into reactor fuel, paired with a European partner, newcleo, that could put up to $2 billion into the effort, all subject to definitive agreements.
None of those last figures are money in the bank. They are indications of interest, which is the polite industry term for a customer saying it is interested.
Where Aurora actually stands
So here is the real scoreboard on July 1. Aurora has cleared three of four DOE steps to start building. It is not going critical this Fourth of July, no matter what the program calendar says. And the plant that sells power to Meta is a 2027-and-later story that still has to pass a final safety review nobody in the program has passed, plus separate NRC licensing for commercial operation somewhere down the line.
But if you want to know whether the AI-nuclear boom is real or just a thick stack of slides, the reactors going critical on the Fourth are the wrong thing to watch. The tell is quieter: who is pouring concrete, building a fuel plant, and lining up 30-year-old uranium for a machine that has to run for years and actually get paid. On that scoreboard, Oklo just moved.
