Research Finder
Find by Keyword
Could A Fleet Of Tiny Reactors Power The Next AI Data Center?
Four US microreactors safely achieved self-sustaining reaction in one month, a pace unmatched in the modern era, while Valar and NVIDIA just showed where that power could go.
07/07/2026
Key Highlights
- Four US startups, Antares Nuclear (June 4), Valar Atomics (June 18), Deployable Energy (June 30), and Aalo Atomics (July 4), each safely brought a new small reactor design to self-sustaining criticality in a single month.
- According to the DOE, this marks an unmatched pace in the modern advanced reactor era, with private U.S. startups driving multiple designs to this milestone rapidly. The feat exceeded exceeded Executive Order 14301's target of three reactors by July 4, 2026, hitting four, with seven more companies (Oklo, Radiant Industries, Terrestrial Energy, Natura Resources, Last Energy, Atomic Alchemy, and Deep Fission) still in the DOE authorization queue.
- On July 1, Valar Atomics and NVIDIA used Ward 250's output to power an NVIDIA Blackwell based system on stage, then announced plans to explore a 30 megawatt AI data center in Emery County, Utah.
- The four reactors represent four distinct technical approaches, sodium heat pipe, helium cooled TRISO, water moderated helium cooled, and sodium cooled, run in parallel rather than converging on one design.
- Valar and other developers are simultaneously challenging the NRC's asserted jurisdiction over DOE authorized microreactors, a case that could widen the licensing pathway for the entire sector.
The News
Four US startups, Antares Nuclear (June 4), Valar Atomics (June 18), Deployable Energy (June 30), and Aalo Atomics (July 4), each brought a new reactor design to criticality in a single month. Criticality is an important physics milestone on the path to energy production, where the atomic reaction becomes self-sustaining. According to the DOE, the first three of these milestones made the US unmatched in achieving criticality in three unique advanced microreactor designs in a single month, with Aalo's reactor then arriving just ahead of the July 4 deadline to bring the total to four. The feat exceeded Executive Order 14301's target of three reactors by July 4, 2026, hitting four, with seven more companies (Oklo, Radiant Industries, Terrestrial Energy, Natura Resources, Last Energy, Atomic Alchemy, and Deep Fission) still in the DOE authorization queue. On July 1, Valar Atomics and NVIDIA used Ward 250's output to power an NVIDIA Blackwell based system on stage, then announced plans to explore a 30 megawatt AI data center in Emery County, Utah. The four reactors represent four distinct technical approaches, sodium heat pipe, helium cooled TRISO, water moderated helium cooled, and sodium cooled, run in parallel rather than converging on one design. Valar and other developers are simultaneously challenging the NRC's asserted jurisdiction over DOE authorized microreactors, a case that could widen the licensing pathway for the entire sector
Analyst Take
Four advanced (and different tech) reactor designs, four different companies, one calendar month. That is the headline the Department of Energy wanted, and it appears to be a defensible one. Decades of nuclear projects measured progress in years; this cycle measured it in weeks, from Antares' sodium heat pipe design on June 4 to Aalo's sodium cooled Aalo-X in the small hours of July 4. Notably, none of the four converged on the same architecture. Different designs (sodium heat pipe, helium cooled TRISO, water moderated helium cooled, and sodium cooled) all launched in parallel under the same policy push. Like a recent U.S. push in quantum computing, this portfolio approach spreads technical risk across bets rather than concentrating it in one design. NVIDIA's decision to stage a live demonstration on the back of Valar's criticality milestone reads as a signal that the industry sees these designs as more than a policy trophy. The obvious skeptical read is that a 100 kilowatt demonstration and an exploratory 30 megawatt data center are small next to the gigawatt scale hyperscalers are used to procuring. Looking at the design math across all four reactors, though, the smaller unit is the point, not the limitation.
What Was Announced
The Ward 250 is a fourth generation, TRISO fueled, high temperature gas reactor that uses pressurized helium rather than water as its coolant, designed for a commercial output of 5 megawatts electric. Valar built and operated the unit at the Utah San Rafael Energy Lab in Emery County, the first Department of Energy authorized reactor built and run outside a national laboratory. On July 1, the company ran Ward 250 at approximately 37% of its test output, or roughly 100 kilowatts of thermal energy, to power an NVIDIA desktop unit built on the Blackwell architecture, hosting a simple web server as evidence the current was real. Antares' Mark-0, by contrast, remains at zero power test configuration, a physics demonstration rather than an electricity producing unit, with the company targeting 2027 for its first electricity production.
Reactor Math: Sizing Microreactors To A 30 Megawatt Data Center
- Valar Ward 250, designed for 5 MWe at commercial scale: approximately six units to reach 30 MW.
- Aalo-X, designed for 10 MWe per unit, clustered into 50 MWe Aalo Pods: three units would exceed 30 MW. Aalo has said it plans a one million square foot factory targeting roughly a hundred reactors a year at full capacity, the clearest aspirational marker of what mass production could mean for this model.
- Deployable Unity, designed as a 1 MWe nuclear battery: approximately thirty units to reach 30 MW.
- Antares Mark-0, sized for remote and military behind the meter loads rather than a stated commercial MWe figure, with electricity production targeted for 2027.
In the Valar case, a proposed Utah facility would pair Valar's generation with NVIDIA's DSX cooling design. The resulting facility aims at closed loop liquid cooling in place of evaporative towers, which the companies frame as a near waterless alternative to conventional data center cooling. Neither company has published a construction schedule or a capital figure, and the facility remains in the exploratory stage. What the math above suggests, however, is that the gap between a criticality demonstration and a functioning AI campus may close through replication of small units rather than years spent scaling up a single design, a materially different build path than the nuclear industry's historical model. An approach where a factory produces a hundred units a year (if Aalo's plan holds) starts to look less like aspiration and more like infrastructure.
Market Analysis
Hyperscalers have already committed to nuclear at gigawatt scale. Microsoft's agreement to restart the Three Mile Island Unit 1 reactor and Meta's commitments spanning TerraPower, Oklo, Vistra, and Constellation are both structured as offtake agreements against existing or restarted utility scale plants with power arriving toward the end of this decade. The four microreactor designs that went critical this cycle represent a different route to the same destination: instead of one large asset tied to the public grid, a data center operator assembles capacity from many identical, factory built units sited directly behind the meter. Aalo's own commercial model, clustering 10 MWe reactors into 50 MWe pods, is essentially the electric utility built from a spare parts catalog rather than a single custom plant, and it is this behind the meter, self contained quality, more than the criticality milestone itself, that points toward a genuinely power independent model for AI infrastructure.
That independence claim rests on more than reactor design. Aalo's parallel collaboration with NVIDIA and Microsoft on automated reactor fleet management points toward the operational layer this model will eventually need, coordinating dozens of small reactors the way a data center operator coordinates racks of servers rather than the way a utility dispatches a handful of large plants. It also rests on the regulatory pathway. Valar, alongside Texas and Utah, is currently challenging the NRC's asserted jurisdiction over DOE authorized test reactors. Should that challenge succeed even partially, the DOE authorization process used to reach criticality this cycle could become a viable route to commercial deployment in its own right, sidestepping a large part of the traditional NRC licensing timeline that has historically kept nuclear projects on multi-year, not multi-month, schedules.
The breadth of the field reinforces the same trend, though it comes with real caveats. Seven additional companies, Oklo, Radiant Industries, Terrestrial Energy, Natura Resources, Last Energy, Atomic Alchemy, and Deep Fission remain in the DOE authorization queue behind this cycle's four. Another point worth emphasis, none of the four winners converged on identical technology. That approach, similar to that used in a recent executive order on quantum computing, functions as a hedge across the sector if one design proves harder to scale or fuel than expected. Set against that hedge, fuel supply remains a real constraint for at least some of the four: Antares depends on HALEU tri-structural isotropic fuel compacts, a supply chain still maturing at commercial scale, while Deployable chose standard low enriched uranium specifically to avoid that exposure. None of the four reactors has yet operated at power for any meaningful duration, so questions about sustained operation, mass production yield against Aalo's stated hundred-unit annual target, and public acceptance of siting multiple units near data centers or industrial loads remain open. The power independent thesis is, for now, a design and regulatory trajectory rather than a proven operating record.
Looking Ahead
It is easy, and we think incorrect, to read this month's pace as a simple political story. The bipartisan ADVANCE Act, signed into law in 2024, already cut NRC licensing fees for advanced reactor applicants and mandated new microreactor guidance, and the executive order reforming the NRC explicitly builds on those provisions rather than replacing them. What has shifted is not bipartisan appetite for advanced nuclear, which has been building since at least 2020, but the mechanism used to accelerate it. The Reactor Pilot Program and the Nuclear Energy Launch Pad rest on this administration's use of DOE's own site authorization authority under a hard deadline, an executive choice rather than a statutory one. That distinction should matter more to how we read 2028 than any partisan framing does. The underlying NRC licensing reforms are now written into law and passed Congress with near unanimous support, which makes them durable regardless of who holds the White House. The DOE's sprint model is not written into law in the same way, and a future administration could simply decline to run another one, leaving the seven companies still in queue to face a considerably slower path than the four that just went critical.
Our analysis of the market suggests the next twelve months will show whether the clustering model translates from criticality demonstrations into actual megawatt hours delivered to a load. Based on what we are observing, Aalo's Project Ascension, a second commercial scale reactor already under construction at the same Idaho site and targeted toward producing electricity in 2027, is the nearest test of whether a 10 megawatt unit can move from zero power physics testing to producing electricity for an on site data center. We will also be tracking whether Valar converts its Utah collaboration with NVIDIA into a signed construction agreement, whether the NRC jurisdiction litigation produces a ruling that widens the DOE pathway for the companies still in queue, and whether Deployable's stated pipeline of letters of intent begins converting into contracted projects. If even one of the four designs reaches sustained power operation within the next year, and if the regulatory question resolves in the developers' favor, the case for a power independent, factory built nuclear fleet moves from design physics to operating fact.
Stephen Sopko | Analyst-in-Residence – Semiconductors & Deep Tech
Stephen Sopko is an Analyst-in-Residence specializing in semiconductors and the deep technologies powering today’s innovation ecosystem. With decades of executive experience spanning Fortune 100, government, and startups, he provides actionable insights by connecting market trends and cutting-edge technologies to business outcomes.
Stephen’s expertise in analyzing the entire buyer’s journey, from technology acquisition to implementation, was refined during his tenure as co-founder and COO of Palisade Compliance, where he helped Fortune 500 clients optimize technology investments. His ability to identify opportunities at the intersection of semiconductors, emerging technologies, and enterprise needs makes him a sought-after advisor to stakeholders navigating complex decisions.