Four American startups have brought their first small nuclear reactors to life, each reaching the point where the reactor can sustain a controlled chain reaction. The companies are chasing a specific goal: generating power on-site for the electricity-hungry data centers behind the artificial intelligence boom. Reaching criticality is a significant technical milestone, though it remains a long way from a fully operational commercial power plant.
The four companies that reached criticality
The startups that hit the milestone are Antares Nuclear, Valar Atomics, Aalo Atomics, and Deployable Energy, each of which achieved criticality within a federally set deadline. According to an overview of the developments, reaching criticality means a reactor can maintain a steady, self-sustaining fission reaction and produce a controlled release of energy. It is the threshold that separates a reactor design on paper from one that is actually operating.
Crucially, criticality is a beginning rather than an end. A reactor that has gone critical has demonstrated its core physics works, but scaling that up into a licensed facility delivering reliable power to paying customers involves years of additional engineering, testing, and regulatory review.
Why data centers are turning to nuclear
The push is being driven by the enormous and growing electricity demand of AI computing. Training and running large AI models requires data centers that consume power on the scale of small cities, and the operators of those facilities are increasingly worried about securing enough clean, uninterrupted electricity. As an examination of the trend detailed, major technology companies have been funding nuclear projects to lock in dependable power rather than relying solely on strained regional grids.
Small modular reactors, or SMRs, are attractive for this because they are designed to be built in factories and deployed close to the point of use. Placing a reactor on or near a data center campus, in theory, delivers steady baseload power without the transmission bottlenecks that slow grid connections, and without the carbon emissions of fossil-fuel generation.
The gap between a milestone and a power plant
For all the significance of reaching criticality, the achievement should not be mistaken for a working commercial reactor. Independent experts have urged caution about the surrounding enthusiasm. According to a critical assessment of the sector, much of the messaging around next-generation nuclear for data centers outpaces what the technology has actually demonstrated, and turning early reactors into affordable, licensed, grid- or campus-scale power sources remains a formidable challenge.
History offers a sobering backdrop: advanced reactor projects have repeatedly run over budget and behind schedule, and the economics of building small reactors at scale are still unproven. A demonstration reactor that achieves criticality has cleared an important hurdle, but the path to routinely powering data centers commercially is longer and less certain than promotional announcements often suggest.
What makes a reactor small and modular
Small modular reactors differ from conventional nuclear plants in both scale and construction philosophy. Where a traditional reactor is an enormous, custom-built facility producing on the order of a thousand megawatts, an SMR is designed to be far smaller and to be manufactured in standardized units in a factory, then shipped and assembled on-site. The idea is that factory production could make reactors cheaper, faster to build, and more consistent in quality than sprawling one-off projects that have historically run years late and billions over budget.
Modularity also allows capacity to be added incrementally, with additional units installed as demand grows rather than betting on a single massive plant from the outset. For a data-center operator, that flexibility is appealing: power can, in principle, be scaled up in step with a facility’s expansion. The reactors reaching criticality in this milestone represent early, small-scale demonstrations of that concept, proving the underlying reactor works even as the manufacturing and commercial models remain to be established.
The scale of AI’s electricity appetite
The urgency behind these projects comes from the sheer amount of power that artificial intelligence consumes. The specialized processors that train and run large AI models draw enormous quantities of electricity, and clustering thousands of them in a single data center can create a demand comparable to that of a small city. Operators want that power to be constant, since AI systems run continuously, and increasingly they want it to be low-carbon to meet corporate climate commitments.
Those requirements are difficult for existing grids to satisfy quickly. Connecting a large new data center to the grid can take years, and in some regions the available capacity is already stretched. On-site nuclear generation offers a way to sidestep those bottlenecks by producing firm power right where it is needed, which is why technology companies have shown growing interest in reactors small enough to sit alongside a computing campus. The four startups reaching criticality are betting that this convergence of digital demand and nuclear supply will define the next phase of both industries.
What the milestone signals for AI and energy
The convergence of AI and nuclear power reflects a broader reality: the computing revolution is running headlong into the physical limits of the electricity system. If AI demand continues to climb, the industry will need vast new sources of firm, low-carbon power, and nuclear is one of the few options that can deliver it around the clock without weather dependence.
Whether these startups can move from criticality to competitive, deployable power plants will shape how much of that vision materializes. For now, four reactors quietly sustaining fission mark an early, concrete step in an effort to marry the digital economy’s appetite for electricity with a new generation of nuclear technology, an ambition that is promising in principle but still has much to prove in practice.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
- More than 60,000 people flee the Spokane area as complex fires overrun 600 structures
- 11 engines built to run well past 200,000 miles
- Card skimmers hidden on gas pumps and ATMs are draining accounts, and here’s the tell
- The FBI says hackers are hijacking outdated home routers, and it named the models to check
