Major technology companies need enormous amounts of electricity to run artificial intelligence workloads, and several are now turning to small modular nuclear reactors to get it. The Nuclear Regulatory Commission has already issued a construction permit for Kairos Power’s Hermes test reactor, a non-light-water advanced design that represents one of the first concrete approvals in a new generation of nuclear technology. At the same time, Energy Northwest is in formal pre-application discussions with the NRC to build up to 12 Xe-100 small modular reactor units at the Hanford Reservation in Washington state. These parallel tracks show that nuclear power for data centers has moved from corporate talking points into active federal licensing proceedings.
Regulatory speed will decide whether nuclear or gas powers AI by 2030
The central question facing tech companies with aggressive carbon-reduction goals is whether the NRC can process advanced reactor applications fast enough to deliver operational reactors before the end of the decade. If licensing reviews stretch past their expected timelines, utilities and data-center operators will face a binary choice: wait for clean nuclear generation or contract new natural-gas plants that can be built on shorter schedules. That tradeoff puts the NRC’s review pace at the center of the AI energy debate.
Kairos Power’s Hermes reactor offers the clearest test case. The NRC granted a construction permit for the Hermes project, making it the first non-light-water advanced reactor to clear that licensing stage. Hermes is a test reactor, not a commercial power plant, but its approval established a regulatory pathway that other advanced designs will need to follow. The speed at which the NRC moved from application to permit on Hermes will set expectations for every SMR developer pitching reactors to hyperscale cloud providers.
The Xe-100, developed by X-energy LLC, is on a different timeline. The NRC lists the design in formal pre-application engagement, meaning the agency and the developer are exchanging licensing white papers and topical reports before a full construction permit application is filed. Pre-application work reduces surprises during formal review, but it also adds months or years before ground can be broken. For companies that need power by 2030, every quarter spent in pre-application dialogue is a quarter closer to defaulting to fossil-fuel backup.
NRC dockets show two distinct reactor projects advancing
The strongest public evidence that SMRs are progressing beyond announcements sits in NRC docket files, not in corporate press releases. Kairos Power’s Hermes reactor has a completed Final Environmental Impact Statement, cataloged as NUREG-2263, which documents the site description, alternatives the agency considered, and its impact findings. That document represents thousands of pages of technical review and public comment, and its completion was a prerequisite for the construction permit. The fact that it exists in final form confirms the project cleared every environmental hurdle the NRC imposes.
Separately, Energy Northwest’s Cascade Advanced Energy Facility would place up to 12 Xe-100 units at the Hanford Reservation, adjacent to the existing Columbia Generating Station. The NRC confirms it is engaged in pre-application activities with Energy Northwest for this project. Siting next to an operating nuclear plant offers practical advantages: the location already has transmission infrastructure, security protocols, and a workforce familiar with nuclear operations. A 12-unit SMR campus would represent one of the largest planned deployments of modular reactors anywhere in the country.
These two projects involve different reactor technologies, different developers, and different stages of NRC review. Hermes uses a fluoride-salt-cooled design; the Xe-100 is a high-temperature gas-cooled reactor. Their simultaneous progress through the NRC system signals that the agency is handling multiple advanced designs at once, though neither project has reached the point of generating commercial electricity. For data-center operators, that means the technology pathway is visible, but the commercial timeline is still uncertain.
Missing contracts and load forecasts leave the 2030 timeline uncertain
For all the licensing activity visible in NRC filings, several pieces of the puzzle are absent from the public record. No signed power-purchase agreements or capacity commitments between these reactor developers and specific AI hyperscalers appear in the NRC dockets reviewed. Corporate announcements have linked companies like Google to Kairos Power, but the regulatory filings themselves do not contain binding offtake contracts. Without those agreements on file, the financial certainty behind these projects is harder to assess from public documents alone.
Quantitative projections tying AI-driven electricity demand directly to these SMR projects are also missing from the cited environmental statements and pre-application materials. Industry analysts have published estimates of data-center load growth, but the NRC filings focus on reactor safety, environmental impact, and site suitability rather than on downstream demand modeling. That gap matters because the business case for building 12 Xe-100 units depends on whether enough customers will commit to buying the power at prices that justify the capital cost and regulatory risk.
Another uncertainty is how quickly AI-related demand will materialize in specific regions. While hyperscale cloud providers are expanding across the United States, the siting of new data centers does not always align with proposed nuclear projects. If the bulk of AI workloads cluster far from Hanford or the Hermes site, long-distance transmission constraints could weaken the argument for co-locating reactors and data centers. Conversely, if operators are willing to place compute campuses near nuclear sites, they must navigate local land-use rules and community acceptance alongside federal licensing.
Financing structures will also influence whether these reactors can realistically serve AI loads by 2030. Test reactors like Hermes are primarily demonstration projects, designed to prove out technology and regulatory processes rather than to supply bulk power. Only after such demonstrations are successful can developers move toward larger commercial units that could anchor long-term supply contracts with technology companies. That sequencing adds years to the timeline between today’s permits and tomorrow’s power purchase agreements.
What tech companies should watch in the NRC process
For technology firms planning multi-decade AI infrastructure, several milestones in the NRC process will be more important than individual press releases. The first is the transition from pre-application engagement to formal licensing for commercial-scale reactors. When a developer submits a full construction permit or combined license application, it signals that design work, site selection, and initial financing are mature enough to justify the expense of detailed review.
The second milestone is the issuance of final environmental impact statements, like the one completed for Hermes. These documents indicate that the NRC has weighed alternatives, evaluated local impacts, and responded to public comments, clearing a major procedural hurdle. For investors and potential offtakers, a final EIS reduces the risk of late-stage delays tied to environmental litigation or community opposition.
The third milestone is the granting of operating licenses or equivalent authorizations that allow reactors to load fuel and generate power. Until that point, even a project with a construction permit remains a promise rather than a resource. For AI planners working on 2030 and 2035 roadmaps, the difference between a reactor under construction and a reactor in commercial operation is the difference between a speculative option and a bankable supply source.
In parallel, tech companies will need to negotiate their own contractual frameworks with utilities and reactor developers. Long-term, fixed-price contracts could help de-risk nuclear projects, but they also lock in costs for data-center operators in a rapidly evolving market. Shorter contracts or more flexible arrangements might preserve agility but make it harder for nuclear developers to secure financing. The balance between these approaches will shape whether SMRs become a mainstream tool for decarbonizing AI or remain a niche solution.
By 2030, the outcome will be visible in the physical grid: either new clusters of advanced reactors feeding power-hungry AI campuses, or a patchwork of new gas plants filling the gap left by delayed nuclear projects. The documents already on file with the NRC show that advanced reactors are moving, but they do not yet guarantee that nuclear will beat gas to powering the next wave of artificial intelligence.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
