The NRC Just Proposed a New Way to License Nuclear Reactors in 18 Months. The Old Way Took a Decade. Here's What Changed.

One hundred and thirty-eight pages. That is the length of the proposed rule the Nuclear Regulatory Commission published in the Federal Register on May 1, 2026, creating an entirely new regulatory part, 10 CFR Part 57, for licensing microreactors. A new part of the Code of Federal Regulations, not an amendment to existing rules and not a guidance document, something the NRC has not created for reactor licensing since Part 52 in 1989.

Read the language carefully, because it is striking for a nuclear regulator. Part 57 uses the phrase "rapid and high-volume deployment" to describe its purpose, and then proceeds to establish the mechanisms that make that phrase operational: general construction licenses for nth-of-a-kind reactors, combined licensing pathways that merge construction permits with operating licenses and manufacturing licenses into a single application, and categorical exclusions under NEPA that could eliminate environmental reviews for standardized designs at previously approved sites. Before this proposed rule, the NRC had never applied "high-volume" language to nuclear power. Solar panels and wind turbines get high-volume permitting; nuclear reactors, until May 1, did not.

Two days earlier, on April 28, the University of Utah's TRIGA research reactor generated electricity for the first time in approximately 50 years and routed it to a high-performance GPU compute node. Power output was modest, just 2 to 3 kilowatts produced via a cold-helium reverse Brayton cycle developed in partnership with Elemental Nuclear Energy Corp and a consortium of 12 universities, which is enough to run a single server rack but not a data center. Still, it was the first demonstrated instance of nuclear-powered AI compute, and Elemental's commercialization target is 2030.

What Part 57 Actually Changes

Nuclear licensing in the United States has operated under two frameworks for decades. Part 50, written in the 1950s, requires separate applications for a construction permit and an operating license, with mandatory public hearings at each stage. A reactor licensed under Part 50 typically spent five to ten years in the regulatory pipeline before generating a single watt. Part 52, introduced in 1989, combined the two steps into a single application but added its own complexity: Inspections, Tests, Analyses, and Acceptance Criteria (ITAAC) that must be verified before operation, with a separate hearing to confirm ITAAC closure. Timeline under Part 52: three to seven years in practice.

Part 57 was mandated by the ADVANCE Act (Pub. L. 118-67, signed July 2024) and Executive Order 14300 (May 2025), which directs the NRC to reach a final licensing decision within 18 months of receiving a complete application. A final rule must be issued by November 23, 2026, and public comments close June 15.

Framework	Typical Timeline	Application Structure	Hearing Process
Part 50 (1950s)	5-10 years	Separate construction permit + operating license	Mandatory hearings at each stage
Part 52 (1989)	3-7 years	Combined license, but ITAAC verification required	ITAAC closure hearing
Part 57 (2026)	~18 months target	Joint CP/OL + manufacturing license	Single hearing; ACRS review only for novel designs
Part 57 nth-of-a-kind	Potentially <12 months	General construction license	Reduced or eliminated

What makes this provision consequential is the general construction license for nth-of-a-kind reactors, which introduces a licensing model nuclear power has never had. Once a microreactor design receives its initial Part 57 approval, subsequent units of the same design can begin construction under a general license before a site-specific construction permit is issued. This is how the NRC licenses nuclear fuel facilities, and it mirrors how the FAA certifies commercial aircraft: prove the design once, then manufacture and deploy without relitigating the engineering for each unit. Applied to microreactors, it means a company could license one 10 MWe design and then deploy dozens of identical units at different sites without submitting dozens of separate applications.

The Data Center Math Nobody Has Run

Here is the calculation that connects Part 57 to the AI infrastructure buildout. A hyperscale data center consumes 100 to 500 megawatts of continuous power, so take a midrange facility at 300 MW. A typical microreactor generates 1 to 20 MWe, and using a 10 MWe unit at 80% capacity factor (conservative for nuclear, which routinely exceeds 90%), each reactor delivers 8 MW of reliable baseload output, meaning 300 MW requires 38 microreactors.

Under Part 50, licensing those 38 reactors would mean 38 separate applications, 38 sets of public hearings, and 38 five-to-ten-year regulatory timelines, a prospect so daunting that nobody would attempt it.

Under Part 57 with nth-of-a-kind general licensing, you certify the design once, a process that takes roughly 18 months, and then deploy 38 units under a general construction license. Regulatory burden drops from 190 cumulative years of application processing (38 × 5 years) to approximately 18 months plus site-specific safety reviews. According to the NRC's own regulatory analysis, the net cost savings amount to $3.76 billion at a 7% discount rate and $11.84 billion at a 3% discount rate across the industry over the rule's lifetime.

Demand is real and growing. Data centers consumed approximately 17% of U.S. electricity in 2025, and that share is expanding as AI training clusters scale to meet insatiable compute requirements from Microsoft, Google, Amazon, and Meta, all of which have signed nuclear power agreements in the past 18 months. Meta's Q1 2026 earnings call raised capex guidance to $125 to $145 billion, with data center power as a primary constraint, and grid transformers have a 3-year backlog while interconnection queues at major utilities exceed five years. Companies that need the most power have the least access to it through conventional channels, which is why a regulatory pathway that could deliver 24/7 baseload power to a single site within two to three years matters more than any individual reactor design.

The NuScale Problem: Faster Licensing Does Not Fix Costs

Consider NuScale Power, because its story is the strongest case against Part 57 optimism. NuScale held the only NRC-certified SMR design in the United States, and its Carbon Free Power Project at Idaho National Laboratory was the most advanced small reactor project in the country. It died in November 2023 because costs ballooned from $5.3 billion to $9.3 billion, a 75% overrun that obliterated the project's economic rationale and sent the subscribers who had committed to buying the power walking away from their contracts.

NuScale's failure was not regulatory, because the NRC had already certified the design; it was a failure of construction economics, supply chain management, and the brutal reality that building nuclear infrastructure at any scale in the United States costs more than almost anywhere else on Earth. South Korea builds nuclear reactors for approximately $2,500 per kilowatt, while the U.S. average for the two AP1000 units at Plant Vogtle, the only reactors completed in the past 30 years, exceeded $13,000 per kilowatt. Part 57 addresses the licensing timeline but does nothing about the cost problem, and the cost problem is what killed NuScale.

The counterargument deserves its full weight, because microreactors are fundamentally different from NuScale's 77 MWe modules. They are designed to be factory-built, truck-transportable, and operator-free, relying on passive safety systems that require no human intervention during abnormal conditions, and the "high-volume" framing in Part 57 explicitly targets standardized, mass-produced units whose economics improve with production scale. A factory that builds 500 identical microreactors per year operates on a different cost curve than a construction site that assembles one bespoke reactor over a decade. But this is a theoretical cost curve built on projected manufacturing efficiencies that no one has demonstrated. No microreactor has been commercially deployed anywhere in the world, no factory production line exists, and the earliest projected commercial deployments are 2027 to 2029 for military and DOD applications, not civilian power.

Who Benefits First

NANO Nuclear Energy, a publicly traded company whose KRONOS mobile microreactor platform was designed specifically for rapid deployment, has seen its stock price rise 18% in the 30 days following the Part 57 announcement, while the Department of Energy has allocated $900 million for SMR development, including $400 million for the Tennessee Clinch River project. Kairos Power, X-energy, and Oklo all have microreactor or small reactor designs at various stages of NRC review, and Part 57 could accelerate each of them.

Companies positioned to move fastest are those with designs already in the NRC pipeline and partnerships with utilities or data center operators who can serve as anchor customers. Google's 500 MW nuclear commitment through Kairos Power, announced in 2024, sits squarely in this category, as does Amazon's agreement with Talen Energy for nuclear-adjacent data center power at the Susquehanna facility in Pennsylvania. These are not hypothetical expressions of interest but contracted power purchase agreements worth billions of dollars, and every month that licensing takes faster reduces the cost of capital on those commitments.

What We Do Not Know

Part 57 is a proposed rule, not a final one, and the comment period runs through June 15, 2026, during which industry groups, environmental organizations, and state regulators will submit feedback that could substantially modify the final version. Provisions allowing applicants to self-define "basic component" and "safety-related" classifications will draw scrutiny from safety advocates who argue this amounts to the regulated writing their own rules, and categorical NEPA exclusions for standardized designs will face legal challenges from environmental groups.

No microreactor has operated commercially. Utah's demonstration generated 2 to 3 kilowatts from a research reactor designed in the 1960s, orders of magnitude below the 1 to 20 MWe range of proposed commercial designs. LCOE projections for microreactors range from $100 to $200 per MWh for early units down to $60 to $80 per MWh at production scale, but these are modeled estimates, not observed costs, and until someone builds unit number one, ships it to a site, connects it to a grid or data center, and reports what the whole thing actually cost, every number in this analysis is provisional. Waste storage, decommissioning liability, and nuclear proliferation risks associated with distributed reactor deployment are beyond the scope of this analysis.

What You Can Do

If you work in data center operations: The Part 57 comment period closes June 15, 2026. Read the 138-page proposed rule, particularly sections on general construction licenses and combined licensing pathways. If your company has power procurement constraints, this rule directly affects your 2030 options. Submit substantive comments to regulations.gov; docket ID NRC-2023-0045.

If you invest in energy or infrastructure: Watch three companies: NANO Nuclear Energy (NNE), Kairos Power, and X-energy, each of which has designs positioned for Part 57 pathways. Cost savings estimates of $3.76B to $11.84B represent value that accrues disproportionately to first movers with NRC-ready designs, so monitor the comment period for signals of weakening provisions that could dilute that advantage.

If you are a policymaker or regulator: Part 57's self-definition provisions, which let applicants determine their own "basic component" and "safety-related" classifications, are the rule's most controversial feature and its most likely point of legal challenge, so review the ACRS (Advisory Committee on Reactor Safeguards) feedback on novel vs. non-novel design determinations, which will shape how much independent safety review each application receives.

The Bottom Line

For the first time in its history, the NRC is proposing to treat nuclear reactors the way the FAA treats aircraft: certify the design, then manufacture and deploy at volume without relitigating the engineering for each unit. Whether that analogy holds depends on whether microreactor companies can solve the construction cost problem that killed NuScale, and that problem has nothing to do with regulation. Part 57 removes the barrier that was always cited as the reason nuclear could not scale. Now the industry has to prove that the barrier was actually the binding constraint, and not a convenient excuse for an economic model that never worked.