Small modular reactors meet the hyperscaler load curve

Why hyperscalers became nuclear buyers in late 2024 #

The chronology is tight. On 20 September 2024, Microsoft and Constellation Energy announced a 20 year power purchase agreement to restart Three Mile Island Unit 1, retired in 2019, with 835 megawatts targeted for return to service in 2028. On 14 October 2024, Google signed a master agreement with Kairos Power for 500 megawatts of advanced reactor capacity from six to seven small modular units, first deliveries by 2030 and full ramp by 2035. Two days later, on 16 October 2024, Amazon Web Services committed to three separate transactions: an investment in X-energy, an offtake agreement for 320 megawatts at Energy Northwest in Washington, and a Memorandum of Understanding with Dominion Energy targeting at least 300 megawatts at the existing North Anna site. In December 2024, Meta issued a public RFP for 1 to 4 gigawatts of new nuclear generation for delivery in the early 2030s.

These four announcements share three features. They are multi gigawatt in stated ambition. They sit outside traditional utility integrated resource planning. They all invoke 24 by 7 carbon free energy procurement standards that wind and solar paired with batteries cannot meet at training cluster load factors.

The buyer logic is straightforward. Hyperscaler carbon accounting moved from annual matching to hourly matching in 2023 and 2024. AI training pushed average data center load factors from 65 to 75 percent typical of cloud workloads to 85 to 95 percent. Higher load factor plus hourly carbon free energy puts a ceiling on what variable renewables plus storage can deliver. Nuclear is the only at scale, dispatchable, zero emission option that closes the gap.

What is actually contracted, what is announced #

The table below distinguishes signed offtake agreements with binding terms from announced pipelines and Letters of Intent. The distinction matters. An LOI does not move a reactor through licensing. A signed PPA with capacity payments does. The contracted column is what vendors can credibly take to NRC, the Department of Energy Loan Programs Office, and project finance.

Three observations come out of the contracted versus announced split. The only operating megawatts in the table are Three Mile Island and Palisades, both existing pressurized water reactors brought back online rather than new construction. The announced pipeline through 2035 is roughly 11 to 13 gigawatts across the four hyperscalers. The contracted, signed offtake is closer to 2 to 2.5 gigawatts. The gap is the SMR commercialization risk. The implicit timing in every announced deal is first delivery between 2028 and 2030 with full ramp by 2035. That requires NRC throughput, HALEU fuel availability, and EPC capacity all clearing together.

Vendor readiness: who can ship and on what calendar #

Eight vendors have credible advanced reactor designs in active US or allied development. NuScale VOYGR cleared NRC Standard Design Approval in July 2024 after UAMPS canceled the original Carbon Free Power Project in November 2023 over cost escalation. Westinghouse AP300 entered NRC pre application engagement in 2024 leveraging the AP1000 design basis. X-energy Xe-100, a high temperature gas reactor under DOE ARDP, is being developed at Energy Northwest and Dow Seadrift. TerraPower Natrium, a sodium fast reactor with molten salt storage, broke ground at Kemmerer in June 2024 on a 4 billion dollar project. Kairos Power received an NRC construction permit for Hermes 1 at Oak Ridge in December 2023, first criticality targeted fiscal 2027. Holtec SMR-300 targets two units at the Palisades site by 2030. Rolls-Royce SMR was shortlisted by Great British Nuclear in October 2024. GE Hitachi BWRX-300 is the most advanced light water SMR with site preparation underway at Darlington Ontario for a 2029 in service date.

The vendor field is wider than 2030 nameplate capacity will support. Industry analysts at the World Nuclear Association and Nuclear Energy Institute project two to four vendors reach commercial operation before 2032. The rest will consolidate, pivot to demonstration only deployments, or stretch dates into the mid 2030s. Hyperscaler procurement decisions through 2025 and 2026 will be the single largest determinant of which vendors clear the survival threshold.

NRC licensing throughput: the binding regulatory constraint #

The Nuclear Regulatory Commission has historically been the slowest moving piece of the US nuclear stack. The Vogtle Units 3 and 4 combined construction and operating license review took roughly 42 months. The first generation of advanced reactor reviews under 10 CFR Part 52 ran longer in many cases. The ADVANCE Act, Public Law 118 to 67, signed 9 July 2024, is the most consequential nuclear regulatory reform since the 1992 Energy Policy Act. The act directs the NRC to develop and implement a risk informed, technology neutral licensing framework with explicit fee reductions for advanced reactors, an aggressive throughput target for combined license reviews, and a mandate to streamline reviews of repeat designs.

The implementation question is whether the NRC can actually deliver that throughput. The agency added roughly 200 staff in fiscal 2025 specifically for advanced reactor reviews, but the institutional bottleneck is reviewer experience, not headcount. The first cohort of approvals under the new framework will set the pace for everything that follows. Kairos Power Hermes 1 demonstration unit, X-energy Xe-100 at Energy Northwest, and Holtec SMR-300 at Palisades are the three pre construction permit reviews that the industry watches most closely as throughput indicators.

The HALEU fuel supply constraint sits behind the NRC question as a parallel binding constraint. Roughly half of the advanced reactor designs require High Assay Low Enriched Uranium, enriched between 5 and 20 percent. The DOE HALEU Availability Program, expanded under the Inflation Reduction Act and the ADVANCE Act to a 2.7 billion dollar envelope, made its first commercial scale shipments from Centrus Energy's Piketon Ohio cascade in November 2023. Centrus Piketon, the only US commercial HALEU supplier, currently produces roughly 900 kilograms per year. The 2030 advanced reactor fleet as currently announced would require 50 to 100 metric tons cumulative through initial cores and reloads. The supply gap is real and it is the single most underpriced risk in the SMR thesis.

Cost: first of a kind versus nth of a kind economics #

Vogtle Units 3 and 4 in Georgia, the only new commercial nuclear units completed in the United States since the 1990s, came online in 2023 and 2024 at a combined project cost of roughly 35 billion dollars against an original 14 billion dollar budget. The Vogtle precedent looms over every advanced reactor cost projection. The Department of Energy August 2024 Pathways to Commercial Liftoff update for advanced nuclear projected first of a kind levelized cost in the 100 to 130 dollars per megawatt-hour range, with nth of a kind targets at 60 to 80 dollars per megawatt-hour after a deployment cohort of roughly 10 to 15 reactors of a given design.

The cost gap between first and nth of a kind is the entire commercial thesis for SMRs. Modular factory fabrication, repeat design certification, and standardized supply chains are supposed to compress the learning curve dramatically relative to large gigawatt class units. The hyperscaler offtake structure is designed to underwrite the first of a kind cost penalty in exchange for guaranteed delivery on a calendar that integrated resource planning processes cannot match. The risk is that cost overruns at the first one or two units of a given design break the offtake economics before the learning curve compresses. The UAMPS cancellation of the NuScale Carbon Free Power Project in November 2023, after the levelized cost estimate rose from 58 to 89 dollars per megawatt-hour during front end engineering, is the cautionary precedent.

Project finance economics shifted decisively in 2024. The DOE Loan Programs Office, recapitalized to roughly 250 billion dollars in remaining authority across the IRA expansions, issued an interim final rule in August 2024 confirming SMR and advanced reactor eligibility for Title 17 loan guarantees. The Palisades 1.5 billion dollar loan that closed in August 2024 is the template. Production tax credits under IRC Section 45U for existing nuclear and Section 45Y for new clean generation provide a 15 to 25 dollar per megawatt-hour subsidy floor that materially de risks the offtake economics. The combined effect of LPO debt at sub Treasury rates plus PTC support is roughly 30 to 40 dollars per megawatt-hour of effective levelized cost reduction. That is exactly the gap between announced first of a kind cost and a credible commercial PPA price.

What hyperscalers, utilities, regulators, and fuel suppliers do next #

For hyperscaler procurement teams, the binding move through 2026 and 2027 is to convert announced pipelines into signed offtake at two or three vendors per buyer rather than concentration in one. The X-energy and Kairos contracts already lean toward a portfolio strategy. The single vendor concentration risk is asymmetric. If one vendor's first of a kind unit goes 80 percent over budget or two years past schedule, an offtake portfolio of one is fatal to the corporate carbon roadmap. An offtake portfolio of three with capacity payment structures that share schedule risk is survivable.

For utilities, the calculus through 2027 is whether to underwrite SMR siting at existing nuclear stations or to host hyperscaler led merchant projects on greenfield campuses. Dominion at North Anna, TVA at Clinch River, and Energy Northwest at Columbia Generating Station are furthest along on existing site siting. Greenfield campus adjacent siting raises grid and emergency planning zone questions that are not resolved.

For NRC and DOE, the 2025 to 2027 throughput question is binary. Either the agency delivers two combined construction and operating licenses for advanced reactor designs by end of 2027, with two more under active review, or the ADVANCE Act reform is judged failed. HALEU is the parallel test. Either DOE and Centrus deliver a 2030 commercial scale supply chain at 5 to 10 metric tons per year with a path to 30 to 50 by 2035, or roughly half the advanced reactor pipeline becomes uneconomic on fuel grounds alone.

For fuel cycle suppliers, the strategic prize is enrichment capacity. Centrus Piketon is the only operating commercial scale US HALEU cascade. Urenco and Orano have announced HALEU expansion plans in the UK, Netherlands, and France. Russian enricher TENEX, historically 20 percent of US enrichment services, is no longer viable under the May 2024 Prohibiting Russian Uranium Imports Act. The structural shortage in the Western HALEU supply chain through 2030 is the most underpriced opportunity in the advanced nuclear value chain. The 2.7 billion dollar DOE HALEU program is a down payment, not a solution.

Sources #

• US Nuclear Regulatory Commission, Advanced Reactor Licensing
• ADVANCE Act of 2024, Public Law 118 to 67
• IAEA Advanced Reactors Information System and SMR Dashboard
• DOE Pathways to Commercial Liftoff: Advanced Nuclear, August 2024 update
• DOE Loan Programs Office, Palisades restart conditional commitment
• DOE HALEU Availability Program
• World Nuclear Association, Small Nuclear Power Reactors
• Nuclear Energy Institute, Advanced Reactors Resources
• IEA, Nuclear Power and Secure Energy Transitions
• Constellation Energy, Three Mile Island Unit 1 restart announcement
• Google, agreement with Kairos Power for advanced nuclear
• Amazon Web Services, nuclear energy investments
• Kairos Power, Hermes demonstration plant
• X-energy, Xe-100 advanced reactor
• TerraPower, Natrium reactor program
• NuScale Power, VOYGR SMR plants
• Holtec International, SMR-300 program
• GE Hitachi BWRX-300 small modular reactor
• Centrus Energy, HALEU production
• Reuters, Microsoft Three Mile Island deal coverage
• Financial Times, hyperscaler nuclear procurement coverage