power for compute
Who can deliver firm power to new load, and how far along each one is.
Data centers need large amounts of power that runs around the clock. Building it is less about inventing new machines than about getting them permitted and connected. Gas turbines are available now. Advanced reactors, fusion and deep geothermal are still proving themselves. Each approach on this page is a different answer to the same problem: firm power, in the right place, soon enough.
- Connection, not generation, sets the pace.
- US interconnection queues held 2,061 GW at end-2025, and new projects now wait more than five years from request to operation. Rule changes at FERC and the grid operators in 2026 and 2027 appear likely to matter more than any single technology.
- Permits are now being issued.
- The NRC approved its first commercial construction permits for advanced and small modular reactors in 2026, in about 16 to 24 months each. The next test is whether those plants are built on schedule and on budget.
- Gas and on-site power fill the gap.
- Until reactors and geothermal arrive at scale, most new firm supply for data centers is gas turbines and fuel cells, often placed behind the meter to avoid the queue. That bridge carries fuel-price and emissions-permitting risk.
- The thesis depends on load growth.
- Developers are signing contracts years before first power. If compute demand slows before those contracts are binding, the later projects lose their buyers first.
what would change the view
- Median request-to-operation time falls below four years in the next Berkeley Lab release, or federal large-load rules let data centers connect within two years.
- Kemmerer or Clinch River slips more than two years, or the next construction permits (Long Mott, KRONOS) stall past 2027.
- Large buyers cancel or shrink binding power contracts, or turbine slot reservations are cancelled at scale.
map
scoreboard Firm megawatts that hold both a permit to build and a right to connect. No public source totals this, so the map shows each project by the furthest gate it has passed. Queue gigawatts count requests, not plants: about 13% of capacity that requested interconnection in 2000 to 2020 reached operation. Zero-power criticality proves physics, not electricity.
One row per project at the furthest stage reached, grouped by approach. Filled dots: listed company or unit of one; hollow: private. Right column: the developer's claimed first-power date, a claim rather than a forecast. Highlighted: Kemmerer (construction permit March 2026, the NRC's first approval for a non-light-water reactor in more than 40 years), Fervo (commercial operation, 30 September 2026) and Crane (restart, after a June 2026 FERC waiver moved its grid connection forward from 2030).
Active US interconnection queue at year-end. The fall since 2023 is mostly withdrawals (more than 750 GW in 2025), not completions. 2014 is derived from Berkeley Lab's description of the 2023 total; hollow.
Median years from interconnection request to commercial operation, by group of projects built. Hollow dots are bounds ("under" or "more than"). Groups overlap, so they are not annual medians.
players
Listed companies with a dated milestone in scope. Private companies are included if they have operated a reactor or fusion machine, are building one under an NRC permit or a DOE program, or own a licensed plant being returned to service. Funding amounts are not shown for anyone.
- GE Vernovalisted, NYSE: GEV
- gas turbines; BWRX-300 small modular reactor · Clinch River construction permit 29 Sep 2026; gas backlog and slot reservations 122 GW (3Q26). Supplies both the near-term gas bridge and a light-water SMR design.
- Constellation Energylisted, Nasdaq: CEG
- restart; existing nuclear fleet · Crane 835 MW, grid connection rights secured by FERC waiver (Jun 2026). Restart targeted 2H27 under a 20-year agreement with Microsoft.
- X-energylisted, Nasdaq: XE
- high-temperature gas reactor; TRISO fuel · Long Mott 320 MWe; NRC completed its final safety evaluation (Oct 2026). Construction permit date not yet set; a second Amazon-backed site in Indiana is at feasibility stage.
- Oklolisted, NYSE: OKLO
- sodium fast microreactor · Aurora-INL up to 75 MWe; DOE preliminary safety analysis approved (Jun 2026). Pursuing DOE authorization at Idaho National Laboratory; operation expected 2028.
- NuScale Powerlisted, NYSE: SMR
- light-water small modular reactor · US460 standard design approval May 2025; no US plant under construction. Framework for up to 6 GW with ENTRA1 and TVA.
- BWX Technologieslisted, NYSE: BWXT
- microreactors; naval reactors; fuel · Project Pele 1.5 MWe in fabrication, to ship to INL. Also fabricates TRISO fuel; Pele first power targeted 2028.
- NANO Nuclear Energylisted, Nasdaq: NNE
- microreactor · KRONOS 15 MWe construction permit application accepted (May 2026). Sited at the University of Illinois; NRC review expected to finish in 2027.
- Terrestrial Energylisted, Nasdaq: IMSR
- molten salt reactor · Two NRC topical reports approved (Jul 2026). Pilot reactor planned with DOE; no construction permit application yet.
- Centrus Energylisted, NYSE: LEU
- HALEU and LEU enrichment · More than 1,900 kg of HALEU produced at Piketon. Signed a DOE task order in July 2026; new commercial capacity expected by 2029.
- Fervo Energylisted, Nasdaq: FRVO
- enhanced geothermal · 33 MW block in commercial operation (30 Sep 2026). About 100 MW Phase I by early 2027; 400 MW Phase II targeted 2028.
- Bloom Energylisted, NYSE: BE
- on-site fuel cells · Framework for up to 2.8 GW with Oracle (Apr 2026). Behind-the-meter supply that avoids the queue; runs on natural gas.
- General Fusionlisted, Nasdaq: GFUZ
- magnetized target fusion · LM26 first plasma Mar 2025. Targets 10 keV and further plasma milestones by 2028.
- TerraPowerprivate
- sodium fast reactor · Kemmerer Unit 1, 345 MWe, construction permit Mar 2026. Molten salt storage can raise output to 500 MW; completion targeted 2030.
- Kairos Powerprivate
- fluoride salt-cooled reactor · Hermes 2 construction permit Nov 2024; ground broken Apr 2026. 50 MWe to TVA for Google; operation by 2030.
- Holtecprivate
- restart; SMR-300 · Palisades 800 MW; fuel reloaded Oct 2026. Holtec has not set a firm restart date; the Energy Secretary said on 29 October that restart could be weeks away. Plans two SMR-300 units at the site by 2031.
- Commonwealth Fusion Systemsprivate
- tokamak fusion · SPARC in assembly; switch-on planned for 2027. Google has agreed to buy power from its planned ARC plant in Virginia.
- Helionprivate
- pulsed fusion (field-reversed configuration) · Polaris operated on deuterium-tritium fuel (2026). Orion plant targets 50 MW to Microsoft by 2028; independent observers question the date.
- Valar Atomicsprivate
- high-temperature gas test reactor · Ward 250 zero-power criticality (Jun 2026). Test unit in Emery County, Utah, under DOE's reactor pilot program.
- Aalo Atomicsprivate
- sodium-cooled microreactor · Test reactor zero-power criticality (4 Jul 2026). Under the DOE pilot program; its Aalo-X plant reactor is in development.
- Antaresprivate
- microreactor · Mark-0 zero-power criticality at INL (Jun 2026). Zero-power demonstration under the DOE pilot program.
resources
- interconnection.fyi
- Independent tracker of US and Canadian interconnection queues and data center projects, built from grid operator filings.
- Utility Dive: data centers and the grid
- Running news coverage of large-load rules, FERC orders and grid events.
- Berkeley Lab, Queued Up
- Annual analysis and free data on queue size, completion rates and wait times.
- NRC advanced reactor projects
- Official status and schedules for each advanced reactor application.
notes
Positions are approximate and reflect editorial judgment, as of 3 November 2026. NRC permits and DOE authorizations are different pathways; test reactors at national laboratories run under DOE and do not need an NRC license. "Critical" here means zero-power criticality; none of the test units produces electricity. Queue figures come from successive Berkeley Lab editions whose coverage and methods change. Company dates are claims, not forecasts, and several have slipped before (SPARC from 2025 to 2027; Palisades from late 2025). Not investment advice.