Why AI and nuclear are suddenly talking
AI may be a digital technology, but the infrastructure behind it is one of the biggest physical energy challenges in the world. Every GPU and server rack depends on electricity, and AI campuses are now planned in hundreds of megawatts, with some approaching a gigawatt or more.
That changes how data centers are sited. Developers used to pick locations for fiber, land, taxes and customers, and assume the utility would eventually deliver whatever power they needed. That assumption no longer holds. Utilities face record interconnection requests, transmission shortages and generation gaps, while electrification and new factories add even more demand. We call this the AI Power Wall.
Nuclear power fits AI's needs unusually well. It runs around the clock, emits no carbon while operating and packs huge output into a small footprint. Small modular reactors (SMRs) push that further. They are smaller, factory-built units that can, in theory, be placed close to large loads like data centers and added in steps as demand grows.
When we first published this article in August, based in part on reporting from Inside Climate News, the headline fact was simple: no commercial SMR was operating in the United States. That's still true. But a lot has moved since then.
What an SMR actually is
The U.S. Energy Information Administration describes SMRs as reactors designed to produce roughly 70 to 350 MW each, compared with about 1,000 MW or more for a traditional large reactor. The idea is to build major components in a factory, ship them to the site and assemble them, the way airplanes are built rather than one-off construction projects.
SMR designs vary widely. Some, like GE Vernova Hitachi's BWRX-300 and Holtec's SMR-300, use proven light-water technology. Others use newer approaches: TerraPower's Natrium uses liquid sodium, Kairos uses molten salt, and X-energy uses high-temperature gas with pebble fuel. About two dozen SMR designs are in development in the U.S., according to an analysis by the Yale Clean Energy Forum. NuScale's designs are the only ones with full NRC design approval so far, though no NuScale plant is under construction in the U.S.
For data centers, the appeal is clear. A few SMR units could, in principle, supply a campus directly with firm, clean power, potentially on or near the same site.
Where U.S. SMRs actually stand
2026 has been the most active year for new U.S. nuclear in decades. In March, the NRC issued a construction permit for TerraPower's 345 MW Natrium plant at a retired coal site in Kemmerer, Wyoming, and construction began in April. Also in April, Kairos Power broke ground on its 50 MW Hermes 2 molten salt reactor in Oak Ridge, Tennessee. It plans to sell power to Google through TVA.
Then, on September 30, the NRC issued a construction permit for TVA's Clinch River SMR near Oak Ridge — the second SMR permit in six months. It's a 300 MW GE Vernova Hitachi BWRX-300, the first construction permit for that design in the U.S. The NRC finished the review in 14 months, four months ahead of schedule. Completion is targeted toward the end of 2034, and TVA says it's still negotiating cost sharing with partners.
A Third Way analysis found that only five of 11 announced commercial advanced nuclear projects had secured offtake, financing and construction contracts: TerraPower Kemmerer, TVA Clinch River, Holtec at Palisades and two X-energy projects. As Canary Media put it, America's nuclear dry spell is over, but most of these plants are still years from producing power.
U.S. Advanced Nuclear Projects to Watch (November 2026)
| Project | Developer / design | Size | Status | Target |
|---|---|---|---|---|
| Kemmerer, Wyoming | TerraPower Natrium (sodium-cooled) | 345 MW | NRC construction permit March 2026; under construction | ~2031 |
| Clinch River, Tennessee | TVA / GE Vernova Hitachi BWRX-300 | 300 MW | NRC construction permit Sept 2026 (review done 4 months early) | ~2034 |
| Hermes 2, Oak Ridge, TN | Kairos Power (molten salt) | 50 MW | Under construction since April 2026; power to Google via TVA | Late 2020s |
| Palisades, Michigan | Holtec SMR-300 (x2) | 600 MW | Awaiting full NRC approval; DOE award | TBD |
| Long Mott, Texas | X-energy Xe-100 for Dow | ~320 MW | Offtake, financing and construction contracts in place; licensing underway | Early 2030s |
| Ohio | Oklo (sodium-cooled) for Meta | 1.2 GW | Early development | 2030s |
Big Tech is already placing nuclear bets
Hyperscalers aren't waiting for SMRs to be proven. They're signing deals now to secure a place in line:
- Google has an agreement with Kairos Power for up to 500 MW of advanced nuclear by 2035, starting with Hermes 2.
- Amazon has backed X-energy, targeting 5 GW of new nuclear capacity by 2039.
- Meta is working with Oklo on about 1.2 GW of reactors in Ohio.
- Microsoft signed a power deal with Constellation to restart Crane (formerly Three Mile Island Unit 1), expected late 2027.
These deals give developers what they need most: committed buyers with deep pockets. That makes financing easier and helps pay for the first, most expensive units. In return, the tech companies lock in future firm, clean power that will be scarce in the 2030s.
The time-to-power mismatch
This is the core problem. An AI data center can be built in about two years. A new reactor, even a small one, typically takes most of a decade from permit to power. The first SMRs now in licensing or construction target the early-to-mid 2030s.
That means SMRs can't solve the power shortage AI faces today. Data centers coming online in 2027 or 2028 need electricity from sources that exist now or can be built quickly: utility power, natural gas, solar and wind with battery storage, existing nuclear plants, restarts like Crane and Palisades, and uprates that squeeze more output from existing reactors.
The fastest nuclear wins in this decade aren't SMRs at all. They're restarts and uprates of plants that already exist. Like geothermal, SMRs are a firm-power solution for the late 2020s and 2030s.
Time-to-Power: How Fast Each Resource Reaches a Data Center
Typical development timelines from project decision to first power. SMR and large nuclear new builds span most of a decade or more; data centers, renewables and existing-nuclear restarts reach power in a fraction of that time.
The cost question
Cost is the biggest open question. The last large U.S. reactors, Vogtle Units 3 and 4 in Georgia, came in years late and billions over budget. The abandoned V.C. Summer project in South Carolina left about $9 billion in losses, much of it passed on to ratepayers. Those experiences are why utilities, regulators and customers are cautious about who pays for new nuclear.
SMRs are supposed to fix this through standardization and factory production. But the first units of any new design are always the most expensive. The savings only come with repetition: building the same design again and again, with a mature supply chain and experienced crews. Until several units of one design are built, the true cost of SMR power is an estimate, not a number. You can test how capital cost and capacity factor shape the price of power with our Nuclear/SMR calculators and the LCOE Calculator.
What still has to go right
For SMRs to become a real part of AI's energy supply, several things need to happen at once:
- Licensing at speed. The NRC is moving faster, finishing Clinch River's review early, but each new design still needs approval.
- Fuel supply. Many advanced designs need HALEU (high-assay low-enriched uranium), and U.S. production is still ramping up.
- Factory manufacturing. The cost savings depend on real production lines, not one-off builds.
- Supply chains and workforce. Reactor vessels, specialty components and skilled nuclear workers are in short supply.
- Financing and risk sharing. First-of-a-kind cost overruns need a clear owner, and it shouldn't be ordinary ratepayers.
- Community acceptance. Nuclear siting still requires local trust, especially next to a data center.
Safety and security standards also can't become secondary just because tech companies urgently need power.
What this means for data center developers
For developers planning campuses today, SMRs are a long-term option, not a near-term solution. A practical strategy looks like this: secure grid power and firm generation now, add battery storage and flexibility to speed interconnection, and reserve land and transmission capacity for future firm resources like SMRs or geothermal.
Some developers are designing campuses with a future nuclear site in mind, so the data center can later shift to on-site or nearby reactors. That fits a broader trend. As we asked in "Is compute the new oil?", the companies that control reliable energy will have the advantage, and a well-designed campus can become an energy asset for its region rather than a drain on it.
Conclusion: promise and patience
SMRs may become one of the most important technologies behind the AI economy. They offer a rare mix of firm power, low emissions, small footprint and modular deployment near large loads. 2026 brought real progress: two construction permits, active construction in Wyoming and Tennessee, and billions in hyperscaler commitments.
But the gap between potential and commercial reality is still wide. No commercial SMR runs in the U.S. today, first-of-a-kind costs are uncertain, and most projects target the 2030s. SMRs won't solve AI's immediate power crunch. The near term belongs to the grid, gas, renewables, storage and existing nuclear.
The bigger question is what happens after the first SMRs are built. If developers move from one-off projects to standardized fleets, the relationship between computing and electricity could change dramatically. Data centers could become anchor customers for fleets of reactors, and new AI energy campuses could combine nuclear, renewables, storage and grid infrastructure. AI is turning tech companies into energy companies. If SMRs cross the commercialization gap, advanced nuclear could become a foundation of AI's energy system in the 2030s and beyond.
This article was originally inspired by Inside Climate News reporting on SMRs and data centers (August 30, 2026). Updated November 2026 with NRC construction permits for TerraPower Natrium and TVA Clinch River, Kairos Hermes 2 construction, and analysis from Canary Media, World Nuclear News and Industrial Info.
Chris Kalowes
Founder, WattThe?!
Energy intelligence. Simplified.
