| Myth | Reality |
|---|---|
| Nuclear power for AI data centers is still a distant, speculative idea | Big tech has already committed to over 10 GW of new nuclear capacity, and multiple SMR projects have secured construction permits or are already under construction |
| SMRs will be operational fast enough to solve today’s power shortages | Even the fastest-moving projects are targeting operational dates years out; SMRs address medium-term capacity, not this year’s grid interconnection queue |
| All nuclear-for-AI deals are the same kind of commitment | They range from direct reactor construction (TerraPower, Oklo) to large capacity-offtake agreements (AWS, Switch) with very different risk and timeline profiles |
Small modular reactors entered 2026 having crossed from concept to construction. China’s Linglong One is on track to become the world’s first land-based commercial SMR, with commercial operation expected in the first half of 2026. In North America, the BWRX-300 at Ontario’s Darlington site became the first SMR actually under construction, and TerraPower’s Natrium plant in Kemmerer, Wyoming secured its Nuclear Regulatory Commission construction permit. None of this happened because of general decarbonization policy — the most significant driver reshaping the SMR industry in 2026 is the explosive, immediate power demand of hyperscale AI data centers, which utilities and traditional grid buildout simply cannot satisfy on the timeline AI capacity requires.
Who is actually committing capital, and to what
Big tech has collectively committed to more than 10 gigawatts of new nuclear capacity, and the specific deals vary considerably in structure. AWS is investing $20 billion in Pennsylvania, exploring new small modular reactors within Talen Energy’s existing nuclear facilities, with plans to be fully operational by spring 2026. Oklo has built a substantial project pipeline that includes a 1.2 gigawatt project in Ohio, a master agreement with Switch for up to 12 gigawatts of capacity, and a 500 megawatt deal with Equinix — backed by more than $10 billion in combined investment across its projects.
| Project / company | Structure | Status / timeline |
|---|---|---|
| China’s Linglong One | State-backed commercial SMR | Commercial operation targeted H1 2026 — first land-based commercial SMR globally |
| BWRX-300 (Darlington, Ontario) | Utility-led construction | First SMR actually under construction in North America |
| TerraPower Natrium (Kemmerer, Wyoming) | Independent developer, NRC-permitted | Construction permit secured; multi-year build ahead |
| AWS + Talen (Pennsylvania) | Hyperscaler investment in existing nuclear facility SMRs | $20B investment; targeting spring 2026 for related capacity |
| Oklo (Ohio, Switch, Equinix deals) | Independent developer with hyperscaler/colo offtake agreements | 1.2 GW Ohio project, up to 12 GW Switch master agreement, 500 MW Equinix deal |
Why SMRs specifically, not just “more nuclear”
SMRs are compact reactors producing up to roughly 300 megawatts of power, designed to be scalable and factory-buildable rather than constructed entirely on-site the way traditional large reactors are. They require far less land — on the order of 50 acres — than a conventional nuclear plant, and critically for a data center operator, they offer uninterrupted baseload power, unlike wind or solar, which is exactly the profile an always-on AI training or inference cluster needs. That combination — smaller footprint, faster (if still multi-year) build timeline, and baseload reliability — is why SMRs specifically, rather than large conventional reactors, have become the preferred nuclear structure for data center power deals.
- 1. Grid interconnection hits its limitPublic grid capacity additions face multi-year regulatory and construction delays, unable to match AI power demand growth.
- 2. Hyperscalers seek dedicated baseload powerAI clusters need continuous, high-reliability power that intermittent renewables can’t guarantee alone.
- 3. SMRs offer a middle pathSmaller footprint and more standardized construction than conventional nuclear, still years faster to site than waiting in the grid queue.
- 4. Capital commits at scaleOver 10 GW in hyperscaler-backed nuclear commitments materialize through direct investment and long-term offtake agreements.
Why SMRs became the nuclear structure of choice for AI power deals
This is a medium-term solution — even the fastest SMR projects operate on multi-year timelines, not a fix for this year’s interconnection queue.
The honest timeline problem
The critical caveat that gets lost in announcement headlines: even the most advanced SMR projects are years from delivering power at scale. Linglong One’s H1 2026 target makes it the first mover globally, but a single first-of-a-kind reactor doesn’t solve an industry-wide power constraint on its own — the BWRX-300 at Darlington and TerraPower’s Natrium plant are both still under construction, meaning the bulk of committed nuclear capacity for AI data centers is a multi-year bet rather than near-term relief. Anyone reading a 10+ gigawatt nuclear commitment headline should understand that as capital committed to a multi-year buildout, not megawatts available today.
Commit to SMR capacity for medium-term buildout What hyperscalers are doing
Right for a company planning data center capacity 3-7 years out, where SMR construction timelines and grid interconnection delays are comparable and dedicated baseload power justifies the capital commitment.
Treat SMR announcements as future capacity, not current relief Necessary framing
A 10+ GW commitment headline reflects multi-year capital deployment, not megawatts available to solve this year’s grid interconnection queue.
Assume any single SMR project de-risks the whole industry’s power problem Overreach
Even a genuine first-mover success like Linglong One is one reactor; industry-wide power constraints require many more projects reaching the same milestone.
Common mistakes
Common mistake
Reading a hyperscaler’s nuclear capacity commitment as equivalent to available power today. Deals like AWS’s $20 billion Pennsylvania investment or Oklo’s multi-gigawatt agreements represent multi-year capital deployment, not power that exists on the grid right now.
Common mistake
Conflating different deal structures as equally reliable. A direct reactor construction project (TerraPower, Oklo’s own builds) carries different execution risk than a capacity offtake agreement with a third-party developer (Switch, Equinix deals) — the headline gigawatt figure doesn’t distinguish between them.
What worked
Operators planning realistic capacity timelines treated SMR commitments as one part of a diversified power strategy alongside grid interconnection requests and on-site generation, rather than betting a single capacity plan entirely on a specific reactor’s construction timeline holding.
Glossary
- Small modular reactor (SMR)
- A compact nuclear reactor design producing up to roughly 300 megawatts, built with more standardized, factory-based construction methods than conventional large reactors.
- Baseload power
- Continuous, always-available electricity generation, as distinct from intermittent sources like wind or solar — the profile SMRs are valued for in powering always-on AI clusters.
- Capacity offtake agreement
- A contract where a buyer (such as a hyperscaler or colocation provider) commits to purchasing power from a reactor project a third-party developer builds and operates, as opposed to building the reactor directly.
- NRC construction permit
- Regulatory approval from the US Nuclear Regulatory Commission required before a nuclear reactor project can begin construction — a key milestone TerraPower’s Natrium plant secured.
Key takeaways
- Big tech has collectively committed to over 10 gigawatts of new nuclear capacity, driven primarily by hyperscale AI data center power demand rather than general decarbonization policy.
- China’s Linglong One is on track to become the world’s first land-based commercial SMR, targeting commercial operation in the first half of 2026.
- The BWRX-300 at Ontario’s Darlington site became the first SMR under construction in North America, and TerraPower’s Natrium plant secured its NRC construction permit.
- AWS is investing $20 billion in Pennsylvania exploring SMRs within Talen Energy’s existing nuclear facilities, while Oklo has built a multi-gigawatt pipeline including deals with Switch and Equinix.
- SMRs offer a smaller land footprint (roughly 50 acres), more standardized construction, and reliable baseload power compared to intermittent renewables — the combination driving hyperscaler preference for this reactor type.
- Even the most advanced projects remain years from delivering power at meaningful scale, meaning multi-gigawatt commitment headlines represent future capital deployment, not near-term relief for today’s grid constraints.
- Deal structures vary significantly in risk — direct reactor construction differs meaningfully from capacity offtake agreements with third-party developers, even when reported under the same headline gigawatt figure.
FAQs
Are hyperscalers actually building nuclear power plants for AI data centers?
Big tech has collectively committed to over 10 gigawatts of new nuclear capacity through a mix of direct investment and long-term offtake agreements, though most projects are still under construction or in early development rather than delivering power today.
What is a small modular reactor and why is it preferred over conventional nuclear?
An SMR is a compact reactor producing up to roughly 300 megawatts, built with more standardized, factory-based methods, requiring far less land (roughly 50 acres) than a conventional plant, while still offering the continuous baseload power that intermittent renewables can’t guarantee.
Which SMR project is furthest along in 2026?
China’s Linglong One is on track to become the world’s first land-based commercial SMR, targeting commercial operation in the first half of 2026, while North America’s BWRX-300 at Darlington became the first SMR actually under construction there.
How much is AWS investing in nuclear power for data centers?
AWS is investing $20 billion in Pennsylvania, exploring building new small modular reactors within Talen Energy’s existing nuclear facilities, with related capacity targeted to be fully operational by spring 2026.
Will nuclear power solve today’s AI data center power shortages?
Not immediately. Even the most advanced SMR projects operate on multi-year construction timelines, so current nuclear commitments represent medium-term capacity additions rather than a fix for today’s grid interconnection backlog.
What is Oklo’s nuclear power pipeline for data centers?
Oklo has built a pipeline including a 1.2 gigawatt project in Ohio, a master agreement with Switch for up to 12 gigawatts of capacity, and a 500 megawatt deal with Equinix, backed by more than $10 billion in combined investment.
Are all nuclear-for-AI power deals structured the same way?
No. Some involve direct reactor construction by the developer (TerraPower, Oklo’s own projects), while others are capacity offtake agreements where a buyer commits to purchasing power from a third party’s reactor — these carry meaningfully different execution risk despite similar headline figures.
What regulatory milestone did TerraPower’s Natrium plant reach?
It secured its Nuclear Regulatory Commission construction permit for the Kemmerer, Wyoming site, a key step before construction can proceed, though the plant remains years from operation.
References
- SMR Intel — “State of Small Modular Reactors 2026 — Annual Intelligence Report”
- iRecruit — “SMR Data Centers: How Small Modular Reactors Power AI Infrastructure (2026)”
- IDTechEx — “Nuclear Small Modular Reactors (SMRs) Market 2026-2046”
Related reading on The Tech Trends: Data Center Power Constraints, Heat Reuse, AI Data Center Buildout, and Sovereign AI Compute.
