Short answer: Nuclear power has become a serious strategic option for AI data centers, but the announced deals do not mean nuclear will soon supply most of their electricity. The strongest near-term cases involve existing reactors or a restart. New small modular reactors (SMRs) remain dependent on licensing, financing, fuel, construction and grid approvals.
Microsoft, Amazon, Google and Meta are using long-term contracts and investments to secure firm, low-carbon power. Those commitments can help projects get financed, but announced megawatts are not the same as operating generation.
Why AI companies are pursuing nuclear power
AI creates a large, continuous electricity load
Training and inference facilities consume substantial power and generally need dependable electricity around the clock. That load profile fits nuclear plants, which are designed to produce steady output. The U.S. Energy Information Administration explains the alignment between data centers and nuclear generation at EIA.
Grid capacity is scarce
Hyperscalers need more than cheap energy. They need large blocks of firm capacity, transmission access, suitable land, cooling and water infrastructure, and a predictable delivery schedule. A direct contract with a generator can provide more certainty than relying only on a congested utility interconnection queue. That is an inference from the announced deals and the grid constraints described by the Department of Energy and Carnegie Endowment.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
Carbon accounting matters
Reactors produce electricity without direct operational carbon-dioxide emissions. That can help companies pursue carbon-free-energy commitments for constant loads that are difficult to match entirely with hourly wind and solar output. But carbon-free is not the same as renewable, and a power-purchase agreement (PPA) does not necessarily mean a particular data center is physically connected to a particular reactor.
The major deals, and what they actually deliver
| Company | Partner and project | Announced scale | Starting point | Timing or status | Main uncertainty |
|---|---|---|---|---|---|
| Microsoft | Constellation, Crane Clean Energy Center (formerly Three Mile Island Unit 1) | About 835 MW reactor; 20-year PPA | Existing reactor being restarted | Expected return in 2028, subject to Nuclear Regulatory Commission approval | Refurbishment, licensing and schedule |
| Amazon Web Services | Talen, Susquehanna station | Up to 960 MW | Operating nuclear station and proposed co-located campus | Contract announced; proposed interconnection arrangement rejected by FERC in November 2024 | Transmission allocation, grid-cost and market rules |
| Kairos Power advanced reactors | Up to 500 MW | Future fleet of advanced reactors | Future deployment; not operating capacity | Licensing, manufacturing, fuel and construction | |
| Meta | TerraPower, Oklo and Vistra projects | Up to 6.6 GW by 2035 | Portfolio of existing and advanced nuclear pathways | Announced agreements, not delivered capacity | Project-by-project permitting, financing and execution |
Microsoft–Constellation: the clearest restart case
Microsoft announced a 20-year PPA with Constellation in September 2024 tied to restarting Pennsylvania’s Three Mile Island Unit 1, renamed the Crane Clean Energy Center. Unit 1 was not the adjacent reactor involved in the 1979 partial meltdown; it shut in 2019 for economic reasons. Constellation describes the project at its announcement.
The reactor is rated at approximately 835 MW. The Department of Energy says it is expected to return to service in 2028, pending regulatory approval, and that a $1 billion DOE loan closed in November 2025. See DOE’s timing summary and the financing announcement.
This is more tangible than a promise to build an unlicensed reactor, but it is not operating capacity today. Public descriptions refer to regional clean-energy matching for Microsoft’s PJM electricity use, not a dedicated wire proving that an individual data center runs directly on Crane output.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
AWS–Talen: existing generation and a grid-policy fight
AWS agreed to purchase up to 960 MW from Talen Energy’s Susquehanna nuclear station. DOE describes a related $650 million transaction involving a co-located data-center campus. The arrangement raised questions about “behind-the-meter” service, transmission charges and whether a large customer could receive preferential access to plant output.
FERC rejected Talen’s proposed interconnection arrangement in November 2024, according to EIA, DOE’s overview at energy.gov, and Carnegie’s analysis at Carnegie Endowment. The case shows that nuclear procurement can affect ratepayers, wholesale markets and transmission policy—not just corporate emissions reporting.
Google–Kairos: an anchor customer for future reactors
Google agreed to purchase up to 500 MW from multiple future Kairos advanced reactors. Google calls it the first corporate agreement to purchase nuclear energy from a fleet of SMRs and says the objective is 24/7 carbon-free electricity for its data centers and offices. The agreement is intended to support commercialization; it does not make 500 MW available now. Details are in Google’s announcement.
Meta’s multi-project portfolio
Meta announced agreements with TerraPower, Oklo and Vistra that could support up to 6.6 GW of nuclear capacity by 2035. The portfolio combines existing plants, advanced designs, grid support and fuel-supply objectives. The 6.6-GW figure is a potential project total, not delivered generation. Meta’s announcement is at about.fb.com.
Rank #3
Amazon’s advanced-reactor strategy
Alongside the Talen procurement, Amazon has supported advanced-reactor development, including X-energy. DOE says widespread commercial availability of advanced reactors is more likely in the 2030s than immediately. An investment or support agreement does not by itself guarantee a commercial reactor on a particular date.
Existing reactors and restarts versus new SMRs
The starting point determines much of the risk:
- Operating reactors: established equipment and licensing, but disputes can arise over market allocation, transmission and customer priority.
- Restarts: an existing site and infrastructure can shorten the path relative to a greenfield project, but inspections, refurbishment and NRC approval remain necessary.
- Large new reactors: based on proven technology, yet exposed to high capital requirements, long construction periods and financing risk.
- SMRs and microreactors: promise standardized production and smaller increments, but first-of-a-kind designs still face licensing, manufacturing, fuel and construction risk.
DOE’s assessment at energy.gov places widespread commercial advanced-reactor deployment mainly in the 2030s. That makes existing plants and restarts the more credible bridge for the current AI buildout.
What a nuclear deal can mean in practice
“Nuclear-powered data center” is not a single legal or physical arrangement. It may mean:
- A campus physically next to a reactor.
- A physical PPA for electricity delivered through the regional grid.
- A utility allocation of nuclear output to a large customer.
- A virtual PPA or purchase of environmental attributes.
- A plan to use a future SMR.
- An investment in a reactor developer without a firm electricity-delivery obligation.
Annual clean-energy matching can reduce a company’s reported net emissions while the data center receives electricity from the wider grid. Hourly matching is more demanding. Neither arrangement proves that a specific reactor is physically supplying a specific server hall at every moment.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsRank #4
How to judge whether a project is genuinely advanced
- Check for a binding contract rather than a memorandum or aspiration.
- Identify whether the generator is operating, restarting, under construction or only proposed.
- Confirm whether the design and site have the required regulatory approvals.
- Look for closed construction financing, not merely a target capital raise.
- Check fuel availability, especially high-assay low-enriched uranium (HALEU) for many advanced designs.
- Verify the transmission and interconnection plan.
- Distinguish a commercial-operation date from a target year.
- Ask who bears cost overruns, delays and replacement-power costs.
- Determine whether the contract provides physical electricity, financial attributes, or both.
- Ask whether the deal adds new generation or reallocates power that already served other customers.
The benefits and limits for hyperscalers
Why nuclear is attractive
- Firm, around-the-clock output at large scale.
- Low direct operational carbon emissions.
- Long-lived assets; DOE says nuclear plants can potentially operate for 80 years or more.
- Potentially more predictable long-term pricing through negotiated PPAs.
- A way to preserve or restart existing plants.
- Support for domestic reactor manufacturing and fuel supply.
What nuclear does not solve
- New construction remains slow and capital-intensive.
- Advanced reactors are not yet a mature, mass-produced product category.
- HALEU and other specialized fuel supply chains are still developing.
- Plants still require transmission, substations, cooling, backup systems and grid balancing.
- Spent fuel remains stored at existing sites while broader disposal pathways develop.
- Water use, land impacts, emergency planning and public acceptance remain material issues.
- Long-term commitments can become expensive if AI demand, chip efficiency or facility plans change.
The ratepayer and grid question
The central policy dispute is not simply nuclear versus renewables. It is whether the largest electricity customers should receive special access to generation and transmission, and who pays for the infrastructure they use.
A co-located campus may reduce some transmission needs, but it can also remove plant output from wholesale markets or require new grid arrangements. Regulators therefore have to consider reliability, cost allocation, congestion and whether smaller customers subsidize service for data centers.
What supplies AI power before new reactors arrive?
DOE expects near-term data-center electricity to come from a mix of existing nuclear, natural gas, coal, wind, solar and other resources. Grid expansion, batteries, efficiency improvements, demand flexibility, uprates and geographically distributed data centers can also contribute. Nuclear may become a larger part of the portfolio without becoming the sole or dominant source.
Demand forecasts are uncertain. Buildout speed, GPU utilization, model efficiency, cooling design, on-site generation and project cancellations can all change the amount of electricity ultimately required. EIA discusses those uncertainties at eia.gov.
Best Value
What this means for investors and infrastructure buyers
For enterprise buyers, a nuclear PPA is a bespoke procurement contract—not a retail electricity plan. Terms can include capacity rights, fixed or escalated prices, environmental attributes, delivery conditions and replacement-power provisions. The reported $650 million Talen-related transaction and $1 billion DOE loan supporting Crane are transaction or financing figures, not comparable tariffs for nuclear electricity.
Potential counterparties include Constellation Energy and Talen Energy for generator-linked procurement, and Kairos Power, TerraPower and Oklo for future advanced-reactor pathways. None should be treated as an immediately available retail product.
Verdict: important option, not inevitability
“Nuclear is inevitable” overstates what the contracts prove. “The deals are just publicity” understates their importance. Hyperscalers are becoming anchor customers and, in some cases, financiers for firm, low-carbon electricity. That can materially improve the prospects for reactor restarts and future projects.
The decisive evidence will be practical: restart approvals, construction permits, closed financing, available fuel, interconnection decisions and megawatts actually entering service. Until those milestones arrive, the announced gigawatts are best understood as strategic options on future power—not as electricity already supplying the AI boom.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
