Skip to main content
CallMePower

The US nuclear comeback is being underwritten by AI data centers, not by climate policy.

By Sasha Updated 9 min read

The US operates roughly 95 GW of nuclear across 54 plants and 31 states, about 19% of total electricity (NRC). Vogtle Unit 4 in Georgia came online 29 April 2024, the first new US reactor in decades. Microsoft signed a deal to restart Three Mile Island Unit 1 (Sept 2024); Amazon contracted Talen Energy\'s Susquehanna nuclear output; Google partnered with Kairos Power on advanced reactors. The Section 45U production credit (IRA 2022) backstops existing reactors at up to $15/MWh. The non-obvious truth: this renaissance is being pulled by AI data-center load, not pushed by climate policy. SMRs remain 5 to 10 years from material grid contribution despite the hype.

~95 GW
Operating US nuclear
19%
Share of US electricity
$15
IRS 45U credit, $/MWh max
Apr 2024
Vogtle 4 online

47 years of US nuclear, from TMI to data centers

Pick a year. See the accident, the policy, or the corporate deal that defines US nuclear right now.

Timeline /

Sources: NRC operating reactor list; Georgia Power Vogtle 4 release (29 April 2024); Constellation Energy news (Sept 2024); Talen Energy news (March 2024); Kairos Power - Google PPA (Oct 2024); IRS Section 45U; TerraPower Natrium project page. Verified May 2026.

Common misconception

"Climate policy is driving the nuclear comeback." Not really.

The common framing of US nuclear in 2026 reads like this: the climate crisis is forcing policymakers to revisit nuclear, the IRS Section 45U credit + EPA GHG rule + bipartisan Advance Act are the policy stack, and a wave of SMRs will arrive shortly to decarbonise the grid. Every clause in that sentence has a kernel of truth. The composite story is wrong about what is actually driving the demand.

The pull is data centers, specifically AI training and inference workloads at hyperscalers. Microsoft, Amazon, Google and Meta need 24/7 carbon-free power to meet their own emissions commitments while powering compute that runs flat-out. Wholesale electricity markets cannot reliably deliver that combination: spot prices are too volatile, capacity supply is too tight, and the wind + solar + battery stack is not a 24/7 product without overbuild. So hyperscalers are signing direct nuclear PPAs at premium prices, often 30 to 50% above wholesale strip prices, to lock in dedicated reactor output.

The IRS Section 45U credit matters too, but it is a backstop for the existing fleet (Diablo Canyon, Palisades restart, the Vermont Yankee-class plants), not the engine of new build. Vogtle Units 3 and 4 are the only new US reactors built this century, and the project ran over budget by roughly 2.5x with a 7-year delay. The next wave of greenfield reactors is being financed off corporate offtake contracts, not federal subsidy alone.

Read the rest of this page for the actual state of play: the operating fleet, the restart projects, the SMR pipeline, and the corporate deals reshaping the merchant nuclear market.

The fleet

What the US operating fleet looks like in 2026.

Three facts cover the operating base. Source: NRC operating reactor units list and EIA nuclear industry overview, May 2026.

1

Size and share

~95 GW of operating capacity across 54 plants in 31 states. ~19% of total US electricity generation. Largest single-source baseload supplier of carbon-free power; second-largest carbon-free source overall after the wind+solar+hydro stack combined.

2

Vintage

Almost all 1970s and 1980s commissioning. Most have NRC licence extensions to 60 years, many in subsequent extension to 80 years. Vogtle 3 and 4 (Georgia, 2023 and 2024) are the only US reactors commissioned this century. Median fleet age in 2026: roughly 42 years.

3

Geography

Concentrated east of the Mississippi: Illinois (11 reactors), Pennsylvania (5), South Carolina (6), Georgia (4 after Vogtle 3+4), New York (3 after Indian Point closure). Western fleet much smaller: Diablo Canyon (CA, 2 units), Columbia (WA, 1), Palo Verde (AZ, 3). Texas has 2 sites (Comanche Peak, South Texas Project).

The detail that surprises most readers. The US nuclear fleet shrank during the 2010s as roughly 13 GW of merchant reactors closed (Indian Point, Vermont Yankee, Three Mile Island Unit 1, Palisades, Duane Arnold, others) because wholesale prices fell below operating cost. The IRS Section 45U credit, created in IRA 2022 and effective from 2024, was the policy backstop that stopped the bleed. Without 45U, several more reactors would have retired by 2025.

The pipeline

Nuclear technology pipeline, status and offtake in 2026.

For each technology track: where the design sits, first commercial deployment estimate, who is buying. Sources: NRC SMR design approval, TerraPower Natrium, Kairos-Google.

US nuclear technology pipeline status and corporate offtake in 2026
Technology / project Status May 2026 First commercial deployment Who is buying
AP1000 (Vogtle 3+4) Operating, 2,200 MW combined 2023 (Unit 3), April 2024 (Unit 4) Georgia Power + Southern Co. ratepayers
TMI Unit 1 restart PPA signed Sept 2024, restart in progress Targeted 2028 Microsoft (Constellation owner)
Palisades MI restart Holtec, DOE loan-backed Targeted 2025-26 MI municipal aggregator + merchant
Susquehanna PA (Talen) Operating, direct-PPA to AWS data campus PPA from 2024 Amazon Web Services
NuScale 50 MW SMR Design approved Jan 2023; UAMPS cancelled Nov 2023 No US deployment scheduled Romania (Doicești) is first global deployment
TerraPower Natrium Under construction in Kemmerer WY Targeted 2030 PacifiCorp + DOE ARDP cost share
Kairos Power KP-FHR Demonstration unit at ORNL; commercial design pending Targeted 2030-2035 Google (500 MW master agreement)
X-energy Xe-100 DOE ARDP; Dow Seadrift TX site selected Targeted late 2020s Amazon investment, Dow industrial offtake

! What "first commercial deployment" means

Dates are sponsor-reported targets, not commitments. The Vogtle 3+4 experience (~2.5x cost overrun, 7 years late) is a reminder that nuclear schedules slip. Apply a 2 to 5 year buffer to any SMR target before treating it as a planning assumption. Restart projects (TMI, Palisades) carry execution risk too, although less than greenfield builds.

The demand pull

Why hyperscalers, not utilities, restarted US nuclear.

Four structural reasons explain why every major 2024 to 2026 nuclear deal carries a tech-company name on the offtake.

A AI training is 24/7 baseload

A frontier model training run is a months-long compute job that cannot pause when wind drops or solar sets. Inference is even more continuous. Hyperscalers need a power profile that wind + solar + battery cannot economically deliver: round-the-clock carbon-free output at a known price for 10+ years. Nuclear is the only mature technology that matches that brief.

B Wholesale markets cannot underwrite the offtake

In PJM and ERCOT spot prices swing from $20 to $5,000/MWh inside a day. No bank will finance a $10 billion reactor based on merchant spot revenue. Direct corporate PPAs at premium fixed prices (often 30 to 50% above wholesale strip) bypass the spot market entirely and produce bankable cash flows. This is why the deals exist outside the regulated rate base.

C Scope-2 emissions accounting rewards 24/7 matching

Google, Microsoft and Meta have shifted from annual REC matching to 24/7 carbon-free energy matching, where every hour of consumption must be matched with carbon-free supply in the same grid region. Solar + battery hedges a few hours; nuclear hedges all of them. The accounting change is one of the under-discussed reasons the deals make corporate sense.

D Utility load growth is below the deal rate

Traditional utility load growth in most US service territories runs 1 to 2% per year. Data center growth in PJM\'s northern Virginia, in ERCOT\'s greater Dallas area, and around Phoenix is double-digit. The marginal MW of new nuclear is being underwritten by a customer whose load is materially larger than the local utility could finance alone.

The corollary: the nuclear comeback is largely orthogonal to climate policy. The 2024 EPA GHG rule helps; the IRS 45U credit helps; the Advance Act streamlines NRC licensing. But the bid that finances the reactor sits in Redmond, Seattle, Mountain View and Menlo Park.

2024 inflection

The four events that flipped US nuclear from contraction to expansion.

2024 to 2025 was the year the trajectory changed. These are the four data points that mark the turn.

29 Apr 2024

Vogtle 4 commercial operation

First new US reactor in decades. Joins Vogtle 3 (July 2023). Combined ~2,200 MW serving Georgia Power + Southern Co. ratepayers.

Sept 2024

Microsoft - TMI restart deal

Constellation announces restart of TMI Unit 1 (renamed Crane Clean Energy Center) under 20-year PPA with Microsoft. Targeted 2028.

Mar 2024

Amazon - Talen Susquehanna

AWS buys the Cumulus data campus adjacent to Talen\'s Susquehanna nuclear plant; PPA effectively powers AI compute with onsite reactor output.

Oct 2024

Google - Kairos master PPA

500 MW master agreement for Kairos Power\'s KP-FHR reactors. First non-utility commercial advanced-reactor PPA at scale.

Three reads on the new numbers

  • A The market structure is bifurcating. Existing fleet sells into wholesale + 45U; new builds and restarts sell direct to corporate offtakers via PPA. The regulated rate base is not the dominant nuclear capital-allocation channel any more, and that has implications for who pays and who benefits.
  • B SMRs are still 5 to 10 years from material grid contribution. NuScale\'s UAMPS cancellation showed that customer aggregation for first-of-a-kind units is hard. TerraPower\'s Natrium project is the closest to commercial deployment and is still targeted for 2030. Treat 2030-2035 as the realistic SMR onset window.
  • C The IRS 45U credit is the load-bearing existing-fleet policy. Without it, several merchant reactors would have retired by 2025. The credit expires after 2032; whether it renews will shape the late-2020s decisions on subsequent licence extensions at Diablo Canyon, Palo Verde and the Pennsylvania merchant fleet.
Insider view

Four reasons the nuclear comeback does not look like a renaissance from inside the industry.

The headlines say "renaissance". The people running the projects describe a different texture: real expansion, narrowly distributed, with execution risk that should temper the optimism.

01

Most of the expansion is restarts, not greenfield

Vogtle 3+4 is the only truly new US reactor build, and it cost ~$35B against a $14B plan. TMI restart, Palisades restart, possibly Duane Arnold are bringing back retired units that already have NRC licences and existing site infrastructure. Restart is faster and cheaper, but it is a one-time inventory: the US closed ~13 GW of merchant nuclear in the 2010s and that is the ceiling.

02

SMR economics still need a first-of-a-kind buyer

NuScale\'s UAMPS cancellation in November 2023 was the visible failure. The underlying issue: a first-of-a-kind SMR costs 2 to 3x what an Nth-of-a-kind unit will cost, and no rational utility wants to be customer #1 at the top of that cost curve. TerraPower\'s Natrium has DOE ARDP cost-share carrying part of the FOAK premium. Without similar federal cost-share or hyperscaler patience, the SMR pipeline stalls.

03

The grid benefits skip the ratepayer

Direct corporate PPAs at TMI, Susquehanna and forthcoming Kairos units dedicate output to the offtaker, not to the merchant market. The reactor exists, the reactor is carbon-free, the reactor generates jobs and tax revenue. But the kilowatt-hours do not flow into the wholesale pool to depress prices for residential customers. The benefit is private; the public-good case is real but secondary.

04

Workforce and supply chain are the binding constraint

The US lost most of its heavy nuclear fabrication base during the 1980s pause. Pressure vessels, large forgings, reactor coolant pumps, fuel fabrication capacity: many components now come from South Korea, Japan or France. Welder and licensed-operator pipelines are similarly thin. Even with finance and customers, a build-out faster than 3 to 5 GW per decade strains the supply chain.

The right read: the US is in a real nuclear expansion, financed by tech-company demand and backstopped by the 45U credit, but the pace is constrained by execution. Public framing as a "renaissance" is forward-looking; the working numbers say steady growth, not a step change.

Your move

Six ways to use this knowledge.

1

Identify if your utility runs nuclear

Use the NRC operating reactor list to see which plants serve your area. Reactors selling into your ISO (PJM, MISO, NYISO) feed your wholesale supply. Reactors on direct corporate PPAs (TMI, Susquehanna) do not.

2

Track restart timelines

TMI Unit 1 targeted 2028. Palisades MI 2025-26. Both face execution risk; treat the published dates as plus 2 years. The actual restart announcements will appear in NRC operating-licence filings.

3

Discount SMR timelines

Apply a 2 to 5 year buffer to any SMR target. TerraPower\'s 2030 Natrium date is the most credible near-term commercial deployment, and even that may slip. Treat 2030-2035 as the realistic onset window for material SMR grid contribution.

4

Watch the 45U expiration

The Section 45U credit expires after 2032 unless extended. Decisions in the late 2020s on Diablo Canyon, Palo Verde and PA merchant fleet licence extensions will be driven by whether 45U continues. Watch IRS guidance.

5

Read the offtake, not the headline

When a nuclear deal is announced, the load-bearing question is who buys the output. Corporate PPA = dedicated to one offtaker. Merchant + 45U = sold into wholesale. Regulated cost-recovery = bill impact for ratepayers in the relevant state.

6

Anchor your own price expectation in EIA data

The US average residential rate in March 2026 was 18.83 cents/kWh (EIA). Nuclear additions of 2 to 5 GW per decade will not measurably change that number; capacity-market dynamics in PJM and ERCOT will. Use the EIA series as your benchmark.

FAQ

Common questions about the US nuclear renaissance.

Around 19% of US electricity generation, from approximately 95 GW of operating capacity across 54 plants in 31 states (NRC operating reactor list). Nuclear remains the second-largest source of carbon-free electricity in the US behind hydro+wind+solar combined, and the largest single-source baseload supplier of clean power.

Vogtle Unit 4 in eastern Georgia, owned by Georgia Power and Westinghouse AP1000 design. It entered commercial operation on 29 April 2024, joining Vogtle Unit 3 which started in July 2023. Combined Vogtle 3+4 output is roughly 2,200 MW (Georgia Power, April 2024). The project cost approximately $35 billion against an initial budget of $14 billion. They are the only new US reactors built this century.

AI training and inference run 24/7 and require carbon-free baseload power, two things that wholesale electricity markets do not reliably provide. Hyperscalers (Microsoft, Amazon, Google, Meta) want to bypass the spot market and lock in dedicated nuclear output at premium prices, often 30 to 50% above wholesale strip prices. Key 2024 deals: Microsoft - Constellation to restart Three Mile Island Unit 1 by 2028 (Constellation Energy news); Amazon - Talen Energy at Susquehanna PA (Talen Energy news); Google - Kairos Power for advanced reactors (Kairos Power news).

An SMR is a reactor with output below 300 MW per unit, designed to be factory-built and shipped to site. The NRC approved NuScale's 50 MW design in January 2023, the first US SMR design certification ever. However the flagship NuScale customer project (UAMPS Carbon-Free Power Project in Idaho) was cancelled in November 2023 when customer subscriptions fell short of breakeven cost. TerraPower's 345 MW Natrium reactor is under construction in Kemmerer WY targeting late 2020s (TerraPower). Kairos Power demonstration units are years out. Realistic timeline for material SMR grid contribution: 2030 to 2035.

The Section 45U Zero-Emission Nuclear Power Production Credit, created by the Inflation Reduction Act of 2022, pays existing US reactors up to $15/MWh of production, starting 1 January 2024 and running through 2032. It is sized to keep marginally economic reactors (Diablo Canyon CA, Palisades MI restart, Three Mile Island restart, others) from premature retirement. Source: IRS Section 45U credit. The credit phases down as wholesale revenue rises; it is structured as a backstop, not a subsidy floor.

Yes and no. Yes: Vogtle 4 is the first new reactor in decades; restart projects at TMI, Palisades and possibly Duane Arnold are underway; hyperscaler PPAs are real; the IRS 45U credit removed the economic cliff for existing fleet; bipartisan policy support is the strongest since the 1970s. No: capacity expansion is small in absolute terms relative to wind + solar + batteries; SMRs are 5 to 10 years from material contribution; the cost overruns at Vogtle (2.5x budget, 7 years late) make new builds hard to underwrite; the entire pipeline is anchored on data-center demand, not utility customer load growth.

The EPA's 2024 Greenhouse Gas Rule for Power Plants targets coal plants (90% CCS or shutdown by 2039) and new baseload gas. Nuclear is the obvious low-carbon baseload replacement and the rule indirectly raises the value of every existing nuclear MW. As of May 2026 the rule's status under the second Trump administration is uncertain, but the economics for nuclear are now driven by tech-company PPAs rather than EPA enforcement.

Indirectly. Most new nuclear MW going on in 2026 to 2030 (Vogtle 3+4 + restarts) is selling to specific corporate offtakers, not into the merchant pool, so it does not directly suppress wholesale prices in your ISO. But by absorbing tech-company load that would otherwise have pulled on wholesale supply, it relieves capacity pressure indirectly. The US average residential rate in March 2026 was 18.83 cents/kWh (EIA); nuclear additions are not a major short-term lever on that number compared to capacity-market dynamics in PJM and ERCOT.

Article reviewed by Cornelia Zavoianu, Selectra energy expert

Written by

Sasha