North American data centers procured nearly 6 GW of nuclear capacity in 2025, followed by Meta’s January 2026 announcement of up to 6.6 GW of nuclear PPAs with TerraPower and other partners [S1][S3].
The 2026 deal set spans operating reactors, uprates, restarts, small modular reactors, and advanced reactors, serving hyperscale AI campuses with firm, carbon-free electricity through contracts extending toward 2035 and beyond [S1][S3][S4].
What nuclear PPAs changed in 2026
A nuclear PPA is a long-term electricity purchase agreement between a data center buyer and a generator or reactor developer, covering output from existing or planned facilities [S2]. In 2025, data centers bought nearly 6 GW of nuclear capacity, establishing nuclear procurement at a scale that moved beyond corporate sustainability offsets [S1].
Meta then announced up to 6.6 GW in January 2026, involving TerraPower and agreements associated with Vistra and Oklo [S1][S3]. The package combines PPAs for existing plants, funding for uprates, and support for advanced reactors serving AI data centers [S3].
The procurement model also differs from the 2021–2024 model, when hyperscalers primarily used solar and wind PPAs to match annual consumption and support ESG goals [S3]. By 2026, hyperscalers had shifted toward securing firm, scalable, and always-available nuclear power for continuous AI workloads, including Meta's 6.6 GW of nuclear PPAs with TerraPower [S1][S3].
Deal structure: existing plants, restarts, and SMRs
Existing nuclear plants provide the fastest route to contracted output because they can supply mature generation without waiting for a new reactor design to complete construction [S2].
Microsoft’s 20-year, $16 billion contract covers the full 835 MW of restarted Three Mile Island Unit 1, now called Crane Clean Energy Center [S4]. Google contracted for up to 500 MW from Kairos small modular reactors plus three 600 MW nuclear projects, while Oracle linked its 1 GW nuclear-powered campus to three reactors [S4].
Amazon committed $20 billion toward Susquehanna-linked capacity, while Meta separately announced a 1.2 GW campus involving Oklo [S4][S5]. A separate project tracker lists Amazon’s Pennsylvania campus at 960 MW and describes Microsoft’s Three Mile Island commitment as 837 MW [S5].
These structures should be compared by delivery timing rather than treated as one market. An operating-reactor PPA can address supply sooner, a restart requires plant and regulatory work, and an SMR PPA carries first-of-a-kind construction barriers [S2].
Pricing is contractual, not a single market rate

No single nuclear PPA price covers the 2026 market because each agreement bundles different assets, volumes, development obligations, service periods, and risk allocations. Microsoft signed a 20-year, $16 billion PPA for the full 835 MW of the restarted Three Mile Island Unit 1, while Amazon's $20 billion commitment toward Susquehanna-linked capacity does not provide enough detail to derive a directly comparable $/MWh price [S4].
The main cost drivers are the generator’s capital and operating costs, the volume committed, the contract term, restart or uprate funding, reactor construction risk, and the allocation of curtailment, delay, and performance risk [S2][S4]. These drivers move price more than the label “nuclear”: the same technology carries materially different development exposure across an operating reactor, a restart, and a first commercial SMR fleet [S2].
For planning, use the disclosed contract value and MW term as a first screen, then model financing, interconnection, transmission, curtailment, and escalation separately. A 20-year fixed-price contract fixes the commercial structure, but it also locks buyer and seller into assumptions about dispatch, fuel, maintenance, load, and regulatory conditions for two decades [S4].
Cost drivers shift from headline purchase price
Purchase price covers only the electricity contract. Total cost of ownership also includes grid interconnection, substation and transmission work, backup or firming capacity, on-site electrical infrastructure, and the cost of maintaining service if the contracted reactor schedule slips [S2][S4].
Existing-unit PPAs generally carry less development risk than new-build SMR PPAs, while restart contracts sit between them because they depend on successful refurbishment, licensing, and commercial operation [S2]. For an operating reactor, fuel, maintenance, capacity-factor performance, and plant life dominate cost. For an SMR, factory-built modules, first-of-a-kind engineering, HALEU fuel availability, licensing, and workforce capacity become material cost and schedule factors [S5].
Nuclear's firm, dispatchable output is identified as a way to meet the continuous power requirements of AI data centers [S2][S5].
The cost comparison is not a PPA quote. SMRs offer uninterrupted energy unlike wind or solar, so procurement teams should compare firm delivered energy and interconnection cost rather than generation cost alone [S5].
Who nuclear PPAs suit, and where they fail

Nuclear PPAs fit data centers with large, continuous loads, firm-power requirements, long planning horizons, and a willingness to manage a multi-decade energy contract [S2]. The market is not limited to owners of the largest AI campuses, because a 500 MW Google allocation, an 835 MW restart contract, and a 1.2 GW Meta-Oklo campus all show demand at industrial scale [S3][S4][S5].
They are less suitable for smaller facilities that cannot absorb fixed contract commitments, near-term deployments that need energization before reactor milestones, or buyers unable to manage curtailment, outage, and regulatory interfaces [S2]. An AI-focused facility requiring 80 MW can exceed the 32 MW associated with a standard data center, but that difference alone does not justify an SMR commitment [S5].
Contract formation is only one part of delivery. The cited risks include nuclear construction cost overruns, cancellation exposure, long interconnection delays of up to a decade, limited HALEU fuel supply, and a thin nuclear-engineering talent pool [S4][S5]. Grid connection remains a practical constraint even when the electricity contract is signed.
Regulatory and delivery risk remain decisive
A signed PPA does not guarantee delivered electricity. Nuclear projects remain subject to licensing, construction, fuel, staffing, interconnection, and commercial-operation milestones, with SMRs facing additional first-of-a-kind barriers [S2]. A 20-year agreement can absorb schedule risk, but it cannot eliminate it, and delayed output can leave a data center dependent on interim grid supply [S4].
Three shutdown U.S. plants were moving toward restart by October 2026, including projects linked with Microsoft and Google, while most commercial SMR deliveries were placed in the early 2030s [S4]. This distinction matters because a 2026 announcement can describe demand that will not become operating load until the 2030s.
The United States government was supporting the restart wave with up to $4.4 billion in loans [S4]. Public financing reduces funding pressure but does not automatically remove construction, licensing, supply-chain, or operational risk.
Market size and the 2035 ceiling

U.S. data-center electricity demand was expected to rise from 17 GW in 2022 to 35 GW by 2030, while nuclear procurement reached nearly 6 GW during 2025 [S1][S5]. The comparison shows why nuclear became material: it addresses firm load while grid additions and interconnection processes can take years [S4][S5].
Deloitte’s analysis caps nuclear’s potential contribution at up to 10% of data-center electricity demand by 2035 [S4]. Nuclear is therefore a premium component of the power portfolio, not a substitute for grid supply, renewable PPAs, storage, dispatchable generation, and demand management.
A January 2026 survey found that nearly three out of four respondents in the region blamed data centers for rising electricity costs [S2]. That political pressure can affect permitting, rate-case outcomes, public acceptance, and the allocation of infrastructure costs, even where the nuclear contract itself is private.
Track plant restart milestones, SMR construction and licensing dates, contracted MW delivered by year, interconnection progress, and adjustment terms for delays and curtailment. The data center capital expenditure breakdown provides the broader cost context, while the decisive signal is whether announced GW becomes commissioned MW on schedule.