Amazon's agreement this week to fund an Indiana nuclear plant that nobody has yet decided to build shows how far corporate nuclear deals now reach ahead of the power they promise. Two weeks earlier, Google signed for upgrades at operating Constellation reactors that start delivering in 2028. Both get called nuclear deals, yet the electricity arrives years apart, and energy buyers who add such megawatts together will carry a supply forecast that looks firmer than it is.
In corporate terms, nuclear procurement means long-term contracts that tie a company's electricity purchases or clean energy claims to a nuclear plant's output. The deals announced over the past 18 months sort cleanly by delivery date. Contracts with operating reactors deliver now and uprates add capacity over several years. Restarts wait on federal licensing, while new small modular reactors (SMRs) wait on a construction decision.
On October 19, American Electric Power (AEP), reactor developer X-energy and Amazon announced plans to evaluate a 960-megawatt (MW) plant built from 12 of X-energy's 80-MW Xe-100 reactors. AEP and X-energy signed a joint development agreement to study feasibility at the former Breed Generating Station site in Sullivan County, Indiana. Amazon is funding development under a separate agreement. Amazon would be the plant's first power customer, the company said. Josh Weissman, Amazon's vice president of data center infrastructure capacity delivery, said in the AEP release that Amazon would "pay our full costs without passing them to others."
Construction depends on the project reaching a final investment decision. AEP chief executive Bill Fehrman said the agreements "will provide all necessary capital protections" for the utility, but the releases leave out how much Amazon is paying and whether its funding is tied to milestones. Indiana is the second site under Amazon's X-energy collaboration after the Cascade Advanced Energy Facility in Washington state, part of a goal of more than 5 gigawatts (GW) of new nuclear capacity by 2039 that fits into Amazon's wider carbon-free power push.
Google's October 6 deal with Constellation shows a nearer-term version. Under a 20-year power purchase agreement (PPA), Constellation plans uprates adding 890 MW across 11 units at six sites in Illinois, Pennsylvania and New Jersey, with the first expected by 2028. A separate 15-year agreement covers 2,700 MW from Constellation's existing fleet in the PJM Interconnection market. Google also committed to demand flexibility, including cutting noncritical consumption during high-stress grid events. Even firm nuclear supply, in that structure, comes with obligations on the buyer's side of the meter.
Restarts can move faster than new construction, though they pass through their own regulatory gate. The Department of Energy (DOE) closed a $1 billion loan in November 2025 for Constellation's restart of the 835-MW Crane Clean Energy Center in Pennsylvania, which stopped operating in 2019 but was never fully decommissioned. Microsoft holds a 20-year agreement for Crane's output, and Constellation is targeting 2027 for its return. The Nuclear Regulatory Commission (NRC) issued a finding of no significant environmental impact in September. Its September 25 notice still lists an exemption and three license amendments under consideration. Other reactors have followed a similar path, with federal money flowing to Palisades in Michigan ahead of its own return.
Delivery dates matter most when set against a buyer's actual load. Suppose a manufacturer plans an expansion that needs power in 2029 and signs with a project that cannot deliver until the early 2030s. The facility still needs electricity in between. That usually means a bridge contract or tariff service at market-linked prices, and a different emissions profile for those years.
Planning around that gap starts with modeling the developer's commissioning date next to a few delay scenarios. The contract itself should settle who buys replacement power if delivery slips, whether financial remedies cover the difference and when the buyer can exit or restructure. Renewable contracts have already shown how delivery shortfalls can land on corporate balance sheets, and nuclear timelines run longer.
Development payments need their own review. Procurement and finance teams should know what each payment secures and which milestone unlocks the next one. They should also know what happens to committed money if a project stops before construction. Public announcements rarely disclose these terms, so the deals described here cannot be judged on them from the outside. They are negotiating priorities for companies entering similar agreements, not findings about how Amazon, Google or Microsoft handled their own exposure.
A third model, which Google signed with Kairos Power and the Tennessee Valley Authority (TVA), runs through the utility. TVA will buy electricity from Kairos's planned 50-MW Hermes 2 reactor in Oak Ridge, Tennessee, which is targeted for 2030. Google will procure the plant's clean energy attributes through TVA for its data centers in Montgomery County, Tennessee, and Jackson County, Alabama. Google describes it as a three-party model in which customers, utilities and developers share the work.
For a buyer weighing a similar structure, the first question is what actually changes hands. It may be physical electricity, environmental attributes, a financial settlement or some mix. Somebody still has to handle transmission, retail service and backup supply. A contract with a nuclear generator does not by itself let a facility run independently of its local grid, which is part of why firm power has moved back to the center of grid planning.
Existing GHG Protocol guidance already lets companies use contracts with nuclear plants under market-based Scope 2 accounting, provided they meet the Scope 2 Quality Criteria. Claims follow electricity delivered during the reporting year. Output from a plant still in development stays a forecast until the reactor runs.
The rules may tighten on timing. A GHG Protocol consultation that opened in October 2025 proposed hourly matching whenever companies use contractual instruments. It also proposed market boundaries based on deliverability, plus a legacy clause for existing contracts. Exemptions would target smaller reporters, leaving the hourly requirement on larger ones. The August 2026 feedback summary says responses will inform the next phase of development, and no final decision has been made. That debate is already reshaping how companies think about renewable energy claims.
An hourly test would favor steady generators. A reactor's output tracks a data center's flat load far more closely than a solar contract with the same annual volume, so nuclear contracts would generally fare well on the hourly half of the test. The open question is deliverability. Many of the plants now under contract will start producing in the late 2020s or the 2030s, possibly under different Scope 2 rules than today's. Buyers signing now can ask whether their contract would still support a claim under a deliverability boundary, and whether a legacy clause would cover it.
