Alphabet, Google’s parent company, has signed a nuclear power supply agreement with Constellation Energy for 890 megawatts across eleven reactor units in three states: Illinois, Pennsylvania, and New Jersey. The commitment exceeds $4.3 billion (roughly €4.04 billion). It is the largest corporate nuclear energy contract in tech industry history, arriving as electricity demand from AI data centers turns energy security into a strategic priority for Big Tech.
- Record deal: Alphabet will buy 890 MW of existing nuclear generation from Constellation Energy, backed by more than $4.3 billion.
- Base-load advantage: Nuclear runs near a 90% capacity factor, far above the 25–35% typical of solar and wind, giving AI data centers constant power.
- Policy ripple effects: The contract could keep nuclear plants open, influence wholesale power markets, and push rival tech firms to secure similar clean-energy supplies.
A Supply Deal, Not a Construction Project
The 890 MW are not new installed capacity but guaranteed production from facilities that already exist. Constellation is the largest nuclear operator in the United States, with a fleet of 21 reactors. The agreement with Google represents a significant portion of output from three of its plants — Limerick in Pennsylvania, Braidwood in Illinois, and Salem in New Jersey — and is equivalent to continuously powering about 670,000 US households. Google will buy the production during the agreed period. Nuclear already works; what changes is who consumes it and at what price.
Google had previously announced its nuclear ambitions for data centers in general terms. This Constellation deal is the concrete version, with figures and identified plants. The size of the commitment — more than $4.3 billion — signals that it is not a pilot project but a strategic infrastructure bet. The shift from the renewable-power purchases of 2019 to this 2026 nuclear agreement is qualitative: earlier deals bought green energy to offset emissions, while this one buys base-load power to guarantee operational continuity.
Why AI Needs Nuclear-Grade Reliability
The electricity demand of frontier AI models is qualitatively different from that of conventional digital services. A standard Google search consumes around 0.3 watt-hours. A query to a large language model such as Gemini Ultra consumes approximately 10 watt-hours, about 33 times more. At the scale of millions of daily queries, that difference becomes structural.
Google’s transition from search engine to generative AI provider multiplies electricity use per interaction by a significant factor. Data centers designed for conventional web traffic lack both the installed power and the supply stability required for training and inference on large models. The decision to buy nuclear energy is partly climate-related — meeting net-zero emissions commitments — and partly pragmatic: it is the only source with enough power density and availability for Google’s AI expansion plans.
Microsoft made a nuclear bet for its data centers in 2023, with an agreement to reopen the Three Mile Island plant in Pennsylvania. Google’s move with Constellation follows the same logic but at a larger scale and across multiple facilities, reducing the risk of relying on a single supply point.
The $4.3 Billion Energy-Policy Moment
An agreement of this size is not only a commercial contract. It has implications for national energy policy. Constellation needs long-term demand certainty to justify investment in maintenance and life extension for its nuclear plants. Google is providing that certainty, with a side effect: keeping nuclear plants online that might otherwise have closed because they were unprofitable in a deregulated electricity market.
Controversy is already emerging. Environmental groups question whether a private corporation should influence decisions about which nuclear infrastructure remains active. State regulators are examining whether private contracts of this kind distort the wholesale electricity market. And technology competitors are now asking whether they should sign similar deals or risk falling behind in access to clean, constant power.
AI needs energy. Energy needs investment. Tech investment is now financing the continued operation of nuclear infrastructure that conventional electricity markets could not sustain on their own. That is neither good nor bad by definition, but it is a transformation of the energy landscape that is likely to keep accelerating.