Google and the Nuclear Power Race

 

 

ENERGY & TECHNOLOGY  |  GLOBAL BUSINESS

Google and the Nuclear Power Race

Alphabet is reportedly nearing a $1 billion-plus nuclear power agreement with Constellation Energy. Signed deals with utilities and generators show a wider shift: hyperscalers are using long-term contracts to support existing plants, upgrades and new capacity as data-centre demand rises.

Hero illustration: AI-generated concept art, not a photograph of a Google data centre or a Constellation facility.

By Murthy ASR  |  Biz Masala  |  6 October 2026

The latest headline is still a negotiation

Bloomberg reported on 6 October that Alphabet, Google’s parent, was close to a multi-year deal to buy nuclear energy from Constellation Energy, worth $1 billion or more. The report cited people familiar with the discussions. Google and Constellation declined to comment, and the report said the exact scale and location were not disclosed. As of publication, this is a reported negotiation, not a public, finalized contract; it should not be described as signed or as a specified megawatt commitment.

The report lands after a series of more concrete Google energy moves. In September, Google agreed to support additional nuclear capacity at Georgia Power’s Vogtle and Hatch plants. In Finland, Google signed a long-term power-purchase agreement with Fortum for electricity from Loviisa. Google also has its earlier agreement with Kairos Power, which aims to develop a fleet of advanced reactors in the United States. Together, these arrangements show more than one route to nuclear power: buying from existing plants, helping expand output, extending plant lives and backing new reactor development.

The demand backdrop is changing

The U.S. Department of Energy’s 2025 update, prepared by Lawrence Berkeley National Laboratory, estimates data centres used 192 terawatt-hours (TWh), or 4.7% of U.S. electricity, in 2024. Its reference case projects 649 TWh, or 11.8%, by 2030. A compounded-uncertainty range spans 521–843 TWh, equal to 9.5–15.3% of projected U.S. electricity use. These are scenarios for all data centres, not AI alone, and they are forecasts rather than measured future demand.

Figure 1. The 2030 values are scenarios in the DOE/LBNL 2025 report. The 2025 update revises the previous 2028 projection.

AI is one contributor to load growth alongside other digital services, manufacturing and electrification. The scale and location of demand matter as much as the national share: data centres cluster in particular utility territories, where new generation, transmission connections and grid capacity need to be planned together.

A deal board across the nuclear supply chain

Buyer and partner	Structure	Scale and status
Google–Constellation, U.S.	Reported multi-year nuclear energy agreement	Bloomberg says $1bn+; near agreement as of 6 Oct. Exact plant, volume and terms not public.
Google–Georgia Power, U.S.	Customer subscription supporting nuclear uprates at Vogtle and Hatch; Google receives nuclear zero-emission credits	About 96 MW of additional grid capacity; agreement subject to Georgia PSC approval. Utility projects about $900m in customer benefits over the units’ lives.
Google–Fortum, Finland	22-year power purchase agreement linked to Loviisa plant life extension	PPA begins in 2028; reaches up to 50% of plant capacity in 2030–49. Google’s separate Finland investment plan is €13bn for data centres and supporting infrastructure in 2027–28.
Microsoft–Constellation, Pennsylvania	20-year PPA tied to restart of Three Mile Island Unit 1, renamed Crane Clean Energy Center	About 835 MW; restart expected in 2028, subject to regulatory approvals. Unit 2, associated with the 1979 accident, is separate and is being decommissioned.
Amazon–Constellation, Maryland	20-year agreement backing Calvert Cliffs expansion and plant investment	690 MW in total, including a 190-MW uprate; new capacity planned for 2030–32. Electricity continues to flow to PJM grid; separate retail supply agreement covers Amazon operations in PJM.
Google–Kairos Power, U.S.	Master plant development agreement for a fleet of advanced reactors	Up to 500 MW targeted by 2035; first project Hermes 2 is planned to provide up to 50 MWe to TVA grid, with operation targeted from 2030.

Table 1. Contract structures differ. Capacity is not annual energy; announced capacity, customer credits and grid electricity are not interchangeable measures.

From buying power to backing the grid

The company announcements do not describe a simple move away from the grid. Nuclear plants generate into regional power systems, and several of these deals explicitly support grid supply. At Georgia Power, Google’s subscription is linked to zero-emission credits from extra nuclear output; the utility says the uprates would add about 96 MW to the grid, with regulatory approval still pending. In Maryland, Constellation says all Calvert Cliffs electricity will continue flowing to the PJM regional grid. Google’s Finnish PPA is a long-term purchase tied to Fortum’s Loviisa generation and life-extension programme.

This distinction matters for how claims about “direct nuclear power” are read. A PPA can contract for a generator’s output or environmental attributes without connecting a data centre to a dedicated power line. Physical supply, commercial settlements, and clean-energy matching are related but separate things. In Georgia, the proposed tariff and customer safeguards will be reviewed by the state regulator; the customer-benefit figure is the utility’s projection, not a realized saving yet.

Figure 2. AI-generated illustration. Several featured arrangements contract for power or energy attributes delivered through a regional grid.

Why utilities are engaging with hyperscalers

Long-term power commitments can give generators a customer and revenue visibility that helps support life-extension investment, uprates or restart work. Google’s Georgia arrangement is designed to fund extra output from existing plants; Amazon’s Calvert Cliffs contract supports a plant upgrade and continued operation; Microsoft’s agreement is tied to the restart of a retired unit; Google’s Finnish contract underpins Fortum investment to keep Loviisa operating through 2050. Each deal allocates risk differently, and most of the headline outcomes remain future or conditional.

That can change the investment case for nuclear utilities, but it does not automatically make a utility stock stable or guarantee higher returns. Plant outages, regulatory approval, capital costs, construction schedules, fuel supply, power prices and contract terms still shape outcomes. In Finland, Fortum says the PPA improves expected returns on assets; in Georgia, projected customer benefits depend on the agreed tariff and regulator approval. These are company forecasts, not universal results for the sector.

The grid-cost question

The DOE/LBNL forecast points to pressure for new supply and transmission as data-centre use expands. Whether households or manufacturers face higher bills depends on where facilities connect, how much new infrastructure is required, who pays for it, how rate structures are designed, and whether new supply arrives on time. Grid crowding is a planning risk, not an automatic result of every corporate power contract. Some agreements are expressly designed to fund incremental capacity or protect non-participating customers; their effectiveness has to be assessed deal by deal and after the relevant regulator reviews the filings.

Nuclear is only one part of the answer. Its continuous generation profile can complement wind and solar, while storage, transmission, demand flexibility, efficiency and other generation sources also contribute to a reliable system. Nuclear life extension and uprates can sometimes use existing sites and grid connections; new reactor builds, including SMRs, add a longer development and licensing pathway. Contracts can support financing, but they cannot remove engineering, regulatory or construction risk.

The Masala Take

Google’s nuclear strategy is widening from an advanced-reactor partnership into a portfolio of long-term supply, capacity and infrastructure arrangements. The reported Constellation deal may become another large contract, but its value, location and megawatt commitment remain undisclosed. Signed agreements in Georgia, Finland, Pennsylvania and Maryland show the emerging model: tech buyers help underwrite nuclear output or plant investment while power generally remains part of the grid. The test is whether promised capacity is approved, built or upgraded, delivered on schedule, and priced in a way that works for customers and power systems.

Sources and data notes

The Constellation negotiation is attributed to Bloomberg’s 6 October 2026 report and remains unconfirmed by the companies in that report. Deal details for other projects are drawn from company announcements and should be read as announced terms, targets or projections. DOE/LBNL forecast estimates are scenarios; they do not attribute all data-centre demand growth to AI.

·       Bloomberg, Google and Constellation near billion-dollar nuclear agreement, 6 October 2026  Open source

·       U.S. DOE/LBNL, United States Data Center Energy Usage Report: 2025 Update  Open source

·       Georgia Power, Google nuclear uprate agreement, 21 September 2026  Open source

·       Fortum, Google Loviisa power-purchase agreement, 9 September 2026  Open source

·       Constellation, Amazon Calvert Cliffs 20-year power agreement, 30 September 2026  Open source

·       Constellation, Microsoft Crane Clean Energy Center 20-year PPA, 20 September 2024  Open source

·       Google, Kairos Power nuclear energy agreement, 14 October 2024  Open source

·       Kairos Power, Hermes 2 groundbreaking, 17 April 2026  Open source