Energy9 min read

Why Big Tech Is Becoming an Energy Company for AI

Microsoft, Google, Amazon, and Meta are moving beyond buying electricity. Their nuclear, geothermal, grid, and power-development deals reveal how energy is becoming a strategic part of the AI business.

Loading the player connects to YouTube (a Google service) and plays the video in YouTube's privacy-enhanced mode. Watch on YouTube instead without loading the player here.

The competition to build artificial intelligence no longer ends with a purchase order for GPUs. It now reaches power plants, utility contracts, nuclear-reactor developers, geothermal wells, batteries, substations, and the regulatory system surrounding the grid.

Microsoft has signed a long-term agreement connected to the proposed restart of a nuclear plant. Google is supporting advanced nuclear reactors and enhanced geothermal energy. Amazon has combined nuclear-development agreements, direct investment, and data-center siting near existing generation. Meta has assembled agreements covering existing plants and proposed advanced reactors.

These companies are not transforming into traditional utilities. They are doing something strategically important: moving upstream to influence where their future electricity will come from, when it may become available, and who will finance the projects needed to produce it.

For the AI industry, energy procurement is becoming part of product capacity.

Why buying electricity is no longer enough

Most businesses treat electricity as a service. They connect a building to the grid, select a supplier where markets allow it, pay the bill, and expect the power system to expand as demand changes.

That model becomes more difficult when a single project requires tens or hundreds of megawatts, operates continuously, and must be built on a technology-company schedule.

Global data-center electricity consumption was approximately 415 terawatt-hours in 2024. The International Energy Agency projects that it will more than double to around 945 TWh by 2030, with AI as the most important driver of the increase (see Sources, [1]).

The challenge is not only producing enough electricity over an entire year. A data center needs capacity at a particular location, a dependable grid connection, and power during every hour its computing equipment is operating.

That is the bottleneck examined in Why AI Data Centers Have a Power Problem Nvidia Can't Solve. The strategic response from Big Tech is the subject here.

Instead of waiting for the conventional power system to deliver every new megawatt, technology companies are signing longer contracts, supporting new generation technologies, investing in developers, and influencing projects years before their data centers begin operating.

Microsoft: using a long-term contract to support a restart

In September 2024, Constellation announced a 20-year power-purchase agreement with Microsoft connected to restarting Three Mile Island Unit 1 in Pennsylvania. The plant was renamed the Christopher M. Crane Clean Energy Center.

The proposed restart would return approximately 835 megawatts of carbon-free generating capacity to the grid, with Microsoft purchasing the energy produced by the renewed plant (see Sources, [2]).

This is significant because the agreement connects a technology buyer's long-term electricity demand to the economics of restarting an existing generating asset. Microsoft is not operating the reactor. Its purchasing commitment gives the plant owner a large customer and a future revenue stream around which the restart can be planned.

However, an announcement is not the same as an operating power plant.

The reactor stopped operating in 2019. Before it can restart, Constellation must restore the plant's operational licensing basis, return components to an acceptable operating condition, make required upgrades, and receive regulatory approvals. The Nuclear Regulatory Commission continues to oversee and review that process (see Sources, [3]).

Microsoft's agreement illustrates one version of the new energy strategy: use a long-term corporate commitment to help preserve or reactivate a large source of firm generation.

Google: creating an order book for new technologies

Google is taking a portfolio approach that includes both advanced nuclear power and enhanced geothermal energy.

In 2024, Google announced an agreement to purchase electricity from multiple small modular reactors being developed by Kairos Power. The initial deployment was intended to come online by 2030, followed by additional reactors through 2035. Google said the overall agreement could enable up to 500 MW of new round-the-clock carbon-free power (see Sources, [4]).

The strategy is about more than acquiring electricity from one plant.

New energy technologies often face a commercial problem before they face a technical one. Developers need customers before they can finance factories, supply chains, licensing work, and repeated deployments. Customers may hesitate to sign contracts until the technology has already been proven at scale.

A large buyer can help break that cycle by creating an order book: a visible pipeline of demand across several future projects. If the first deployment works, later units may benefit from accumulated manufacturing and construction experience.

Google has used a similar early-customer role with enhanced geothermal energy. Its project with Fervo Energy began delivering carbon-free electricity to the Nevada grid in 2023. The system applies horizontal drilling and subsurface-monitoring techniques to access geothermal resources that would be difficult to develop with conventional wells (see Sources, [5]).

Nuclear and geothermal are technically different, but they solve a similar procurement problem. Both aim to supply dependable carbon-free electricity beyond the hours when solar or wind generation happens to be available.

Google's strategy is therefore not a bet on one power plant. It is an attempt to help commercialize energy technologies that could later be repeated across several regions.

Amazon: combining siting, investment, and development

Amazon's nuclear strategy operates across several layers.

Its 2024 announcement included an agreement with Energy Northwest to support four advanced small modular reactors in Washington. Amazon said the first phase was expected to provide approximately 320 MW, with an option to expand the project to 960 MW.

Amazon also invested in X-energy, whose reactor design is intended for the Energy Northwest project. According to Amazon, that investment included manufacturing capacity that could support more than five gigawatts of future projects using X-energy's technology.

In Virginia, Amazon and Dominion Energy agreed to explore an SMR project near the existing North Anna nuclear station, potentially adding at least 300 MW. Amazon also described its previously announced arrangement to locate a data-center facility next to Talen Energy's nuclear plant in Pennsylvania (see Sources, [6]).

These are three different tools:

  • Locate computing near an existing source of generation.
  • Become an early customer for proposed generating capacity.
  • Invest in the company and manufacturing capability needed to build the technology.

The combination shows why power strategy now affects data-center strategy. Where Amazon places future computing infrastructure can depend on where reliable electricity already exists or where new generation has a credible path to completion.

Meta: building a multi-project nuclear portfolio

Meta's approach demonstrates the scale this competition is reaching.

In January 2026, Meta announced agreements involving Vistra, TerraPower, and Oklo that, together with an earlier Constellation agreement, could support up to 6.6 GW of new and existing nuclear capacity by 2035 (see Sources, [7]).

The portfolio covers several kinds of projects.

Meta's agreement with Vistra includes more than 2.1 GW from operating nuclear plants in Ohio, along with proposed increases in output at plants in Ohio and Pennsylvania. The objective is partly to preserve and expand existing generation.

Its TerraPower agreement supports two proposed Natrium units capable of providing up to 690 MW, with rights associated with as many as six additional units. Across eight potential units, Meta described up to 2.8 GW of generating capacity, plus built-in storage capacity.

The Oklo agreement supports a proposed advanced-nuclear campus in Ohio that may eventually provide up to 1.2 GW.

The word potential matters throughout these announcements. Some capacity already exists. Some depends on plant upgrades. Some requires entirely new reactor technologies, construction programs, fuel supplies, financing, and regulatory approvals.

The 6.6 GW figure is therefore a pipeline, not a statement that 6.6 GW of new power is currently operating for Meta.

Even with that qualification, the scale is revealing. Meta is planning energy supply on the same long time horizon as its future AI infrastructure.

Four ways Big Tech is moving upstream

The company announcements reveal four recurring strategies.

Long-term power-purchase agreements

A long contract can give a generator predictable revenue while giving the buyer more certainty about future energy supply and pricing. It transfers some risk, but it does not guarantee that a proposed project will be completed.

Direct investment

Investing in a reactor or energy-technology developer can help finance engineering, manufacturing, and early deployment. It also exposes the investor to technology and execution risk beyond an ordinary electricity contract.

Project development commitments

An agreement covering several proposed plants can create demand before a technology reaches broad commercial scale. Successful repeated deployment could reduce later costs, but first-of-a-kind projects commonly face difficult schedules.

Strategic data-center siting

Locating computing near generation can reduce dependence on finding an entirely new source of power elsewhere. The project still needs an acceptable interconnection arrangement, transmission capacity, regulatory approval, and a plan for reliability.

These approaches do not make the technology companies utilities. Generation owners, grid operators, and regulators remain essential. But the buyers are becoming much more involved in shaping what gets built.

Nuclear is not the only part of the energy race

Nuclear projects receive attention because they can provide large amounts of firm, low-carbon electricity. Their challenge is time: restarts, plant expansions, and new reactor designs can require years of regulatory and construction work.

AI demand is growing before many proposed nuclear projects can arrive.

The IEA expects renewable generation to supply nearly half of the additional global electricity demanded by data centers through 2030. It also expects natural gas and coal together to meet more than 40% of that growth, while nuclear becomes more important toward the end of the decade and after 2030 (see Sources, [8]).

That means the practical power portfolio can include solar, wind, gas generation, existing nuclear plants, geothermal energy, transmission, and storage.

Batteries can shift electricity between hours and help manage peaks. Microgrids can coordinate local generation, storage, and data-center demand. Natural-gas plants can provide dispatchable power, but they create fuel-price and emissions exposure. Renewable projects can be developed relatively quickly in favorable locations, but continuous AI workloads also need transmission, storage, flexible demand, or other firm resources.

There is no single technology that solves every location, schedule, reliability, cost, and emissions requirement.

Who benefits from the new energy competition

When technology companies move upstream, economic opportunity moves with them.

Owners of existing nuclear plants gain customers willing to sign long contracts supporting continued operation, restarts, or output increases. Advanced-reactor developers gain potential anchor customers before their technologies reach large commercial fleets.

Geothermal developers can apply drilling, subsurface engineering, pumps, and monitoring systems to a new source of round-the-clock generation. Utilities and transmission developers must connect large new loads and generation projects.

Manufacturers of transformers, switchgear, cables, turbines, generators, batteries, and cooling systems supply the equipment between the power plant and the computing rack. Engineering, construction, and project-finance companies help turn announcements into operating assets.

These energy suppliers join the wider infrastructure businesses mapped in The Hidden Economy Behind AI Data Centers.

The opportunity does not guarantee attractive returns. A company can operate in a growing market and still lose money through cost overruns, excessive debt, project cancellation, technological failure, or an overvalued share price.

The useful question is not only which sector has demand. It is which company can deliver operating capacity on time and earn an acceptable return from doing so.

What these agreements cannot guarantee

Large energy announcements often combine existing power, proposed additions, options, development targets, and projects expected many years in the future. Those categories should not be treated as equivalent.

Several risks remain:

  • Regulatory approval can delay or prevent a project.
  • First-of-a-kind technology can cost more and take longer than expected.
  • A signed agreement may support a project without guaranteeing its completion.
  • Grid and transmission constraints can remain even when generation is available.
  • Fuel, equipment, construction labor, and financing can become bottlenecks.
  • Electricity-demand forecasts can change as AI hardware and software become more efficient.
  • Communities and regulators may reject projects whose costs or local impacts appear unfair.

A power contract also does not necessarily create a private electrical path between one named plant and one data center. In many arrangements, generation and consumption remain connected through a regional grid. The commercial contract and the physical electricity flow are related, but they are not always the same thing.

That distinction matters when judging claims about reliability, emissions, and who ultimately pays for grid expansion.

Energy strategy is becoming AI strategy

The first stage of the AI infrastructure race was dominated by access to advanced chips. The next stage requires companies to transform those chips into dependable computing capacity.

Use the AI Data Center Power & Cost Calculator to see how GPU count, power draw, utilization, overhead, and PUE can translate into facility load and electricity cost.

At hyperscale, those calculations stop being an ordinary utility bill. They influence where data centers are built, which generating projects receive financing, and how far into the future technology companies must plan.

Microsoft is supporting the proposed return of an existing nuclear plant through a long-term purchase agreement. Google is helping create an order book for advanced nuclear and geothermal projects. Amazon is combining development agreements, investment, and strategic siting. Meta is constructing a multi-project portfolio covering existing and proposed nuclear generation.

None of these strategies eliminates the grid. They show how valuable access to the grid has become.

The companies that once competed mainly for processors are now competing for megawatts. AI's new energy war is not simply about buying more electricity. It is about securing the infrastructure capable of producing it.

Sources & References

  1. [1]Energy and AI — Executive summaryInternational Energy Agency
  2. [2]Crane Clean Energy CenterCommonwealth of Pennsylvania
  3. [3]Christopher M. Crane Clean Energy CenterU.S. Nuclear Regulatory Commission
  4. [4]New nuclear clean energy agreement with Kairos PowerGoogle
  5. [5]A first-of-its-kind geothermal project is now operationalGoogle
  6. [6]Amazon signs agreements for innovative nuclear energy projectsAmazon
  7. [7]Meta Announces Nuclear Energy Projects, Unlocking Up to 6.6 GWMeta
  8. [8]Energy supply for AIInternational Energy Agency