Data Center Power Startups: The Energy Companies Racing to Power AI

Artificial intelligence runs on electricity, and lots of it. Training and running AI models in large data centres has made power one of the biggest constraints on the AI boom, and technology companies are signing deals with energy startups that would have seemed unlikely a few years ago: fusion developers, small nuclear reactor makers, geothermal drillers and battery companies. This guide covers the private companies racing to power AI, how much electricity data centres need, the deals that have been signed, valuations and the risks.

The table ranks private energy companies in our database that are positioned to supply power or storage for data centres, by sourced valuation. Some leading companies have not disclosed valuations, and two that were private in 2025, Fervo Energy and X-energy, went public in 2026. Each ranked company links to a full profile.

The ranking

How much power AI needs

Data centres' electricity use rose 17% in 2025, far faster than overall global demand, and the International Energy Agency expects it to roughly double by 2030, with AI-specific demand tripling, according to its April 2026 analysis as reported by W.Media and Let’s Data Science. Individual AI campuses now plan for hundreds of megawatts to several gigawatts, comparable to the output of large power stations. In many regions, getting a grid connection for a new data centre can take years.

The constraint is not just energy but timing and reliability. AI data centres want power around the clock, and they want it soon, while new transmission lines and large power plants take many years to build. Shortages of gas turbines, transformers and other equipment have added delays.

Why tech companies are turning to new energy startups

Large technology companies have committed to clean energy goals, and solar and wind alone cannot provide steady power at all hours without large amounts of storage. That has pushed them toward "clean firm" power, carbon-free electricity available whenever needed, such as nuclear, geothermal and, eventually, fusion. Tech companies signed around 40% of corporate renewable power purchase agreements in 2025 and increasingly back nuclear and advanced geothermal, according to the IEA analysis. Many developers are also building on-site natural gas generation as a bridge.

Fusion

Helion Energy, valued at US$15.5 billion in 2026, has an agreement to supply Microsoft with power from a fusion plant, an ambitious target. Commonwealth Fusion Systems has an agreement with Google to buy power from its first planned plant in Virginia. Pacific Fusion and stellarator developer Type One Energy, which is working with the Tennessee Valley Authority, are earlier. Fusion has not yet produced net electricity for the grid, so these deals are long-term bets. See our fusion energy guide.

Advanced nuclear fission

Small and advanced fission reactors could arrive sooner. Kairos Power has an agreement to supply Google with up to 500 megawatts from its salt-cooled reactors by 2035 and was reported in 2026 to have raised about US$643 million including government funds, according to Dealroom. Valar Atomics and Radiant are building smaller reactors. X-energy, backed by Amazon, raised about US$1.02 billion in an April 2026 IPO, according to Bloomberg Law. Our nuclear energy startups guide covers the sector in depth.

Geothermal

Enhanced geothermal systems drill deep and use techniques from the oil and gas industry to tap heat almost anywhere. Fervo Energy, which signed agreements with Google and others, priced an upsized IPO at US$27 a share in May 2026, raising about US$1.89 billion, according to Axios. Geothermal offers round-the-clock power with a small land footprint, and its success has drawn investment into other developers.

Batteries and grid flexibility

Batteries do not generate electricity, but they can store cheap power and release it when needed, and they can be installed far faster than power plants. Base Power installs batteries in homes and uses them as a distributed power plant for the grid. Redwood Materials, the battery recycler, launched Redwood Energy to turn used electric-vehicle batteries into storage for data centres, and Google joined its US$425 million Series E in January 2026, according to TechCrunch. Form Energy makes iron-air batteries designed to store power for days. Data centres can also act as flexible loads, reducing use at peak times, which makes it easier to connect them to strained grids.

Computing companies that build their own power

Some AI infrastructure companies have moved into energy themselves. Crusoe builds data centres sited near cheap or stranded energy, and others pair data centres with on-site generation. These approaches blur the line between energy companies and computing companies.

Policy and regulation

Governments are trying to speed up power supply for AI. In the US, executive actions in 2025 aimed to accelerate nuclear licensing and federal permitting, the Department of Energy offered sites and support for reactors and data centres, and regulators debated how to allocate grid costs between data centres and other customers. Utilities have proposed special tariffs for large loads. Elsewhere, Ireland and parts of Europe have limited new data centre connections because of grid constraints.

How to evaluate a data center power startup

Ask: when could it deliver power, and how firm are its customer agreements? Has it demonstrated its technology at meaningful scale? What will its power cost per megawatt-hour compared with gas or grid power? Does it have a site, permits and a grid connection? How much capital will it need before generating revenue? And who bears the risk if a project is delayed?

Risks

• Timing mismatch: AI demand is growing now, while many new energy technologies will not deliver at scale until the 2030s.
• Technology risk: fusion and some advanced reactors are unproven commercially.
• Cost overruns: first-of-a-kind energy projects routinely exceed budgets.
• Demand uncertainty: efficiency gains in AI chips and software could slow electricity growth.
• Concentration: many startups depend on a few large technology buyers.

What to watch

Watch for new power purchase agreements between tech companies and startups, reactor and fusion milestones, geothermal project results, battery deployments at data centres, grid connection reforms and how the recently listed Fervo and X-energy perform. Our climate tech guide and AI infrastructure guide cover related companies.

Key takeaways

AI's electricity needs have created a new market for energy startups. Fusion developers such as Helion and Commonwealth Fusion Systems, nuclear companies such as Kairos Power, geothermal pioneer Fervo and battery companies such as Base Power and Redwood are all signing deals with technology giants. Delivering reliable power on time and at a competitive price will separate winners from long-term bets.

The deals between tech giants and energy startups

A pattern has emerged in how technology companies support new energy sources. Rather than building power plants themselves, they sign long-term agreements to buy electricity, or the environmental attributes of electricity, from specific projects, often years before those projects exist. These agreements give startups a credible customer, which helps them raise money and secure permits. Google has backed geothermal, advanced nuclear and fusion through such deals; Microsoft has agreed to buy fusion power from Helion and supported the restart of an existing nuclear plant; Amazon has invested directly in X-energy and committed to buy its power; and Meta has signed nuclear agreements. Some deals include equity investments, as with Google's participation in Redwood Materials' Series E.

For investors, these agreements are among the most important signals in the sector, but their terms vary widely. Some are binding purchase commitments; others are options or frameworks that depend on the technology working and costs falling.

The bridge: gas and existing plants

Because most new clean technologies are years away, much of the power for AI data centres built in the mid-2020s comes from natural gas plants, existing nuclear plants and grid power. Developers have ordered gas turbines faster than manufacturers can build them, and some data centres run on-site generation until grid connections arrive. This bridge period is why startups that can deliver sooner, such as battery companies and geothermal developers, have attracted strong interest.