Meta has announced a nuclear power package that tells you a great deal about where the AI industry is heading. The company says it has agreements with Vistra, TerraPower and Oklo to secure power for its data centre expansion, with the Prometheus campus in New Albany, Ohio sitting at the heart of the story. That matters because this is not a side project or a branding exercise. It is a large, deliberate attempt to lock in electricity for AI systems that will need vast amounts of it for years.
The headline number is striking: Meta says the combined package supports up to 6.6 gigawatts of new and existing clean energy by 2035. But the more interesting detail is buried inside the structure of the deals. The phrase “up to eight” does not mean eight reactors are suddenly breaking ground. It comes from the TerraPower portion, where Meta is funding two Natrium units targeted for 2032 and has rights to up to six more aimed at 2035.
The short version: Meta has bought itself a long runway for power, not a quick fix.
Why Meta is turning to nuclear power for AI
The logic is straightforward. AI data centres need dependable power at a scale that wind and solar alone cannot always provide on a predictable timetable. Meta’s framing around Prometheus makes that plain enough. This is about long-term power certainty for a major computing cluster, not just offsetting emissions on paper or polishing a sustainability narrative.
That is also why nuclear keeps returning to the conversation. Existing reactors can provide round-the-clock generation, and advanced nuclear companies keep pitching future plants as a way to serve energy-hungry industrial loads. Meta’s earlier move points in the same direction. In June 2025, it announced a 20-year agreement with Constellation for about 1.1 gigawatts from the Clinton plant in Illinois. This latest package is larger, broader and more ambitious, but it follows the same underlying concern: AI growth is running into the hard limits of electricity supply.
The Vistra agreement is the most immediate part of the package
If you want the least speculative element, it is the deal with Vistra. Meta has agreed to a 20-year power purchase arrangement tied to the Perry, Davis-Besse and Beaver Valley nuclear plants. Purchases are set to begin in late 2026, and the arrangement is linked to support for 20-year licence renewals.
That is an important distinction. This part of the strategy is not mainly about inventing the future. It is about keeping existing nuclear generation in the system and directing its output towards Meta’s needs over a long period. For a company trying to plan data centre capacity with fewer unpleasant surprises, that kind of stability has obvious appeal.
The Vistra deal is the practical spine of the package: existing reactors, a long contract and power that arrives first.
It also shows how these announcements should be read. The package mixes near-term supply with much longer-term bets. Lumping them together makes the number look enormous, but the delivery profile is uneven by design.
TerraPower is where the ‘up to eight nuclear plants’ story comes from
The most eye-catching part of the announcement is Meta’s agreement with TerraPower. This covers funding for two Natrium units targeted for delivery by 2032, plus rights to as many as six additional Natrium units targeted for 2035. That is the basis for the “up to eight” framing now circulating in headlines.
It is not a trivial detail. There is a real difference between backing two units and holding rights on six more. Options create a path for expansion, but they are not the same thing as fully committed builds. Permitting, financing, site work, supply chains and construction all still matter, and in nuclear they matter a great deal. Meta is clearly trying to secure first-mover access if TerraPower’s model works at scale. That is a strategic position, not a guarantee of eight finished plants.
Seen that way, the TerraPower piece says more about how big Meta thinks its future power needs will be than it does about what will definitely be operating by the mid-2030s. The company is reserving room. It is buying optionality while the rest of the market is still arguing over whether the grid can keep up.
Oklo adds another Ohio bet around the Prometheus buildout
The agreement with Oklo is tied to a proposed 1.2 gigawatt advanced nuclear technology campus in Pike County, Ohio. Meta describes its support as helping finance development, including early procurement activity. Again, the pattern is familiar: secure a position now in exchange for a better chance of dependable supply later.
Ohio matters here because Prometheus is not an abstract concept. Meta has pointed directly to New Albany as a key use case, which makes the state a focal point in the company’s AI energy planning. Oklo’s proposed campus fits that geography, even if the usual development hurdles still apply. A proposal is not a completed power station, and advanced nuclear companies still have to move from ambition to execution.
Meta is not waiting for reactors to exist before trying to claim their output.
That may be the clearest lesson in the whole announcement. The scramble is no longer just for chips, land and cooling. It is for electricity contracts that can survive an AI buildout measured in decades rather than quarters.
What the 6.6 gigawatt figure really tells us
Meta says the total package supports up to 6.6 gigawatts of new and existing clean energy by 2035. That is the number meant to signal scale, and it does. But it also compresses very different kinds of commitments into one neat figure: an operating fleet under a 20-year purchase agreement, a proposed advanced nuclear campus in development, and TerraPower units split between funded targets and future rights.
There is nothing dishonest about that, but investors, policymakers and anyone following AI infrastructure should read it carefully. The announcement is best understood as a layered power strategy. One layer brings electricity sooner through existing plants. Another tries to finance future supply around a specific AI hub. The last reserves the right to go much bigger if advanced nuclear technology reaches commercial reality on time.
That is why the package feels both substantial and unfinished. Meta has shown what large AI companies are willing to do for energy certainty. It has not shown that the hardest parts of nuclear development suddenly become easy because a tech company is willing to sign a large cheque.
The real story is about timing, not just scale
The easy version of this story is that Meta is helping build up to eight nuclear plants. The harder, more useful version is that the company is trying to match AI demand with a portfolio that stretches from late 2026 into the mid-2030s. Some of that power looks tangible. Some of it is still contingent. All of it reflects a simple reality: the next phase of AI competition will be shaped by who can secure power at industrial scale and hold onto it.
That makes the announcement worth watching, but for a reason more practical than the headlines suggest. Meta is not merely buying energy. It is buying time, position and a claim on capacity that many rivals will soon want as badly as it does.