Meta just made one of the biggest bets yet that nuclear will power the AI boom.
The company has signed longâterm deals with three nuclear providersâVistra, Oklo and TerraPowerâfor more than 6 gigawatts (GW) of electricity to run its data centers in the U.S.
Thatâs utilityâscale power, on par with several large nuclear plants combined.
Three suppliers, two strategies
Metaâs announcement bundles three very different kinds of nuclear deals:
- Vistra â a major U.S. power producer selling output from existing reactors
- Oklo â a young small modular reactor (SMR) startup
- TerraPower â Bill Gatesâcoâfounded SMR and advanced nuclear company
The deals follow a request for proposals Meta issued in December 2024, where it asked for partners that could add 1â4 GW of new capacity by the early 2030s. Much of this power will move through PJM Interconnection, the massive grid that spans 13 MidâAtlantic and Midwestern states and is already crowded with data centers.
Vistra: immediate impact from legacy plants
Vistraâs 20âyear power purchase agreement is the fastest way for Meta to get lowâcarbon baseload power.
- Meta will buy 2.1 GW from two existing Ohio nuclear plants: Perry and DavisâBesse.
- As part of the deal, Vistra will add capacity at both of those plants plus its Beaver Valley plant in Pennsylvania.
- Those upgrades will deliver an additional 433 megawatts (MW), scheduled to come online in the early 2030s.
Electricity from reactors that are already built and operating is generally among the cheapest firm power resources on the grid, which means the Vistra tranche is almost certainly the lowestâcost piece of Metaâs nuclear package.
Financial terms werenât disclosed, but the economics here are straightforward: keep existing plants running harder and longer, and sell that dependable output to hyperscalers that canât afford to have their AI clusters blink.
Oklo: Meta backs a highârisk SMR startup
The second deal is a straight bet on an unproven SMR player.
Meta will buy 1.2 GW from Oklo, which is promising to start sending power to the grid as early as 2030.
- Oklo went public via SPAC in 2023.
- It already has a large agreement with data center operator Switch.
- But it has struggled to get its reactor design approved by the U.S. Nuclear Regulatory Commission (NRC).
For Meta, thatâs a clear regulatory risk. For Oklo, itâs the opportunity itâs been waiting for.
Okloâs Aurora Powerhouse reactors each produce 75 MW. To meet Metaâs 1.2 GW contract, the company will need to build more than a dozen reactors at a site planned in Pike County, Ohio.
The broader SMR thesis is on the line here: build lots of smaller, standardized reactors and drive down cost via manufacturing scale. Itâs a âplausible hypothesis,â as the TechCrunch piece notes, but one that still hasnât been tested in the real world.
Okloâs own target costs land in the $80â$130 per megawattâhour range for later units. Early builds are expected to be higher.
TerraPower: advanced nuclear plus builtâin storage
The most technically ambitious piece of Metaâs plan comes from TerraPower, the advanced nuclear startup coâfounded by Bill Gates.
TerraPower aims to deliver electricity to Meta starting around 2032.
Its design:
- Uses molten sodium to move heat from the reactor to the generator.
- Integrates an energy storage system where superheated salt can be held in insulated tanks when demand is low and tapped when demand spikes.
Each reactor:
- Generates 345 MW of electricity.
- Comes with storage able to deliver an additional 100â500 MW for more than five hours.
TerraPower is already working with GE Hitachi on its first commercial plant in Wyoming and appears to be navigating the NRC process more smoothly than some rivals.
Under the Meta deal:
- The first two reactors dedicated to Meta would provide 690 MW.
- Meta has rights to buy six more, for a total of 2.8 GW of nuclear capacity and 1.2 GW of storage.
If exercised, that would make Meta one of TerraPowerâs anchor customers for a whole fleet of advanced reactors.
TerraPowerâs longâterm cost target: $50â$60 per megawattâhour for later plants. Again, firstâofâaâkind units will be more expensive.
AI is forcing a rethink of energy strategy
The throughâline here is AI.
Training large models and running inference at scale have turned data centers into some of the hungriest power users on the grid. At the same time, tech companies have made climate commitments that make it harder to simply fall back on gasâfired generation.
Nuclear offers something wind and solar canât on their own: 24/7 carbonâfree baseload that maps neatly to alwaysâon AI workloads.
But nuclear also comes with:
- Regulatory risk (especially for startups like Oklo).
- Construction risk and delays for firstâofâaâkind plants.
- Uncertain SMR economics, where the promised cost curve is still theoretical.
By splitting its bets across existing reactors, early SMRs, and more advanced designs with storage, Meta is effectively building a diversified nuclear portfolio for the 2030s and beyond.
And with more than 6 GW now spoken for, itâs sending a clear signal to the rest of Big Tech: if you want to keep scaling AI, youâll need to lock down serious, longâduration powerâfast.



