Meta’s decision to lock in nuclear power for its Prometheus AI supercluster isn’t just an energy deal—it’s a sign that the limits of AI growth are no longer theoretical. Power, not talent or chips, is quickly becoming one of the defining constraints of the AI race, and Meta is moving early to secure it.
The company has signed agreements with three nuclear power providers—Vistra, TerraPower, and Oklo—to supply electricity for the massive data center it’s building in New Albany, Ohio. Together, the projects are expected to add roughly 6.6 gigawatts of power by 2035, far more than the total electricity demand of some U.S. states. Prometheus, which Meta says will come online in 2026, is a cornerstone of Mark Zuckerberg’s push toward increasingly advanced AI systems.
At a basic level, the logic is straightforward. Modern AI models require enormous amounts of continuous electricity to train and operate. Data centers already strain regional grids, particularly in places like the PJM interconnection, which serves much of the Midwest and Mid-Atlantic and has become saturated with new data center demand. Buying power on the open market is no longer enough. Meta’s response is to go directly to the source.
What’s more revealing is how Meta is doing it. Part of the deal relies on existing nuclear plants operated by Vistra in Ohio and Pennsylvania—among the cheapest and most reliable sources of baseload power available. But Meta is also making a long-term bet on next-generation nuclear startups. Oklo and TerraPower are developing small modular reactors, or SMRs, that promise consistent power with faster deployment and, eventually, lower costs through mass manufacturing. Whether those economics hold up remains unproven, but Meta’s commitments give those companies a rare opportunity to scale.
This approach marks a shift in how AI companies think about infrastructure. Energy is no longer a background operating cost—it’s a strategic asset. By securing dedicated nuclear capacity, Meta reduces its exposure to grid congestion, price volatility, and regulatory delays. In effect, it’s building a customized energy pipeline for AI.
That creates a competitive dynamic that’s hard to ignore. Power is finite, and nuclear capacity in particular is scarce. When a company like Meta locks in long-term supply, it effectively removes that energy from the broader market. Rivals without similar arrangements may find themselves competing not just for chips and engineers, but for electricity itself. Access to power becomes a new kind of moat—one that favors companies with the capital, political relationships, and long-term planning horizon to negotiate deals like these.
It also helps explain why nuclear has emerged as the preferred option. Unlike solar or wind, nuclear delivers steady, round-the-clock power—exactly what massive AI clusters need to operate efficiently. And unlike fossil fuels, it aligns with corporate climate commitments. Meta, Amazon, and Google have all publicly supported expanding nuclear generation, signaling that this isn’t a one-off move but part of a broader industry shift.
There are broader implications, too. Nuclear power is tightly regulated, politically sensitive infrastructure. Meta’s ability to secure multiple agreements—some involving plants still years from completion—highlights how closely AI development is becoming intertwined with public policy and national energy strategy. As AI grows more central to economic and geopolitical competition, control over the energy that fuels it starts to look like a form of strategic leverage.
Environmental questions also linger. Nuclear power is carbon-free at the point of generation, but expanding capacity takes time and comes with its own political and social tradeoffs. Betting on nuclear doesn’t eliminate AI’s environmental footprint so much as relocate it deeper into the energy system. Whether that tradeoff proves sustainable will depend on how quickly new capacity can actually be delivered.
Looking ahead, Meta’s move is likely a preview of what’s coming next. Other AI labs and hyperscalers will almost certainly pursue similar deals, intensifying competition for reliable power. Energy efficiency will matter more, not less, as electricity becomes a limiting factor. And infrastructure decisions—where data centers are built, how they’re powered, and who controls that power—will shape the future of AI as much as model architecture or training techniques.
In that sense, Meta’s nuclear agreements reflect a maturing AI industry. The next phase of competition won’t be driven solely by algorithms and compute budgets, but by access to real-world resources: energy, land, cooling, and regulatory approval. To understand where AI is headed, it’s no longer enough to watch benchmarks and research papers. Increasingly, the most important signals are coming from power markets, reactor projects, and long-term infrastructure deals.
This analysis is based on reporting from CNBC.
Image courtesy of Unsplash.
This article was generated with AI assistance and reviewed for accuracy and quality.
