Meta’s nuclear power push reshapes AI energy race

The arrangements, involving power supply and development pathways with Vistra, Constellation, TerraPower and Oklo, aim to underpin Meta’s expanding network of AI-focused data centres, often described by the company as “superclusters”. Executives argue that nuclear generation offers the only realistic combination of scale, reliability and carbon-free output capable of supporting next-generation AI systems that operate continuously and consume vast amounts of electricity.

Meta’s push places it at the forefront of a broader shift among technology firms, which are confronting the limits of renewable-only strategies as data centre loads climb sharply. Training large language models and running inference at global scale requires power that is not just clean, but firm and predictable. Nuclear plants, with capacity factors far higher than wind or solar, are increasingly viewed as the backbone for such operations.

Under the framework of the deals, Meta is not directly building or owning nuclear plants but is locking in long-term access to output and supporting development pipelines. Vistra and Constellation operate some of the largest nuclear fleets in the United States, while TerraPower and Oklo represent a newer generation of companies developing advanced reactors and small modular reactors, or SMRs. These technologies promise lower upfront costs, enhanced safety features and the flexibility to be sited closer to industrial loads, including data centres.

The scale involved is striking. A commitment of up to 6.6 GW rivals the output of several large conventional power stations and would be sufficient to supply millions of homes. For Meta, the figure reflects projections that its AI compute requirements will multiply over the coming decade as models grow larger, more capable and more deeply embedded across consumer and enterprise products.

Industry analysts say the move signals a decisive turn in how technology companies approach energy security. For years, corporate clean-energy strategies were dominated by power purchase agreements tied to wind and solar farms. While those contracts helped accelerate renewable deployment, they often relied on grid balancing from fossil fuels when generation dipped. Nuclear power, by contrast, aligns more closely with the always-on nature of hyperscale computing.

Regulators and policymakers are watching closely. Nuclear projects face lengthy approval processes, complex financing structures and public scrutiny around safety and waste management. Supporters counter that advanced reactor designs and life-extension programmes for existing plants can be deployed faster than building new large-scale facilities from scratch, especially when anchored by long-term corporate demand.

The partnerships also highlight a changing relationship between utilities and corporate customers. Rather than acting solely as ratepayers, companies like Meta are becoming strategic counterparties that can de-risk large infrastructure investments. Guaranteed demand from a single, creditworthy buyer can make financing easier and accelerate timelines, particularly for emerging reactor technologies that have struggled to move beyond demonstration stages.

Competition is intensifying across the sector. Other technology firms are exploring similar pathways, weighing nuclear options alongside investments in grid-scale batteries, hydrogen and geothermal power. The difference with Meta’s approach lies in its breadth, spanning both legacy nuclear operators and next-generation developers, effectively hedging technological and regulatory risks.

Environmental groups have offered mixed reactions. Some welcome the emphasis on carbon-free baseload power as a means to curb emissions from the fast-growing digital economy. Others warn that nuclear expansion should not divert attention from efficiency gains, demand management and the rapid scaling of renewables paired with storage. Meta has said that energy efficiency remains a priority, but acknowledges that efficiency alone cannot offset the explosive growth in compute demand.