AI demand accelerates nuclear power shift with long-term investments and modular reactors

The artificial intelligence boom is reshaping power markets with a speed that has caught even seasoned energy investors off guard. For hyperscale operators, the bottleneck is no longer just chips, talent or land; it is finding enough dependable, carbon-free electricity to keep vast data centres running around the clock. That shift is pushing Big Tech into a role once reserved for utilities and large industrial off-takers: the long-term backer of nuclear generation.

What makes this moment significant for industrial decarbonisation is not simply the return of interest in atomic power, but the way demand is being re-written by the digital economy. AI workloads are far more electricity-intensive than conventional cloud computing, and the industry’s appetite for uninterrupted supply is difficult to match with variable renewables alone. Batteries can smooth short fluctuations, but they do not yet solve the challenge of providing firm power at the scale needed for multi-hundred-megawatt campuses. Nuclear, and particularly smaller reactors designed for modular deployment, is increasingly being presented as the closest fit.

The clearest sign of that transition is the willingness of technology companies to commit capital decades in advance of output. Microsoft’s 20-year power purchase agreement with Constellation Energy to help restart Unit 1 at Three Mile Island in Pennsylvania has become a reference point for the sector. The reactor, which shut in 2019 for economic reasons, is being brought back with an estimated $1.6 billion investment, underpinned by Microsoft’s demand. The deal illustrates how highly prized already-built nuclear assets have become when compared with the cost, timing and execution risk of new build.

Alphabet has taken a different route, backing the development of advanced modular reactors rather than leaning on legacy capacity. Its agreement with Kairos Power is intended to support up to 500MW of nuclear generation by the early 2030s, a wager on first-of-a-kind technology that uses the strength of a balance sheet to help absorb licensing and manufacturing risk. Amazon, meanwhile, has adopted a multi-pronged strategy. It has anchored a $500 million investment in X-energy and agreed with Energy Northwest to develop 320MW across four small modular reactors in Washington state. The company has also pursued proximity to existing nuclear generation, including a large data centre site adjacent to the Susquehanna plant in Pennsylvania, reflecting a growing interest in so-called power-behind-the-meter arrangements that reduce exposure to grid congestion.

Small modular reactors are central to this narrative because they match the way data centre demand grows. Rather than waiting for a single giant plant to be completed, operators can add capacity in stages, aligning new generation with each step in compute expansion. That modularity is a practical advantage for campuses that may begin at 100MW and grow several times over. It also has industrial relevance beyond the technology sector, because the same attributes that appeal to cloud operators , compact footprint, factory-style fabrication and shorter deployment timelines , are increasingly attractive to heavy industry searching for low-carbon baseload supply.

For uranium producers and fuel-cycle companies, the implications are substantial. The technology firms are not buying uranium directly, but their long-term power contracts improve the visibility that utilities and reactor developers need before they commit to fuel supply deals. That matters even more for advanced reactor designs that require high-assay low-enriched uranium, or HALEU, a specialised fuel that remains in short supply. As the Western world works to build non-Russian enrichment capacity, the race to secure fuel has become as strategically important as the race to secure reactor orders.

The broader market is also being re-rated by the perception that Big Tech demand is unusually price-insensitive. Unlike traditional utilities, which tend to buy nuclear power through slower and more conservative procurement cycles, hyperscalers are making commitments linked to their own rapid growth trajectories. That creates a new class of demand for uranium miners and developers, one that is less tied to the old utility calendar and more closely aligned with the expansion of AI infrastructure.

There are still major obstacles. Licensing remains slow and technically demanding, especially for first-of-a-kind reactor designs, and cost overruns remain a persistent risk in nuclear construction. Fuel supply is another pressure point, with restrictions on Russian enrichment accelerating the search for Western alternatives such as Urenco and Orano. Even so, the direction of travel is clear: the data centre sector’s need for 24-hour carbon-free power is helping to turn nuclear from a long-discussed option into a funded industrial strategy.

For mining and decarbonisation professionals, the significance lies in timing. The first commercial wave of small modular reactors may not reach full scale until the early 2030s, but the financing commitments are being made now. In effect, the AI build-out is shortening the adoption curve for nuclear and creating a more durable floor under uranium demand. What began as a power problem for the digital economy is becoming a long-term investment case for the entire nuclear fuel cycle.