Big Tech's Nuclear Bet for AI Data Centers, Examined

What's happening

Over the past year, the biggest names in cloud computing have rushed to attach their AI ambitions to nuclear power. Amazon announced small modular reactor (SMR) agreements in October 2024, framed publicly as a step toward carbon-free energy 1. Amazon also expanded a power purchase agreement with Talen Energy to draw electricity directly from the Susquehanna nuclear plant for a data center campus built next door in Pennsylvania 2. Microsoft struck a 20-year deal with Constellation Energy to revive Unit 1 at Three Mile Island and dedicate its output to Microsoft's mid-Atlantic data centers 4. Google partnered with Kairos Power to site an advanced nuclear plant in Tennessee, targeting electricity delivery to its southeastern U.S. data centers by 2030 5. Meta went further still, signing three separate deals with TerraPower, Oklo, and Vistra to supply its Prometheus AI data center in New Albany, Ohio, with power described as enough to serve roughly 5 million homes 7. Even Equinix has said it's exploring multiple alternative-energy partners, with nuclear featuring prominently in that outreach 5. Reuters describes this collectively as Big Tech reshaping the funding landscape for next-generation nuclear technology, giving reactor developers both capital and a credible path to revenue 6.

The appeal is straightforward: AI data centers run continuously and cannot tolerate the intermittency of solar and wind. Nuclear plants generate power around the clock regardless of weather, which is why IAEA head Rafael Mariano Grossi has argued that only nuclear energy satisfies the combined needs of low-carbon generation, constant reliability, high power density, grid stability, and scalability 3. The International Atomic Energy Agency itself frames the trend as tech companies chasing two goals simultaneously — meeting surging demand and satisfying sustainability commitments — and turning to SMRs alongside renewables to do it 9.

Beyond conventional reactor projects, more unconventional ideas have surfaced. One proposal, first reported by Bloomberg, floated repurposing the decommissioned nuclear reactors from the USS Nimitz for land-based AI data center power, with backers estimating conversion costs of $1 million to $4 million per megawatt — far below the $8 million to $10 million per megawatt MIT researchers estimated for building a new large reactor from scratch 10.

Where the reporting agrees

Every outlet in this roundup agrees on the basic shape of the story: AI-driven electricity demand is pushing hyperscalers toward nuclear power, and the deals now span the industry's biggest players — Amazon, Microsoft, Google, and Meta — each pursuing SMRs, restarted legacy reactors, or both 12457. There is also broad agreement on the underlying rationale: nuclear's steady, weather-independent output matches the always-on demands of AI computing in a way intermittent renewables cannot 389. And multiple sources converge on the observation that these corporate commitments are doing real financial work for the nuclear industry — turning speculative reactor technology into something investors and developers can treat as bankable 36.

Where it doesn't

The coverage diverges most sharply on tone and skepticism. The Bulletin of the Atomic Scientists is openly critical, framing Amazon's carbon-free messaging as something the public shouldn't simply accept, and it raises pointed concerns about the cost and accident risk of restarting old reactors — treating that option as something companies can only pursue if they get the public to look away from those risks 1. Forbes and the IAEA, by contrast, largely present the nuclear pivot as a rational, even necessary, response to grid realities, quoting Grossi's case for nuclear's unique suitability rather than interrogating the risk side 39. Data Center Knowledge takes a more conditional middle position, suggesting nuclear could become central to hyperscale power planning only if a number of pieces fall into place — and warning that a single high-profile incident near a major campus could reverse the current optimism just as quickly 8.

The reporting also diverges on specifics that are hard to reconcile because they simply cover different deals: Amazon's Talen/Susquehanna agreement 2, Microsoft's Three Mile Island revival 4, Google's Kairos Power partnership in Tennessee 5, and Meta's three-way Ohio deal 7 are each singular transactions rather than competing accounts of the same event, so there's no factual contradiction — just a fragmented landscape of parallel bets. The Nimitz reactor-reuse proposal stands apart as the only source raising it, and its cost comparison against MIT's new-build estimate is presented without independent verification in the available coverage 10.

The read

Taken together, the sources support a clear pattern rather than a single contested fact: Big Tech's nuclear turn is real, well-funded, and industry-wide, but the coverage splits along a predictable line between outlets treating it as sound energy policy and outlets warning it doubles as a public-relations shield for costly, risk-laden infrastructure decisions. The disagreement isn't about what's happening — it's about whether anyone should be reassured by it.