AI's Nuclear Renaissance: Why Microsoft, Google, and Amazon Are Betting Billions on Nuclear Power for AI Data Centers

In September 2023, Microsoft signed a deal with Constellation Energy to reopen Three Mile Island Unit 1 — the reactor at the same site as the infamous 1979 partial meltdown — to power Microsoft's AI data centers in Pennsylvania. In December 2023, Google signed a deal with Kairos Power to purchase electricity from small modular reactors (SMRs). Amazon invested in X-energy and other nuclear startups. In 2026, these deals are no longer novelties — they are the leading edge of what analysts are calling a nuclear renaissance driven entirely by the AI industry's unprecedented energy hunger. Understanding why this is happening requires understanding the physics of AI computing and why the AI industry, of all industries, has become nuclear power's most enthusiastic new customer.

Why AI Data Centers Need Nuclear Power: The Physics Problem

A single NVIDIA H100 GPU — the primary chip used to train and run large AI models — consumes approximately 700 watts of power. A rack of 8 H100s consumes more electricity than an average American home. A hyperscale AI training cluster containing 100,000 GPUs consumes approximately 100 megawatts — enough electricity for about 80,000 homes. The AI industry is building dozens of these clusters simultaneously. The cumulative power demand is extraordinary: the International Energy Agency estimates that data center electricity consumption will more than double by 2030, driven primarily by AI. The US electric grid, already strained in many regions, cannot absorb this demand from traditional sources without significant new power generation.

• Solar and wind cannot solve AI's power problem alone: solar and wind are intermittent — they generate power only when the sun shines and the wind blows. AI data centers require 24/7/365 power availability with extremely high reliability (99.999% uptime is standard). Solar and wind require battery storage or gas backup to achieve that reliability, significantly increasing cost and complexity.
• Natural gas is available but carbon-intensive: tech companies have made aggressive carbon-neutral commitments. Microsoft, Google, and Amazon all have pledges to be powered by 100% clean energy, with net-zero targets. Large-scale natural gas use for AI data centers conflicts with these commitments.
• Nuclear provides what AI needs: nuclear power plants run 24/7 at near-100% capacity factors, generate zero direct carbon emissions, and produce some of the most reliable electricity in the grid. A nuclear plant that is operating produces power regardless of weather, time of day, or season. For AI data centers, nuclear is essentially the perfect power source — if the economics and public perception challenges can be managed.

The Nuclear Deals Reshaping Energy in 2026

• Microsoft / Three Mile Island: Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1. The reactor, which had been shut down in 2019 for economic reasons, reopened in September 2024. It now generates approximately 835 megawatts of carbon-free electricity, primarily for Microsoft's AI infrastructure. This was the first major US nuclear restart in decades and proved the commercial viability model for subsequent deals.
• Google / Kairos Power: Google signed an agreement to purchase electricity from small modular reactors (SMRs) to be built by Kairos Power, with first delivery targeted for 2030. SMRs are smaller, factory-built versions of traditional nuclear reactors (50–300 megawatts vs. 1,000+ megawatts for a conventional plant) that can be sited closer to data centers and built faster than conventional nuclear.
• Amazon / X-energy and others: Amazon has invested in multiple nuclear startups and signed agreements to support nuclear development near its data center clusters. Amazon Web Services' data center regions in Virginia, Ohio, and the Pacific Northwest are all near or exploring nuclear power procurement.
• Oracle's nuclear ambitions: Oracle CEO Larry Ellison stated publicly that Oracle plans to power a cluster of 'a little over a gigawatt' of data center capacity with nuclear energy. A gigawatt of nuclear power serves roughly 800,000 homes — illustrating the scale of power demand these AI facilities represent.
• Oklo and other advanced reactor startups: Sam Altman (OpenAI's CEO) is a major investor in Oklo, a startup developing small modular reactors. The alignment between the AI industry's power needs and nuclear startup investment is not coincidental — it is a vertically integrated energy strategy.

Small Modular Reactors: The Technology That Makes the AI-Nuclear Connection Practical

Traditional nuclear power plants are massive, expensive, and slow to build — a conventional 1,000-megawatt plant costs $6–$10 billion and takes 10–15 years to build in the US. Small modular reactors (SMRs) are designed to change this economics equation: smaller (50–300 MW), factory-manufactured in standardized modules, faster to deploy, and potentially sitable directly next to industrial facilities including data centers. The NRC approved NuScale Power's design in 2022 — the first SMR approval in US history. Multiple other designs are in the approval pipeline. The commercial deployment of SMRs in the late 2020s would give AI companies the ability to effectively power their own data centers with dedicated nuclear capacity.

The Environmental Paradox: Is AI's Nuclear Turn Actually Green?

Nuclear power produces essentially zero operational carbon emissions. Lifecycle carbon analysis — including uranium mining, plant construction, and decommissioning — gives nuclear one of the lowest carbon footprints per kilowatt-hour of any energy source, lower than solar and wind in most lifecycle analyses. From a pure carbon perspective, AI companies choosing nuclear over natural gas is environmentally beneficial. The complications: nuclear waste storage (the US still lacks a permanent high-level nuclear waste repository), the very small but non-zero risk of accidents, and the opportunity cost argument (that the capital going to nuclear could build even more renewable capacity per dollar). The tech industry's position is that nuclear's reliability and 24/7 availability make it a necessary complement to intermittent renewables for baseload data center power, not a competitor.