Cloud Giants AWS and Azure Turn to Nuclear Power to Secure AI Data Center Capacity

The artificial intelligence industry has a fundamental physics problem. Training frontier large language models and serving billions of generative queries requires an uninterrupted, massive supply of electricity that traditional infrastructure was never designed to handle. For years, Silicon Valley relied on purchasing renewable energy credits to offset its carbon footprint, allowing tech giants to claim net-zero operations on paper. However, this accounting maneuver masked a stark physical reality: intermittent solar and wind power simply cannot provide the 24/7 baseload generation required to keep hyperscale data centers running without fail. As the AI arms race accelerates, the illusion of a purely renewable cloud is fracturing, forcing the industry to seek out energy sources that are both entirely carbon-free and relentlessly reliable.[1]

Global electricity demand from data centers is projected to surge from 460 terawatt-hours in 2024 to over 1,300 terawatt-hours by 2035, driven almost entirely by the computational density of AI workloads. A single modern hyperscale AI facility can easily consume between 300 and 500 megawatts of continuous power—roughly the equivalent of the total electricity consumption of a mid-sized American city. Facing severe grid constraints, multi-year transmission delays, and their own ambitious net-zero climate pledges, the world's largest cloud providers have realized that the only viable solution to their energy bottleneck is a technology invented in the 1950s. Nuclear power has transformed overnight from a heavily regulated, stagnant sector into the centerpiece of Silicon Valley's infrastructure strategy.[1]

Over the past year, major technology companies have quietly contracted for more than 10 gigawatts of new nuclear capacity in the United States alone. This unprecedented capital influx represents a fundamental shift in corporate strategy. Cloud giants like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud are no longer operating merely as software and infrastructure providers; they are actively reshaping the global energy market. By funding the revival of dormant nuclear plants and bankrolling the development of next-generation reactor technologies, these companies are vertically integrating their power supply chains. They are transitioning from passive energy consumers relying on public utilities to active infrastructure developers willing to underwrite the immense financial risks of atomic energy.[6]

Microsoft has opted for a strategy prioritized around speed and proven infrastructure, making headlines with a historic move to resurrect a symbol of America's complicated nuclear past. In a landmark $16 billion agreement with Constellation Energy, the software giant committed to purchasing 100 percent of the output from the reactivated Three Mile Island Unit 1 reactor in Pennsylvania. The 835-megawatt facility, which operated safely for decades and was completely separate from the infamous 1979 partial meltdown of the adjacent Unit 2, is slated to reopen in 2028. Upon its revival, the facility will be rebranded as the Crane Clean Energy Center, dedicated entirely to fueling Microsoft's expanding AI clusters.[5]

By securing a massive 20-year power purchase agreement, Microsoft provided Constellation Energy with the exact revenue certainty required to justify the billion-dollar capital investment needed for the restart. This pragmatic approach bypasses the decade-long regulatory hurdles, supply chain bottlenecks, and construction delays typically associated with building new nuclear facilities from scratch. For Microsoft, restarting an existing, proven reactor offers a relatively rapid path to securing nearly a gigawatt of firm, carbon-free power. It is a calculated bet that the fastest way to solve the AI energy crisis is to leverage the heavy industrial infrastructure that the previous generation left behind.[1]

Amazon, conversely, is pursuing a strategy focused on massive scale and direct, physical control over its power supply. AWS initiated its nuclear pivot by acquiring a sprawling data center campus adjacent to the Susquehanna Steam Electric Station in Pennsylvania from Talen Energy for $650 million. Rather than stopping at the initial purchase, Amazon subsequently announced a staggering $20 billion investment plan to convert the site into a dedicated, state-of-the-art AI campus. This facility will be powered directly by the 2.5-gigawatt conventional nuclear plant, piping electrons straight from the reactor to the server racks without ever touching the public grid.[5][6]

This 'behind-the-meter' direct-connection model allows Amazon to completely bypass congested regional transmission grids. By co-locating its computing infrastructure directly alongside firm generation, AWS avoids the transmission bottlenecks, substation shortages, and utility upgrade delays that are increasingly stalling data center deployments across the country. Cementing this strategy, Amazon expanded its relationship by signing a long-term agreement with Talen Energy to purchase up to 1,920 megawatts of carbon-free electricity through 2042. In an era where proximity to power is becoming a critical competitive advantage, Amazon is ensuring its AI infrastructure has priority access to the ultimate baseload source.[1][3]

Beyond tapping into existing conventional reactors, the cloud giants are heavily subsidizing the experimental frontier of atomic energy: Small Modular Reactors (SMRs). SMRs are advanced nuclear reactors designed to be manufactured in centralized factories and shipped to sites for modular assembly. Theoretically, this approach will drastically reduce the exorbitant capital costs, bespoke engineering requirements, and crippling construction timelines that have plagued traditional large-scale nuclear projects. By shrinking the physical footprint and standardizing the components, the tech industry hopes to turn nuclear reactors into scalable, deployable products rather than multi-billion-dollar bespoke civil engineering megaprojects.[4][5]

Amazon has aggressively positioned itself at the forefront of this SMR push, committing over $500 million to lead an investment round in X-energy, a developer of advanced high-temperature gas-cooled reactors. In Washington state, AWS is partnering directly with the regional utility Energy Northwest to deploy up to 12 of X-energy's modular reactors at the new Cascade Advanced Energy Facility. This ambitious project aims to generate 960 megawatts of power by the early 2030s. By backing multiple SMR modules, Amazon eliminates the single points of failure inherent in massive conventional plants, building a resilient, distributed power network tailored specifically for industrial-scale computing.[2][3]

Google is taking a similarly forward-looking approach, prioritizing deep technological innovation over immediate deployment. The search giant signed a milestone agreement with Kairos Power to deploy a fleet of experimental molten-salt SMRs, targeting 500 megawatts of capacity between 2030 and 2035. Unlike conventional reactors that rely on highly pressurized water for cooling, Kairos's novel design uses molten fluoride salt combined with ceramic, pebble-type fuel. This system operates at low pressures and offers inherent, physics-based safety advantages, theoretically making meltdowns impossible. Google's willingness to back a technology that has never been commercially deployed underscores the industry's desperation for scalable clean energy.[5]

Fascinatingly, the convergence of artificial intelligence and nuclear energy is becoming deeply symbiotic. While nuclear power is required to fuel AI data centers, artificial intelligence is now being actively deployed to accelerate the notoriously sluggish nuclear development process. Microsoft and Nvidia recently launched a joint initiative utilizing generative AI and digital twin simulations to streamline the complex nuclear lifecycle, from initial permitting through to daily operations. The tech giants are effectively using their own computational breakthroughs to solve the infrastructure bottlenecks holding back their physical expansion.[2]

Historically, licensing a new reactor design in the United States involves tens of thousands of pages of regulatory submissions, fragmented engineering data, and years of manual bureaucratic review. By applying specialized large language models to draft these massive documents and conduct automated gap analyses, developers can drastically reduce regulatory friction. Aalo Atomics, an Austin-based SMR startup, recently reported reducing its permitting process workload by an astonishing 92 percent using Microsoft's generative AI tools. This acceleration not only shaves years off deployment timelines but saves an estimated $80 million annually in administrative and legal overhead.[2]

Furthermore, Nvidia's Omniverse platform is being utilized to create high-fidelity 4D and 5D digital twin simulations of nuclear plants long before ground is ever broken. This technology allows engineers to virtually build the entire facility, adding time scheduling and cost tracking to standard 3D spatial models. By tracking physical progress against the digital plan, construction teams can catch schedule collisions early and model the downstream effects of design changes. The goal is to optimize the construction process virtually, mitigating the risk of the multi-billion-dollar cost overruns that have historically crippled the Western nuclear industry.[2]

Despite the unprecedented influx of tech capital and AI-driven optimization, significant uncertainties remain. The commercial viability of Small Modular Reactors is entirely unproven at scale. Factory fabrication facilities must still be built, entirely new supply chains for advanced fuels like High-Assay Low-Enriched Uranium (HALEU) must be established, and the U.S. Nuclear Regulatory Commission must approve novel reactor designs that deviate from decades of water-cooled precedent. The tech industry is betting billions on a timeline that assumes these experimental reactors will transition smoothly from laboratory prototypes to commercial deployment without the delays that typically plague nuclear engineering.[2][6]

Cost remains another major, unresolved variable. While Microsoft's strategy of restarting Three Mile Island is estimated to cost a relatively efficient $1,900 per kilowatt of capacity, new SMR construction could run anywhere between $6,000 and $12,000 per kilowatt, reflecting the high capital costs of first-of-a-kind deployment. These elevated capital requirements mean that nuclear energy will carry a significant premium over natural gas generation. This premium is currently justifiable only by the hyperscalers' strict zero-carbon mandates and their desperate, price-inelastic need for absolute 24/7 reliability.[5]

Ultimately, the technology industry's embrace of nuclear power represents a pragmatic, high-stakes acknowledgment of physical limits. The digital economy cannot expand indefinitely on a fragile, weather-dependent grid. By vertically integrating their energy supply chains, bypassing traditional utilities, and underwriting the next generation of atomic technology, cloud providers are taking their destiny into their own hands. They are ensuring that the artificial intelligence revolution will be powered by the most energy-dense, reliable fuel source on the planet—fundamentally altering the trajectory of global energy infrastructure in the process.[1]