Big Tech’s 18-month nuclear capacity contracts exceed the US total of the prior 20 years; SMR commercialization enters a critical window.
In early 2026, the first direct nuclear power purchase agreements (PPAs) between Big Tech and reactor operators are closing, marking a historic shift in how the world’s largest companies power artificial intelligence. Global data center electricity demand is projected to double by 2027, with AI workloads responsible for over 60% of that growth. Faced with grid interconnection queues stretching eight years and wholesale power prices surging 267% near hyperscale facilities, Microsoft, Amazon, and Google are bypassing public utilities entirely—signing direct 20-year PPAs to restart shuttered nuclear plants and bankroll small modular reactor (SMR) startups. This strategic pivot from efficiency tweaks to on-site generation marks a structural shift in both energy markets and AI infrastructure, with profound implications for electricity pricing, carbon targets, and the geography of compute.
The United States interconnection queue has ballooned to over 2,600 gigawatts as of early 2026, with average wait times approaching five years and project withdrawal rates nearing 80%. According to the International Energy Agency, 20% of planned data center projects globally are at risk due to grid congestion. In PJM, the largest U.S. wholesale electricity market, failure to connect new generation cost consumers an estimated $7 billion in a single capacity auction. For every $1 billion in transmission investments delayed, consumers lose $150–$370 million per year. Nearly half of U.S. AI data centers planned for 2026 are delayed, creating a 7-gigawatt gap that now bottlenecks $650 billion in hyperscaler capital expenditure. The grid interconnection crisis has emerged as the single greatest structural barrier to deploying new energy and data center capacity.
Microsoft is restarting the former Three Mile Island Unit 1 reactor—now renamed the Crane Clean Energy Center—via a $1.6 billion refurbishment with Constellation Energy. The 835-megawatt facility is expected online by 2027, with Microsoft signing a 20-year PPA covering 100% of its output. The company has also inked a power purchase agreement with fusion startup Helion Energy, targeting commercial operations later this decade. Microsoft’s nuclear power purchase agreements represent the largest corporate commitment to baseload nuclear in history.
Amazon has secured 1.92 gigawatts of capacity from the Susquehanna nuclear plant in Pennsylvania, acquiring the Cumulus Data Center campus for direct access to the facility. The company is also partnering with Energy Northwest and X-energy on SMR development, aiming to deploy up to 300 megawatts of advanced nuclear capacity by 2030. Amazon Web Services (AWS) has stated that nuclear power is essential to meeting its 2040 net-zero carbon goals while supporting AI workloads that require 300–500 megawatts per facility—comparable to a mid-sized city.
Google signed the first corporate small modular reactor agreement with Kairos Power in 2025, targeting 500 megawatts of capacity by 2030. The deal uses a novel “order book” model where Google commits to purchasing power from multiple SMR units as they come online, providing the revenue certainty needed for Kairos to secure financing and regulatory approvals. Google’s corporate SMR agreements are being closely watched as a template for future tech-nuclear partnerships.
Meta has announced plans to secure up to 6.6 gigawatts of nuclear capacity across partnerships with Vistra, Oklo, and TerraPower. The social media giant’s aggressive push into nuclear reflects the reality that one-third of data centers are expected to be fully off-grid by 2030, according to industry analysts. Meta’s off-grid data center strategy includes colocation with nuclear plants and direct transmission lines.
Small modular reactors are moving from theoretical promise to commercial reality in 2026, supported by $1.3 billion in equity funding in 2025, Department of Energy-backed deployments, and the first North American SMR receiving final approval. Companies like X-energy, Kairos Power, Oklo, and TerraPower are leading the race, with Big Tech providing both capital and guaranteed offtake. SMRs offer distinct advantages, including factory fabrication, shorter construction timelines (three to five years versus seven to ten for large reactors), and scalable deployment ranging from 50 MW to 300 MW per unit. However, levelized cost of electricity (LCOE) estimates for first-of-a-kind SMRs range from $100–$180/MWh, significantly higher than existing nuclear ($30–$60/MWh) or renewables ($20–$50/MWh). SMR economics remain a key challenge, with analysts projecting costs to decline only after more than 10 gigawatts of cumulative deployment.
The strategic pivot to nuclear carries profound implications. Wholesale power prices near hyperscale facilities have surged 267% since 2020, driven by demand from AI data centers that now consume over 500 terawatt-hours annually—exceeding France’s total electricity consumption. By bypassing public grids, Big Tech avoids contributing to these price spikes, though critics argue that residential ratepayers may end up subsidizing the grid infrastructure that tech giants continue to rely on for backup and peaking power. On carbon targets, nuclear’s greater-than-90% capacity factor and zero emissions make it an attractive complement to intermittent renewables. Microsoft, Amazon, and Google have all reaffirmed their 2030 net-zero goals, relying on nuclear to provide firm, round-the-clock carbon-free power that solar and wind cannot guarantee. Nevertheless, the nuclear-versus-renewables debate intensifies as some environmental groups oppose diverting investment from solar and wind toward nuclear projects.
“Electricity supply, not chip supply, has become AI’s binding constraint,” said Benjamin Rossi, energy analyst at the Global Energy Institute. “The interconnection queue crisis means that even if you have the capital and the GPUs, you cannot power them without a direct line to a baseload plant. Nuclear is the only option that scales to 500 MW per facility with 90%+ uptime.” Rossi notes that the trend is creating a parallel energy economy, with hyperscalers effectively becoming their own utilities. “We are seeing the privatization of the grid’s most valuable customers. This has profound implications for utility business models and regulatory frameworks.”
Why are AI data centers turning to nuclear power? AI data centers require massive, continuous power—300–500 MW per facility—that intermittent renewables like solar and wind cannot reliably provide. Nuclear offers a greater-than-90% capacity factor, zero emissions, and 20-year PPA stability, making it the only scalable baseload option for hyperscalers facing grid interconnection delays of five to eight years.
Which Big Tech companies are signing nuclear PPAs? Microsoft (Three Mile Island restart, Helion fusion), Amazon (Susquehanna 1.92 GW, X-energy SMRs), Google (Kairos Power SMR), and Meta (up to 6.6 GW across Vistra, Oklo, TerraPower) have all signed or announced nuclear power purchase agreements in 2025–2026.
What is the interconnection queue crisis? The U.S. interconnection queue has grown to over 2,600 GW of projects waiting for grid connection, with average wait times of five years and 80% withdrawal rates. This backlog stems from insufficient transmission capacity, regulatory bottlenecks, and transformer shortages, delaying nearly half of planned AI data centers.
How do small modular reactors differ from traditional nuclear plants? SMRs are factory-fabricated reactors rated at 50–300 MW per unit, designed for shorter construction periods (three to five years), lower upfront capital, and scalable deployment. They utilize advanced cooling technologies (molten salt, sodium, helium) and passive safety systems, but currently carry higher LCOE than large reactors or renewables.
Will nuclear power help Big Tech meet carbon targets? Yes. Nuclear provides firm, carbon-free electricity around the clock, complementing intermittent renewables. Microsoft, Amazon, and Google all include nuclear in their 2030 net-zero strategies. Critics, however, argue that investment in nuclear diverts resources from faster-to-deploy solar and wind, and note that nuclear waste and proliferation risks remain unresolved.
The pivot to nuclear is reshaping where AI infrastructure is built. Data centers are collocating with existing nuclear plants in Pennsylvania, Virginia, and the Midwest, while SMR startups are planning deployments near major load centers. This trend may concentrate compute capacity in regions with established nuclear infrastructure, potentially exacerbating geographic inequalities in AI access. As the first direct nuclear PPAs close in early 2026, the message is clear: energy supply, not chip availability, has become the defining constraint for the next era of artificial intelligence.