Energy
Big Tech Nuclear Energy AI Data Centers — The $10 Trillion Power Grab
AI data centers now consume over 10% of US electricity. Big Tech isn't waiting for the grid — they're buying nuclear plants. Here's the full picture.
What Just Happened
Big Tech nuclear energy AI data centers are no longer a future strategy. They are happening right now, at a scale that is reshaping the entire US energy industry.
AI data centers consumed over 10% of total US electricity in Q1 2026 — up from approximately 4% in 2023. Goldman Sachs projects that figure will grow 160% by 2030. A single large-scale AI data center running 50,000 NVIDIA H100 GPUs draws 150 to 200 megawatts continuously — equivalent to powering 150,000 average American homes, around the clock, forever.
The US electricity grid was not built for this. And Big Tech isn't waiting for it to catch up.
In the span of 18 months, the seven largest AI companies in the world — Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI — have stopped buying power from utilities and started building their own. Microsoft restarted Three Mile Island under a 20-year, $16 billion agreement. Meta contracted up to 6.6 gigawatts of nuclear capacity across multiple deals. Amazon signed a 17-year, 1.92 gigawatt agreement with Talen Energy and a $50 billion partnership to deploy small modular reactors with X-energy. Google signed the largest corporate nuclear power purchase agreement in history with Kairos Power and acquired Intersect Power for $4.75 billion to vertically integrate into electricity generation entirely.
Big Tech has collectively contracted more than 10 gigawatts of new US nuclear capacity in the past year alone.
To put that in perspective: 10 gigawatts is roughly 10% of all nuclear power currently operating in the United States. These companies aren't just buying electricity. They are becoming the anchor tenants of a new nuclear industry — and in doing so, they are fundamentally changing how AI infrastructure gets built and who controls the energy that powers it.
The energy trade is now the AI trade. And the race to secure power has only just begun.
Big Tech Nuclear Energy AI Data Centers — The Deals in Full
The scale of what has been contracted in the past 18 months requires concrete numbers to understand.
Microsoft committed to a 20-year, 837-megawatt power purchase agreement with Constellation Energy to restart Three Mile Island — the Pennsylvania plant that became synonymous with nuclear disaster after its 1979 partial meltdown. The unit being restarted was not involved in that accident. It was shut down in 2019 due to operating losses. Microsoft's 20-year commitment provided the revenue certainty needed to justify the $16 billion restart investment. The plant is expected to come back online in 2028 under the name Crane Clean Energy Center, with Microsoft taking 100% of its output. Microsoft has now contracted 34.7 gigawatts of clean power total, surpassing Amazon as the largest corporate clean energy buyer on earth.
Meta's nuclear strategy is the most aggressive in the group. In January 2026, Meta signed agreements with Vistra to purchase power from the Davis-Besse and Perry reactors in Ohio — more than 2.1 gigawatts of operating nuclear capacity. Those deals, combined with earlier agreements, give Meta access to up to 6.6 gigawatts of nuclear capacity. For context, nuclear power costs Meta between $141 and $220 per megawatt hour — compared to $50 to $60 for gas, wind, or solar. Meta is paying two to four times the market rate for electricity because nuclear offers something renewables cannot: reliability. Nuclear plants run at over 92% capacity factor. Solar runs at around 25%. Wind runs at around 35%. AI data centers need power 24 hours a day, 7 days a week, with no weather dependencies.
Amazon's nuclear strategy is the most forward-looking. AWS signed a 17-year, 1.92-gigawatt agreement with Talen Energy for power from the Susquehanna Steam Electric Station in Pennsylvania, with broader investment near $20 billion at that site. Amazon also led a $500 million financing round for X-energy, which is developing a gas-cooled small modular reactor, and signed a $50 billion partnership to deploy 960 megawatts of SMRs. Amazon is not just buying nuclear power — it is funding the development of the next generation of nuclear technology.
Google signed a 500-megawatt development agreement with Kairos Power for small modular reactors — the first large-scale corporate SMR fleet deal in US history. Google also signed a 615-megawatt power purchase agreement that facilitated the restart of a previously decommissioned nuclear plant, and acquired Intersect Power for $4.75 billion, vertically integrating into electricity generation rather than simply contracting for offtake.
Why Nuclear and Not Solar or Wind
The shift to nuclear is not ideological. It is engineering.
AI data centers have a power profile that is fundamentally incompatible with intermittent renewable energy. A data center running frontier AI workloads cannot throttle down when the sun isn't shining or the wind isn't blowing. The compute is always on, the cooling is always running, and the power draw is constant. Solar's 25% capacity factor and wind's 35% capacity factor mean that relying on renewables alone would require massive battery storage infrastructure — infrastructure that does not yet exist at the required scale and would add enormous cost.
Nuclear's 92% capacity factor means it runs almost all the time. It produces no carbon emissions. Its fuel costs are stable regardless of global commodity prices. And critically, a single nuclear plant can produce 800 megawatts to 1.6 gigawatts of power from a relatively small physical footprint — density that no renewable installation can match.
The small modular reactor bets from Google, Amazon, and Oracle are the longer-term play. SMRs are compact, factory-built nuclear reactors designed to be deployed faster and cheaper than conventional nuclear plants. They don't yet exist at commercial scale in the United States — the earliest deployments are projected for the early 2030s. But the corporate contracts being signed today are providing the revenue certainty that nuclear startups like Kairos Power, X-energy, Oklo, and TerraPower need to complete their reactor designs and begin construction.
Big Tech is not just buying nuclear power. It is financing the commercial nuclear renaissance.
The Grid Independence Strategy
The nuclear deals are part of a broader strategic shift that has received almost no coverage outside of energy industry publications.
By early 2026, roughly 30% of new AI data center capacity is being designed to operate at least partially independent of grid infrastructure — up from effectively zero a year earlier. The hyperscalers are not just securing power purchase agreements. They are acquiring generation assets, building dedicated transmission lines, and in some cases designing data centers that are physically co-located with power plants to eliminate transmission losses entirely.
Amazon's data center campus at the Susquehanna nuclear site is the clearest example. AWS built a data center adjacent to a nuclear power plant, the first AI data center designed from the ground up around a dedicated nuclear power supply. There is no grid intermediary. Power flows directly from the reactor to the servers.
Google's acquisition of Intersect Power takes this further. Rather than contracting for power, Google now owns generation capacity — solar, wind, and storage assets that feed directly into its data center operations. The $4.75 billion acquisition was described internally as a vertical integration move, bringing energy production inside the company's operational boundary rather than treating it as an external input.
S&P Global has described this shift as hyperscaler procurement now driving US power investment rather than the other way around. For most of the past century, utilities built power plants and industrial customers bought from them. The AI infrastructure buildout has inverted that relationship. Tech companies are now the anchor customers underwriting the construction of new generation capacity — and increasingly, they are becoming the operators of that capacity as well.
What This Means for the Energy Industry
The implications of Big Tech's nuclear strategy extend well beyond the technology sector.
US power usage is projected to climb at least 30% by 2030, with most of the new demand coming from data centers. The grid infrastructure that serves the United States was built for a different era — one of stable, predictable industrial demand and gradual residential growth. The concentrated, dense, always-on power draw of AI data centers is a fundamentally different load profile than anything the grid was designed to handle.
The nuclear deals being signed today are effectively privatizing the buildout of new baseload power capacity that the entire grid will benefit from. When Microsoft restarts Three Mile Island, the power goes to Microsoft's data centers — but the plant's existence stabilizes the grid, supports jobs in the surrounding region, and provides optionality for future capacity expansion that would not exist if the plant remained shuttered.
For the nuclear industry specifically, Big Tech's intervention has been transformational. Nuclear power was a declining industry five years ago — aging plants, high operating costs, competition from cheap natural gas and renewables, and a public image still haunted by Three Mile Island and Fukushima. Corporate power purchase agreements have changed the economics entirely. They provide the long-term revenue certainty that makes operating existing plants profitable and building new ones financially viable.
The SMR startups — Kairos, X-energy, Oklo, TerraPower, Commonwealth Fusion — are the long-term beneficiaries. Each corporate offtake agreement they sign moves them closer to the financing they need to build their first commercial reactors. The AI infrastructure boom has done more for the commercial nuclear renaissance in 18 months than two decades of government policy and advocacy.
The energy trade is the AI trade. And the companies that secure the most reliable, lowest-cost, carbon-free power over the next decade will have a structural advantage in AI that no amount of GPU procurement can replicate.