Why Microsoft, Google, and Amazon Are Betting Big on Nuclear Energy
The world's most powerful tech companies aren't just buying electricity anymore — they're reshaping the entire energy industry to power AI.
Something strange happened in September 2024. Microsoft signed a deal to restart Three Mile Island — yes, that Three Mile Island, the Pennsylvania plant synonymous with America’s worst nuclear accident — and barely anyone batted an eye. ☢️ A few weeks later, Google announced a contract to buy power from small modular reactors that don’t even exist yet. Then Amazon piled in with its own nuclear deals. The nuclear industry, which had been essentially written off for a generation, suddenly had the most valuable companies on earth as its biggest cheerleaders.
This is not a coincidence. It is not greenwashing. It is the direct result of one uncomfortable fact: artificial intelligence is a power-hungry beast, and solar panels and wind turbines cannot feed it fast enough.
The hunger that started it all
Every time you send a message to ChatGPT, you consume about 2.9 watt-hours of electricity. That sounds trivial until you realize a standard Google search uses 0.3 watt-hours — ten times less — according to the International Energy Agency. Now multiply ChatGPT’s consumption across hundreds of millions of queries per day, every single day, and you start to see the scale of the problem. 🔌
Goldman Sachs Research projects that global data center power demand will surge 165% by 2030 compared to 2023 levels — the rough equivalent of adding an entirely new top-10 electricity-consuming country to the grid. The bank estimates this will require roughly $720 billion in grid spending through the end of the decade. That figure alone explains why tech companies stopped waiting around for someone else to solve the energy problem.
The core issue with renewables is timing. Solar panels work approximately six hours a day on average. Wind turbines manage about nine. Data centers, on the other hand, run 24 hours a day, 365 days a year, with zero tolerance for interruption. Goldman Sachs analysts found that wind and solar could serve roughly 80% of a data center’s power needs if paired with storage — but some form of always-on baseload generation is still required to close the gap. Nuclear is the only zero-carbon technology that provides that at scale.
Private capital for advanced nuclear is reflecting this reality. Goldman Sachs notes that private capital raises for advanced nuclear have surged 13 times compared to 2023 levels. That is not a rounding error. That is a seismic shift in investor sentiment.
Are you surprised that it took a tech boom to rehabilitate nuclear’s reputation after decades of public skepticism?
Microsoft goes back to the future
Microsoft’s move is the most dramatic of the bunch, and honestly the most interesting to think about. 🏭 In September 2024, the company signed a 20-year power purchase agreement with Constellation Energy to restart the dormant Unit 1 reactor at Three Mile Island in Pennsylvania — a facility shut down in 2019 purely for economic reasons, not safety ones.
The deal, valued at approximately $16 billion, will deliver over 835 megawatts of electricity starting around 2028. Microsoft takes 100% of the plant’s output. The plant, now rebranded the Crane Clean Energy Center, will power Microsoft’s AI data centers across the mid-Atlantic region.
Bobby Hollis, Microsoft’s vice president of energy, described the appeal bluntly in an interview with Trellis: “There is this really untapped resource of nuclear energy that is existing or that has exited the grid recently because the economics have pushed them off.” That is a remarkably direct statement. Microsoft is not chasing nuclear because it is trendy. It is chasing nuclear because it is available, reliable, and carbon-free.
The company is not stopping at Three Mile Island either. Microsoft has separately:
Signed an agreement with fusion startup Helion Energy dating to 2023
Invested in nuclear startup Oklo, chaired by OpenAI’s Sam Altman
Explored small modular reactor and microreactor options for its global network
Committed to pilot SMR projects for its data center infrastructure
Bill Gates, Microsoft’s co-founder, invested more than $1 billion of his own capital in TerraPower, an SMR startup working in partnership with Warren Buffett’s PacifiCorp. So the company’s nuclear ambitions run deeper than any single deal. This is a bet on the entire technology category, not just one power plant.
Google and Amazon go modular
Google took a slightly different path, and it may be the more consequential one in the long run. ⚡ In October 2024, the company struck what IEEE Spectrum called the first corporate deal of its kind: a contract with Kairos Power to purchase 500 megawatts of electricity from small modular reactors, with the first unit targeted for 2030 and additional deployments through 2035.
Kairos Power uses TRISO fuel — golf ball-sized graphite spheres packed with uranium, carbon, and oxygen — and coolants that operate at lower pressures than conventional reactors, which makes them inherently safer and simpler to build. Michael Terrell, Google’s senior director of energy and climate, said the company is drawn to nuclear because it provides “steady, straight through, 100% power, 24 hours a day, 365.” Hard to argue with that.
In May 2025, Google doubled down, committing early-stage capital to Elementl Power for three U.S. reactor sites totaling 1.8 gigawatts. Google is no longer just a power buyer — it is positioning itself as a financier of nuclear development.
Amazon’s approach is equally aggressive, though more diversified:
A $650 million campus purchase adjacent to Pennsylvania’s Susquehanna nuclear plant
A deal with Energy Northwest and developer X-Energy to build small modular reactors in Washington state, with a combined capacity of 320 megawatts in the first phase and options to scale to 960 megawatts
A memorandum of understanding with Dominion Energy to explore an SMR near its Virginia data centers
A target of 5 gigawatts of nuclear capacity by 2040
More than $1 billion invested in nuclear projects and technologies over the past year alone
AWS CEO Matt Garman put it plainly: “Nuclear is a safe source of carbon-free energy that can help power our operations and meet the growing demands of our customers.” That is the kind of sentence the nuclear industry has waited two decades to hear from a Fortune 10 company.
The SMR question: promise or pipe dream?
Small modular reactors are at the center of nearly every tech-nuclear deal, and they deserve honest scrutiny. 🔬 The concept is genuinely appealing: reactors of up to 300 megawatts, built with modular components that can be factory-manufactured, assembled on-site, and scaled incrementally as demand grows. Instead of spending $20 billion on a single massive plant — see Georgia’s Vogtle project, which took years longer and cost billions more than projected — companies could deploy multiple smaller units, adding capacity as needed.
The appeal for data centers specifically is real. A data center can “plug in” additional nuclear modules as its computing demands grow. The first SMR comes online; a year later, a second unit attaches. This is how you match nuclear’s baseload reliability with a tech company’s fluctuating growth trajectory.
The challenges, though, are also real:
Regulatory timelines in the U.S. are long. The NRC licensing process for new reactor designs runs years, not months, and the first commercial SMRs are mostly targeted for the early-to-mid 2030s
Economics at small scale have historically been unfavorable — larger reactors benefit from economies of scale that smaller ones cannot easily replicate
Construction risk is significant: NuScale Power, once the leading U.S. SMR developer, scrapped its flagship project in 2023 after cost estimates climbed far beyond initial projections
Very few SMRs exist today — the International Atomic Energy Agency counted just two operating examples as of 2024
The Trump administration’s executive orders supporting nuclear deployment and the bipartisan ADVANCE Act passed by Congress are designed to speed up NRC licensing and reduce regulatory friction. Whether that translates into reactors being built on schedule and on budget remains the open question. 📈 Goldman Sachs sees potential for just three nuclear plants to be operational by 2030 — a modest number relative to the 10+ gigawatts of contracts already signed.
What tech money does change, though, is demand certainty. Developers building SMRs now have long-term purchase agreements from Microsoft, Google, and Amazon. That is the single biggest historical obstacle to nuclear finance — the absence of a buyer willing to commit for 20 years — effectively removed.
What the next decade actually looks like
The honest answer is: messy, expensive, and probably later than advertised. 🌍 Restarting Three Mile Island is more achievable than building brand-new reactors from scratch, which is why Microsoft’s deal is the clearest near-term bet. The Crane Clean Energy Center should deliver power by 2028, giving Microsoft a concrete result within four years.
SMR deployments are further out. Amazon’s X-Energy project targets the early 2030s. Google’s Kairos Power deal aims for 2030 for its first unit. These timelines assume regulatory, supply chain, and financing conditions cooperate — a lot of assumptions.
The broader energy picture matters too. Goldman Sachs analysts are clear that the answer to data center power demand is “and,” not “or.” These companies are still investing heavily in renewables:
Meta and Amazon contracted over 20 gigawatts of clean power in 2024 alone
BloombergNEF tracked nearly 6 gigawatts of hybrid solar-plus-storage agreements in 2025
Goldman Sachs forecasts 40% of new data center power demand will be met by renewables
Nuclear is the baseload anchor in a portfolio that still leans heavily on wind and solar. It is not replacing renewables — it is filling the gap they cannot fill. The Department of Energy’s Advanced Nuclear pathway report suggests U.S. nuclear capacity could triple from roughly 100 gigawatts today to 300 gigawatts by 2050. Whether that projection holds depends on whether the SMR industry can actually deliver its technology at a cost that makes sense.
The SMR market is estimated to be worth somewhere between $150 billion and $300 billion between now and 2040. The winning developers will have access to some of the most reliable long-term customers in business history. Tech companies do not cancel 20-year power agreements.
So here is the question worth sitting with: if Microsoft, Google, and Amazon — companies that spent years building their reputations on renewable energy commitments — are now signing contracts to restart dormant nuclear plants and fund reactors that don’t exist yet, what does that tell you about what it actually takes to power the AI age? And more practically: if the SMR promises do finally come true in the 2030s, which developers are best positioned to deliver?