AI's massive power demand is reviving an industry that's been stalled for 40 years: nuclear power

[Panther21]

Microsoft is paying to restart Three Mile Island's sister reactor, now renamed the Crane Clean Energy Center, in a 20 year, $16 billion contract to buy every watt of electricity it produces once it comes back online in late 2027, with all of that power feeding Microsoft's AI data centers rather than homes or factories. It's part of a broader pattern, every major AI company, Microsoft, Google, Amazon and Meta, has now signed at least one nuclear deal in 2026, together committing to more than a dozen agreements worth nearly 10 gigawatts of capacity, enough to power roughly 7 million homes. Meta alone has signed deals worth up to 6.6 gigawatts, Google has struck agreements with two nuclear developers plus a separate 1,800 megawatt contract, and Amazon has put $700 million into X-energy to build up to a dozen small modular reactors

The appeal is straightforward, nuclear plants run constantly regardless of weather, exactly what an AI data center needs since it can't power down at night the way solar or wind generation naturally does. IAEA director Rafael Mariano Grossi put it plainly, only nuclear energy can meet the combined needs of low carbon generation, round the clock reliability, high power density and true scalability. Global electricity demand is projected to grow by more than 10,000 terawatt hours by 2035, with data centers driving over a fifth of that growth in wealthy countries by 2030, and the queue of projects waiting to connect to the US grid has swollen past 2,600 gigawatts with average wait times nearing five years

Critics argue this is mostly well funded publicity rather than a near term solution. Researchers writing in the Bulletin of the Atomic Scientists this month argued nuclear power and small modular reactors will be unable to meaningfully boost data center electricity supply anytime soon, since a single new reactor costs billions and takes the better part of a decade even under ideal conditions, and Georgia's Plant Vogtle, the last large scale US reactor completed, finished seven years late and $36.8 billion over budget. The counterpoint is that restarts like Three Mile Island are meaningfully different from new builds, returning to service in two to three years using existing infrastructure rather than a decade of new construction, meaning near term power genuinely is coming from restarts even while the newer reactor bets remain years away from producing a single watt

The AI Boom Is Making Nuclear Power Bankable Againwww.forbes.com

[CosmicRay17]

The restart versus new build distinction is the most important nuance in this whole story, lumping Three Mile Island's 2027 timeline in with a brand new reactor that won't produce power until the 2030s badly misrepresents how near term this actually is

[Sparrow]

Plant Vogtle finishing seven years late and $36.8 billion over budget is exactly the kind of cautionary example that makes the skeptics' case genuinely credible rather than just reflexive nuclear pessimism

[ThreadNecro11]

Big Tech becoming nuclear power's most important financial backer after 40 years of the US government and utility industry failing to make new reactors affordable is a remarkable shift in who actually drives energy infrastructure now

[Bussin99]

The 2,600 gigawatt grid connection queue with five year average wait times explains so much of why these companies are financing reactors directly rather than just buying power from existing utilities

[Sentry]

Grossi's round the clock reliability argument makes intuitive sense for why nuclear specifically beats solar and wind for this exact use case, a data center genuinely cannot pause overnight the way other power demand can flex around intermittent generation

[FinnBalor99]

Genuinely torn here, the restart argument is compelling and near term, but the newer small modular reactor bets from Meta, Amazon and Google really do seem to be exactly the kind of unproven technology the skeptics are right to question

[BetaMyles75]

Good, balanced piece that takes both the bulls and the skeptics seriously rather than picking one side, the honest answer really does seem to be it depends heavily on restart versus new build timelines

[LuckySentinel]

This being the first time in 40 years nuclear has become bankable again is a huge deal regardless of how the specific reactor timelines shake out, that changes the whole trajectory of the industry going forward

[Highland Dylan]

The comparison between AI growth curves and nuclear reactor growth curves moving in tandem for the first time is a striking way to frame just how tightly these two industries have become linked

[DataStream Luca]

Worth remembering all of this power is explicitly earmarked for AI data centers rather than homes or general grid capacity, that's a real tradeoff in how this new nuclear buildout actually benefits the broader public versus a handful of tech giants