Nuclear Power Was Supposed to Be Dead. AI Just Brought It Back to Life.

You don't have to read everything — just the right thing. 1440's daily newsletter distills the day's biggest stories from 100+ sources into one quick, 5-minute read. It's the fastest way to stay sharp, sound informed, and actually understand what's happening in the world. Join 4.5 million readers who start their day the smart way.

Join for free today!

Twenty years ago, nuclear energy was treated like an embarrassing relative the energy industry no longer wanted to invite to dinner. Today, every major tech company on Earth is racing to plug their AI data centers into reactors, the U.S. government is targeting a fourfold expansion of nuclear capacity, and a Finnish-American startup just signed a deal to put miniature nuclear reactors inside server farms. The comeback is so abrupt that most people have not caught up to what is happening. Today's edition walks you through the deals, the numbers, and most importantly, the four big counter-arguments people have against nuclear, with the actual evidence on each one. Buckle up.

Here is the headline number that explains everything else. Goldman Sachs estimates that global data center electricity demand will rise by 160 percent by 2030. The Electric Power Research Institute projects that U.S. data centers alone will consume up to 9 percent of the entire country's electricity by 2030, more than double the current share. Meta, Microsoft, Google, Amazon, and Oracle have all looked at that math, looked at the existing grid, and reached the same conclusion: they cannot get there with natural gas, solar, and wind alone. The numbers do not work. What does work, by elimination, is nuclear.

That conclusion has triggered a wave of deals over the last six months that most casual observers have missed entirely. NANO Nuclear Energy signed a memorandum of understanding with Supermicro on May 6, 2026, to deploy micro modular reactors directly inside AI server farms. Terrestrial Energy partnered with Riot Platforms to integrate Generation IV molten salt reactors with large-scale data centers in Texas and Kentucky. Holtec is preparing to restart the shuttered Palisades nuclear plant in Michigan, the first decommissioned U.S. nuclear plant ever to be brought back online, while also building two new 340-megawatt small modular reactors on the same site. The Tennessee Valley Authority and Holtec each received 400 million dollars in cost-shared federal funding earlier this year to accelerate advanced light-water SMR deployment. China's Linglong One small modular reactor is scheduled to begin commercial operations in the first half of 2026, making it the world's first commercial onshore SMR.

The U.S. policy backdrop is even more aggressive. Recent executive orders aim to expand U.S. nuclear capacity from 100 gigawatts today to 400 gigawatts by 2050. The Department of Energy is actively working to remove permitting barriers and is making federal land available for co-located data center and reactor projects. More than 40 gigawatts of small modular reactor capacity is already being positioned globally for industrial users, primarily hyperscalers. This is the largest coordinated nuclear buildout the world has attempted since the 1970s, and it is happening because the alternative is not having enough electricity to run modern AI.

Carbon Credits' full overview of 2026 as the nuclear comeback year: https://carboncredits.com/2026-the-year-nuclear-power-reclaims-relevance-with-15-reactors-ai-demand-and-chinas-expansion/

Highways Today on the Terrestrial Energy and Riot Platforms partnership: https://highways.today/2026/05/08/nuclear-power-ai-infrastructure/

This is the loudest objection and, by the actual evidence, the weakest one. The data on energy-related deaths per terawatt-hour of electricity produced has been compiled across multiple peer-reviewed studies, most prominently by Oxford University's Our World in Data project. The headline finding is genuinely surprising for most people. Coal causes roughly 24.6 deaths per terawatt-hour, almost entirely from air pollution. Oil causes more than 18. Natural gas causes nearly 3. Nuclear causes roughly 0.03 deaths per terawatt-hour, including all deaths from Chernobyl and Fukushima.

To put that in plain numbers, nuclear energy results in 99.9 percent fewer deaths than brown coal, 99.8 percent fewer than coal, 99.7 percent fewer than oil, and 97.6 percent fewer than gas. It is statistically as safe as wind, and only marginally less safe than solar. In an average town of 150,000 people powered entirely by coal, you would expect at least 25 premature deaths per year, mostly from air pollution-related heart and lung disease. The same town powered by nuclear would, on average, experience a death attributable to the energy source once every 33 years.

The intuition gap here is fascinating. Coal kills slowly and invisibly through air pollution, so the deaths do not register as energy-related. Nuclear accidents are rare but dramatic, so they dominate the perception. The actual mortality math is the opposite of what most people assume. Modern SMR designs are also fundamentally different from the gigawatt-scale plants of the 1970s and 1980s. They use passive safety systems that rely on physics rather than active human intervention, meaning a power failure or operator error cannot cause a meltdown the way Three Mile Island or Chernobyl did. The reactor physically cannot enter a runaway state.

Our World in Data's comprehensive safety analysis: https://ourworldindata.org/safest-sources-of-energy

Canary Media's chart-based breakdown of deaths per energy source: https://www.canarymedia.com/articles/fossil-fuels/which-power-sources-are-most-deadly-hint-not-solar-and-wind

This one has historical merit and is the most legitimate concern. Traditional gigawatt-scale nuclear plants have a long track record of cost overruns and delays. The Vogtle 3 and 4 reactors in Georgia ran roughly seven years late and 17 billion dollars over budget. The first planned U.S. SMR, NuScale's project in Idaho, was cancelled in 2023 because rising costs killed the customer base. Critics like Stanford's M.V. Ramana have argued for years that SMRs are "an idealized projection rather than a proven reality."

The counter-evidence is that the economics of SMRs are structurally different from traditional reactors in ways that directly attack the historical cost problem. Traditional nuclear plants are bespoke megaprojects built on-site over a decade, which means every plant is essentially a one-off prototype with all the budget risk that implies. SMRs are designed to be factory-built in modules and shipped to the site. Once a factory line is operational, each subsequent unit gets cheaper because manufacturing learning curves apply the way they do to airplanes or cars. Holtec's current SMR construction costs are estimated at 12 to 15 million dollars per megawatt, which would put a 680-megawatt plant at 7 to 10 billion dollars. That is still expensive in absolute terms, but it is roughly half the per-megawatt cost of the Vogtle overruns, and the cost is expected to drop as production scales.

There is also a more honest version of this argument worth acknowledging. Solar and wind are now genuinely cheaper per kilowatt-hour for intermittent generation. The nuclear pitch is not that it competes with solar on cost, it is that solar and wind cannot deliver the constant, 24-hour-a-day baseload that AI data centers need. Server farms running large language model inference cannot tolerate the kind of supply gaps that come from a cloudy week or a calm month. The choice is not nuclear versus solar. It is nuclear versus natural gas for baseload, and on that comparison nuclear is increasingly competitive on price and dramatically better on emissions and air-quality deaths.

ScienceDirect's analysis of SMR economics: https://www.sciencedirect.com/science/article/abs/pii/S0149197025003877

Circle of Blue on the Holtec Palisades restart economics: https://www.circleofblue.org/2026/water-energy/a-nuclear-shift-buoyed-by-billions-and-the-waters-of-the-great-lakes/

The waste question is real, persistent, and deserves a direct answer. The United States has, after 70 years of commercial nuclear power, still not established a permanent deep geological repository for spent nuclear fuel. Yucca Mountain was politically killed in 2009. Spent fuel currently sits in cooling pools and dry casks at reactor sites across the country. A 2022 PNAS study by Stanford and University of British Columbia researchers found that some SMR designs may actually produce more voluminous and reactive waste per megawatt-hour than traditional reactors, because their compact cores leak more neutrons into surrounding materials.

The counter-evidence has three parts. First, the absolute volumes of nuclear waste are tiny compared to fossil fuel waste streams. All of the high-level nuclear waste ever produced by the U.S. commercial nuclear industry, across 70 years and roughly 20 percent of national electricity generation, would fit on a single football field stacked about 10 yards high. The annual coal ash output of the United States, by contrast, is roughly 70 million tons, much of which contains arsenic, lead, mercury, and other toxic heavy metals that leach into groundwater. Coal ash spills like the 2008 Kingston disaster in Tennessee have caused billions in environmental damage. The relative scale of these waste streams is not close.

Second, modern reactor designs are specifically engineered to address the waste problem. Generation IV reactors, including the molten salt designs Terrestrial Energy is building, can actually consume existing spent nuclear fuel as their input, reducing the long-term waste burden rather than adding to it. Some of these designs reduce the radioactive lifetime of the waste from hundreds of thousands of years to a few hundred years. Third, the political and regulatory framework for waste storage is finally moving. The Department of Energy is actively pursuing consent-based siting for an interim storage facility, and Finland's Onkalo deep geological repository began operations in 2024, providing a proven template that the U.S. can follow.

The PNAS study on SMR waste characterization: https://www.pnas.org/doi/10.1073/pnas.2111833119

Carbon Credits' guide to SMR waste handling: https://carboncredits.com/the-ultimate-guide-to-small-modular-reactors/

The objection here is that even if SMRs work, they will not come online fast enough to meet AI's electricity demand. The 2030 deadline that everyone is racing toward leaves only four years to deploy gigawatts of new capacity, and the Nuclear Regulatory Commission's licensing process has historically been measured in decades.

The counter-evidence is mixed but encouraging. China's Linglong One demonstrates that an SMR can go from groundbreaking to commercial operation in roughly seven years, which is fast by nuclear standards but slow by data center standards. The U.S. response has been to streamline the regulatory pathway dramatically. The NRC has been issuing construction permits and operating licenses for SMR designs at an accelerated pace, the Department of Energy is co-locating projects on federal land where permitting is faster, and the ADVANCE Act passed in 2024 specifically reduced licensing fees and timelines for advanced reactor designs. Restart projects like the Palisades plant in Michigan and the Three Mile Island Unit 1 restart that Microsoft contracted with Constellation Energy are faster still, because the physical infrastructure already exists.

It is fair to say that not every SMR project announced today will be online by 2030. It is also fair to say that the first wave, including the Holtec Palisades units, the TVA SMR project, and the NANO Nuclear and Terrestrial Energy industrial deployments, is on track for the early 2030s. That timeline aligns roughly with the second wave of AI infrastructure buildout, when the current generation of data center projects backed by natural gas peaker plants begins to retire or expand. The question is not whether nuclear will arrive in time for the first AI buildout. It will not. The question is whether it will arrive in time for the second, and on current evidence the answer is yes.

The IEEE Spectrum overview of the hyperscaler nuclear pivot: https://spectrum.ieee.org/nuclear-powered-data-center

The UN News deep dive on AI and the new nuclear age: https://news.un.org/en/story/2026/01/1166768

Even if you have no professional stake in energy or AI, the nuclear comeback affects you in three concrete ways. First, your electricity bill. The hyperscalers building captive nuclear capacity are doing so in part because public utility rates have become unpredictable. As that captive capacity comes online, public grid demand from data centers eases, which should put downward pressure on residential rates over time. Second, your air. Every megawatt of nuclear baseload that replaces a megawatt of natural gas peaker capacity is a meaningful reduction in regional air pollution and the asthma, heart disease, and premature deaths that come with it. Third, your political conversations. Nuclear is becoming one of the rare technologies with genuine bipartisan support, and the policy debates over the next five years are going to be increasingly serious. Understanding the actual numbers, rather than the inherited intuitions from the 1970s, is going to matter.

The honest summary is this. Nuclear energy is not a perfect solution. It is expensive, the waste problem is unresolved at the policy level, and the historical track record on cost overruns is real. But it is also dramatically safer than the fossil fuels it competes with for baseload, it produces no carbon emissions during operation, and it is the only currently scalable technology that can deliver 24-hour-a-day reliable electricity at the scale AI is demanding. The combination of those three facts is why every major tech company and the U.S. government have all reached the same conclusion at roughly the same time, and why the nuclear renaissance is finally moving from talk to construction.

We will keep tracking this and bring you the next chapter as it lands. Stay charged out there.