The Nuclear Renaissance
Bill Gates wants his nuclear reactor running three years early. TerraPower got the first commercial reactor permit in over
Bill Gates wants his nuclear reactor running three years early. TerraPower got the first commercial reactor permit in over a decade in February 2026. The Natrium reactor in Kemmerer, Wyoming was supposed to open in 2030. Gates now pushes for 2027.
Artificial intelligence changed the timeline. Training large language models requires data centers consuming hundreds of megawatts continuously for weeks. Solar and wind can’t provide that reliably. Batteries cost too much. Nuclear can deliver concentrated power 24/7 for decades.
Meta ordered eight nuclear reactors for data centers. Google signed deals for 500+ megawatts from multiple reactor companies. Microsoft restarted Three Mile Island. Amazon invested in nuclear startups.
Big Tech discovered nuclear because AI requires more electricity than renewables can supply. That single fact revived an industry left for dead after Fukushima.
Fukushima to Climate Solution
The Fukushima disaster in 2011 killed nuclear momentum globally. Germany shut down reactors immediately. Japan halted its entire nuclear program. New plant construction stalled everywhere. Environmental groups declared victory.
Climate change forced governments to reconsider. Wind and solar generated clean electricity but only when weather cooperated. Batteries couldn’t store enough power for grid stability at reasonable cost. Natural gas stayed cheap but contributed to emissions.
Nuclear delivers baseload power running constantly regardless of conditions. It generates massive electricity continuously for 40-60 years with zero carbon emissions during operation.
Public opinion flipped by 2024. Younger environmentalists accepted nuclear as necessary for decarbonization. Polls showed majority support for new nuclear plants in developed countries.
Small Modular Reactors Versus Giant Plants
Traditional nuclear plants generate 1,000+ megawatts from enormous custom-built reactors. They take 10-15 years to construct on-site. Every plant is unique. Costs run $8-15 billion.
Small modular reactors generate 50-300 megawatts from factory-manufactured units assembled on-site. The first one is expensive but the tenth costs half as much through manufacturing learning curves. Construction takes 3-5 years.
Safety systems use passive physics – gravity and convection – instead of powered pumps. When Fukushima’s tsunami knocked out power, pumps failed and the reactor overheated. SMRs shut down automatically without external power.
Gates Funded This for 18 Years
Bill Gates founded TerraPower in 2008. The company spent nearly two decades developing sodium-cooled fast reactor technology before breaking ground on anything.
Gates invested billions personally. The federal government added more through Department of Energy programs. This was patient capital on a timeline no venture firm would accept.
Kemmerer, Wyoming got chosen deliberately. A coal plant shut down there in 2020, eliminating 300 jobs. The nuclear reactor replaces those jobs while reusing existing transmission infrastructure. Politically smart: you can’t oppose clean energy that brings back jobs.
The Natrium design uses molten sodium as coolant instead of water. Sodium transfers heat more efficiently at higher temperatures. The system includes molten salt thermal storage – a giant battery storing heat. This lets the reactor adjust output quickly to match grid demand.
Construction started in 2023 with 2030 completion targeted. Gates now wants 2027 because AI data centers can’t wait three more years. TerraPower accelerated permitting, ordered long-lead equipment earlier, and tripled construction staffing.
The reactor will generate 345 megawatts. That powers roughly 300,000 homes or several large data centers running AI training continuously.
Meta Wants Eight Reactors
Meta announced plans in early 2026 to order eight small modular reactors dedicated to powering data centers. The company didn’t specify which SMR design or vendor.
Mark Zuckerberg’s AI strategy requires this much dedicated capacity. Training Llama 4 and future models consumes electricity that can’t come from intermittent sources. Meta concluded that owning nuclear reactors made more sense than competing for limited grid capacity.
This represents a major shift. Tech companies previously relied on grid electricity supplemented by renewable energy credits. They’d claim carbon neutrality by buying solar and wind credits while actually consuming whatever the grid provided – coal, gas, nuclear mixed together.
Nuclear gives them actual 24/7 clean baseload power they control. No credits needed. No dependence on weather. Guaranteed capacity for 40+ years under direct ownership or long-term contract.
Other hyperscalers reached identical conclusions. Google signed deals with X-energy and other SMR companies for 500+ megawatts. Amazon invested in multiple nuclear startups. Microsoft took the fastest path: restart an existing reactor.
Three Mile Island Returns
Three Mile Island’s Unit 1 reactor shut down in 2019 for economic reasons, not safety problems. The famous 1979 partial meltdown happened in Unit 2, which closed permanently afterward. Unit 1 operated successfully for 45 years without incident.
Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart Unit 1. The reactor will generate 835 megawatts exclusively for Microsoft data centers in Pennsylvania.
Restarting costs roughly $1.6 billion. Building new capacity of 835 megawatts would cost $8-10 billion. Economics strongly favor restart over new construction when existing reactors can operate safely.
Other shut-down reactors may follow this pattern. Palisades in Michigan got federal support for restart evaluation. Diablo Canyon in California extended operations after planned closure. Several retired reactors are being reconsidered as electricity demand spikes.
The irony is thick. Three Mile Island became synonymous with nuclear danger. Now it’s restarting to power artificial intelligence. Public perception changed completely in 47 years.
Regulatory Approval Accelerated
The Nuclear Regulatory Commission approved TerraPower’s Natrium design in February 2026. First new commercial reactor design approval in over a decade.
The NRC historically took 5-7 years reviewing new designs. TerraPower’s approval took four years. Still long but faster than historical processes. Pressure from DOE and bipartisan Congressional support pushed things along.
The Infrastructure Investment and Jobs Act of 2021 included $6 billion for existing nuclear plants and $2.5 billion for advanced reactors. The Inflation Reduction Act added production tax credits making nuclear electricity economically competitive with natural gas.
Without these subsidies, new nuclear couldn’t compete on cost alone. With them, nuclear pencils out for utilities and tech companies needing reliable baseload power.
Bipartisan political support emerged by 2024. Republicans liked nuclear for energy independence, manufacturing jobs, and competition with China. Democrats needed it for climate goals and grid stability. Both parties supported it for AI competitiveness.
This political alignment didn’t exist in 2015 when nuclear was partisan and controversial. By 2025 it became consensus. China competition and AI power demands created strange bedfellows.
China and Russia Never Stopped
While the US essentially halted nuclear development for 40 years, China and Russia kept building. China operates 57 reactors with 22 under construction. Russia operates 38 reactors and exports technology globally.
China’s CAP1400 and Hualong One designs compete internationally. Russia’s VVER-1200 reactors and floating nuclear plants sell to developing countries needing electricity. Both countries treat nuclear as strategic technology for energy security and geopolitical influence.
The US ceded nuclear leadership from 1980-2020. Now it’s scrambling to catch up. TerraPower, NuScale, X-energy, and other American SMR companies received billions in government support to compete globally.
This became national security priority under Biden and continued under subsequent administrations. Whoever provides nuclear technology to developing countries gains political influence. Letting Russia and China dominate nuclear exports was considered strategic failure.
The competition intensified in 2024-2025. US SMR companies bid against Chinese and Russian state-backed firms for projects in Poland, Czech Republic, Philippines, and other countries seeking energy independence from fossil fuels.
Cost Overruns and Delays
Small modular reactors promise lower costs through factory production and manufacturing learning curves. Promises don’t always match reality.
NuScale, the most advanced US SMR company, canceled its first project in Utah in 2023 when costs doubled from initial estimates. Customers walked away. The company still seeks buyers but faces justified skepticism.
TerraPower’s Natrium is first-of-a-kind. Costs will be high. Gates invested billions personally. Federal subsidies covered billions more. This is essentially a demonstration project proving the technology works at commercial scale.
If it succeeds and operates reliably, subsequent reactors should cost less. The tenth Natrium reactor might cost half what the first one did. That’s the manufacturing learning curve theory.
But traditional nuclear projects went catastrophically over budget. Vogtle Units 3 and 4 in Georgia cost $35 billion versus $14 billion projected. They finished seven years late. SMR proponents claim standardization and factory production avoid these problems.
The first SMRs completing between 2027-2030 will prove whether that’s true. If TerraPower, NuScale, or X-energy deliver on-time and on-budget, orders will flood in. If they don’t, SMR momentum collapses.
Waste Storage Unresolved
Nuclear generates radioactive waste staying dangerous for thousands of years. Modern reactor designs don’t fix this.
The US lacks permanent waste storage. Yucca Mountain in Nevada was supposed to solve it. Nevada’s Congressional delegation killed the project. Spent fuel sits in temporary storage at reactor sites indefinitely.
Modern SMR designs reduce waste volume. Some can reprocess spent fuel. But they still create waste requiring isolation for 10,000+ years.
The nuclear industry argues total waste volume is tiny – all US nuclear waste ever produced fits in a football field at 10 yards deep. Critics argue volume doesn’t matter when material stays lethally radioactive for 300 generations.
Nuclear expansion proceeds anyway because climate change feels more urgent than waste storage to policymakers.
TerraPower’s Wyoming reactor will answer whether SMRs work as promised by 2027-2028. Either it operates reliably and nuclear becomes standard tech infrastructure, or it doesn’t and the renaissance stalls.
Sources:
NPR – Three Mile Island Restart
