Microsoft and Google Race to Buy Nuclear Power as SMRs Become Wall Street's New Bet
0xBroomberg
Microsoft, Google, and Meta have signed nuclear-power purchase agreements spanning 20–25 years, betting on nuclear energy to deliver round-the-clock, zero-carbon electricity for AI data centers; small modular reactors are the centerpiece of this revival, but fuel shortages and an unproven commercial timeline remain the biggest unknowns.
Why are tech giants suddenly buying nuclear power?
AI data centers need round-the-clock, zero-carbon electricity. Solar and wind are intermittent; gas-turbine deliveries now take years. Nuclear fills the gap.
The Electric Power Research Institute projects that data centers could account for 17% of U.S. electricity consumption by 2030, up from roughly 5% today. This means → demand is not growing linearly — it could more than triple in just a few years.
Public opinion is shifting too: a Pew Research Center survey shows 59% of U.S. adults now support expanding nuclear power, up from 43% a decade ago.
Who signed what?
Microsoft signed a 20-year power-purchase agreement with Constellation Energy to restart the Three Mile Island plant in Pennsylvania, targeting 2027.
Google signed a 25-year deal with NextEra Energy to restart the Duane Arnold plant in Iowa by 2029.
Meta signed separate agreements with Oklo and TerraPower for AI data-center power. Holtec International is restarting the Palisades plant in Michigan, expected to complete this year.
What is the government pushing — and what is controversial?
The Trump administration pledged over $80 billion for reactor construction based on Westinghouse designs and signed an executive order compressing new-build licensing to 18 months — half the previous timeline.
The Department of Energy launched a reactor pilot program with 11 projects. At least four — Antares Nuclear, Valar Atomics, Aalo Atomics, and Deployable Energy — reached "criticality" before the deadline. In plain terms = the reactors sustained a nuclear chain reaction, but that is a technical milestone, not commercial power.
The controversy: the administration is pushing to revise the Nuclear Regulatory Commission's long-standing radiation-limit guidelines, raising broad safety concerns.
What exactly is an SMR, and how close is it to reality?
An SMR — small modular reactor — typically generates no more than 300 MW, far below the 1,000 MW-plus of a conventional reactor. Components are factory-built and assembled on site, theoretically cutting construction time and cost.
The two furthest along: NuScale Power has received NRC design certification; Bill Gates–backed TerraPower has a construction permit in Wyoming. But no SMR company holds an operating license yet.
Industry consensus puts the earliest grid connections in the early 2030s, and those first units will be expensive, first-of-a-kind systems. This means → even if the technology works, cost reductions from scale are still years away.
Why does history make the market nervous?
The most recently completed U.S. nuclear project — Vogtle Units 3 and 4 in Georgia — ran seven years late and finished at more than double the original budget.
This reflects the biggest risk in nuclear construction: not the technology itself, but schedule blowouts and cost overruns — exactly what SMRs promise to solve but have yet to prove.
The administration's target is 400 GW of nuclear capacity by 2050, four times the current installed base. Vogtle's legacy still weighs on investor appetite.
How serious is the uranium fuel gap?
The U.S. imports most of its enriched uranium. In 2024, Russia was still the largest supplier, accounting for roughly one-fifth. Sanctions waivers on Russian uranium expire in 2028, making alternative supply chains urgent.
The U.S. has only one commercial-scale enrichment facility, operated by Urenco, which announced plans in June to expand capacity by nearly 50%.
Many SMRs require HALEU — high-assay, low-enriched uranium with up to 20% uranium-235 concentration — and only two countries can produce it at commercial scale today. The government has allocated $1.8 billion to General Matter and a Centrus Energy subsidiary to build domestic capacity, but a supply chain built from scratch may not align with the nuclear expansion timeline.
Content is for reference only, not financial advice.