Global RaceData center-linked nuclear offtake agreements have grown from roughly 25 GW in late 2024 to nearly 45 GW today, and almost none of that growth came from wind or solar. The reason is simple math. ⚡ A single AI training cluster can draw as much power as a mid-sized city, hyperscaler capital spending is on track to hit $600 billion to $700 billion in 2026 alone, and solar-plus-battery still can’t guarantee power at 3 a.m. on a cloudy week. Nuclear can, and small modular reactors promise to deliver it without a decade-long permitting fight for a gigawatt-scale plant. 🔋
What’s interesting isn’t that Big Tech wants nuclear power. It’s that Google, Amazon, Meta, and Microsoft have each landed on a different answer for how to get it. Some are buying equity in reactor developers outright. Others are restarting mothballed giants while hedging with SMR options on the side. Here’s how each company is actually placing its bet. 📊
Google bets on Kairos Power for round-the-clock power
Google signed the tech industry’s first multi-reactor SMR purchase agreement in October 2024, committing to buy power from a fleet of Kairos Power small modular reactors totaling up to 500 MW, with the first unit online by 2030 and the rest following through 2035. Kairos’s KP-FHR design cools TRISO fuel pebbles with molten fluoride salt, running at low pressure instead of the high pressure a conventional reactor needs, a genuinely unusual engineering bet for a company better known for search ads. 🧪🔥
The deal stopped being theoretical in mid-2025. Kairos, Google, and the Tennessee Valley Authority struck an agreement under which Kairos’s Hermes 2 plant in Oak Ridge will feed up to 50 MW into the TVA grid serving Google data centers in Montgomery County, Tennessee, and Jackson County, Alabama, making TVA the first US utility to buy power from a Generation IV reactor. Google senior director Michael Terrell called nuclear part of the company’s effort to “commercialize a broad portfolio of advanced clean electricity technologies.” 🌍
What makes Google’s approach distinct:
One reactor developer, one long-term relationship, rather than a scattered portfolio
A named utility offtaker (TVA) instead of a purely behind-the-meter arrangement
Molten salt cooling, a genuine departure from the light-water reactors most utilities already operate
A firm 2030 first-power target, unusually specific for a first-of-a-kind design
Which strategy do you think ages better: Google’s single-partner bet or the multi-vendor approach the next three companies are running?
Amazon spreads its wager across X-energy, Energy Northwest, and Dominion
Amazon has taken the opposite approach from Google: diversify early and often. 🎯 The company holds an equity stake in X-energy and backs the deployment of more than 5 GWe of X-energy’s Xe-100 reactors across the US by 2039. The first project, called Cascade, is a four-unit, 320-MWe plant with Energy Northwest in central Washington, targeting construction by 2030, a genuinely aggressive timeline for a first-of-a-kind design. 🏗️
Amazon didn’t stop there. It signed a separate memorandum with Dominion Energy to explore a 300-MW SMR near Dominion’s existing North Anna nuclear station in Virginia, where power demand is projected to climb 85% over the next 15 years. And in June 2026, Amazon layered a 17-year, $18 billion power purchase agreement with Talen Energy on top of its existing arrangement, drawing up to 1,920 MW from Talen’s existing Susquehanna plant while the two companies explore building new SMRs within Talen’s Pennsylvania footprint. ⚙️ Every project, deal, company, and regulatory event we cover is logged, verified, and searchable in SMRbrief Pro, which is genuinely the only way to keep Amazon’s overlapping nuclear commitments straight in your head.
The logic behind Amazon’s spread-the-risk model:
Equity in X-energy gives Amazon a seat at the table on reactor design and supply chain decisions
Existing large reactors (Susquehanna) cover near-term load while SMRs mature
A separate Dominion relationship hedges against any single developer’s schedule slipping
Amazon also joined a strategic collaboration with Korea Hydro & Nuclear Power and Doosan Enerbility aimed at mobilizing up to $50 billion for Xe-100 supply chain buildout
Meta swings the biggest bat yet: 6.6 GW across three partners
Meta waited longer than its peers to make a nuclear move, then made the largest one on record. 🐘 On January 9, 2026, Meta announced agreements covering up to 6.6 GW of nuclear capacity through 2035, split across three very different partners: Vistra for output from existing operating plants, plus TerraPower and Oklo for next-generation SMR capacity. It’s the largest corporate nuclear commitment in US history, full stop. 📈
The structure is deliberately layered. Vistra’s existing reactors, including Davis-Besse in Ohio, supply power now and give Meta a reason to back license extensions that keep those plants running past 2037. TerraPower’s Natrium design, already under construction in Kemmerer, Wyoming, and Oklo’s Aurora microreactor fill in the next decade as they clear licensing. Meta paired the nuclear push with a separate May 2026 deal alongside Enbridge for 365 MW of solar and a 1,600 MWh battery in Wyoming, suggesting the company sees nuclear and storage as complements rather than either-or bets. 🔌
Why the three-partner blend matters:
Combines firm power today (Vistra) with next-generation capacity later (TerraPower, Oklo)
Spreads regulatory risk across two very different SMR technologies: sodium-cooled fast and Oklo’s compact fast reactor
Sets a template other hyperscalers are likely to copy rather than negotiate from scratch
Signals confidence that at least one of the two SMR developers clears NRC licensing on schedule
Microsoft hedges between a restarted giant and a portfolio of new reactors
Microsoft moved first, and its bet looks the most conservative on paper even though it’s anything but small. 🏢 In September 2024, Microsoft signed a 20-year power purchase agreement for the full 835 MW output of Constellation’s Three Mile Island Unit 1, since renamed the Crane Clean Energy Center. Constellation is investing $1.6 billion to bring the plant back online, now targeted for 2027, a year ahead of the original schedule. ⚛️
That restart isn’t Microsoft’s whole nuclear strategy, just its most visible piece. The company has also committed more than $1 billion toward a broader roughly 2.5 GW nuclear portfolio that includes relationships with X-energy, Dominion, and Talen, spreading its bet across an existing large reactor and a bench of SMR developers at the same time. Constellation CEO Joe Dominguez said the Crane restart puts the company “on track to make history ahead of schedule.” 🧊
The pattern in Microsoft’s approach:
One large, low-technology-risk restart locks in firm power for the early 2030s
A parallel SMR portfolio keeps optionality open if any single next-gen design stalls
Pennsylvania’s PJM grid gets the near-term supply while newer designs work through licensing elsewhere
The $1.6 billion restart cost is a known number, unlike first-of-a-kind SMR construction budgets
Four companies, four different mixes of existing reactors and unproven SMRs, and one shared bet that nuclear beats waiting on the grid. Which of these four strategies would you put your own capital behind?