$16 billion from Microsoft. Over $20 billion from Amazon. Up to 6.6 gigawatts from Meta. The money is real. The nuclear reactors to justify it exist mostly on PowerPoint slides.

Every major tech hyperscaler has signed at least one nuclear power deal for AI data center capacity. Thirteen announced projects commit over 9.8 GW of nuclear capacity — more capital committed to US nuclear than any prior decade in history, according to SMR Intel’s March 2026 tracker. The nuclear-for-AI thesis is no longer theoretical.

The problem is what “executed” actually requires: enriched uranium that doesn’t exist in domestic supply, welders and machinists who retired without being replaced, and a supply chain currently controlled by the two countries Washington considers its primary strategic rivals.

The Fuel Problem in One Number

Approximately 90% of US advanced reactor developers have chosen High-Assay Low-Enriched Uranium (HALEU), enriched to between 10% and 20%, because it allows smaller designs and longer refueling cycles — exactly the profile Big Tech needs. The United States currently produces one metric ton of HALEU per year. Centrus Energy, the sole domestic producer, runs a single facility. The Department of Energy projects demand could reach 50 metric tons annually by 2035.

That gap is not a rounding error. It is the entire story.

Russia exported $20.5 billion of enriched uranium between 2013 and 2023, according to the World Nuclear Industry Status Report — nearly double the next largest exporter, the United Kingdom, at $10.6 billion. A US law banning Russian uranium imports took effect in August 2024, with waivers through the end of 2027 to keep fuel flowing while domestic alternatives are supposedly built.

“Russia has a strong interest in keeping this business going because it generates substantial revenue,” said Sebastian Stier, author of the Russia Nuclear Interdependencies chapter in the 2025 World Nuclear Industry Status Report. Rosatom reported overseas project revenue rose 10% year-over-year to $18 billion in 2024.

The Workforce That Isn’t There

Even if the fuel materialized tomorrow, there is no one to build the reactors.

A March 2026 report commissioned by the Nuclear Scaling Initiative and funded by the Bezos Earth Fund identified a “critical lack of nuclear-qualified machinists, welders, inspectors, nondestructive examination specialists, and project managers.” The pipeline leaks at every stage: low apprentice wages, competition from other industries, childcare difficulties, and a “looming retirement wave” all drain potential workers before certification.

The report recommends “dramatic investments” in centralized, standardized training — essentially recreating the nuclear workforce buildout of the 1950s and 1960s, but faster, while competing with construction booms in every other sector.

The Chicken, the Egg, and the Order Book

Then there is the structural deadlock. Suppliers will not expand capacity without demand signals. Customers will not place orders without a supply chain in place. The NSI report calls this a “chicken-and-egg problem” and proposes an order book approach: multiunit commitments of 10 to 30 reactors that justify supplier investment and reduce customer risk.

This has precedent. When orders for the four AP1000 reactors at Vogtle and V.C. Summer were placed simultaneously, prices on certain key materials dropped as much as 25 to 30 percent, according to the report. The NSI received $3.5 million from the Bezos Earth Fund earlier this year to facilitate a US order book for “new reactor builds of mature design.”

The irony of Jeff Bezos funding the institutional fix while Amazon bets billions on unproven next-generation designs is not lost on anyone paying attention.

The Competition Is Not Waiting

Russia and China have led or financed more than 57 of the 63 nuclear power reactors that began construction worldwide in the past decade — over 90% of global nuclear construction, according to industry data cited by Intellinews.

Rosatom is building 19 reactors abroad, from Turkey to Bangladesh, often with full financing and decades-long fuel supply agreements. China’s nuclear firms, backed by Belt and Road Initiative money, are expanding across Asia, the Middle East, and Africa.

Western companies like EDF, KHNP, and Westinghouse remain active but are “increasingly outpaced by the scale and state financing available to their Russian and Chinese counterparts.”

Start With What Exists

Some developers are pragmatists. GE Hitachi, Westinghouse, and Aalo Atomics have opted for LEU+ — fuel enriched between 5% and 10% that existing domestic facilities can produce, or at least upgrade to produce, rather than build from scratch. Holtec’s SMR-300 is designed to support both standard LEU and LEU+.

“We know we want to get to market fast, and we know we need to scale up, to build hundreds of reactors, and we can’t do that with HALEU for many years, because the U.S. is still pumping money into that HALEU machine, trying to figure out how to crack the code,” Aalo Atomics chief technology officer Yasir Arafat told Reuters Events.

The NSI report recommends focusing first on Gen III+ designs — the AP1000, the SMR-300, the BWRX-300 — before pursuing the Gen IV reactors that dominate Big Tech’s deal sheets. Rebuild the foundation before designing the penthouse.

Microsoft will be first to receive nuclear power specifically because it chose to restart an existing reactor — Three Mile Island Unit 1 — rather than build new. That $16 billion, 20-year power purchase agreement with Constellation Energy is the safest bet in the portfolio precisely because it requires the least that doesn’t yet exist.

The rest of the deals demand concrete, steel, enriched uranium, and skilled workers in quantities the United States does not have and cannot quickly conjure. The bottleneck is not software. It is not regulation. It is the physical world, and it does not scale on command.

Sources