Every hyperscaler earnings call this year has mentioned nuclear power. Microsoft is restarting Three Mile Island. Amazon is buying into small modular reactors. Google signed the first-ever corporate SMR fleet deal. Meta committed to up to 6.6 gigawatts across four different reactor partners. If you’re evaluating where to put your workload for the next decade, it’s a reasonable question to ask: does this actually change anything for me, or is it a press release?
The honest answer is both — and the difference matters enormously depending on which specific deal you’re looking at.
The headline number, and the number underneath it
Hyperscalers have committed to 9.8 gigawatts of nuclear capacity across 13 separate deals as of mid-2026. That sounds like a real answer to the power scarcity driving up colocation prices across every major market. But only 1.92 GW of that committed capacity is actually operational today — meaning roughly 80% of everything Big Tech has announced on nuclear power has not yet generated a single kilowatt-hour for them. One analysis calls this the “Nuclear Readiness Gap,” and pegs it at 80.4%.
That gap isn’t evenly distributed, and it’s the single most important thing to understand before you take any nuclear announcement at face value.
The deals, and how far each one actually is from real power
| Company | Capacity | Partner / Source | Status |
|---|---|---|---|
| Amazon | 1.92 GW | Talen Energy’s Susquehanna plant | Operational since June 2025 — existing reactor, not new |
| Microsoft | 835 MW | Constellation / Three Mile Island Unit 1 restart | Targeted H2 2027 — existing reactor, restart in progress |
| Amazon | Up to 960 MW | X-energy Xe-100 SMRs, Energy Northwest | 2030s — new construction, not started |
| 500 MW | Kairos Power SMR fleet | First reactor ~2030 — new construction, not started | |
| Meta | Up to 6.6 GW | TerraPower, Oklo, Vistra, Constellation | 2032–2035 — mostly new construction, not started |
Notice the pattern: the two deals actually delivering power right now — Amazon’s Talen/Susquehanna arrangement and Microsoft’s Three Mile Island restart — both involve existing, already-built nuclear plants, either already running or being brought back online. Every deal involving a genuinely new small modular reactor design is sitting somewhere between “not yet under construction” and “targeted for 2030 at the earliest.” That’s not a technicality — it’s the entire ballgame for anyone trying to figure out whether this affects their power situation in the next five years or the next fifteen.
Why the SMR timelines deserve real skepticism
History gives good reason to expect these 2030-plus dates to slip further, not arrive early. Russia’s KLT-40S small reactor took 13 years from construction start to generating electricity, against an expected 3. China’s twin high-temperature gas-cooled reactors took more than double their promised 50-month timeline. In the US, the Department of Energy’s own 2020 goal of having advanced reactors operational within 5–7 years has already shifted to 2031 — an eleven-year slip from the original target, and there is currently no small modular reactor under construction anywhere in the United States.
The investment amounts also don’t yet match the ambition. A single Natrium reactor — the design Meta and TerraPower are betting on — is estimated to cost $9.4 billion for 345 megawatts. Amazon’s $500 million investment in X-energy and TerraPower’s $650 million fundraise are real money, but they cover only a small fraction of what a single commercial-scale reactor actually costs to build. And critics note some agreements are looser than the press coverage implies — Amazon’s arrangement in one case secures only the right to purchase electricity, without pricing terms or a firm obligation to buy.
Need Expert Guidance?
Talk to a Data Center Expert
21+ years of hands-on experience in data center design, operations & infrastructure. Book a quick discovery call to discuss your project.
📞 Book a Discovery CallThe scale problem nobody’s press release mentions
Even in the best case where every announced SMR gets built on schedule, the math is humbling. A single SMR produces less than 0.3 GW. Individual large AI data center campuses now require 5 to 7.65 GW of power on their own. You would need something like twenty to twenty-five SMRs, all built successfully and on time, to power one hyperscale campus — against a global track record where no Western SMR has yet reached commercial operation at all. Nuclear is not going to relieve the power-constrained colocation market in the next three to five years, full stop. It’s a decade-plus supply-side bet, not a near-term fix for the grid interconnect queues currently running four-plus years in markets like Northern Virginia.
The cost reality, if it does arrive
Even once built, nuclear isn’t cheap relative to the alternatives. First-of-a-kind SMR projects are estimated at $80–150 per megawatt-hour, with vendors targeting $60–80/MWh once designs mature (the IEA’s advanced nuclear estimate sits at $63.10/MWh). Compare that to solar at $30–50/MWh and wind at $25–45/MWh, and nuclear’s real selling point isn’t cost — it’s the thing solar and wind can’t offer, which is dispatchable, round-the-clock baseload power that doesn’t depend on weather. That’s genuinely valuable for a data center, which is exactly why hyperscalers are paying the premium — but it means nuclear power arriving on your grid won’t necessarily translate into cheaper colocation rates even once it’s live.
What this actually means if you’re evaluating a site or provider today
Two practical takeaways matter more than the headline gigawatt figures. First, when a provider or a region touts a “nuclear-powered” facility, ask specifically whether the power comes from an existing operating plant (like Talen’s Susquehanna arrangement) or a planned SMR that hasn’t broken ground — those are fundamentally different claims wearing the same marketing language, and only one of them affects your power reliability today. Second, treat every nuclear-adjacent site-selection pitch with a 2030-or-later delivery date as a long-horizon hedge for your organization’s future capacity planning, not a factor that changes anything about the power-constrained market you’re actually buying into right now. The grid interconnect queues, the record-low vacancy, and the per-kW price increases covered in our colocation pricing analysis are the reality for the next several years regardless of how many gigawatts get announced this earnings season.
Understanding which power claims are real today versus real in 2032 is exactly the kind of due-diligence distinction covered in the power infrastructure chapter of the Data Center Buyer’s Guide — so you’re evaluating a provider’s actual power position, not their press release.
Written from 15+ years running data center design, operations, and project management.
Sources
- Axis Intelligence — Nuclear Energy for Data Centers 2026: 9.8 GW Committed, SMRs by 2029
- SMR Intel — Every Nuclear-Powered Data Center Deal: Google, Amazon, Meta & Microsoft (2026)
- Bulletin of the Atomic Scientists — Data centers powered by next-gen nuclear? Don’t fall for Big Tech’s PR
Written by
Raajeev Ratra
Data Center Infrastructure Expert | 15+ Years in DC Design, Operations & Project Management
Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.