On April 21, 2026, Kairos Power broke ground on the Hermes 2 Demonstration Plant in Oak Ridge, Tennessee — the first commercial-scale Gen IV reactor in the United States to hold an active Nuclear Regulatory Commission construction permit. On May 13, 2026, the NRC issued an order extending the latest date for completion of Hermes 1 — the 35 MWth low-power test reactor on the same Oak Ridge campus — from December 31, 2026 to April 30, 2029. A 28-month slip on the small reactor. A first concrete pour on the bigger one. Same campus. Eight days apart.
If you read only the headline on the May 13 order, the conclusion writes itself: advanced nuclear is late again, the schedules don’t hold, and the data-center crowd should keep buying gas turbines. That conclusion is wrong, and the reason it is wrong is the reason Hermes 2 is in the ground.
What the 28-Month Extension Actually Means
The original Hermes 1 construction permit, issued December 12, 2023, set a latest completion date of December 31, 2026. On March 24, 2026, Kairos submitted a license amendment request to push that date out. Eight weeks later the NRC said yes. The agency’s stated grounds match Kairos’s own filing: first-of-a-kind construction techniques and design features required non-nuclear demonstrations before they could be executed on the licensed facility. Translation — Kairos chose to prototype the construction process itself before pouring nuclear-grade work on the real plant.
That is not slippage in the way that Vogtle 3 and 4 slipped. Vogtle was a fixed-design, well-understood AP1000 that lost six years and roughly $20 billion to QA failures, rework, and a contractor bankruptcy. Hermes 1 is a 35 MWth research reactor that was designed to be the place where Kairos finds out where the design is wrong. That is what first-of-a-kind means in nuclear regulation: you are paying the learning cost on a small, instrumented, low-consequence platform before you pour the commercial plant. The 28-month extension is the receipt for that decision.
The harder question — the one the May 13 order does not answer — is whether the lessons from Hermes 1 will be ready in time to inform Hermes 2 construction. They will be partially. Kairos has structured the program so that the major construction techniques (modular reactor vessel installation, salt-loop fabrication, online refueling architecture) are demonstrated on Hermes 1 before they are repeated on Hermes 2, but the two builds will overlap substantially. That is a calculated risk, and it is the right one if you believe — as Kairos’s customers do — that the bottleneck on advanced nuclear is not technology, it is iteration count.
The KP-FHR Design — Why the Physics Tolerates the Schedule
The Kairos Power fluoride salt-cooled high-temperature reactor (KP-FHR) is what makes the program survivable. The fuel is TRISO — tristructural-isotropic coated particles, roughly the diameter of a poppy seed, with a uranium oxycarbide kernel wrapped in a porous carbon buffer, two pyrolytic carbon layers, and a silicon-carbide shell. About 16,000 of these particles are pressed into a 40 mm graphite pebble. The fuel can survive 1,600°C without melting, and the silicon-carbide layer retains essentially all fission products to that temperature.
The coolant is Flibe — a 2:1 molar mix of lithium fluoride and beryllium fluoride. Flibe freezes at 459°C and boils at 1,430°C, so the entire primary loop operates near atmospheric pressure with a working window almost a thousand degrees wide. Outlet temperature is 650°C, which is high enough to drive a steam Rankine cycle at competitive thermal efficiency and high enough to be useful as process heat for industrial customers if the AI buyers ever blink.
The reason this matters for the schedule story is the safety case. A KP-FHR cannot suffer a loss-of-coolant pressure transient — there is no pressure to lose. It cannot suffer a hydrogen explosion from zirconium-water reaction — there is no zirconium and no water in the core. The fuel form is engineered to retain fission products at temperatures higher than the reactor can physically reach in any credible accident. The defense-in-depth case is largely passive, which is what allowed the NRC to license a Gen IV reactor for construction in the first place — the first non-light-water reactor permitted in the U.S. in more than fifty years. A delay on the schedule does not put any of that physics at risk, because none of the physics changes when the calendar does.
Why Hermes 2 Is Already in the Ground
Hermes 2 is a 28 MWe / two-unit configuration (up to 50 MWe total) coupled to a steam-Rankine power conversion train. Kairos signed a power purchase agreement with the Tennessee Valley Authority and Google in 2024; the first MWh from Hermes 2 are to land on TVA’s grid in 2030, with Google taking the environmental attributes against its data center load in Tennessee and Alabama. The broader Google master agreement covers a 500 MW fleet across six to seven reactors by 2035.
Kairos broke ground on Hermes 2 on April 21, 2026 — before the May 13 Hermes 1 extension — because the commercial program does not wait for the demonstration. It cannot. The hyperscaler buyers have a 2030-window problem of their own.
The Nuclear-for-AI Demand Curve
U.S. data center power demand is on track to hit roughly 76 GW in 2026, up from about 50 GW in 2024, with AI workloads driving more than 40% of the incremental load. Over the trailing twelve months, the four major hyperscalers — Microsoft, Amazon, Google, Meta — have signed power purchase commitments covering more than 10 GW of new and restarted nuclear capacity. Microsoft contracted the Three Mile Island restart for 835 MW over 20 years. Amazon expanded its Talen Energy PPA at Susquehanna to 1,920 MW. Meta issued an RFP for 1 to 4 GW of new nuclear in late 2025. Google’s Kairos commitment is the only one of the four built around a fleet of advanced Gen IV reactors rather than restarted or expanded Gen II LWRs.
Nuclear’s structural advantage in this procurement cycle is the capacity factor. The U.S. nuclear fleet ran at 92.7% in 2024, against approximately 25% for utility-scale solar and 35% for onshore wind. The AI training cluster does not negotiate with the duck curve. The 24/7 firm power requirement is not a preference — it is a thermal-design constraint of the GPU racks, which lose efficiency and reliability when cycled with grid intermittency. Hyperscalers are paying nuclear PPAs at premiums to spot power because the alternative is overbuilding gas turbines and battery storage to firm up renewables, and the math does not favor that build-out at scale.
The Competitive Landscape
Kairos is one of four U.S. advanced reactor programs with material 2030-window deliverables. NuScale’s VOYGR pulled out of its lead Utah project in late 2023 and is now repositioning around industrial offtake; its 77 MWe LWR module is licensed but has no customer site under construction. TerraPower’s Natrium broke ground on the non-nuclear portion of its Kemmerer, Wyoming sodium-cooled fast reactor in 2024, but the company is still in pre-application review with the NRC on the actual reactor permit. X-energy’s Xe-100, an HTGR using TRISO pebbles like Kairos but cooled by helium gas, holds environmental clearance for the Dow Seadrift project in Texas and is targeting first criticality late this decade.
What separates Kairos in May 2026 is not that its technology is better than X-energy’s or TerraPower’s on a paper comparison — there are reasonable engineering arguments on each side. It is that Kairos is the only one of the four with two reactors physically under NRC-licensed construction in the same week. That is the iteration count Kairos has been buying with each Hermes-class build. The 28-month slip is the price of admission. The April 21 groundbreaking is the receipt.
What the Engineer Should Take Away
If you are sitting on the operations side of a refinery, a chemical plant, or a data center campus and you are watching the nuclear-for-AI cycle with a healthy operations engineer’s skepticism, here is what matters about the May 13 order. Advanced reactor schedules will slip. They will slip in ways that are visible (NRC orders, FOAK design changes) and ways that are not (component qualification, supply-chain delays on HALEU, weld procedure development on materials nobody has welded before). The schedules will not slip in ways that compromise the physics. The KP-FHR’s safety case rests on TRISO fission product retention up to 1,600°C and on a coolant that does not require pressure to function. None of that changes when the calendar does.
The hyperscalers know this. Their PPAs are priced and structured around it. Google did not buy 500 MW of paper from Kairos in 2024 because they believed Hermes 1 would be online in 2026 — they bought it because they believed the Kairos iteration cycle was real and the 2030 first-power date on Hermes 2 was the right number to anchor against. The May 13 extension does not move that number. It validates how Kairos got there.
For the engineer running compliance, training, and process safety on the conventional fleet that has to keep running through this build-out, the takeaway is different but adjacent. The same advanced-reactor program that took 28 months of FOAK extension to demonstrate construction techniques will require an operating culture and a documentation stack that does not exist yet. The 29 CFR 1910.119 PSM frame that governs the petroleum and petrochemical fleet will need its Gen IV analogue, and the gap between “advanced reactor operating procedures drafted in 2027” and “advanced reactor PHAs defensible at NRC inspection in 2032” is exactly the documentation problem AI-assisted engineering tools were built to close.
The Forward Look
The Hermes 1 extension does not change Porritt Inc.’s thesis on advanced nuclear, and it should not change yours. The KP-FHR continues to hold the strongest licensed-and-under-construction position of any Gen IV design in the United States. The Google PPA continues to anchor the largest dedicated hyperscaler offtake for an advanced reactor program. The 2030 first-power date on Hermes 2 continues to be the correct date to mark on the calendar. The May 13 order tells you the regulator is willing to extend on documented FOAK grounds — it does not tell you the program is in trouble. The April 21 groundbreaking tells you the program is not.
If you are building, financing, training, or compliance-assuring anything that depends on a 2030 nuclear deliverable, that is the right shape of the picture. Two reactors, one campus, eight days, two NRC documents. The math still works.
Timothy Porritt is founder of Porritt Inc., building AI-powered tools for process safety, engineering compliance, and industrial operations — including AI co-engineers for refinery PSM, NEXUS CAD for advanced-reactor and chemical-plant design, and NORMEX Standards AI for RAGAGEP modernization. Based in Salt Lake City, Utah. Reach the team at porrittinc.com/contact.