Nuclear's scarce input is construction capacity; Kairos just bought some with equity

A developer with no operating revenue cannot buy EPC capacity the way an independent power producer buys a turbine, so when Kairos Power sought Samsung C&T for the construction team on its 50MW demonstration reactor at Oak Ridge, Tennessee, it paid the Korean firm in the only currency it holds: a stake in a reactor program a hyperscaler has already committed to buy power from. Kairos has handed Samsung C&T a place on that team and is taking up to $100 million from the Korean engineering firm in return, Data Center Dynamics reported on September 22, with the money arriving partly as an equity investment in Kairos and partly as in-kind engineering services, subject to regulatory approvals, and Samsung joining the integrated engineering, procurement and construction team Kairos is assembling alongside partners the report does not name.

The trade answers a constraint that has become familiar across the data center buildout: delivery limits have moved past steel and labor into supplier slots, and those slots increasingly get priced in equity rather than in contracts. A first-of-a-kind design also carries a first-of-a-kind supply chain, which is what the integrated EPC team is meant to solve, and Samsung is taking the developer's paper rather than a fee for the first unit, putting part of its money in as engineering labor rather than cash — a position in a program rather than a contract on a demonstration.

Samsung's willingness to hold equity in Kairos rather than simply bill it suggests the contractor expects a stream of units, not a single build. That expectation runs straight through Google.

Kairos broke ground at the Hermes 2 Demonstration Plant in April, and the unit is slated for completion by 2030; it would be one of the first Generation IV reactors built, if not the first, though the coverage hedges the claim and anyone underwriting the schedule should hedge it too. The design, the Fluoride Salt-Cooled High-Temperature Reactor, uses fluoride molten salt as a coolant and a ceramic, pebble-type fuel to carry heat to a steam turbine, a departure from the water-cooled Generation III fleet that makes up the operating nuclear base.

Google's position is far larger than the demonstration unit: power from Oak Ridge flows onto the Tennessee Valley Authority's grid to support Google data centers under a PPA signed in August of last year, which puts an incumbent utility with its own generating fleet in the middle of a hyperscaler's clean-power claim instead of a behind-the-meter host. Google backed Kairos in 2024 with a corporate agreement to purchase up to 500MW across six or seven reactors, and dividing that ceiling by the reactor count implies commercial units somewhere between 70MW and 83MW, larger than the demonstration unit itself.

The permission that matters at Oak Ridge belongs to a utility that is neither the buyer nor the builder, a variant on a rule this buildout keeps teaching: grid access, not generation, carries the underwriting risk. It is also a reminder that a hyperscaler's clean-power claim gets routed through someone else's dispatch desk before it reaches a server rack.

Kairos is not alone in courting this buyer class: Crusoe has partnered with Aalo Atomics on what could be the first nuclear-powered AI data center, at Idaho National Laboratory, while Data4, Oracle, Switch and Equinix have all put their weight behind small modular reactor technology. What the buyers appear to have concluded is that waiting in an interconnection queue is a worse bet than funding the supply side directly, through PPAs, through equity, and increasingly through both in one transaction.

The $100 million is disclosed; the ownership percentage it buys is not, and neither is the price in the Google PPA. That gap has run through project announcements in this sector for a while, with one inversion at Oak Ridge: the offtake is contracted and the customer is named, so the unpriced item has moved all the way to the developer's cap table. Until the equity math surfaces, Samsung's commitment looks more like a position taken in a reactor franchise than a construction contract priced at market.

Mike Laufer, Kairos's chief executive and co-founder, tied the investment and the partnership to the delivery model at the center of the company's commercialization strategy, and to the cost and schedule certainty it wants to sell to customers. Alex CW Lee, who heads energy solutions at Samsung C&T, described the collaboration as groundwork for broader commercialization of the reactor. Both statements address capability rather than dates, which leaves 2030 standing as the only schedule in the transaction.

The points that matter next are specific: which partners fill out the rest of Kairos's EPC roster, what ownership share Samsung's money buys, and whether the regulatory approvals the investment is conditioned on arrive on a timeline that keeps 2030 intact. Google's 500MW option converts into supply only if the demonstration converts into a construction cadence, and Kairos has now sold part of itself to improve the odds that it does.