Samsung’s Kairos Deal Highlights the Long Wait Behind AI Nuclear Power

Samsung C&T’s newly reported support for Kairos Power may help build a Google-linked reactor, but it does not put new electricity on the grid sooner. The project remains targeted for 2030 and holds construction permits, not operating authorization. That gap is the central finding of a review of five recent AI-linked nuclear arrangements: signed capacity can mean delayed purchases from existing plants, future upgrades, construction-stage generation, or reactor development that has years of work ahead.

The review covers Google’s agreements with Fortum and Kairos, plus Meta’s agreements with Vistra, Oklo and TerraPower. It classifies each arrangement by the status of the underlying resource, disclosed commercial structure, remaining dependencies and stated delivery timing. The result is a bounded comparison, not a market-wide count: all five arrangements are linked to AI infrastructure, but they are not equivalent claims on electricity available now.

Samsung adds execution support, not delivered power

Samsung C&T will make an equity investment in Kairos and provide engineering services worth up to $100 million, according to TechCrunch. Hermes 2 is Kairos’s first commercial-scale reactor and its first deployment under the company’s deal with Google. The project is intended to supply up to 50 MW to the Tennessee Valley Authority grid for Google-linked data centers in Tennessee and Alabama.

Yet the support does not change the project’s delivery clock. Kairos began construction in April 2026 and targets 2030 for Hermes 2. The Nuclear Regulatory Commission issued construction permits for the facility in 2024; permission to operate remains a separate requirement. The deal therefore improves the project’s financing and engineering backing without turning contracted prospective generation into usable electricity.

Operating generation still arrives on contract schedules

Meta’s 20-year Vistra power purchase agreements cover 2,176 MW of operating generation from Ohio plants and 433 MW of future output increases at three plants in Ohio and Pennsylvania. Purchases begin in late 2026, while the full 2,609 MW is expected by 2034. That makes the operating portion the closest arrangement in the sample to near-term contracted supply, while leaving a meaningful share of the total dependent on uprates.

Google’s 22-year agreement with Fortum concerns the operating Loviisa plant in Finland, but it begins at smaller capacity in 2028 and reaches 50% of the plant’s capacity from 2030 through 2049. The PPA provides revenue certainty for a life-extension program, for which Fortum says substantial investment decisions remain pending. Google’s planned Finnish infrastructure construction runs from 2027 to 2028, but public disclosures do not say when individual facilities will open or what power each will require. A precise facility-by-facility lag cannot be calculated.

Future-reactor capacity comes with different commercial rights

• Kairos has a grid power purchase arrangement, with Hermes 2 positioned as the first 50 MW of its Google deal.
• Oklo describes Meta prepayment and development funding for a planned Ohio campus; pre-construction and site characterization are slated for 2026, and the first phase is targeted as early as 2030.
• TerraPower says Meta will fund development of two Natrium units and receive energy rights for up to six additional units. The companies planned to identify a site for the first dual-reactor unit in coming months.

The timing becomes clearest against Meta’s own infrastructure schedule. Meta targeted its 1 GW Prometheus AI cluster for 2026. Oklo’s first phase is targeted for 2030, a disclosed gap of at least four years, and TerraPower’s initial units are projected no earlier than 2032, at least six years later. These are date comparisons, not proof that either project was meant to exclusively power Prometheus: Meta describes the projects as supplying grids that support its regional operations.

The next proof point is operational

This distinction does not make the commitments trivial. Fortum says Google’s PPA supports the economics of Loviisa’s life extension, and Vistra says Meta’s agreements give it certainty to pursue license extensions and uprates. Future-reactor agreements can similarly supply developers with funding or customer backing. But only Vistra and Fortum disclose 20- and 22-year terms in this sample, while public materials do not provide complete allocation, pricing, remedies or conditions for the agreements.

The practical test now is progress rather than pledges: Loviisa investment decisions, Hermes 2’s construction and operating authorization, a site for TerraPower’s initial units, and regulatory and construction milestones for Oklo. Samsung’s support is evidence that Kairos has added project backing. It is not evidence that the 50 MW is already available. Across this sample, the most accurate reading is a portfolio of delayed operating-plant purchases and longer-dated bets on nuclear generation still being built.