The Three-Party Model Driving Advanced Nuclear Commercialisation

For the first time in the modern nuclear era, the entities absorbing the financial risk of building unproven reactors are not power utilities. They are artificial intelligence companies and the Pentagon.

That inversion sits at the heart of a structural shift in how advanced nuclear gets financed. The Kairos Power and Google fleet agreement, the June 2026 criticality milestone at Antares Nuclear, and remarks by Kairos Power’s Candice Yu at the inaugural Los Alamos Nuclear Forum on 25 August 2026 all point to the same conclusion: the commercial model being tested right now will decide whether advanced nuclear scales across the grid or stays a bespoke arrangement for a handful of deep-pocketed buyers.

Here is how to read who is actually paying for the first wave of advanced nuclear, and what that tells you about where the sector heads next. This is a structural lens on a market in formation, not a news recap.

Why the old financing model broke down for advanced nuclear

Rate-regulated utilities are not built to be the risk-absorber for first-of-a-kind (FOAK) reactors. The reason is regulatory, not technological.

Utilities recover their costs through rates approved by state commissions, and those commissions apply a prudence test: every major expenditure has to be justified as a reasonable bet on behalf of captive ratepayers. Signing a long-term offtake agreement for a reactor design that has never operated commercially is precisely the kind of commitment that test is designed to scrutinise.

The prudence standard for utility cost recovery requires that every major capital expenditure be demonstrably reasonable at the time it was made, a threshold that first-of-a-kind reactor commitments struggle to clear because no comparable commercial operating record exists against which state commissions can benchmark the decision.

The scale problem compounds the prudence problem. According to Candice Yu, director of business development at Kairos Power, unlocking the cost reductions that make advanced nuclear financeable requires “committed order books” of 5-10 deployments of a single reactor design. A single utility, acting alone, cannot credibly commit at that scale.

Hyperscalers and defence customers can. Their profiles differ from utilities in three ways that matter:

• Developers build and own the FOAK reactors, carrying the technology risk.
• Hyperscalers and defence buyers provide the demand signal and absorb development risk through long-term contracts, drawing on non-regulated revenues and a genuine need for 24/7 firm baseload power.
• Utilities handle grid integration and transmission, gaining firm zero-carbon capacity without betting their balance sheets on first-time technology.

The distinction turns on how each party values the power. Wholesale markets price electricity as a commodity. AI data centres and military installations place additional value on carbon-free attributes and energy resilience that those markets do not fully compensate, which lets them rationalise above-market early-stage costs.

The read for investors is this: the entities that control the offtake commitments control who gets built first. This financing gap is not a temporary inefficiency waiting to self-correct. It is a structural feature of regulated energy markets, which means the three-party model is not a clever workaround. It is the only viable path for FOAK advanced nuclear at scale.

How defence procurement sidesteps the financing barrier

The military reaches the same destination by a different road. Through Other Transaction (OT) authority, a procurement mechanism that bypasses traditional acquisition processes, the Department of Defense (DoD) can treat FOAK reactor capital as a national security investment rather than a rate-based asset.

Because defence programmes run on appropriated federal funds, the cost-recovery calculus is entirely different from a utility’s balance sheet exposure. There is no state commission to satisfy and no ratepayer to protect. The reactor is funded as mission-critical energy resilience, which removes the prudence constraint altogether.

The three-party model in practice: Google, Kairos, and TVA

The theory becomes concrete in the Kairos Power and Google agreement. In October 2024, the two signed a Master Plant Development Agreement under which Kairos develops, constructs, and operates a series of advanced small modular reactor (SMR) plants, while Google commits to long-term power purchase agreements (PPAs) across a fleet rather than a single facility.

The fleet structure is the whole point. The agreement targets up to 500 MWe of carbon-free electricity for Google data centres by 2035, delivered across an implied six to seven reactors. Standardised construction across that many units is what generates the cost reductions; one plant cannot produce them.

The first live test is Hermes 2. In August 2025, Google, Kairos Power, and the Tennessee Valley Authority (TVA) announced plans to install the reactor on TVA’s grid by 2030, with TVA purchasing up to 50 MW under a PPA to serve Google data centres in Tennessee and Alabama.

The detail most investors will miss is TVA’s positioning. TVA is not the risk-taker here; it is the integration layer, buying firm capacity and handling grid connection without betting its balance sheet on unproven technology. That is exactly what makes the model replicable for other utilities willing to move quickly.

A critical turning point Candice Yu characterised nuclear deals involving major technology companies as “a critical turning point” for the advanced nuclear market. Los Alamos Nuclear Forum, 25 August 2026.

The practical takeaway is that the fleet commitment, not any single plant, is the mechanism that unlocks supplier cost reductions. When you see a single-plant announcement, the question to ask is whether a fleet framework sits behind it. If it does not, the economics that make FOAK financeable are probably absent.

What the defence track adds that hyperscalers cannot

The defence track is not a parallel version of the same story. It contributes something commercial customers structurally cannot: technical and regulatory validation.

That validation arrived on 4 June 2026, when Antares Nuclear’s Mark-0 reactor achieved zero-power criticality at Idaho National Laboratory (INL) under the Department of Energy’s (DOE) Reactor Pilot Program. It was the first advanced reactor in the United States to reach criticality in more than 40 years.

The path to that milestone and beyond runs through a sequence of defence contracts:

1. April 2025: The Defense Innovation Unit (DIU) named Antares eligible for its Advanced Nuclear Power for Installations (ANPI) programme.
2. April 2026: The Department of the Air Force and DIU selected Antares for prototype deployment at Joint Base San Antonio, with a 2027 demonstration and 2028 deployment target.
3. June 2026: The Mark-0 reactor reached zero-power criticality at INL.
4. August 2026: The US Army and DIU awarded Antares an OT agreement to build and operate microreactors at Fort Bragg, North Carolina, under the Janus programme.

Following the criticality milestone, Antares was selected for the DOE’s Nuclear Energy Launch Pad programme, which facilitates authorisation to operate reactors outside national laboratories.

NRC licensing precedents established through existing plant extensions are shaping how the commission approaches first-of-a-kind advanced designs: the evidentiary and procedural standards applied in the Diablo Canyon licence extension are being cited by developers and regulators alike as reference points for what a standardised commercial licensing pathway might look like.

What defence uniquely delivers is de-risking before commercial operators ever encounter the technology: proving out the regulatory pathway on military sites, assembling federal and private funding in combination, and embedding national laboratory expertise throughout development.

For investors tracking the sector, this reframes how to read defence news. Military contract milestones function as leading indicators for commercial viability, because the defence track is doing the technical and regulatory work that lowers the cost of the first commercial operator’s bet.

The blended capital model defence enables

The defence route assembles a capital stack no commercial transaction can build alone. Appropriated federal funds, DOE cost-sharing, and private equity combine in a way that spreads FOAK risk across public and private balance sheets simultaneously.

According to Matt Griffin, head of nuclear affairs at Antares Nuclear, getting these deployments to scale demands a specific combination of resources: federal funding alongside private capital, close collaboration with national laboratories, and a structure in which government absorbs a meaningful share of the risk. A purely commercial deal has access to none of the federal appropriations or laboratory infrastructure that make this stack work, which is why the defence track is a distinct engine rather than a duplicate of the hyperscaler model.

The capital landscape and the constraints that determine who wins

The demand signal is credible. By the start of Q3 2025, nuclear fission companies had raised a record $1.3 billion in equity funding. Over $10 billion in US federal investment has flowed into new nuclear since 2020, and major technology companies have committed to around 10 GW of new capacity.

The market opportunity is sized accordingly.

The headline number The US nuclear capacity expansion market is sized as a $57 billion opportunity through 2035.

The capital, in short, is largely committed. The harder question is which constraints threaten to decouple that commitment from actual deployment. The obstacles are downstream of the reactor technology:

• HALEU supply: Most advanced designs rely on high-assay low-enriched uranium (HALEU). US demand is projected at roughly 50 metric tons annually by 2035 against heavily constrained domestic production.
• Precision manufacturing: Limited domestic capacity for specialised components.
• Craft labour: A shortage of skilled nuclear construction workers.
• Waste management: Some SMR designs produce more voluminous and chemically reactive waste than conventional reactors, and there is no centralised organisation to handle it for small operators.
• Multi-agency regulatory sequencing: Approvals across DOE, the Nuclear Regulatory Commission (NRC), the Department of Transportation, and DoD limit replicability.

Chris Stanek, director of nuclear energy programmes at Los Alamos National Laboratory, notes that early reactor designs are among the least efficient. Stephanie Booth, R&D manager at Sandia National Laboratories, says research into new fuel configurations aimed at lowering waste volume is ongoing.

Sorting the tractable from the intractable is where the analysis earns its keep. Precision manufacturing and craft labour can scale with investment inside the 2030 window. The HALEU supply gap is the single constraint most capable of decoupling the financial commitments already made from actual reactor deployment. For investors, domestic enrichment capacity is the canary: watch it to judge whether 2030 targets are real or aspirational.

The GAO’s 2026 findings on HALEU supply chain gaps document the specific production shortfalls, governance fragmentation, and enrichment bottlenecks that separate the financial commitments already on the books from the fuel volumes required to honour them.

Whether this model scales beyond its founding customers

The honest read is that the three-party model works for the entities that built it. Whether it becomes a standardised template for the broader grid remains open.

The historical parallels cut both ways. The current arrangement mirrors early government-anchored nuclear programmes, where federally operated experiments became subsidised demonstration plants before private deployment. It also echoes the renewable energy PPA playbook, where hyperscalers underwrote early wind and solar. Neither precedent resolved cleanly into broad commercial replication, and internationally the pattern is spreading, with data centres already acting as direct offtakers for firm power outside the United States.

For the model to generalise, three conditions have to hold:

1. Multiple FOAK projects operating simultaneously, proving the fleet economics at scale.
2. A standardised regulatory pathway outside national laboratories, which the DOE Launch Pad function is intended to provide.
3. A HALEU supply chain that reaches genuine commercial scale.

The scale of the ambition is large. The global unrisked SMR pipeline reached 47 GW in early 2025, up 14 GW quarter-over-quarter, with roughly 53% planned for the United States and an estimated $360 billion of investment required.

The honest conclusion is that the model has proven it can get FOAK reactors built. The jump from bespoke anchor-customer deals to a broadly replicable grid-scale template requires conditions, fuel supply, regulatory standardisation, and multiple simultaneous operating plants, that are not yet in place.

What a failure to generalise looks like for the sector

The downside scenario is not collapse; it is confinement. Advanced nuclear keeps attracting capital and produces operational plants, but utility-scale replication does not arrive before 2035, leaving sector-wide returns concentrated among participants in the anchor-customer deals.

That outcome would still mark progress against the pre-2024 baseline, when almost none of this existed. But it would disappoint the investment theses currently priced into the broader sector, which assume the model generalises rather than stays bespoke. For investors, the distinction between “this works for Google and the Pentagon” and “this scales to the grid” is the line between a genuine catalyst and a premature extrapolation.

Reading the advanced nuclear sector as it actually is in September 2026

The genuine achievement is structural and durable. FOAK risk has been reassigned away from utilities that could never rationally carry it and onto entities that can absorb it. That change is real and unlikely to reverse.

What remains unresolved is whether fuel supply catches up, whether regulatory pathways standardise outside military sites, and whether fleet commitments translate into sequential, on-schedule completions. The near-term proof points are Hermes 2, targeted for 2030 installation, and the Antares demonstration at Joint Base San Antonio in 2027, with multi-FOAK commercial operation broadly expected around 2035.

Watch these milestones as they arrive:

• Hermes 2 NRC licensing progress ahead of its 2030 target.
• The Antares Joint Base San Antonio demonstration in 2027.
• Domestic HALEU enrichment capacity announcements.

Candice Yu’s framing points to who benefits: the utilities best positioned are those willing to pursue flexible contracting and move with speed. The structural shift is real, but treating current hyperscaler and defence commitments as proof that grid-scale advanced nuclear arrives by 2030 extrapolates further than the evidence supports. A calibrated read separates the progress already locked in from the conditions still required, which is the basis for sizing a position to match the actual state of the evidence.

Nuclear capacity tripling by 2050 is the macro frame inside which the three-party FOAK model occupies an early but critical position: the analysis of what tripling actually requires in terms of capital deployment, fuel supply, and workforce development contextualises why the current anchor-customer deals are necessary preconditions rather than the end state.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors, and forward-looking statements are speculative and subject to change based on market developments and company performance.

—