NUCLEAR / SMR

Firm carbon-free power is the AI build-out's scarcest input

55/100 conviction

▼ -7 weakening

Last reviewed 16 July 2026 · our own structural assessment, not a price target

The thesis

Firm, 24/7 carbon-free power is a growing structural constraint on data-centre expansion — but honestly, this is two theses in one. The near-term leg (existing-fleet restarts, uprates and PPAs) is executing with real capital, though on slipping timelines. The SMR leg is genuine long-duration optionality — mostly 2030+, with unproven first-of-a-kind economics and a HALEU fuel bottleneck. Nuclear is one of the clearest LONG-duration answers, not the quickest or cheapest fix for the 2026–2030 crunch, where the IEA expects renewables and gas to do most of the near-term work.

The case for

Demand is real and growing: the IEA (Energy and AI, Apr 2025) projects data-centre electricity roughly doubling to ~945 TWh by 2030. The near-term action is in the existing fleet: Microsoft's 20-year PPA with Constellation to restart Three Mile Island Unit 1 (Crane Clean Energy Center, ~835 MW, originally targeted 2028 and since accelerated toward 2027), and Talen's expanded agreement to supply Amazon up to 1,920 MW from Susquehanna through 2042. The marquee proof point is Palisades (Michigan): a finalised $1.52B DOE loan guarantee (agreement executed 31 Jul 2024), and as of early July 2026 Holtec reports the major restart work closed out and the site in final pre-startup — positioned to be the first-ever US recommissioning of a retired reactor. On the SMR leg, TVA's BWRX-300 at Clinch River is the first US BWRX-300 construction-permit application accepted for NRC review (docketed July 2025), and it runs on conventional light-water fuel — far less exposed to the HALEU bottleneck than Oklo, Kairos or X-energy.

The case against

The near-term leg's marquee proof point has not happened. Palisades (Michigan) targeted an October 2025 restart and HAS NEVER RESTARTED: the NRC's daily Power Reactor Status Report shows it at zero power on 16 July 2026 — and at zero on every one of the 311 days it has been listed since 9 September 2025. It appears in the report at all only because Holtec rescinded its certifications of permanent cessation on 25 August 2025 — a regulatory return to operating status, not a physical one. Holtec's own 2 July 2026 update says its focus 'remains on safely completing the remaining work required as we prepare for fuel load' — fuel had not yet been loaded. Note also a category correction we previously got wrong: Palisades is a 1971-vintage Combustion Engineering pressurized water reactor of ~793 MWe net, permanently shut on 20 May 2022 and now being recommissioned — a CONVENTIONAL restart, not an SMR. Holtec has separately proposed two SMR-300 units (~680 MWe combined) at the same site, docketed by the NRC on 27 February 2026; nothing is permitted or built. Read the hyperscaler deals carefully: the Talen-Amazon PPA reaches its full 1,680-1,920 MW no later than 2032 and shifts to a front-of-the-meter, grid-connected arrangement. 'Nuclear is the clearest answer' is overstated: the IEA has renewables meeting roughly half of data-centre demand growth to 2030 (about +450 TWh) and natural gas adding about +175 TWh of near-term firming, with first SMRs arriving around 2030. The SMR leg carries first-of-a-kind cost risk — NuScale/UAMPS cancelled its Carbon Free Power Project in Nov 2023 after IEEFA (2024) estimated per-kW cost rose from $9,964/kW (2015) to $21,561/kW (2023) — plus a fuel bottleneck: domestic commercial HALEU enrichment is not expected at scale until around 2029-2030 under DOE's $2.7B January 2026 task orders. Most hyperscaler SMR 'deals' remain LOIs/MOUs, not firm offtake.

What would change our mind

Re-scored 62 → 55 on 2026-07-14 after a fresh source and counter-case review. The cut reflects new disconfirming evidence: Palisades' restart has slipped past its Oct-2025 target (still not generating as of July 2026), partially triggering the thesis's own falsifier. The number is an honest blend of a stronger near-term existing-nuclear leg (restarts/uprates/PPAs — high conviction on value, but timeline-slippery) and a weaker SMR leg (2030+, first-of-a-kind cost + HALEU fuel risk, mostly non-binding deals). The two legs have very different risk profiles; a single blended score should be read with that in mind (a case for splitting this into two theses).

Where the exposure sits

Near-term firm-nuclear exposure sits with existing-fleet operators signing hyperscaler PPAs: Constellation Energy (NASDAQ:CEG, Crane/TMI restart), Talen Energy (NASDAQ:TLN, Amazon 1,920 MW), Vistra (NYSE:VST), and privately-held Holtec (Palisades). SMR/optionality exposure sits with earlier-stage names — Oklo (NYSE:OKLO, pre-revenue), NuScale (NYSE:SMR), GE Vernova/Hitachi (BWRX-300 at Clinch River), and X-energy and Kairos (private). Utility offtakers include TVA (federal). Note the SMR names are pre-revenue and volatile — the exposure profile differs sharply between the two legs.

Show our work — the evidence

The key evidence behind the score, dated and sourced — supporting and disconfirming. This is how the conviction is built, and what we re-check as it moves.

ContextTalen–Amazon 1,920 MW nuclear PPA runs through 2042 but ramps to full volume no sooner than 2032 and was restructured to a grid-connected front-of-the-meter arrangement.Company disclosureTalen Energy 8-K · 11 June 2025
ContextPalisades has never restarted: the NRC's daily Power Reactor Status Report shows it at zero power on 16 July 2026, and at zero on every one of the 311 days listed since 9 Sep 2025. Holtec (2 Jul 2026) says focus 'remains on safely completing the remaining work required as we prepare for fuel load'. Palisades is a 1971 Combustion Engineering PWR (~793 MWe net) shut on 20 May 2022 — a conventional restart, not an SMR.Official dataUS NRC — Power Reactor Status Report · 16 July 2026
AgainstUnder DOE's $2.7B January 2026 HALEU/LEU enrichment task orders, domestic commercial HALEU capacity is expected around the end of the decade — Centrus targets first new commercial capacity by ~2029 and General Matter's Paducah plant by 2030 — a fuel bottleneck for advanced reactors (Oklo, Kairos, X-energy).Official dataUS Department of Energy — $2.7B uranium enrichment task orders · 5 Jan 2026
AgainstIEA: renewables meet roughly half of data-centre demand growth to 2030 (~+450 TWh), natural gas adds ~+175 TWh of near-term firming, and nuclear begins to contribute materially only toward the end of the decade, with first SMRs online around 2030.Official dataIEA — Energy and AI · 1 Apr 2025
ContextNuScale/UAMPS terminated the Carbon Free Power Project (Nov 2023); per IEEFA (2024), estimated cost rose from $9,964/kW (2015) to $21,561/kW (2023), with target project cost climbing from ~$3B to $9.3B.Analyst estimateIEEFA · 1 May 2024
ForHoltec announced Palisades restart reached a watershed moment on 2026-07-02: major restart projects successfully closed out, site transitioned from large-scale work to final routine maintenance, testing, inspection and operational-readiness activities (~5,000 individual activities remain) before startup. Turbine-generator placed on turning gear; new fuel-handling system installed and tested.Company disclosureHoltec International / GlobeNewswire · 2 July 2026
ForIEA projects data-centre electricity roughly doubling to ~945 TWh by 2030.Official dataIEA — Energy and AI · 1 Apr 2025
ContextDOE closed a loan guarantee of up to $1.52 billion to Holtec Palisades to finance the restart of an 800MW nuclear station — the first-ever recommissioning of a retired nuclear power plant in the US.Official dataUS Department of Energy · 9 July 2026

Tags mark how each fact is sourced: Company disclosure (the company said it) · Official data (a government or agency figure) · Analyst estimate (a third party's number) · Reported (via the press).