LONGActiveMed3Y

Nuclear Is the Only Power Source That Scales for AI Data Centers

By MenFem Editorial·Energy — Nuclear·13 April 2026·Methodology·

ai-infrastructureenergypowernuclearConstellation

Key Points

AI data centers need 24/7 baseload power — renewables can't deliver reliability at scale
Constellation Energy restarting TMI Unit 1 under 20-year Microsoft PPA — first nuclear restart driven by AI demand
SMR regulatory pathways clearing — NuScale, Kairos, TerraPower all advancing
Projected 30-50% increase in US electricity demand from AI data centers by 2030
Nuclear plant owners signing 20-year PPAs at premium rates — long-duration value lock-in

AI data centers need 24/7 baseload power at enormous scale. Renewables are intermittent. Natural gas faces emissions targets. Nuclear is the only carbon-free baseload power source that can scale to meet the projected 30-50% increase in US electricity demand driven by AI. Constellation Energy is restarting Three Mile Island Unit 1 under a 20-year PPA with Microsoft. NuScale and other SMR developers have regulatory pathways clearing. The nuclear renaissance isn't driven by climate policy alone — it's driven by AI compute demand creating a power buyer willing to pay premium prices for guaranteed baseload. The thesis is straightforward: AI companies need reliable, large-scale power. Nuclear is the only technology that delivers all three attributes (reliable, large-scale, carbon-free) simultaneously. The companies that own nuclear capacity or are building next-generation nuclear plants are positioned to capture premium power purchase agreements for the next two decades.

Research Log

11 Sept 2026

Re-underwritten 2026-09-11 (Connor's ruling: sharpen the falsifier). The old invalidation named "the bulk of newly contracted baseload going to gas" with no threshold, date or source. Replaced with a >50%-by-GW / 2028-12-31 / LBNL + SemiAnalysis condition, plus the Crane restart as a hard leg. Evidence: SemiAnalysis "What is So Hard About Behind-The-Meter Power" 2026-09-10 (75 GW firm BTM orders, ~20 GW in Q2 2026, recip gas engines expected to take the lion's share); NRG Q2 2026 call (1.2 GW CCGT for a hyperscaler); LBNL Queued Up 2026 (gas queue +86% to 253 GW). On this evidence the condition is currently being met. Conviction MEDIUM and status ACTIVE unchanged; grading stays Connor's at the review date. Previous invalidation text: "The reason for holding this stops being true if the buyers pick something else — the bulk of newly contracted data-centre baseload going to gas, or to renewables-plus-storage that solved intermittency. 'Only nuclear scales' would then be wrong however good the physics argument reads. It also stops being true if the Three Mile Island restart fails or is abandoned: this call names it as the proof case, and a proof case that does not prove is a thesis break, not a delay."

source: docs/research/thesis-review-2026-09-11.md

7 Sept 2026

Sources rebuilt primary-first (catalogue II item 69): 3 primary, 1 secondary kept.

source: docs/plans/markets-work-catalogue-2026-09-07.md

6 Sept 2026

Re-underwritten 2026-09-06. The review proposes INVALIDATED: LBNL Queued Up 2026 shows active gas in the US interconnection queue up 86% to 253 GW while solar, storage and wind all fell, GE Vernova sold out to 2031, and Constellation itself serves ~1.16 GW of CyrusOne load from a gas plant while its own nuclear output fell (its 2026-09-06 dossier). The TMI/Crane leg has NOT fired (FERC waiver and NRC fuel licence granted, 2H-2027). Connor: verdict held for his own ruling in /admin/market-calls — the row stays ACTIVE with this evidence logged. The restatement the evidence would support: nuclear is the only firm, clean, 24/7 source at scale and cannot be added fast enough this decade, so gas is taking the load.

source: docs/plans/call-reviews-2026-09-06-thematic.md

6 Sept 2026

Sources filled from the shelf and the 2026-09-06 review docs (catalogue item 3). 3 entries.

source: docs/plans/markets-work-catalogue-2026-09-06.md

Bull Case

Nuclear becomes the premium power source for AI infrastructure. Constellation, NextEra, and SMR developers sign multi-decade PPAs at rates that justify massive capital investment. Nuclear plant operators re-rate from utility valuations to infrastructure growth valuations.

Bear Case

Regulatory delays push nuclear restarts and new builds beyond 2030. Meanwhile, solar + battery storage achieves reliable baseload at lower cost. AI demand growth slows or distributes geographically, reducing the concentration of power demand.

What would prove this wrong

Wrong if, by 2028-12-31, a majority (>50% by GW) of firm US datacenter baseload contracted since 2026-04-13 is non-nuclear, measured on LBNL "Queued Up" plus SemiAnalysis's behind-the-meter tracker. Also wrong if the Crane (TMI-1) restart is abandoned or slips past 2028. On evidence to 2026-09-11 (75 GW firm BTM orders, gas expected to take the lion's share; gas queue +86% to 253 GW) the condition is currently being met, so the call is failing.

Catalysts

Three Mile Island Unit 1 restart timelineProduct Launch

Constellation Energy's restart of TMI Unit 1 for Microsoft. Concrete construction and licensing milestones validate the nuclear-for-AI thesis.

SMR regulatory approvalsRegulatory

NRC design certifications and construction permits for small modular reactors. Each approval de-risks the technology path.

Risk factors

Regulatory and permitting delaysHigh

Nuclear permitting timelines remain the biggest risk. NRC review processes can add years. Public opposition at specific sites can block construction.

Cost overruns on new buildsMedium

Nuclear construction historically runs over budget. Vogtle Units 3-4 came in at 2x original cost. SMRs claim to solve this but are unproven at scale.

Renewable + storage closes the gapMedium

If battery storage costs continue falling and grid-scale storage solves intermittency, the nuclear advantage narrows.