The mismatch is the whole trade. AI buildouts run on deployment time (quarters). Nuclear runs on licensing time and construction time (years). That is why the headlines about “SMRs powering AI” often read like near-term certainty, but the grid reality is that the quickest firm clean megawatts are still coming from the existing fleet.
The practical menu for “firm clean MW sooner” is not exotic:
1) Keep reactors online longer (license renewal / subsequent renewal).
2) Get more MW out of them (uprates).
3) Restart a small number of recently closed units where economics and politics line up.
SMRs still matter, but mostly as an orderbook for the 2030s unless permitting, supply chains, and financing compress dramatically.
[wnr_snapshot title=”AI + nuclear: what actually delivers MW — at a glance” items=”Fastest firm clean MW|Uprates + renewals|NRC-approved uprates already added 8,030 MWe;Existing fleet advantage|High utilization|U.S. nuclear ran at 92.7% capacity factor (2022);SMR reality|Long-dated option|First deployments hinge on licensing + FOAK execution;Restart pathway|Big blocks, still years|Restarts can return ~0.8–1.0 GW but require approvals + refurb;Contract truth|Physical vs credits|Credits don’t solve local congestion or peak capacity”]
[wnr_math title=”Quick math: turning “24/7 clean” into reactor math” items=”1 GW 24/7 load|8.76 TWh/yr|1,000 MW × 8,760 hours;Nameplate needed @ 92.7% CF|~1,079 MW|1,000 ÷ 0.927;835 MW unit @ 92.7% CF|~6.78 TWh/yr|835 × 0.927 × 8,760;NRC uprates energy|~65.2 TWh/yr|8,030 × 0.927 × 8,760;Vogtle capex density|~$14,318/kW|$31.5B ÷ 2.2 GW”]
What the “nuclear handshake” is really buying
Hyperscalers don’t buy nuclear because it is trendy. They buy it because it is one of the few scalable low-carbon sources that naturally matches a 24/7 load profile without massive storage buildouts. That makes nuclear the clean-power equivalent of “firm capacity,” not just “annual MWh.”
But there’s a catch: many deals marketed as “24/7 clean power” are actually contracts for attributes (credits) rather than physical electrons delivered into the load zone. That can satisfy corporate accounting while leaving the local grid still short on deliverable capacity at peak.
[wnr_table title=”What counts as “24/7 clean” in contracts” headers=”Structure|What the buyer gets|Why it’s used|Hidden catch” rows=”Physical PPA (existing reactor)|Power + attributes|Immediate, firm profile|May not be deliverable to the datacenter node; still grid-dependent;Book-and-claim credits|Clean attributes matched over time|Flexible across regions|Does not fix local congestion or peak capacity;Development PPA (SMR orderbook)|Future offtake pathway|Pulls forward supply chain|Delivery depends on licensing, fuel, FOAK execution;Co-location / tight coupling|Better local firmness|Reduces some congestion risk|Can trigger cost-shifting and regulatory fights”]
Where near-term MW actually comes from
Uprates and renewals are dull, but they are “real MW” on timelines that can beat greenfield projects. Uprates are particularly underrated because they stack across a fleet: small upgrades repeated across many units can equal multiple large reactors’ worth of capacity.
[wnr_table title=”Timeline-first view: the real menu for firm clean MW” headers=”Lever|Typical lead time|What it delivers|Primary gating item” rows=”Uprates (existing reactors)|~1–3 yrs|Incremental MW|NRC approval + equipment upgrades;License renewal / subsequent renewal|~2–5 yrs|Keeps MW online decades|Aging management + NRC review;Restarts (recent closures)|~3–6 yrs|~0.8–1.0 GW blocks|Refurb + approvals + politics;SMR first deployments|~6–10+ yrs|New 50–300 MW blocks|Licensing + FOAK execution + fuel;Large new builds|~10–15+ yrs|1+ GW blocks|Financing + project execution”]
Bottom line
The “AI nuclear” story that matters for the next 3–7 years is mostly existing fleet preservation and incremental upgrades, with a handful of restart-scale step changes. SMRs are the longer-dated bet.
Sources (primary)
• U.S. EIA (2023 explainer): U.S. nuclear capacity factor (92.7% in 2022) and industry context — https://www.eia.gov/energyexplained/nuclear/us-nuclear-industry.php
• U.S. NRC: Reactor power uprates (171 approved; 8,030 MWe added) — https://www.nrc.gov/reading-rm/doc-collections/fact-sheets/reactor-power-uprates.html
• U.S. NRC: License renewal overview and framework — https://www.nrc.gov/reactors/operating/licensing/renewal/overview
• Lazard (2024): LCOE+ and nuclear cost framing; Vogtle capex reference — https://www.lazard.com/media/2ozoovyg/lazards-lcoeplus-april-2024.pdf
Leave a Reply