Nuclear restart vs SMR vs traditional-large: hyperscaler PPA routes
Hyperscalers pursuing 24/7 carbon-free energy commitments have three practical nuclear routes: restart of previously-retired units (Microsoft/Constellation TMI reference), small modular reactor new-build (Kairos-Google, X-energy-Amazon), and traditional large-scale nuclear (AP1000, ESBWR). Each has materially different timelines, financing stacks, regulatory pathways, and vendor + operator counterparties. Understanding the trade-offs is a specific investment + capital-planning workstream.
01Bottom-line verdict
02Comparison table
| Restart | SMR (small modular reactor) | |
|---|---|---|
| Timeline to first MW | 2025-2028 (specific candidates: TMI Unit 1, Palisades, Duane Arnold reviewing) | 2029-2032 for first commercial deployments (Kairos, X-energy) |
| Reference deal | Microsoft/Constellation TMI 20-year PPA (~835 MW) | Google/Kairos + Amazon/X-energy (SMR pilot orders) |
| Capacity per deal | 500-1500 MW (unit-scale) | 50-300 MW per unit; multi-unit configurations common |
| CapEx per MW | Lower (existing asset restart) | Higher (first-of-a-kind risk premium) |
| Financing stack | Utility BS + hyperscaler prepay + DOE LPO + IRA credits + refinance | More equity, less debt-eligible early; DOE LPO + IRA credits + hyperscaler prepay |
| Regulatory pathway | NRC restart process (specific to existing site) | NRC design certification + first-site license |
| Vendor + operator | Existing operator (Constellation, Talen, etc.) | SMR OEM (Kairos, X-energy, TerraPower, GE Hitachi, Rolls-Royce, Westinghouse AP300) |
| First-of-a-kind risk | Low (restart of proven unit) | High (early commercial deployments) |
| Hyperscaler off-take structure | Long-term PPA (20+ years), often BTM-adjacent structure | Multi-decade PPA + potential equity + first-refusal rights |
03Decision framework
Choose Restart when
- Need power in the 2025-2028 window. Only restart delivers on this timeline
- Working with an existing utility operator. Constellation, Talen, PSEG et al have restart-ready assets
- Preference for proven-asset financing profile. Restart risk profile matches conventional infrastructure debt
- Regional constraint (must be at specific existing site). Restart is site-specific by definition
Choose SMR (small modular reactor) when
- Long-term horizon (2029+ commissioning acceptable). SMR timelines fit multi-year capital planning
- Multiple hyperscaler-scale sites needed. SMR standardisation supports multi-site deployment
- Willing to underwrite first-of-a-kind risk premium. Equity + DOE LPO structure fits FOAK
- Site flexibility (SMR can be sited more broadly). SMR footprint + safety profile widens siting options
04Deep-dive research
05Primary sources
06Frequently asked
Why not just build more traditional-large nuclear?
Traditional-large has longer timelines (7-12+ years construction), higher first-of-a-kind risk in Western markets (Vogtle 3+4 cost overruns), and vendor concentration constraints. Better suited to utility-anchored plays with hyperscaler supplemental off-take than pure hyperscaler-anchored primary off-take.
What went wrong with the Amazon/Talen BTM at Susquehanna?
The behind-the-meter arrangement required FERC-approved ISA amendment. FERC rejected on cross-subsidisation concerns. Restructured as front-of-meter with different regulatory + economic profile. See Case Study 03.
Are SMRs actually going to deliver by 2029?
The most-advanced (Kairos + X-energy) have NRC design certification progress + specific first-site commitments. TVA Oak Ridge Kairos site + Dow Seadrift X-energy site are the leading candidates. First delivery 2029-2031 is a reasonable expectation with normal-course execution risk.
What financing structures work for each route?
Restart: utility BS + hyperscaler prepay + DOE LPO + IRA credits. SMR: more equity (FOAK), DOE LPO structured for demonstration projects, hyperscaler prepay + potential equity. Traditional-large: utility PPA + hyperscaler supplemental off-take. See Financing IV.
| Dimension | Nuclear restart | Small modular reactor (SMR) | Traditional nuclear PPA |
|---|---|---|---|
| Landmark deal (2024-2026) | Microsoft / Constellation (TMI Unit 1, 2028 target) | Google / Kairos Power (500 MW, mid-2030s target) | Amazon / Talen (Cumulus, existing Susquehanna capacity) |
| Capacity per site | 500-900 MW (restart of single unit) | 50-300 MW per module (multi-module common) | 500-1,300 MW (existing operating capacity) |
| Commercial operation date | 2027-2029 typical | 2029-2035 typical (first-of-a-kind risk) | Immediate (existing capacity) |
| Permitting complexity | NRC restart application (moderate) | NRC new-build (high; first-of-a-kind) | Existing NRC operating license |
| PPA price ($/MWh, indicative) | $80-120 (site-specific) | $100-150 (first-of-a-kind premium) | $60-90 (existing operating cost basis) |
| Technology risk | Low (operational track record) | High (first commercial deployment) | Very low |
| Grid interconnection | Existing (site infrastructure preserved) | New (dedicated substation build) | Existing |
| Fit for AI DC baseload | Strong (mid-decade capacity) | Strong-conditional (late decade) | Strongest (immediately dispatchable) |