The highest-return nuclear investment is increasing output from reactors that are already licensed, cooled, and connected to the grid. The binding constraint is the finite stock of sites with uprate headroom. That stock cannot expand materially by 2028–2032, so scarcity rent flows first to fleet owners, then to nuclear engineering labor and specialized equipment suppliers.
Google’s October 2026 agreement with Constellation Energy demonstrates the model. A 20-year power purchase agreement finances 890 MW of incremental output through uprates at 11 reactors across six plants in Illinois, Pennsylvania, and New Jersey. Constellation plans to invest more than $4.3 billion, with the first additional capacity scheduled for 2028.
A separate 15-year contract covers 2,700 MW of existing generation, bringing contracted output to 3,590 MW. The 890 MW represents new capacity. The remaining contract provides a long-term revenue floor for the existing fleet and supports the balance sheet funding the uprates.
Time-to-Power
The economic driver is delivery speed. AI data centers need firm, carbon-free power within the current investment cycle. Large greenfield reactors ordered today are unlikely to deliver before the mid-2030s. Uprates can begin delivering in 2028.
For the 2028–2032 window, the practical alternatives are gas turbines, storage, or additional output from existing plants. Uprates gain value from combining speed, dispatchability, and clean generation. Hyperscalers are paying to secure firm power before grid scarcity intensifies.
How Uprates Work
The $4.3 billion program covers physical and digital modernization, including:
- High- and low-pressure steam-turbine upgrades
- Steam-generator and feed-water improvements
- Advanced digital control systems
- Reliability and life-extension work
The program aligns with the Department of Energy’s UPRISE initiative, which encourages private investment in higher output from existing civilian reactors.
The largest expansions include approximately 348 MW at Limerick Generating Station in Pennsylvania and an estimated 97 MW at Salem Generating Station in New Jersey. Constellation owns 43% of Salem alongside operator PSEG Nuclear. Roughly $2 billion is allocated across Braidwood, Byron, LaSalle County, and Quad Cities in Illinois.
Headline capital per incremental megawatt overstates the uprate’s marginal cost because the spending also extends plant life and improves reliability.
| Program | Headline capital | New MW | Headline $/kW | Delivery |
|---|---|---|---|---|
| Google–Constellation | >$4.3B | 890 | ~$4,800 | First capacity in 2028 |
| Amazon–Constellation, Calvert Cliffs | >$3B | ~190 | ~$15,800 | 2030–2032 |
| Vistra, DOE conditional loan | Up to $4.2B | 433 | ~$9,700 | Through 2034 |
Calvert Cliffs’ $3 billion program covers improvements across the full 1,790 MW plant and supports relicensing. The Vistra financing also preserves nearly 4 GW of existing generation. The Google–Constellation program provides the clearest uprate cost benchmark, although $4,800/kW remains an upper bound on the incremental cost.
Cost and Timing Advantage
Vogtle Units 3 and 4 cost approximately $35 billion for 2.2 GW, or about $16,000/kW. At a 10% hurdle rate over 40 years, a greenfield reactor’s capital charge is approximately $125/MWh at $10,000/kW and $188/MWh at $15,000/kW, excluding construction-period interest.
An uprate costing $4,800/kW produces a capital charge of roughly $60–79/MWh at an 8–12% hurdle rate over 20 years, assuming no residual value after the power purchase agreement expires.
| Variable | Uprate | Greenfield large reactor |
|---|---|---|
| Overnight capital cost | ~$4,800/kW upper bound | $10,000–16,000/kW |
| Capital charge | $60–79/MWh over 20 years | $125–188/MWh over 40 years |
| First capacity | 2028 | Mid-2030s |
| New site and interconnection | Unnecessary | Required |
| Greenfield licensing | Unnecessary | Required |
Uprates avoid new sites, interconnection queues, and greenfield licensing. Siemens Energy estimates that steam-side upgrades alone can raise unit output by 7–10%. Delivery speed is therefore as important as capital efficiency.
Source of the Scarcity Rent
PJM’s 2028/29 capacity auction left the system approximately 6,800 MW below its reliability requirement. A price collar caps merchant capacity revenue through the 2029/30 delivery year. Bilateral power purchase agreements allow buyers to pay for firm supply outside the auction and transfer scarcity value directly to generators.
Announced uprates total approximately 1,600 MW, equal to roughly one-quarter of the 2028/29 shortfall, although delivery is spread across several years. On 890 MW, each year of lead time over greenfield construction could produce approximately $106–180 million in capacity revenue, plus energy margin.
Assuming the full $4.3 billion is attributed to the 890 MW uprate, with a 93% capacity factor, 20-year recovery period, and $75 million in annual fuel and incremental operating costs:
| Hurdle rate | Required annual revenue | Bundled revenue | Energy price required after capped capacity revenue |
|---|---|---|---|
| 8% | $513M | $71/MWh | $56/MWh |
| 10% | $580M | $80/MWh | $65/MWh |
| 12% | $651M | $90/MWh | $75/MWh |
The 2028/29 PJM auction cleared at the $325/MW-day cap, while the uncapped simulation reached $554.72/MW-day. At the cap, 890 MW generates about $106 million annually, covering 16–21% of required revenue. At the simulated uncapped price, annual capacity revenue rises to approximately $180 million, covering 28–35%.
Energy margin and the PPA premium over forward prices determine Constellation’s return.
Distribution of the Rent
| Factor | Supply response through 2032 | Rent capture |
|---|---|---|
| Licensed, cooled, interconnected sites with uprate headroom | Fixed | Highest; fleet owners monetize scarcity through long-term PPAs |
| Licensing and outage-execution engineering | Slow | High and persistent through wages, contractor margins, and backlog |
| Turbine, generator, transformer, and nuclear-qualified suppliers | Constrained until capacity expands | Moderate and temporary |
| Cooling-system retrofits | Relatively elastic | Low |
| Fuel | Scales with power output | Small share of incremental economics |
Existing sites capture Ricardian rent because their supply is fixed. Engineering labor retains pricing power because workforce availability limits execution. Specialized equipment suppliers benefit while manufacturing capacity remains constrained. Cooling retrofits and fuel capture less value because their supply responses are more elastic.
Limits on the Opportunity
The NRC has approved about 8,030 MWe of uprates since the 1970s. The 1,600 MW announced in 2026 equals approximately 20% of that 45-year total. Reactor uprates generally remain below 20% of original licensed capacity.
Installation must occur during scheduled refueling outages, creating a hard sequencing constraint. Once economical uprate capacity is exhausted, additional firm nuclear supply must come from relicensing, plant restarts, or new construction at progressively higher cost.
The rent also has a limited window. PJM capacity pricing remains capped through mid-2030, while 20-year PPAs monetize anticipated post-collar scarcity today. Additional supply during the 2030s would reduce that scarcity value.
Variables to Monitor
- NRC amendment requests and approvals for the 11 reactors
- Delivery of the first uprated capacity in 2028
- PPA pricing disclosed in Constellation filings
- The December 2026 PJM auction for the 2029/30 delivery year
- Post-collar PJM market design
- Turbine, transformer, and nuclear-services backlogs
- Refueling-outage execution and schedule slippage
Conclusion
Existing nuclear fleets offer the strongest near-term economics because they combine fixed site scarcity, faster delivery, lower capital intensity, and established grid access. Long-term PPAs convert those advantages into contracted revenue before additional supply can relieve the shortage.
The decisive variable is the PPA premium over forward energy prices. Until that figure is disclosed, returns cannot be measured precisely. The underlying scarcity rent remains anchored in two physical constraints: finite uprate headroom and fixed refueling-outage calendars.