A DC Atlas Frontier
Module 4 of 6
The Economics
The costs LCOE ignores are the ones that decide it. A 20-year total cost of ownership model for a 500 MW campus, including the honest finding that first-of-a-kind nuclear still costs more than the grid.
Built on DC Atlas's asset-level data across ~8,000 tracked facilities.
The number that rewired Constellation Energy's entire corporate strategy was not the one you would expect. It was not the 835 megawatts of the Three Mile Island reactor. It was not the $1.6 billion restart cost. It was $785 million. That is the estimated annual revenue from a single 20 year power purchase agreement with Microsoft. The agreement is priced at roughly $112 per megawatt hour. The same region's spot market sells that electricity for approximately $60. Microsoft is paying nearly double the going rate for grid power, and every analyst who has examined the deal structure agrees it is rational.
Understanding why that premium is rational, and what it implies for the economics of nuclear powered data centres broadly, is the subject of this module. We are going to build the cost comparison that most analysis gets wrong, not a simple levelised cost of energy chart but a full 20 year total cost of ownership model that captures the line items conventional LCOE misses entirely: capacity charges that now exceed $60 million per year for a 500 megawatt campus, interconnection costs measured in billions, backup generation fleets, carbon exposure, and the quantifiable economic value of price certainty in a market where electricity forward contracts carry larger risk premiums than at any point in the past decade.
Why Is LCOE the Wrong Metric for a Data Centre?
Levelised cost of energy is the metric everyone uses and almost everyone misapplies. LCOE divides the total lifetime cost of building and operating a power plant by its total lifetime energy output, producing a single dollar per megawatt hour figure. It is useful for comparing technologies on a narrow basis. It is misleading when used as the sole metric for evaluating data centre power strategy, because it excludes the costs that actually dominate the decision.
Lazard's June 2025 LCOE+ report, the 18th edition and the definitive benchmark for institutional investors, presents the following unsubsidised ranges.
| Technology | LCOE Range (Unsubsidised) | LCOE with Tax Credits | Trend |
|---|---|---|---|
| Utility-Scale Solar | $38 to $78/MWh | As low as $24/MWh | Down 84% since 2009 |
| Onshore Wind | $37 to $86/MWh | As low as $15/MWh | Up 49% since 2020 |
| Gas Combined Cycle (new build) | $106 to $146/MWh | N/A | 10-year high |
| Nuclear (Vogtle basis) | Based on $15,200/kW capital | ~$50/MWh with 40% ITC (2nd plant) | Learning curve projected |
| Solar + Storage | $50 to $131/MWh | Lower with ITC | Wide range by duration |
The headline numbers suggest that solar and wind are dramatically cheaper than nuclear. On a pure LCOE basis, they are. But LCOE does not answer the question a data centre operator is actually asking, which is: what does it cost to guarantee 500 megawatts of 99.995 percent reliable, carbon free electricity, 24 hours a day, 365 days a year, for 20 years, without depending on a grid that cannot deliver it?
That question requires a total cost of ownership framework, and the answer changes everything.
What LCOE Misses
The total cost of operating a grid connected data centre at scale includes at least six categories of expense that LCOE ignores entirely. Each of these costs has increased substantially since 2023, and together they transform the competitive calculus.
| Cost Category | 500 MW Campus (Annual) | 20-Year NPV at 5% | Included in LCOE? |
|---|---|---|---|
| PJM Capacity Charges | $60.1M/year | $750M+ | No |
| Transmission and Distribution | $15 to $25M/year | $190 to $310M | No |
| Interconnection Construction | One-time $200 to $500M | $200 to $500M | No |
| Backup Generation (diesel fleet) | $8 to $15M/year | $100 to $190M | No |
| Carbon Offset / REC Purchases | $5 to $20M/year | $60 to $250M | No |
| Grid Energy Charges | $175 to $260M/year | $2.2 to $3.2B | Yes (partially) |
| Total Grid-Connected Cost | $263 to $380M/year | $3.5 to $5.2B | — |
The cost of grid power that nobody models
500 MW data centre campus · 20-year NPV at a 5% discount rate. Only the grey band is captured by LCOE — the five costs stacked above it are not.
PJM capacity is stated in the source as “$750M+” — a floor, charted here at $750M. On-site nuclear eliminates the capacity charge, the interconnection queue cost and the carbon/REC cost outright, and cuts transmission and backup generation. It does not, however, make the campus cheaper: this chart shows grid cost only, and the module’s 20-year model concludes that with first-of-a-kind capital, on-site nuclear is more expensive than grid-connected operation for a first project.
The capacity charge alone is extraordinary. The current PJM clearing price is $329 per megawatt day. At that price, a 500 megawatt campus pays $60.1 million annually just for the right to draw power from the grid, before consuming a single kilowatt hour. The 2027/28 auction cleared even higher, at the FERC price cap of $333.44 per megawatt day. Even at that cap, the auction left a 6,623 megawatt reliability shortfall. Over 20 years at a 5 percent discount rate, the net present value of capacity charges alone exceeds $750 million. On site nuclear generation eliminates this line item entirely.
Why Is New-Build Gas No Longer the Cheap Option?
The assumption that natural gas combined cycle plants provide cheap backup or baseload power is obsolete. Lazard's 2025 analysis documents that new build gas CCGT has reached a 10 year high LCOE of $106 to $146 per megawatt hour, driven by turbine manufacturing constraints, rising equipment costs, extended construction timelines, and elevated cost of capital. Goldman Sachs notes that turbine shortages and rising costs "are expected to continue driving steep LCOE increases for gas technologies in the near term."
These bars are not like for like: solar, wind and gas are LCOE range midpoints, while the two nuclear figures are PPA contract prices, not LCOE. The nuclear fleet's operating cost is a separate metric shown in “The Operating Cost Advantage” below and must not be read as a nuclear LCOE.
This creates a binary choice for industrial buyers seeking firm power. You can purchase renewable energy with storage at $50 to $131 per megawatt hour and accept the intermittency, grid integration costs, and the reality that solar generates zero power during evening training runs. Or you can purchase firm nuclear power at $70 to $112 per megawatt hour with 24/7 availability and 20 year price certainty. The fact that new gas generation now costs more than nuclear PPAs on an unsubsidised basis is the single most important shift in the competitive landscape since 2023.
What Did Microsoft Actually Pay for Three Mile Island Power?
The Constellation and Microsoft Three Mile Island restart, now renamed the Crane Clean Energy Center, provides the most detailed public case study of nuclear economics for a data centre customer. The structure is worth examining precisely because Constellation has disclosed enough in SEC filings and investor presentations for analysts to reconstruct the economics.
Jefferies analyst Paul Zimbardo stated explicitly that "that is not cheap for electricity. It is a healthy premium to what you would pay if you just bought grid power." The $100 to $112 per megawatt hour price is approximately double the regional spot rate for renewable energy. Microsoft is paying this premium for three things that renewables cannot provide at any price: 24/7 firm capacity, 20 year price certainty, and genuine carbon free generation that does not depend on annual matching accounting tricks.
“Just the Unit 1 reactor alone could add $1.70 per share in incremental earnings and $445 million in net income. The operational risk to bring it back online in 2028 appears manageable.”Morgan Stanley, Constellation Energy price target increase to $313
The Meta and Constellation Clinton deal, announced in February 2026, provides a second pricing benchmark. Jefferies estimated the Clinton virtual PPA at approximately $70 per megawatt hour. That is a roughly $20 premium to Illinois energy plus capacity market compensation. It is also approximately $15 higher than regional wind PPAs. The fleet average operating cost is $31.76 per megawatt hour. Selling at $70 against that cost, Constellation earns a spread of $38.24 per megawatt hour to cover capital recovery, interconnection, risk premium, and equity returns.
The Operating Cost Advantage
The operational economics of the existing nuclear fleet are far more favourable than most investors realise. The Nuclear Energy Institute's February 2025 report documents that the US fleet average total generating cost in 2023 was $31.76 per megawatt hour, a 39.9 percent reduction from the 2012 peak of $52.83.
The cost structure reveals why nuclear operating economics improve with scale. Fuel costs run to $5.32 per megawatt hour. That represents just 17 percent of total costs. Fuel costs have declined 43.7 percent since 2012. Capital additions run to $7.06 per megawatt hour, and have declined 51.1 percent as post Fukushima safety upgrades were completed. The dominant cost is operations, at $19.38 per megawatt hour. That cost is overwhelmingly labour driven: staffing, security, maintenance, and regulatory compliance. This means multi unit sites benefit from enormous economies of scale. Multi unit facilities average $29.53 per megawatt hour. Single unit plants average $41.62 per megawatt hour. That is a 29 percent cost advantage from shared overhead.
The Learning Curve Question
First of a kind nuclear construction is expensive. This is not in dispute. Vogtle Units 3 and 4 cost approximately $34 billion for 2,234 megawatts, roughly $15,200 per kilowatt in 2023 dollars. The Darlington BWRX-300 carries a first unit budget of approximately CAD 6.1 billion for 300 megawatts. NuScale's cancelled CFPP saw costs escalate past $20,000 per kilowatt before termination.
The economic case for SMRs rests on whether sequential deployment produces cost reductions comparable to other manufacturing industries, or whether nuclear specific regulatory and construction constraints hold costs to historical patterns. The evidence is mixed but trending positive.
Idaho National Laboratory's April 2025 peer reviewed analysis, directly evaluating the Vogtle experience, projects that with a 40 percent Investment Tax Credit applied, a second AP1000 plant could achieve an LCOE as low as $50 per megawatt hour. That figure would make new build nuclear competitive with subsidised renewables. The INL analysis explicitly concludes that the Vogtle experience, "despite significant cost and schedule overruns, a major bankruptcy, and a pandemic, demonstrates that the US nuclear industry can still develop new supply."
The IEEFA counterargument deserves honest acknowledgment. Their May 2024 analysis documents that "the US nuclear industry has never shown a positive learning curve" historically, noting that NuScale's costs escalated from $9,964 per kilowatt in 2015 to $21,561 in 2023. However, the IEEFA critique is primarily directed at small modular reactors, not at the sequential deployment of proven designs like the AP1000 or BWRX-300. The distinction matters. South Korea achieved a 23 percent cost reduction over decades of continuous construction of standardised designs. The question is whether the US can replicate that discipline.
How Much Does the Inflation Reduction Act Cut Nuclear Costs?
The Inflation Reduction Act created two tax credit mechanisms that materially alter nuclear project economics. Understanding the structure and scale of these credits is essential for any investment analysis.
For a representative 600 megawatt nuclear facility operating at 93 percent capacity factor (conservative relative to Constellation's 96.8 percent), the numbers are substantial.
Under the §45Y production tax credit at the enhanced rate with both bonus adders, the effective credit reaches 1.815 cents per kilowatt hour. Applied to annual output of 4,873 gigawatt hours, this produces approximately $88.5 million per year for 10 years. The present value at a 5 percent discount rate is approximately $683 million.
Under the §48E investment tax credit at the enhanced 30 percent rate with both bonus adders, the credit reaches 50 percent of qualified investment. For a facility with $5.75 billion in capital costs, this produces approximately $2.875 billion in upfront credit value.
The DOE Lending Machine
The Department of Energy's Loan Programs Office, now operating as the Office of Energy Dominance Financing, holds 191 active applications requesting cumulative financing of $297.7 billion as of January 2026. The office expects to issue more than $50 billion across all programmes in fiscal year 2026, with $750 million in new credit subsidy earmarked specifically for SMR and advanced reactor construction.
The Palisades nuclear plant restart in Michigan established the template. In September 2024, the DOE finalised a $1.52 billion loan guarantee to Holtec International, the first closing through the Energy Infrastructure Reinvestment programme for a nuclear project. That precedent, combined with the $1 billion loan supporting the TMI restart for Microsoft, demonstrates that the federal lending mechanism is functional, scalable, and available to qualified applicants.
For a data centre operator evaluating nuclear, the DOE loan guarantee changes the capital structure fundamentally. A $5 billion SMR project with an 80 percent DOE guarantee requires the developer to raise only $1 billion in equity and non guaranteed debt. The government backing reduces the cost of debt capital, extends available tenors, and enables project finance structures that would be impossible for a technology with no commercial operating history.
Is On-Site Nuclear Cheaper Than the Grid Over 20 Years?
When you assemble every cost component into a 20 year total cost of ownership framework for a 500 megawatt campus, the on-site nuclear stack totals $4.9 to $7.4B — set against the $3.5 to $5.2B grid-connected total established above. The comparison produces results that contradict the LCOE narrative.
| Cost Component | On-Site Nuclear (20yr NPV) |
|---|---|
| Energy Charges | $1.8 to $2.4B (at $70 to $90/MWh) |
| Capacity Charges (PJM) | $0 (eliminated) |
| Transmission and Distribution | $0 to $50M (interconnection only) |
| Interconnection Queue Cost | $0 (no queue) |
| Backup Generation | $30 to $60M (reduced need) |
| Carbon Costs / RECs | $0 (zero emission) |
| Nuclear Capital (net of ITC) | $2.5 to $4.0B |
| Nuclear O&M (20yr) | $400 to $600M |
| Decommissioning Fund | $150 to $300M |
| Total 20-Year NPV | $4.9 to $7.4B |
The honest conclusion is this: on a pure present value basis with first of a kind capital costs, on site nuclear is more expensive than grid connected operation for the first project. The nuclear capital cost, even net of ITC credits, dominates the comparison. But this analysis contains three assumptions that change dramatically over time.
First, PJM capacity prices are projected to continue rising. The Brattle Group's April 2025 analysis projects that if gas generation additions are limited, capacity prices could reach $1,300 per megawatt day, nearly four times the current level. At $1,300 per megawatt day, the 20 year capacity charge NPV for a 500 megawatt campus exceeds $3 billion.
Second, the nuclear capital cost applies only to the first unit. OPG projects a 33 percent reduction by the fourth BWRX-300. INL projects that a fourth sequential AP1000 deployment could achieve $6,200 per kilowatt. At nth of a kind costs with ITC credits, the nuclear capital component drops by 40 to 55 percent.
Third, the nuclear option provides something the grid connected alternative cannot price: 20 year price certainty. Every forward electricity contract carries a risk premium that increases with duration. Academic research consistently finds that forward power prices systematically exceed expected spot prices by $4 to $7 per megawatt hour at even short maturities, with the premium expanding dramatically at the multi year horizons relevant to data centre investment decisions.
“If gas-fired generation additions are limited, the PJM capacity price might have to be $1,300/MW-day. If the VRR curve price cap is 1.5 to 1.75 times that, the price could rise to nearly $2,300/MW-day in scarcity.”Brattle Group, PJM CONE Analysis, April 2025
The Price Certainty Premium
Microsoft did not pay $112 per megawatt hour because nuclear electricity is worth $112 in any given hour. Microsoft paid for the elimination of a risk that no other technology can eliminate: the possibility that electricity costs spike during the exact years when AI infrastructure investment is highest and demand growth is fastest.
The academic evidence on this premium is unambiguous. Research by Borenstein documented that simple hedging strategies using forward contracts can eliminate more than 80 percent of bill volatility for large industrial customers. The Oxford Institute for Energy Studies found that in the context of the energy transition, the price required for generators to provide long term hedges "can be multiples of the expected value of prices." Stanwich Energy's 2025 analysis concluded that "fixed rate contracts now carry higher premiums than ever before" due to shifting energy policies, rapid demand growth, and heightened grid reliability concerns.
For a 500 megawatt data centre campus with a 20 year investment horizon, the value of price certainty is not abstract. It is the difference between a financial model where operating costs are known with precision and one where the single largest variable input, electricity, can fluctuate by 300 percent in a single auction cycle, as PJM capacity prices demonstrated between 2024 and 2027.
Module 5 maps the regulatory and siting pathway that determines whether the economics described here can actually be realised on a timeline that matters, with a specific focus on Illinois as the most promising jurisdiction in the country for nuclear data centre development.
Questions this module answers
- What would it cost to power a hyperscale data centre campus with an SMR?
- For a 500 megawatt campus, on-site nuclear carries a 20 year NPV of $4.9 to $7.4B, against $3.5 to $5.2B for grid-connected power. The nuclear capital component alone is $2.5 to $4.0B net of ITC, plus $400 to $600M of O&M and $150 to $300M for the decommissioning fund. Against that it eliminates $750M+ of PJM capacity charges and $200 to $500M of interconnection queue cost.
- Is on-site nuclear cheaper than grid power?
- No — not for the first project. On a pure present value basis with first of a kind capital costs, on site nuclear is more expensive than grid connected operation, at $4.9 to $7.4B against $3.5 to $5.2B over 20 years for a 500 megawatt campus. The case rests instead on rising grid costs, on nth of a kind capital reductions of 40 to 55 percent, and on 20 year price certainty.
- Why is LCOE the wrong metric for a data centre?
- LCOE excludes the costs that dominate the decision. For a 500 MW campus those excluded items — PJM capacity charges at $60.1M/year, transmission and distribution at $15 to $25M/year, one-time interconnection of $200 to $500M, backup diesel at $8 to $15M/year and carbon or REC purchases at $5 to $20M/year — take total grid-connected cost to $3.5 to $5.2B over 20 years. LCOE answers what a plant costs to run, not what it costs to guarantee 500 megawatts of firm, carbon free power for 20 years.
- What did Microsoft actually pay for Three Mile Island power?
- Roughly $100 to $112 per megawatt hour under a 20 year power purchase agreement for the 835 megawatt reactor, now the Crane Clean Energy Center. That is an estimated $785 million in annual revenue for Constellation, against a regional spot market of approximately $60. Jefferies analyst Paul Zimbardo called it "a healthy premium to what you would pay if you just bought grid power."
- How much do PJM capacity charges cost a 500 MW data centre?
- $60.1 million per year at the PJM clearing price of $329 per megawatt day, paid purely for the right to draw power before consuming a single kilowatt hour. Over 20 years at a 5 percent discount rate the net present value exceeds $750 million. On site nuclear generation eliminates this line item entirely.
- How much does the Inflation Reduction Act cut the cost of a nuclear project?
- The §48E investment tax credit at the enhanced rate with both bonus adders reaches 50 percent of qualified investment, producing approximately $2.875 billion for a 600 megawatt facility with $5.75 billion in capital costs. The alternative §45Y production credit is worth approximately $88.5 million per year for 10 years, a present value of approximately $683 million. Projects cannot claim both.