"Power-Hungry Beast" Reshapes the Logic of Power Asset Revaluation: Can Tech Giants' 20-Year Long-Term Agreements Fundamentally Resolve the Nuclear Power Economics Dilemma?

On October 6, 2026, Constellation Energy (Nasdaq ticker CEG, the largest commercial nuclear reactor operator in the United States) and Google announced the signing of a 20-year strategic clean energy cooperation agreement. Following the announcement, Constellation Energy's stock price rose approximately 12% that day.

The capital market's excitement is understandable. But an easily overlooked detail is that the true structure of this deal differs considerably from the "1 billion dollars" in media headlines. Bloomberg's initial report of "at least 1 billion dollars" was the figure at the time the deal first came to light, while the officially announced agreement amount is 4.3 billion dollars. The essence of the 4.3 billion dollars is Constellation Energy's capital investment to upgrade its 11 nuclear power units, with Google providing revenue certainty for this investment through a 20-year power purchase agreement.

To understand this deal, one cannot look only at the level of "Google bought nuclear power." What truly merits scrutiny is: when tech giants provide certainty of cash flow for nuclear power assets through long-term Power Purchase Agreements (PPAs, a type of long-term contract in which buyer and seller directly agree on electricity price and term), is the business model of American nuclear power undergoing a fundamental transformation? And can this transformation resolve nuclear power's own economic challenges?

890 Megawatts of Incremental Capacity, 11 Units, and a "No New Build" Pathway

Let us first examine the specific contents of this deal.

Constellation Energy will conduct systematic upgrades to 11 operating nuclear power units located in Illinois, Pennsylvania, and New Jersey, including replacing turbines, steam generators, and digital control systems, with the goal of releasing 890 megawatts (MW, megawatt, 1 megawatt equals 1,000 kilowatts) of net incremental generating capacity without altering the reactor core safety boundaries. The first batch of upgrade work is expected to be completed by 2028, with the new electricity to be integrated into the PJM Interconnection (the largest regional grid organization in the United States, covering 13 states and Washington, D.C., serving 67 million people).

In addition, the two parties also signed a 15-year, 2,700-megawatt power supply agreement, under which Google will procure electricity from Constellation Energy's broader generating assets within the PJM market. Combined, the two agreements lock in approximately 3,590 megawatts of power supply for Google.

Why choose "upgrade" rather than "new build"? This is precisely the most critical engineering economics judgment of this deal. The Levelized Cost of Energy (LCOE, a metric measuring the average cost of power generation over the full lifecycle) for a new nuclear power plant in the United States is as high as 154 dollars/megawatt-hour, compared to 63 dollars/megawatt-hour in China and 190 dollars/megawatt-hour in France. New-build projects also face grid connection queues of nearly a decade and multi-billion-dollar cost overrun risks. By contrast, Power Uprates on existing units have extremely low marginal costs and rely entirely on existing plant sites and transmission corridors, bypassing the bottleneck of new transmission line approvals.

Exchanging 4.3 billion dollars for 890 megawatts of incremental capacity works out to approximately 483 dollars per kilowatt——a figure far below the thousands of dollars per kilowatt in capital expenditure for new nuclear builds. From an engineering perspective, this is the "most cost-effective" pathway.

PJM Capacity Prices Surge 11-Fold, Tech Giants Forced to "Bring Their Own Power"

Why is Google willing to make a 20-year long-term commitment to pay for nuclear power? The answer lies in the structural changes in the American electricity market.

PJM grid capacity prices have surged more than 11-fold since 2024. The 2025/2026 capacity auction clearing price soared from 28.92 dollars/megawatt-day the previous year to 269.92 dollars/megawatt-day. Goldman Sachs predicts that actual electricity demand from U.S. data centers will rise from approximately 31 gigawatts in 2025 to approximately 108 gigawatts by 2030, and nearly all power grids will face a structural bottleneck of insufficient reserve capacity.

PJM's proposed "Bring Your Own Power" scheme requires data center customers connected to the grid to arrange their own power sources, or else they may be remotely disconnected during peak periods. Against this backdrop, tech giants locking in nuclear baseload power is no longer a "green image project" but a strategic necessity to ensure that computing expansion is not constrained by power supply.

This trend is not unique to Google. Microsoft signed a 20-year power purchase agreement with Constellation Energy as early as 2024 to support the restart of Unit 1 at the Three Mile Island nuclear power plant in Pennsylvania, expected to provide 835 megawatts of power. Amazon signed a 20-year agreement with Constellation Energy on September 30, 2026, while Meta signed a 1,121-megawatt nuclear virtual power purchase agreement with Vistra. With all four major tech giants entering the fray, nuclear power is being repositioned from a "marginalized baseload power source" to "infrastructure for the computing economy."

From "Policy Transfusion" to "Market Hematopoiesis": The Shift in Nuclear Power Valuation Logic

The deeper impact of this deal on the nuclear power industry lies in the fact that it has changed the valuation anchor for nuclear power assets.

For the past several years, the U.S. nuclear power industry has relied primarily on federal government loan commitments and tax subsidies to sustain the operation of existing units. But this "policy dividend-driven" model has a fundamental flaw: the continuity and timing of subsidy delivery depend on political cycles, and the secondary market's ability to price a single policy dividend has always been limited.

Google's 4.3 billion dollar long-term agreement provides a different logic: a commercial closed loop driven by private capital. The 20-year power purchase agreement provides Constellation Energy with a transparent and stable forward revenue model, directly offsetting the financial leverage risk of nuclear power overhauls and equipment replacement. BMO Capital estimates that these nuclear power purchase agreements, once fully operational, will contribute approximately 2.15 dollars in earnings per share and approximately 685 million dollars in free cash flow before growth, a significant gain compared to the company's current diluted earnings per share of 10.33 dollars.

This means that the pricing logic for nuclear power assets is shifting from "marginal cost competition in the wholesale electricity market" to "discounted cash flow certainty from long-term contracts." Constellation Energy's 2024 agreement with Microsoft, its September 2026 agreement with Amazon, and now this cooperation with Google have cumulatively locked in approximately 1,100 megawatts of incremental nuclear capacity. Against the backdrop of high PJM capacity prices and tight power supply, nuclear power operators are transforming from passive "price takers" into "scarce baseload asset providers" with pricing power.

NRC Approval, Execution Progress, and Valuation Overextension

But the other side of the deal is risk.

First is the uncertainty of regulatory approval. The systematic power uprates of 11 nuclear power units still require license amendment approval from the U.S. Nuclear Regulatory Commission (NRC, the federal agency responsible for nuclear reactor safety regulation). Although the NRC has pushed to compress the review cycle for extended power uprates to approximately 12 months and in February 2026 simplified the environmental review process for advanced reactors through a new categorical exclusion rule from the Department of Energy, batch approval involving 11 units still carries timing risks.

Second is execution risk. Constellation Energy needs to complete large-scale upgrade and renovation work while ensuring reliable unit operation. The nuclear power industry's history of cost overruns and schedule delays is not encouraging——the grid connection date for the restart of Three Mile Island Unit 1 has already been postponed from the original plan to 2031 due to delays in PJM grid engineering.

Third is valuation overextension. After the deal was announced, several institutions lowered their target prices. BMO Capital cut its target price from 379 dollars to 350 dollars, Scotiabank sharply reduced its from 441 dollars to 355 dollars, and Goldman Sachs maintained a "neutral" rating and a target price of 305 dollars. As of October 8, Constellation Energy's stock closed at 285.07 dollars, implying potential upside of approximately 19.8% based on the consensus target price of 341.53 dollars from 22 analysts. The broad downward revision of analyst target prices reflects market concerns that "good news is already fully priced in."

The Real Variable in the Nuclear Renaissance Is Not Technology, But Business Model

The Google-Constellation Energy deal may carry more weight than it appears on the surface.

It did not build a single new reactor, did not break through any physical limit, and did not even change the technical parameters of any unit——what it did was bind "certain electricity demand" and "certain electricity supply" together with a 20-year contract. In the U.S. electricity market, a system that primarily operates on short-term bidding mechanisms, this kind of long-term certainty is itself the scarcest resource.

Looking back, the Google-Constellation Energy deal is less a clean energy procurement than a business model experiment forced into existence by AI-era electricity demand. Behind the numbers——4.3 billion dollars, 890 megawatts, 20 years——lies a key step in American nuclear power's shift from reliance on policy subsidies to long-term certainty provided by computing capital. It did not build a single new reactor, yet it gave existing units a reason to continue upgrading; it did not solve all of nuclear power's economic challenges, yet it provided a replicable pathway of "tech giants pay, nuclear operators expand capacity." When the endpoint of computing is electricity, and the endpoint of electricity is long-term contracts, the real variable in the nuclear renaissance may never have been in the reactor, but in who is willing to pay in advance for the certainty of the next twenty years.