On October 6, Google and Constellation Energy announced the largest power contract the hyperscaler build-out has produced so far: 3,590 MW across two agreements. The headline contract is a 20-year power purchase agreement for 890 MW of new nuclear capacity, to be added through uprates of 11 existing Constellation reactors in Illinois, Pennsylvania, and New Jersey, all inside PJM. The second is a 15-year supply agreement for another 2,700 MW from Constellation’s existing PJM generation. Constellation said it will make more than $4.3 billion of new investments in its fleet as part of the arrangement, with the first uprated reactor delivering additional power in 2028. Shares rose more than 13% on the day, to $304.04 by mid-morning, making Constellation one of the S&P 500’s top gainers.
Read it as a capital-allocation story first and a nuclear story second. What Google is really buying is a commitment by Constellation to spend billions on its own plants, with a creditworthy buyer attached to the incremental output. For equity investors, that structure — growth capex with contracted revenue — is the cleanest earnings-growth mechanism a merchant generator can offer. It is worth understanding precisely what the two contracts are, what the numbers do and do not imply, and where the risks sit.
Two contracts, two different economics
The deal has two legs, and they do different work. The 20-year PPA covers 890 MW of new capacity: incremental megawatts that do not exist yet, to be created by upgrading existing reactors to run harder. Under this leg, Google’s contract is the economic justification for $4.3 billion-plus of capital spending. Without the PPA, there is no obvious reason to commit that capital; with it, the spending becomes contracted revenue, the kind of cash flow that supports a valuation multiple.
The 15-year agreement covers 2,700 MW of power that already exists. It is not tied to a specific generation source, and no new megawatts result from it. Google described it as providing long-term revenue certainty for Constellation’s operating plants — which is another way of saying it converts merchant exposure into contracted cash flow. For shareholders, this is arguably the larger leg in revenue terms: 2,700 MW of around-the-clock nuclear output locked in with an investment-grade counterparty for fifteen years, in a market where capacity prices are sitting at regulatory caps. The first leg builds new value; the second stabilizes existing value.
Note what was not disclosed. The companies did not publish the price Google will pay under either agreement. Any analyst model of the earnings impact is working with assumptions, not facts — the one genuinely knowable figure is the investment commitment, not the revenue it will earn. That asymmetry should discipline any attempt to price the deal from the public record.
The uprate as growth capex
The $4.3 billion commitment deserves its own frame. In a regulated utility, capital spending of this kind goes into rate base: the utility petitions its commission, the cost is recovered from customers over decades, and the shareholder earns an allowed return. Constellation is not that kind of company. It is a merchant generator — the country’s largest nuclear operator, and following its acquisition of Calpine earlier this year, a major owner of natural-gas-fired plants and geothermal as well. Its nuclear output is overwhelmingly its identity and its margin.
For a merchant generator, uprate spending is growth capex funded by shareholder capital with no guaranteed recovery. That is the risky way to read it. The less risky way is to notice the contract attached to it: a 20-year PPA with Google, one of the most creditworthy companies on earth, for the full 890 MW of incremental output. Contracted growth capex is a fundamentally different proposition from speculative growth capex. The former behaves like a regulated investment without needing a regulator; the latter is a bet on future power prices.
This is the template investors have now seen twice in a month. On September 30, Constellation announced a 20-year PPA with Amazon covering 690 MW, including 190 MW of new capacity at the Calvert Cliffs plant in Maryland, backing more than $3 billion of Maryland infrastructure investment through 2032. The Vistra–Meta–DOE structure covered in this series two weeks ago is the same template with federal debt as the senior layer: corporate offtake, contracted revenue, nuclear hardware. The hyperscalers are funding the existing fleet’s expansion, and the capital markets are treating each announcement as evidence the template repeats.
The per-megawatt math, with a warning attached
Divide the announced investment by the announced capacity and you get roughly $4.8 million per MW of uprate capacity for the Google deal. Independent analysis of the Amazon deal put that one above $15 million per MW. The gap invites a conclusion — that Amazon overpaid or Google got a bargain — and the conclusion is premature, for two reasons.
First, the $3 billion-plus figure attached to the Amazon package explicitly includes broader Maryland infrastructure investment, not just the 190 MW of uprate hardware; the arithmetic is not comparing like with like. Second, and more importantly, neither package discloses the contract price. An uprate investment is a cost; the PPA is the revenue. Knowing one without the other tells you nothing about the return. An analyst who tells you the Google deal earns Constellation a particular spread is forecasting, not reading.
There is one honest comparison available. Recall the arithmetic trap from the Vistra piece: the honest statement is what the public record supports, and nothing more. What the record supports here is that Constellation has committed to spending $4.3 billion-plus, that the revenue covering it is a 20-year contract with Google, and that the equity market — shares up double digits on the day — judged the combination to be value-creating. Everything else is model.
What the contracted base is worth
Step back from the uprates and look at what Google is underwriting across the whole arrangement. At the company’s second-quarter filing, Constellation had signed 920 MW of long-term PPAs with investment-grade customers in the quarter alone, at an average duration of 18.5 years and expected to be fully ramped by 2032. Add the Amazon and Google agreements, and independent analysis puts Constellation’s long-term contracted nuclear book above 5 GW. Roughly 30% of the company’s clean baseload output is now contracted under long-term agreements, according to the second-quarter earnings call.
For a merchant generator, contracted share is the single most important quality metric investors track. Every megawatt moved from spot exposure to a 15- or 20-year contract with an investment-grade buyer removes earnings volatility — and earnings volatility is what keeps merchant multiples below regulated ones. The 2,700 MW leg does this at scale: it takes a slice of Constellation’s existing PJM output and converts it from market-exposed to contract-locked, in the tightest large power market in the country. That conversion has a real, if unquantifiable from the public record, effect on what multiple the market is willing to pay for the company’s earnings.
The fundamentals underneath the contracts were already strong. Second-quarter adjusted operating earnings came in at $2.55 per share, up from $1.91 a year earlier, on strength from the Calpine addition, higher PJM capacity prices, and the commercial business. Management raised full-year guidance to $11.50–$12.50 per share. The contracts announced since then do not touch 2026 earnings — uprate capacity begins arriving in 2028 — but they lengthen the runway behind the guidance trajectory the market is already pricing.
The rate-base versus merchant distinction
One structural point matters for comparing Constellation to the utilities investors usually hold. A regulated utility that wanted to uprate a reactor would file a rate case: the spending would enter rate base, customers would pay for it through rates, and the shareholder would earn the allowed return, typically in the 9–10% range on equity. The return is modest but nearly certain, once approved.
Constellation’s version skips the regulator entirely. No commission approves the $4.3 billion; no rate case determines its recovery. The “allowed return” is negotiated directly with Google in the PPA price, and whatever Constellation can extract above its cost of capital is the equity return. This is the merchant bargain: higher potential returns in exchange for higher risk — construction risk, since the uprates are not yet built; licensing risk, since the NRC must approve power-level increases; and counterparty concentration risk, since an increasing share of the contracted book now depends on a handful of hyperscalers.
Investors who hold both regulated utilities and Constellation should be clear-eyed about which risk they are buying. The PPA de-risks the revenue line the way a rate order de-risks a utility’s — but the buyer can renegotiate, default, or shrink, where a commission’s rate order, for all its politics, is harder to unwind. Investment-grade counterparties make that risk remote, not nonexistent.
The capacity-premium question
The day’s trading answered one question and left another open. Constellation’s double-digit gain — peers Talen and Vistra rose nearly as much — showed the market reads hyperscaler contracting as a sector-wide repricing event: firm, contracted, carbon-free megawatts are worth more than the market thought. But on October 8, two days later, BMO Capital cut its Constellation price target from $379 to $350 while keeping an Outperform rating — a reminder that even friendly analysts are recalibrating what the growth is worth after the run-up.
This is the capacity-premium question in its purest form: how much should investors pay today for megawatts that begin delivering in 2028, backed by contracts whose prices they cannot see? The market’s instinct has been to pay up — the stock roughly tripled from its lows before these deals — and the logic is coherent: scarce, contracted, firm supply in a market clearing at price caps deserves a scarcity premium. The counterargument is equally coherent: paying for 2028 earnings in 2026 assumes execution on hardware, licensing, and counterparties that are all still ahead of the company. The premium is a bet on the template holding.
Risks worth naming
The template’s risks are specific and checkable, which makes them worth listing rather than gesturing at. Execution risk: 11 units across three states is a program, not a project, and the first uprate is not expected until 2028. Constellation has a strong operating record — a 93% nuclear capacity factor in the second quarter across the owned fleet — but uprate construction is outage work, and outages can slip.
Regulatory risk: each uprate requires an NRC license amendment demonstrating the plant can operate safely at higher power. The NRC has approved more than 140 uprate applications over the decades, adding roughly 6,000 MW to the U.S. fleet, so the pathway is well-trodden — but well-trodden is not guaranteed, and the amendments are not yet filed in the public record for these units.
Counterparty and concentration risk: the contracted book’s growth is increasingly tied to four names — Google, Amazon, Meta, Microsoft — in a single sector. That sector’s demand thesis is the load-growth story itself: AI data centers. If the demand thesis weakens, the buyers’ appetite for 15- and 20-year power commitments weakens with it. The contracts are assets; the concentration is a risk.
Policy risk: the nuclear production tax credit, currently projected at $50.88 per MWh by 2030, underpins the economics of keeping merchant nuclear plants running. Tax policy can change, and merchant nuclear margins are sensitive to it.
What to watch
Four milestones will tell investors whether the equity story is delivering. First, the definitive agreements and any subsequent disclosures of terms — particularly whether future earnings materials quantify the incremental earnings contribution, which Constellation has not yet done. Second, the NRC license amendment filings for the 11 units and their approval timeline. Third, the 2028 delivery: whether the first uprate comes online on schedule, and at what cost. Fourth, the cadence of further deals — another hyperscaler contract would confirm the template is a pipeline, not a one-off.
None of this is investment advice, and nothing in this analysis recommends buying or selling any security. What the Google–Constellation announcement demonstrates, in investor terms, is a mechanism: merchant nuclear capex, de-risked by long-term corporate offtake, in the country’s tightest power market. The mechanism worked for Amazon in September, for Meta’s Vistra deal in January, and for Google twice in October. Until the uprates actually deliver megawatts in 2028, the equity story is a story about a story — but it is, at least, a story with contracts behind it.