Constellation Energy

IThe business

Constellation makes electricity and sells it. It has two businesses that fit together. One is a fleet of power plants. The other is a commercial arm that sells power and gas to large customers. It is not a utility: nobody guarantees it a return, and it owns no wires. It earns what the market pays, plus what contracts and government programs add.

It was Exelon’s generation arm until February 2022, when it was spun off. Exelon kept the wires and Constellation took the plants.¹³ On 7 January 2026 it bought Calpine, the largest American owner of gas plants, for $21.8B of stock and cash, and took on $12.6B of Calpine’s debt.¹⁴ It now describes itself as the largest private-sector power producer in the world, with about 55 GW.³ That is the company’s claim, and its own table supports it: 298M megawatt-hours against 250M for NextEra in the 2024 data it uses, adjusted for Calpine.⁷

A word on units. A megawatt-hour, or MWh, is what a mid-sized home uses in about a month. A gigawatt, or GW, is a large reactor. A plant that runs all year at one GW makes about 8.8M MWh. Power prices are quoted per MWh. Capacity, which pays a plant for being available at the peak, is quoted per megawatt per day.

How it really earns money

The most useful thing management has done is publish a kit. In March it split earnings into two parts and gave the inputs for one of them.⁷ Base earnings are what it calls visible: contracts, nuclear at the tax-credit floor, gas at a historic minimum, and average trading margins. Enhanced earnings are everything above that, such as market prices over the floor, or volatility.

I rebuilt Base from the published tools to test it. For 2026: 53M MWh of Illinois credits at $34.09, 101M at the $44.75 floor, 36M contracted at $70, about 127M MWh of gas at spark margins of $15 to $25, capacity, commercial margin, less fuel, operating cost, depreciation, $700M of interest and 26% tax. A spark margin is the power price less the gas it burns. That comes to $6.86 a share against the company’s $6.65 to $6.75 C (calculated).⁵ So the kit works. An outsider can price this company on the back of an envelope, which is rare in commodity businesses.

Then look at the split. Base is about $6.70 a share in 2026. Guidance for the whole is $11.50 to $12.50.⁵ About 40% of this year’s profit is therefore not in the visible part. It comes from power and spark prices that are above the floors today. Management says the Enhanced share falls to 30% to 35% by 2029 as Base grows to $11.40 to $11.90.⁵

The floor

The 2022 Inflation Reduction Act gave existing nuclear plants a production tax credit, the PTC, through 2032. It pays up to $15 a MWh and falls by 80 cents for each dollar of a plant’s receipts over a threshold of $26.⁵ In 2026 the result is that a plant’s revenue can’t really fall below about $44.75 a MWh. Above that the credit is zero and the plant keeps the market price. It is a free put option from the government, on about 147M MWh a year from 2028, indexed to inflation.⁵

The floor is not income today. Power is above it, so in the first half of 2026 nuclear PTC revenue was $25M.¹⁴ Management says only 5% of the nuclear fleet is in the zone where the credit pays.⁵ The floor protects the bad years, and it stops on 31 December 2032. Everything on this page turns on that date.

Illinois, New York and the contracts

About a third of the nuclear fleet has had other support. Illinois pays carbon mitigation credits, or CMCs, to Braidwood, Byron and Dresden through May 2027, and zero-emission credits to Clinton and Quad Cities on the same date.⁷ The CMC is a two-way deal: when the market is high Constellation pays customers back. In the first half of 2026 the average CMC was negative $16.05 a MWh, so the plants paid ComEd.¹⁴ From June 2027 those plants move to the PTC floor or to contracts, which is why 53M MWh of credits become 23M next year and none after.⁵

New York’s zero-emission program was extended in January 2026 through 2049, with combined market and credit revenue that the state’s order puts at $72.01 a MWh in 2030/31, rising to $115.75 in 2048/49.⁷ That is the longest-dated price in the portfolio, and it is one reason contracted revenue rises from $70 a MWh to $88 by 2030 in the company’s tools.⁵

Then come the technology customers. Microsoft signed 20 years for the restarted Crane reactor, the former Three Mile Island Unit 1, 835 MW. Meta signed 20 years for Clinton from June 2027.³ In September Amazon signed 20 years for 690 MW at Calvert Cliffs, including a 190 MW uprate, with over $3B of investment.⁹ An uprate is more output from an existing reactor. On 6 October Google signed 20 years for 890 MW of uprates at 11 units, with over $4.3B of investment, and a separate 15-year agreement for 2,700 MW across the PJM fleet.⁸ Google calls the second a financial structure that keeps plants operating. Neither party gave a price.⁸¹⁹ That silence matters, and I come back to it.

The customer side

Constellation served about 275M MWh of electric load in 2025 with Calpine included, about three quarters of it retail, and about 190M MWh of commercial and industrial load, roughly double the next supplier. It says it serves over 80% of the Fortune 100.⁷ That desk does three jobs. It sells the fleet’s output forward. It earns a margin of about $4.30 a MWh on power and $0.40 to $0.45 per dth on gas.⁵ And it gives Constellation relationships a plant owner would not otherwise have, which is partly why Google and Amazon deal with it directly.

Where the power goes

Half of its output comes from PJM, the 13-state grid from Illinois to New Jersey; 23% from ERCOT in Texas; 9% from New York; 8% from California; the rest elsewhere.⁷ PJM is where the argument is. Data-center growth has run into a grid that closed plants for a decade. Its last two auctions cleared at the price cap and still fell short of the reliability target, the latest by 6,831 MW.¹⁰¹¹

Each $100 a megawatt-day on one GW is about $36M a year. That is real money on 15,700 MW of cleared nuclear and 3,175 MW of gas in the latest auction.²⁰ But nuclear capacity revenue counts in the PTC’s gross receipts, so where a unit sits near the floor, the gain is partly handed back.⁵

IIThe moat

A durable business keeps others from taking its profits. There are two questions here. Can anyone copy these assets? And does owning them let you set the price? The answers are no, and no.

Copying first. The Google release sets uprates against the multi-year timelines of building from scratch, and says 890 MW equals one large new reactor.⁸ Constellation’s plants hold operating licences running from 2029 to 2054, most extended or being extended to 60 and in some cases 80 years.⁷ It filed in the second quarter to take Ginna and Nine Mile Point Unit 1 from 2029 to 2049.² A site with a licence, cooling water, trained staff, security and a grid connection is close to irreplaceable. Connection alone has become an asset: the Crane restart needed 760 MW of connection rights moved from the Eddystone gas units, because PJM’s study suggested upgrades running as late as December 2030.¹⁴²¹

Management says the fleet would cost over three times its enterprise value to build new, and that a new combined-cycle gas plant costs about $2,500 to $3,000 per kW against the $960 it paid for Calpine’s.⁵⁷ Those are company figures, and replacement cost is only a ceiling on what a plant is worth if someone actually needs to build one. Still, the asset sales give an honest market check. The Justice Department forced it to sell plants, and LS Power paid $5.0B for 4.4 GW, about $1,140 per kW, and $860M for 606 MW at Brazos Valley, about $1,420.¹⁴⁵ So a forced seller got more than the purchase price.

Pricing is another matter. Electricity is the same product whoever makes it. Three things set its price: gas at the margin, the capacity auction, and policy. Nuclear earns whatever the marginal plant earns. And when that gets high enough to hurt voters, politics steps in. On 16 January 2026 the White House’s energy council and PJM’s governors urged PJM to protect households from capacity prices and to hold a backstop auction, and Constellation’s shares fell 9.8% that day C (calculated).³¹ PJM then filed, and FERC approved, a cap of about $325 and a floor of about $175 for two more auctions, cutting the cap for 2028/29 from about $550.¹¹¹⁰ PJM’s market monitor opposed Constellation’s own Crane waiver in April, saying it could shift costs to others, and lost.²²²¹ A business whose price can be changed by a phone call from a governor has a lot less protection than one with a patent.

Evidence it is widening

The contracts. In March about a quarter of clean firm output was under long-term agreement, after about 36M MWh a year of 2029 output were committed in twelve months. By August management said about 30% of expected clean output by 2032.⁷⁵ Then came Amazon and Google. My own sum puts Google’s 2,700 MW at about 22M MWh a year and Amazon’s existing 500 MW at about 4M, which would take it above 40% by 2032 if those deals start by then E (estimate). A 20-year contract with an investment-grade buyer changes what kind of asset this is. Its bonds say so: in January it sold $800M of 40-year notes at under 6%.⁷

Uprates are the second point. Constellation says it has a path to add or extend about 9,350 MW, with about 1,100 MW of uprates at the first stage.⁷⁵ These are new megawatts at sites that already have everything else. On 27 October it got a conditional commitment of up to $5B from the Department of Energy for them.²³ Government capital at Treasury-like spreads, if it closes, makes it harder for anyone else to compete with this fleet for the same buyers. The $1.0B Crane loan is priced at 0.375% over Treasuries.¹⁴

Evidence it is narrowing

Buyers are learning to do without. PJM’s Bring Your Own Power proposal and FERC’s rules for colocated load aim to make data centers bring supply of their own.⁸¹⁴ That can be good for Constellation, as Google shows. It can also mean new gas behind the fence that never pays for anything Constellation owns. New supply responds to high prices in time: PJM has fast-tracked large new plants and a backstop auction.¹¹²⁴ And the 2,700 MW with Google looks like a hedge. A hedge at a fixed price protects the seller in a fall and gives away the rise. Without a price I can’t tell how much of the scarcity value was sold.

My judgement: the assets can’t be replaced, and the price of what they make is not under the owner’s control. That is a real but incomplete form of protection. It is better than a commodity producer’s, since supply can’t grow quickly. It is worse than a toll road’s, because the toll is a political decision. The long contracts are slowly turning the second kind of business into the first, and I would pay more for that if I could see the prices.

IIIThe competitive advantage

Four listed American companies sell power from plants they own into open markets. Constellation is by far the largest, and the only one where nuclear is most of the output.

On enterprise value per kW, Constellation is priced well above a gas fleet. Take the gas at a price like LS Power’s, say $1,250 per kW on 28.2 GW, or $35B. The other $79B, about $3,029 per kW on 26.2 GW of nuclear and other plants, is what the market pays for the rest E (estimate). That is a rough way to see it, and a fair one, since it says most of the premium sits in the nuclear.

Where Constellation wins

Operations first. Its nuclear capacity factor, the share of possible output it delivers, was 94.7% in 2025 and has stayed within a point of that for a decade.³¹⁷ It says that beats the industry by about four points every year since 2013, which it puts at over 8M extra MWh, about a reactor’s worth.³⁷ Refuelling outages took 22 days in 2025 against an industry 38 in the comparison it uses.⁷ This is the one number that belongs to management and not to the market, and it has held up through a spin-off and an acquisition.

Scale second. A nuclear operator with 25 reactors’ worth of spare parts, procedures and staff can do things a one-site owner can’t. Fuel is the plain example: it reports a diversified book of long-term contracts into the 2030s, with forward uranium prices below spot.⁷ The balance sheet is the third. It is rated BBB+ and Baa1, and Moody’s moved its downgrade trigger to 25% FFO to debt from 30%, a sign of how the agency sees the contracted mix.⁷ Long buyers want a counterparty who will be there in 2046.

Where it loses

Pure exposure. Smaller fleets may move more on each dollar of power price, though I haven’t modelled Vistra or Talen. Texas is the second. The company’s own reference spark spread for ERCOT North fell to $16 for 2026 and $22 for 2027 from $26 and $28 in March.⁵⁷ That matters more now that Calpine has made ERCOT a quarter of its capacity.⁷ Consumer brands are the third, where NRG, with about 8M residential customers, is much bigger.²⁷

Which leaves the point that a rival could make: Constellation paid up for a gas fleet just as the gas story was getting more crowded, and the nuclear is what people came for. On the other hand, gas is what fills the hours nuclear can’t flex. The combined-cycle fleet ran at a 55% capacity factor in 2025, so there is room for more output if data centers arrive.⁷ Management sells that as optional upside. I treat it as a modest part of Enhanced.

IVManagement

Joe Dominguez, 63, has been chief executive since the spin-off in 2022, and became chair as well on 4 August 2026 when Robert Lawless retired. Charles Harrington is now lead independent director.³²⁹ Shane Smith became finance chief at the Calpine close, moving up from treasury; Dan Eggers, who held the job, moved to a role on finance and the data economy; Andrew Novotny, who ran Calpine, runs power operations.³⁰¹⁴ Kathleen Barrón, the strategy chief, said she would retire in mid-2026.³⁰ That is a lot of change at once around a large integration, and it comes with one person now in both top jobs.

How they are paid

Dominguez’s reported pay for 2025 was $17.1M: $1.5M of salary, $12.1M of stock awards and a $2.6M bonus, about 140 times the median employee’s $122,610.¹² The bonus is 70% operating earnings and 30% operational: capacity factor, dispatch, renewable capture and customer satisfaction. It paid 112.69% of target. The long-term award is two thirds performance shares and one third time-vested units. Performance shares run three years and pay on free cash flow before growth, two thirds, and on share return relative to a custom peer group, one third, with a cut if credit ratings fall.¹² The proxy’s own pay-versus-performance table puts what he actually earned in 2025 at $78.8M, because his unvested stock awards rose in value.¹²

Follow what that pays them to do. Earnings and capacity factor reward running the plants well, and the record there is excellent. Free cash flow before growth leaves out growth spending by definition, and for pay it also leaves out working capital. So the $7.3B going into uprates costs nothing against that measure. Relative share return is measured against other companies over three years, a window that can sit wholly inside a boom in power stocks.¹² No part of the plan measures return on capital. For a company about to spend heavily on uprates, on prices it does not disclose, that is the gap I would most like closed. The proxy also says in-flight targets were raised for Calpine, which is the right instinct.¹²

A smaller point. Executives who hold 200% of their ownership target have performance shares paid in cash, not stock.¹² The guideline for Dominguez is six times salary. In February 2026 a Form 4 shows 191,817 shares vesting, 83,104 withheld and 92,993 going back to the company at $272.15, which fits the cash rule and comes to about $25M C (calculated).³¹ I don’t object to that. It does mean the more he succeeds, the less of his pay stays in the stock.

How much of their own money is in

The proxy shows Dominguez with 157,074 shares and units as of March, about $41M at the price and about 28 times salary C (calculated).¹² Directors and officers as a group hold under 1%. The 39 Forms 4 I read from July 2025 to October 2026 show no open-market sale by a filer, and one open-market purchase: Roger Crandall, a new director, bought 1,500 shares at $278.62 in August.³² Novotny’s Forms 144 show about $14M of stock sold or planned for sale in August and October, under a plan set in May for 118,915 shares.³³¹⁴

Holders outside the company tell a similar story of quiet selling after the deal. ECP, a former owner of Calpine, had 22.0M shares in the March proxy and 14.4M at 30 June.¹²³⁴ In June holders sold 11M shares in a secondary, and the company took 2M of them for about $558M.³⁵ The second half of their lock-up ends on 30 June 2027, one to watch.³⁵

What they did with capital

Four and a half years is not a full cycle for this company, because the cycle it has seen has mostly gone one way. The test before that is hard to use: as Exelon Generation it closed Three Mile Island in 2019 for economic reasons, a plant it now pays to restart.³ What I can do is list the choices since 2022.

The Calpine deal deserves a closer look. It was announced in January 2025 when the shares were near $305.¹ It closed with shares valued at $354.58, so the recorded price rose to $21.8B from the $16.4B reported at the time.¹⁴³⁷ Paying in expensive stock is what a sensible seller of stock does. At $262.38 those 50M shares are worth about $13.1B, so the deal looks cheaper than it did at close. Add the debt and take off the sales, and Constellation paid about $28.5B for about 23 GW and a retail book, about $1,239 per kW before counting the retail business C (calculated). Its own slide uses $960 per kW, on the announcement figures.⁷ That is well under the cost of a new plant, on the company’s own figure.

The judgement I’d make: excellent operators, opportunistic and able dealmakers, and a pay plan that doesn’t yet ask the question that matters most from here, which is what each new dollar earns.

VFinancials

The history here needs handling with care, and three things make it hard to read. Nothing before 2022 is a standalone company. 2016 to 2021 are Exelon Generation’s statements, with costs and plant closures of that era.¹⁷¹³ Calpine appears only from 7 January 2026. And the revenue line mostly reflects fuel and power prices passing through, so it says little.

Over those ten years net income came to $13.0B and revenue to $211B, an average margin of 6.1% C (calculated). Operating income was low or negative until 2023, then jumped to $4.4B in 2024. What changed was the PTC, capacity prices and a better market. Adjusted operating earnings were $8.67 a share in 2024 and $9.39 in 2025.³

Why the cash flow statement misleads

Reported operating cash flow was negative from 2021 to 2024, −$5.3B at the worst.¹⁵ Taken at face value this was a company burning cash, and it wasn’t. Under a receivables facility, cash collected on sold receivables was booked in investing. From 2020 to 2024 that added $30.2B C (calculated).¹⁵³ Put it back and the ten-year cash after all capital spending is $15.9B, about $3B above net income. The facility ended on 31 December 2024, so 2025’s $4.2B is the first clean figure.³ After $2.9B of capital spending and $0.4B of stock pay, that leaves $0.9B C (calculated).

Other items also distort the picture. Nuclear decommissioning trusts swing GAAP profit with the stock market: in the second quarter net income was $513M and adjusted earnings $920M, with $340M after tax of unrealized hedge losses and $221M of decommissioning gains between them.² Calpine adds a non-cash amortization of contracts, $149M after tax in that quarter.² So adjusted earnings are the better guide to profit here. Stock pay stays in them, since it runs through operating cost: $385M in 2025.³

This year

In the first half revenue was $18.6B, up 45%, with $5.0B from Calpine.²¹⁴ GAAP net income was $2.1B, or $5.88 a share. Adjusted earnings for the second quarter were $2.55 a share, up from $1.91, and guidance rose in August to $11.50 to $12.50.² Operating cash flow was $1.6B against $2.5B of capital spending, and the half included about $920M of refunds to state programs for 2024’s PTC.¹⁴ TTM revenue is $31.3B and TTM GAAP net income $3.5B C (calculated). On a pro forma basis, as if Calpine had been owned all year, 2025 revenue was $36.8B, operating income $5.7B and net income $4.0B.⁴

The balance sheet

On 30 June debt was $24.7B, of which $5.2B was short term, against $0.7B of cash.² So net debt was $24.0B. Sale proceeds will take off about $4.5B after tax. The $860M Brazos Valley sale closed in October, and the $5.0B sale of the PJM plants is expected by year end.⁷³⁸ Corporate debt had an average life of about 12.5 years at about 5.25% in March.⁷ The balance also holds $11.5B of goodwill, nearly all new with Calpine.²

There’s an asset that doesn’t appear in most summaries. Decommissioning trusts hold $20.5B against asset retirement obligations of $12.6B and a $5.9B payable tied to units where surplus belongs to customers.² So the owner’s trusts look about $2.0B over on those book figures C (calculated). I don’t add that to value, since those obligations are long-dated and estimates, but it means the cost of closing the plants is mostly funded, which is unusual for a liability this size.

Where the cash goes in 2026 and 2027

Management’s plan from March: $0.8B of cash, $8.4B of free cash flow before growth and $4.5B of asset sales give $13.6B. It sets that against $5.0B of buybacks, $3.9B of growth, $3.4B of debt reduction and $1.3B of dividends.⁷ Capital spending is $5.7B in 2026 and $4.7B in 2027. That splits into about $3.3B and $3.2B to stand still, baseline work and nuclear fuel, and $2.4B and $1.5B for growth.⁷ Now add the Google and Amazon uprates. Their $7.3B was not in that March plan as far as I can tell, and most of it falls after 2027 on my phasing E (estimate). The DOE commitment could fund up to $5B.²³ If it closes, the buyback survives. If not, uprates and buybacks compete, and something has to give, which is a good reason to read the Q3 deck, expected on 9 November.³⁹

VIValuation

First, how this goes badly wrong

Before any numbers I asked what would most likely ruin this investment, and worked back. Three things can.

• The price of power becomes a political decision. It partly is already. PJM’s cap sits at $325 when its own process would have cleared far higher, a White House council and a set of governors have pressed it, and a backstop auction has been accepted but suspended to 28 February 2027, subject to refund.¹⁰²⁴³ Household bills are going up: the average American price in August 2026 was 5.1% above a year earlier.⁴⁰ A company earning most of its Enhanced profit from high prices depends on voters not minding. Windfall rules, a lower cap, or a state moving back to regulated generation would each cut the top of the range. None needs a surprise. The 10-K lists price caps among its risks.³
• The floor ends and the buyers don’t replace it. Base earnings rest on 147M MWh at the PTC floor. Congress wrote it to end in 2032. If by then new gas, or lower AI demand, has pulled power back to something like $40 to $48 with capacity, the fleet earns less in 2033 than the floor paid in 2032. Each $10 on those megawatt-hours is about $20 a share of value E (estimate). And it would arrive just as the $7B of uprates are finished.
• One reactor, every reactor. The worst event is an accident, here or abroad, that leads to a fleet-wide order. Constellation carries $500M of liability cover per site, the rest is shared across the industry, and the 10-K says a major incident could exceed its resources.³ I can’t price that. It has very low odds, and the loss would be far beyond any case on this page. It belongs on the list because 60% of output sits under the same regulator.

None of these needs Constellation to run its plants badly. All are outside its control. That shapes my method: I want to pay for what contracts and the floor protect, and pay less for everything that needs the price of power to stay high after 2032.

What an owner can really take out

Management’s measure is free cash flow before growth: $8.4B over 2026 and 2027, and $11.5B to $13.0B over 2028 and 2029.⁵ That is after baseline capital spending and nuclear fuel, so it does count the cost of standing still, which is about $3.2B a year against about $2.9B to $3.2B of depreciation and fuel amortization C (calculated).⁷⁵ That’s a decent match. Here depreciation is a fair guide to true upkeep, which is not always so.

It isn’t quite owner cash. It leaves out stock pay, about $0.3B a year after tax, and the costs of joining Calpine, $475M in 2026 and $125M in 2027 before tax.³⁵ It also leaves out collateral swings. Take those off and an owner gets about $3.7B a year in 2026 and 2027, or $10.37 a share E (estimate). Against the guidance midpoint of $12.00, about $4.3B, that is 86 cents on the dollar. Against the $3.5B of TTM GAAP net income it is a little more. So reported profit is a reasonable guide here, which it is not for many companies on this site.

That $3.7B is 4.0% of the market value. Spend the growth capital too, which you must if you want the contracts, and my base leaves about $1.6B a year for 2026 and 2027, and under $5B a year until the uprates finish E (estimate). In the first half operating cash flow didn’t cover capital spending, so the $2.2B of buybacks was funded with debt, ahead of the sale proceeds.¹⁴⁵

What the price implies

At $262.38 the market values the shares at $93.1B and the whole company at about $114B. That’s about 22 times the 2026 guidance midpoint, 39 times 2026 Base earnings and 22.5 times 2029 Base C (calculated).⁵ Nasdaq’s consensus runs $12.12 for 2026, $13.25 for 2027, $16.74 for 2028 and $19.55 for 2029, on seven, seven, four and two estimates, so 13.4 times the last of those C (calculated).¹⁸ My base has $17.53 in 2029, about 10% under.

So what has to be true? Take my base to 2032, then change only the open price. The price needs about $89 a MWh in 2033, with capacity, rising 2.5% a year. In round numbers that’s 40% above where I estimate today’s fleet mix sits. On the reference prices in the August deck, Illinois is about $55 with capacity, the Mid-Atlantic about $81 and New York about $59, for 2027. Weighted by where its nuclear ran in 2025, that’s about $63 E (estimate).⁵⁶ So the price assumes the whole uncontracted fleet earns more in 2033 than its best region does today. Some of that is possible, and PJM’s uncapped result would point that way. It’s not my central guess.

What I can’t judge

I can’t tell you the price of power in 2033. Nobody can. Three inputs matter, and I don’t know any of them: what uncontracted power earns when the PTC ends, whether a cap survives, and what Google and Amazon are paying. The first two move my value most. The third moves it least, by about $17 either way across the range below, but it is the one most likely to come out soon, through the Base tools. So I show the range, and my $218 is best read with it.

The discount rate

I use 8.75%, and it applies to cash before interest, so it has to blend the cost of shares and debt. The ten-year Treasury yields 4.94% and Damodaran’s implied equity premium is 4.20%.⁴¹⁴² My beta for Constellation on month-end returns since 2022 is 1.13, which gives 9.7% for the shares. The last three years alone give 2.03 and 13.5%, because the stock has traded like an AI stock C (calculated).¹⁴³ I don’t think that reflects the plants. Corporate debt costs about 5.25%, or 3.9% after tax, and it is about a fifth of enterprise value.⁷ Together that is 8.6%, and I round up a little for the politics.

It looks lower than the 10.5% on my Broadcom page and the 9.0% on Berkshire. The comparison isn’t like for like, since those are rates on equity cash flows and Constellation’s is a blended rate on cash before interest. On equity my base return is 7.5%, and that is the number to set beside theirs.

Three cases

The model starts from the company’s tools and adds what they leave out. It takes the published 2026 to 2030 volumes and prices for contracted nuclear, floor-priced nuclear and gas, and it reproduces 2029 Base earnings within about 3% before I add new deals C (calculated).⁵ To that I add the signed contracts: 920 MW from the second quarter, 2,700 MW with Google and 500 MW at Calvert Cliffs converting existing output, and 1,080 MW of uprates as new output, at prices none of them publishes. The company’s own slide puts a 1 GW nuclear contract at a premium of $20 to $50 a MWh over the tax-credit floor, which is about $51 in 2030, so about $71 to $101.⁵ I use $70, $80 and $90 for bear, base and bull, which are $19, $29 and $39 over the floor, and test $65 to $95 below. The growth program earns the 10% that management calls its minimum. Enhanced margin follows the company’s 30% to 35% in my base. After 2032 the floor stops, and uncontracted nuclear takes one price. Costs and capital rise 2% a year, fuel 2.5%. Tax is 26%. Stock pay is a cost, inside operating cost. Cash for 2026 counts half a year, because net debt is from June.

Two cross-checks. Before uprate spending, my base has $11.9B of free cash flow over 2028 and 2029, inside its $11.5B to $13.0B range E (estimate).⁵ And it has $12.06 of adjusted earnings for 2026, against guidance of $11.50 to $12.50 E (estimate).²

The weighted value on 25%, 50% and 25% is $219. A stress case, with no weight, puts open power at $40 in 2033, near today’s Illinois energy price with no capacity at all. It is worth $100, and returns −1.0% a year at the price. Even that one keeps every signed contract, which is the sense in which this company has become safer: the worst case here is a smaller business, and not a broken one, unless the third danger above comes true.

The margin of safety

My margin-of-safety price is $184: the point at which my $138 bear value is three quarters of the price paid. It’s the same rule as on my other pages. At $184 the bear case returns 6.8% a year and the base 12.2% E (estimate). That is 22% below the lowest close of the past year, $236.50 in July, and about where the shares closed on 8 April 2025.¹ So it is not an absurd level for this stock in a bad month. There is none at $262.38.

Other uses of the same dollar

The ten-year Treasury pays 4.94%, with no risk of a reactor. My base here returns 7.5% a year, so about two and a half points more, and my bear 2.4%, a little under half of the Treasury E (estimate).⁴¹ The closest comparison on this site is Berkshire, which owns a large power business, Berkshire Hathaway Energy.⁷ My Berkshire page has the shares returning about 7.0% a year at its price, with a bear value only 27% under it. Constellation’s base is about the same return with a bear 47% under the price. For the steadier owner of American power assets, the same dollar buys about the same return and much less to lose.

For exposure to AI power demand, the question is what else sells to the same buyers. TSMC is 9.4% on my base and 5.3% on my bear, and NVIDIA 9.7% and 5.5%, on their own prices and rates. Each makes more of the money from the same data centers. What they don’t have is a federal floor and 20-year contracts with investment-grade buyers. American Express, which sits level with my base value, shows what a well-understood franchise costs today. Within power, I have no page on Vistra or Talen, so I can’t say whether a dollar does better there.

Quality, and price

These are two questions. As a business this looks good, and getting better: irreplaceable plants, the best operator in its field, a government floor for six years, and twenty-year contracts arriving from four of the largest companies on earth. Against that, a product whose price is set by others, and a pay plan that doesn’t ask about returns. At $262.38, the price asks for more than I can underwrite after 2032. This looks to me like a good business at a full price: not an absurd one, since the bull is 14% above and a rate of 7.9% gets there. The market may well be right to pay it. Buyers are signing 20-year contracts at prices I can’t see, and those prices are the best evidence anyone could have. The company illustrates $20 to $50 over the floor. Even at the top of that, my base gives only $242, so the contracts alone do not close the gap. The open price after 2032 does.

The last month

On 6 October the shares rose 12.2% to $300.40 on Google, and by the latest close they had given back all of it and more.¹ I found no filing that explains that. They fell 5.5% on 14 October and 5.7% on 22 October, and I could not find a clear cause for either in the sources I could open. This stock doesn’t move on one document.

Where this could be in 5 and 10 years

The value at each date is my model’s present value at that date, at 8.75%, of the owner cash after it, less $21.2B of net debt and other claims, per today’s 355M shares and in that year’s dollars. Cash paid out is owner cash from November 2026, all of it paid out. The return starts at $262.38 and ends at the 2036 value E (estimate).

The base holds at five years if owner cash reaches $5.4B in 2031, from $1.1B in 2026, as the uprate spending ends. It holds at ten if open nuclear power earns $66 in 2033 and rises 2.5% a year, and owner cash reaches $7.7B in 2036. At 8.75% the value rises about 4.5% a year. Of the 7.7% return, about 2.6 points is that rise and the rest is cash: $162 a share over the decade.

The bear is a stall. New contracts stop at 33 million MWh, owner cash sits near $5B, and open power earns $48. The bull is the reverse. 85 million MWh is contracted at $90, open power earns $80 and owner cash reaches $10.5B. Most of the gap is the price of power after 2032, and how much output is contracted. My returns run about 0.2 points above the cases table, which I put down to dating the cash and the exit differently.

The most likely way my base is wrong is that open price. At $45 instead of $66, base value in 2036 is about $263, not $341, and the return falls from 7.7% to 5.2% E (estimate). At a 9.5% discount rate the value is $298 and the return 6.7%.

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three things cut in Constellation’s favour and rarely come up. The floor rises with inflation, and in steps. The August print of 2.8% lifted the 2030 floor by $1.00 and added $0.30 to Base earnings in that year. At 3% inflation the floor reaches $56 in 2032, not $51.88, which management puts at $1.55 a share.⁵ So this is one of few large companies whose earnings rise by rule if inflation returns. Second, New York’s program now runs to 2049 on more than 3,000 MW, with combined revenue of $72.01 a MWh in 2030/31 and $115.75 in 2048/49.⁷ That is a state-backed price more than twice the PTC floor, for twenty years past the date this page worries about. Third, the 10-K says Clinton’s uprate is expected to qualify for the separate technology-neutral credit.³ If the Google and Amazon uprates do too, which I can’t confirm, they earn more than my model allows.

The bull case

In plain words: America needs more firm power than it can build, the fleet can’t be copied, and buyers with the deepest pockets in the world will sign for decades to get it. PJM has now come up short of its reliability target twice at the cap.¹⁰¹¹ Management says its fleet would cost over three times its enterprise value to build, and that 147M MWh a year remain to be contracted in 2029.⁵ Joe Dominguez said on 6 October that the Google deal can “serve as a model” for technology companies and the energy industry.⁸ The Energy Department has conditionally offered up to $5B.²³

The Street leans this way: Nasdaq’s means have earnings rising 61% from 2026 to 2029, to $19.55, though that last figure rests on two estimates.¹⁸ Among well-known investors the record is thinner than the story suggests. Coatue held 4.63M shares at 30 June, worth $1.15B, down from 5.91M at the end of 2025, and I found no statement from the firm on why.⁴⁴⁴⁵ Third Point reported 475,000 shares at the end of 2025 and none three months later.⁴⁶ The June 13F filings of Altimeter, Atreides, Duquesne, Appaloosa, Situational Awareness and Berkshire show no position, nor do Pershing Square’s filings for 31 March and 30 June.⁴⁹ ARK’s ARKK and ARKQ funds list no Constellation in their holdings files of 30 October, and I found no ARK model or comment on it.⁵⁰

Against reality

Today’s numbers support the near part. Capacity is priced at the cap through May 2029. Guidance rose in August. Over 40% of clean output looks contracted by 2032 once the October deals are counted E (estimate). If they do that, and deliver Crane in 2027 and the uprates from 2028, my base earnings of about $18 a share in 2029 are within reach, and owner cash of about $7B by 2032 follows.

The price needs more. If they get over half of output contracted by 2036 at prices inside the company’s $71 to $101 range, and keep Enhanced margin near $2.5B, and open power after 2032 reaches about $89 a MWh, the price is possible. Today the same mix is about $63 on forward prices E (estimate). So the gap is in the third leg, and it’s about $26 a MWh, or about $1.5B a year of extra margin starting now.

It helps to multiply it out. Say PJM stays short with capacity near the cap through 2030: 60%. Contracts cover over half of output at prices inside that $71 to $101 range: 50%. Open power after 2032 clears $75 with no PTC: 40%. Spark margins stay firm in Texas and California: 65%. No fleet-wide nuclear event, and Crane starts on time: 85%. Those are my guesses, and they overlap, since strong demand helps most of them. Multiplied, they give about 7% E (estimate). Allowing for the overlap, perhaps 20% to 25%, which is the weight I give the bull. Each step looks plausible. All five together are less likely than they feel.

What a 10x would take. Ten times $262.38 is $2,624, about $931B of market value. At 20 times earnings that needs about $47B of net income, eleven times the 2026 guide. Constellation’s fleet makes about 300M MWh a year, on the company’s own table.⁷ At that output the profit alone would have to be about $155 a MWh after tax, more than twice the price of power at PJM West in 2027.⁵ American customers paid $61.4B for all their electricity in August 2026, a peak month.⁴⁰ Net income would equal over three weeks of that bill every year C (calculated). Or look at value. NextEra, the largest American power company I know of, is worth about $176B on its latest share count, and tenfold would make Constellation over five times that.⁴⁷²⁸ By 2036 it would take 26% a year for ten years from $262.38 C (calculated). The stock did rise over nine times from $42.00 to $403.95 between 2022 and 2025, but that began at a market value near $14B and with power at the bottom of its range.¹ From here tenfold is close to impossible, and I would not build a case on it.

Where I come down

The Google and Amazon deals moved me: before them my base had less contracted volume and a thinner path past 2032. What didn’t move me is the market’s price after 2032, since no contract yet tells me what uncontracted power earns once the floor goes. What would raise my values most is evidence on open power after 2032: Congress extending the credit, or a Google price near the top of the company’s range, which would suggest buyers pay well above the floor.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Crane. The restart is targeted for 2027, with the NRC’s fuel approval and FERC’s waiver in hand.² PJM’s study showed upgrades as late as December 2030, and FERC said the rights transfer could allow full operation before the end of 2030, which is not the same as 2027.¹⁴²¹ A restart of a closed reactor has not been done often, and it is the one part of this story with construction risk.
• Illinois after May 2027. The CMC and ZEC programs end. The move to the PTC or contracts is in the plan, but Illinois energy is priced far below PJM West, and those are the plants that most need a buyer.⁵
• Calpine integration. The company has not published a synergy figure that I could find. Integration costs are $475M in 2026 and $125M in 2027.⁵ Two different cultures now share one fleet, one trading book and a new structure at the top.
• Trading and collateral. A large commercial book needs liquidity. Cash fell to $0.7B at June and short-term debt rose to $5.2B, with a $7.0B commercial paper program behind it.²¹⁴ A price spike can call for collateral fast, as in 2022 and 2023.¹⁵ A three-notch downgrade would be needed before cash collateral use is restricted.¹⁴
• Nuclear fuel. The Russian uranium ban, and Russia’s own limits, tighten enrichment. The company says it is covered well into the 2030s, and it is building inventory.⁷³ Costs in the tools rise from $5.78 to $6.48 a MWh in one year.⁵
• Holders’ stock. Half of ECP’s lock-up runs to 30 June 2027.³⁵ That’s a known seller at a known date.
• Governance. Chair and chief executive are now one person.²⁹ A new director, Alan Armstrong, joined on 1 January and resigned on 23 March; the filing gives no reason.⁴⁸
• Contract terms. Long contracts protect the downside and can sell the upside for twenty years. If power in 2040 is worth far more, the holders of these contracts will be Google and Amazon.