SYNTHOS RESEARCH

Constellation Energy CEG

Utilities · Independent Power Producers · Synthos Deep Dive · 2026-08-04

$276.75
Watch
—
—

The Overview

Constellation Energy generates electricity and sells it. What makes it unusual is what it generates with: it owns more nuclear power stations than any other company in America, and it sells their output into competitive markets rather than to a regulated customer base at a fixed return.

For most of the last two decades that was a mediocre business. Nuclear plants are expensive to run and cheap electricity from natural gas kept prices low. Several were closed.

Then artificial intelligence arrived. Data centres need enormous amounts of electricity, they need it to be available every second of every day, and their owners have promised it will be carbon-free. There is exactly one existing technology that does all three, and almost nobody is building new ones. Microsoft has contracted the entire output of the restarted Three Mile Island reactor. Our own research library flagged this in mid-2024 and was right — the shares roughly tripled.

Then they fell 34%.

Two things changed the company itself. On 7 January 2026 it bought Calpine, which owns about 23 gigawatts of gas-fired power stations and a big business selling electricity directly to households and companies. Constellation paid 50 million of its own shares plus $4.5 billion of cash and took on about $12.6 billion of Calpine's debt. Revenue in the first quarter jumped 64%. The competition authorities required it to sell six power stations as a condition, and the deadline to sign those sale agreements is 4 September — a month away.

On 14 July, every one of Constellation's power stations in the mid-Atlantic grid won a three-year-forward auction at $325 per megawatt per day, starting June 2028. That is essentially guaranteed money, and it is why analysts expect profits to jump 28% in 2028.

At $267.25 the shares cost about 20 times what analysts expect the company to earn in 2027. Our estimate of fair value is $300, about 12% above the price. Analysts on average say $355, and the most pessimistic of them says $296 — still above today's price. We are more cautious than all of them.

Why we are saying Watch rather than Buy: 12% is not enough for a business this volatile, the company reports results in two days, the divestiture deadline is a month away, and the financial data supplied for this company is riddled with errors — it reports a gross profit larger than its revenue in one quarter, and shows negative operating cash flow in three of the last four years for a business that plainly generates cash. We would buy this at $240.

Putting a number on it: our fair-value estimate is $300 against a current price of $276.75 — real upside if our numbers are right.

Our summary metrics

Downside Risk (lower = safer)7/10High

"Rated 7 — a genuinely strategic asset base carrying integration, merchant-price and timing risk simultaneously, in the middle of a 34% drawdown. The supports: approximately 32,400 MW of owned generation before Calpine and roughly 23 GW added with it; every PJM plant cleared the 2028-2029 capacity auction at $325 per megawatt-day, effective 2028-06-01, which is a contracted revenue floor three years out; an Inflation Reduction Act nuclear production tax credit floor that the knowledge base puts near $43.75 per megawatt-hour; approximately $3.5 billion of remaining share-repurchase authority after the June buyback; and a trailing return on equity of 20.1%. Against that, five risks. FIRST, the Calpine integration is seven months old, purchase accounting is provisional, and the company assumed approximately $12.6 billion of Calpine debt — $7.6 billion corporate and $5.0 billion of project financing. SECOND, a hard regulatory deadline: the DOJ resolution requires definitive agreements to divest five PJM generating assets, one ERCOT asset and the Gregory Power minority interest within 240 days of closing, BY 2026-09-04 — thirty-one days from this dive. Five PJM assets totalling approximately 4.4 GW are under agreement with LS Power; $5,735 million sits in assets held for sale. THIRD, this is a merchant generator: outside contracted capacity and PTC floors, its earnings are power prices. FOURTH, the stock is 33.8% below its high with a 33.8% maximum drawdown and a MINUS 21.6% twelve-month return, and it fell 2.36% today — the de-rating is not demonstrably over. FIFTH, results are released in two days."

Growth Quality7/10High

"Rated 7 — very large, acquisition-driven, and with a genuine second step scheduled for 2028. March-quarter revenue was $11,122 million against $6,788 million a year earlier, UP 63.9%, almost entirely Calpine, which closed 2026-01-07. Adjusted EPS from the `earn_cal` actuals runs $2.14, $1.91, $3.04, $2.30 and $2.74 across the last five quarters — a trailing $9.99 — with beats of MINUS 1.8%, plus 3.8%, MINUS 2.3%, plus 0.9% and plus 7.9%. Note that record: two misses in five quarters, and the beats are small. This is not a company that consistently surprises. Consensus wants $11.734 in FY2026 (11 analysts), $13.335 in FY2027 (13) and $17.085 in FY2028 (11) — 13.6% then 28.1% growth. The FY2028 step is the interesting one and it is not arbitrary: the PJM 2028-2029 capacity auction cleared at $325 per megawatt-day effective 2028-06-01, with 9,800 MW of nuclear clearing in COMED alone, 4,325 MW in EMAAC and 1,575 MW in MAAC, and capacity revenues for nuclear units count toward the production-tax-credit gross-receipts calculation. That is a legislated and auctioned revenue step, not a forecast. What holds this at 7: the FY2027 estimate range is $11.866 to $14.314, a 20.6% spread, and FY2028's is $14.532 to $23.119, a 59% spread — the street does not agree on the size of the step; and the growth to date is bought, not compounded."

Exponential Potential6/10High

"Rated 6 — the highest exponential score in this batch, and it rests on a genuine structural repricing of an asset class rather than on a technology curve. The argument, which the knowledge base makes in all three of its surviving claims: artificial-intelligence data centres need enormous quantities of firm, carbon-free power on short notice; the only existing asset that supplies it at scale is the merchant nuclear fleet; and that fleet cannot be replicated, because nobody is permitting new large reactors on a five-year horizon. `business_breakdowns` (2024-06-01, speaker John Dullis) put it concretely — an IRA price floor near $43.75 per megawatt-hour, green premiums for data centres, tight short-power markets, and 'put that in your DCF, worth a lot more.' Constellation is the largest owner of that fleet, and the Three Mile Island restart contracted to Microsoft is the proof of concept both surviving voices cite. Calpine adds the complementary asset: roughly 23 GW of dispatchable gas that firms the intermittent supply data centres cannot use alone, plus a retail platform serving about 62 TWh. Against that: this is still a power company. Its output is a commodity; its capacity revenue is set by an auction; its production-tax-credit floor is legislated and therefore repealable; and the repricing the knowledge base identified has already happened once and given back a third. A 6: a real, scarce, non-replicable asset positioned in front of a genuine demand shock, wrapped in a business whose price is set by auctions and politics."

Fair value$300 $210–$358
What does “fair value” mean?

Fair value is Synthos’s estimate of what one share is worth today, based on our model of the company’s future cash generation and the risks to it. It is not a price target or a prediction of where the stock trades next quarter — it’s the price where we think risk and reward are balanced. Above it, you’re paying for outcomes better than our base case; below it, the market is offering a margin of safety.

The model’s inputs and weightings are proprietary — they’re the product. What we publish is the output, plus an independent, public-math cross-check further down the page (“Check our number”) so you can judge it for yourself.

What do the 5 tiers mean? (Core · Tactical · Watch · Hold · Avoid)
Buy — CoreOwn it as a foundation — start or add now, size it for years, let dips be gifts.
Buy — TacticalGood price + confirmed trend + a defined exit — buy the setup, not a marriage.
WatchWe want the business, just not at this price/setup — act only when the listed trigger hits.
HoldFine to keep if you own it — no reason to buy more; new money does better elsewhere.
AvoidDon't own it — the problem is the business or the expectations, so a cheaper price won't fix it.

The Road Ahead

What we expect to matter in each window, and the evidence that would prove us wrong.

Short term 0-6 months

No differentiated view
Driver
"A stock in a 34% drawdown, reporting in two days, with a regulatory deadline a month later. Constellation closed 2026-08-04 at $267.25, DOWN 2.36% or $6.46, having opened at the day high of $277 and traded down to $263.53 — a weak intraday pattern. That leaves it 33.8% BELOW its 52-week high of $403.95 and only 13.0% above its low of $236.50, 1.7% above a 50-day average of $262.68 and 12.9% BELOW a 200-day average of $306.75. RSI is 56.4 and MACD is +1.98, both mildly constructive, but the maximum drawdown from peak over the trailing year is 33.8% — the second-deepest in this batch after KKR's — and the twelve-month return is MINUS 21.6% against SPY's PLUS 24.3%, a 46-point deficit. Three-month is MINUS 16.8% against SPY's plus 7.6%. The 50-day has stabilised and the price sits just above it, which is the only constructive technical fact available. Two hard dates sit inside the window: second-quarter results on 2026-08-06, and the 2026-09-04 deadline for definitive agreements on the DOJ-mandated divestiture of five PJM generating assets, one ERCOT asset and the Gregory Power minority interest."
What we’re watching
"The 2026-08-06 print against consensus adjusted EPS of $2.40 and revenue of $7,789M. This will be the second full quarter with Calpine and the first clean year-on-year comparison of the combined company against a standalone base. Within it: whether Calpine synergies are quantified; whether purchase accounting is finalised, since the March-quarter allocation was provisional; the fate of the $5,735 million of assets held for sale and the LS Power agreement covering approximately 4.4 GW of PJM generation; and any guidance revision for FY2026 against the $11.734 consensus. Separately and with a harder deadline: the DOJ resolution requires definitive divestiture agreements by 2026-09-04. And watch the buyback — the company repurchased 2,000,000 shares for approximately $558.0 million on 2026-06-02 out of a secondary offering by former Calpine holders whose 50% lock-up expired on 2026-06-30, with the remaining 50% locked to 2027-06-30. A second tranche of selling by those holders is a known future supply event."
Confidence
Low

Medium term 6-24 months

Tailwind
Driver
"The medium term contains the clearest scheduled earnings step of any name in this batch. Consensus has adjusted EPS at $11.734 in FY2026, $13.335 in FY2027 and $17.085 in FY2028 — 13.6% then 28.1% — and the FY2028 acceleration has an identifiable, dated, auctioned source. On 2026-07-14 the company disclosed that every one of its PJM plants cleared the 2028-2029 capacity auction at a Capacity Performance price of $325 per megawatt-day, effective 2028-06-01, with 9,800 MW of nuclear clearing in COMED, 4,325 MW of nuclear plus 2,225 MW of fossil in EMAAC, 1,575 MW of nuclear plus 150 MW of fossil in MAAC, and further volumes in BGE and RTO. Capacity revenues for nuclear units count toward the gross-receipts calculation for the production tax credit. That is contracted revenue with a start date, not a projection. Alongside it: Calpine synergies, a retail platform serving approximately 62 TWh annually that Constellation did not previously own at that scale, and approximately $3.5 billion of remaining repurchase authority against a $95.97 billion market capitalisation. The counterweight is that the street cannot agree on the magnitude — the FY2028 estimate range is $14.532 to $23.119, a 59% spread — and that a merchant generator's earnings between capacity auctions are power prices."
What we’re watching
"Whether Calpine integrates. Purchase accounting was still provisional at the March quarter; approximately $12.6 billion of Calpine debt was assumed, of which $5.0 billion is project financing sitting at asset level; and the DOJ divestitures shrink the acquired fleet by roughly 4.4 GW in PJM alone. Whether the PJM capacity price holds in subsequent auctions — $325 per megawatt-day is at the high end of the range PJM has cleared, and capacity prices mean-revert. Whether the nuclear production tax credit survives legislatively; the knowledge base's own claim identifies it as a price floor near $43.75 per megawatt-hour, and a floor set by statute can be moved by statute. Whether additional data-centre power purchase agreements are signed and at what premium — the Three Mile Island restart contracted to Microsoft is the template and the 2027 reopening is the next milestone. Whether the former Calpine shareholders' remaining locked shares, released 2027-06-30, are sold into the market and whether the company absorbs them as it did in June. And whether the balance sheet, now carrying materially more debt than the fiscal-2025 statements show, is deleveraged."
Confidence
Medium

Long term 2+ years

Tailwind
Driver
"Long-run, Constellation owns something that cannot be built: the largest fleet of operating merchant nuclear reactors in the United States, now paired with roughly 23 GW of dispatchable gas. The structural argument is the one the knowledge base makes and it is sound. Artificial-intelligence data centres require firm, carbon-free, always-available power in quantities that intermittent renewables cannot supply alone and on timelines that new nuclear construction cannot meet. The existing fleet is therefore scarce in a way it has never been, and the evidence is contractual rather than theoretical: Three Mile Island is being restarted specifically to supply Microsoft, ahead of a 2027 reopening. Add an IRA production-tax-credit floor, capacity revenue auctioned three years forward at $325 per megawatt-day, and a retail platform that lets the company sell the output directly, and the shape of a durable franchise is visible. Against that stand the permanent facts of the sector: power is a commodity, capacity prices are set by an auction that mean-reverts, the tax credit is a statute, nuclear plants have long-tail operational and decommissioning risk, and the political economy of large industrial users paying premium prices while retail bills rise is not settled. Joseph Dominguez is chief executive; approximately 15,339 employees before Calpine."
What we’re watching
"Whether the AI power-demand thesis converts into signed long-term contracts at premium prices, or remains a narrative that lifts the multiple and then gives it back — which is what has happened over the last twelve months. Whether the merchant nuclear fleet's operating performance holds; capacity factors in the high nineties are the whole economics and a single extended outage is material. Whether the production tax credit survives a change of administration. Whether PJM capacity prices stay near $325 per megawatt-day or revert toward the double digits they cleared at earlier in the decade. Whether new nuclear — small modular reactors or otherwise — ever arrives at a cost that erodes the scarcity value of the existing fleet, which on any realistic timeline is a 2030s question. Whether the retail platform acquired with Calpine proves to be a durable channel or a low-margin commodity business. And whether the political reaction to data centres bidding power away from households produces regulation that caps what Constellation can charge them."
Confidence
Low

Exponential Potential

Exponential Potential6/10High

"Rated 6 — the highest exponential score in this batch, and it rests on a genuine structural repricing of an asset class rather than on a technology curve. The argument, which the knowledge base makes in all three of its surviving claims: artificial-intelligence data centres need enormous quantities of firm, carbon-free power on short notice; the only existing asset that supplies it at scale is the merchant nuclear fleet; and that fleet cannot be replicated, because nobody is permitting new large reactors on a five-year horizon. `business_breakdowns` (2024-06-01, speaker John Dullis) put it concretely — an IRA price floor near $43.75 per megawatt-hour, green premiums for data centres, tight short-power markets, and 'put that in your DCF, worth a lot more.' Constellation is the largest owner of that fleet, and the Three Mile Island restart contracted to Microsoft is the proof of concept both surviving voices cite. Calpine adds the complementary asset: roughly 23 GW of dispatchable gas that firms the intermittent supply data centres cannot use alone, plus a retail platform serving about 62 TWh. Against that: this is still a power company. Its output is a commodity; its capacity revenue is set by an auction; its production-tax-credit floor is legislated and therefore repealable; and the repricing the knowledge base identified has already happened once and given back a third. A 6: a real, scarce, non-replicable asset positioned in front of a genuine demand shock, wrapped in a business whose price is set by auctions and politics."

What could take this further than the base case — and where the market may be underpricing it. Full forward-growth and acceleration math in Deeper analysis, §4 below.

Deeper analysis

Technicals, fundamentals, valuation, and the full expert-claim evidence panel — the detail behind the numbers above.

Check our number Market-implied growth ≈ 32%/yr This isn’t how we calculate fair value — it’s public math you can verify yourself. To justify today’s $277, earnings would have to compound roughly 32% a year for 10 years (9% discount rate). Analysts forecast ~19%/yr, so the market is pricing in MORE than what the Street expects.

⚠ EARNINGS BANNER — RESULTS ARE RELEASED IN TWO DAYS

Constellation Energy reports second-quarter 2026 results on 2026-08-06, two days after this dive is struck. Vendor consensus is adjusted EPS $2.40 (against $1.91 in the year-ago quarter, +25.7%) and revenue $7,789M (against $6,101M, +27.7%) — the growth almost entirely the Calpine acquisition, which closed 2026-01-07.

This will be the second full quarter of the combined company, and it matters more than a normal print for three reasons: purchase accounting for Calpine was still provisional at the March quarter; $5,735 million of assets are held for sale under a DOJ-mandated divestiture programme; and the deadline for definitive divestiture agreements is 2026-09-04, thirty-one days away.

The company has MISSED consensus in two of its last five quarters (2025-05-06: $2.14 against $2.18; 2025-11-07: $3.04 against $3.11), and its beats have been small. This is not a name with a reliable positive-surprise record, and the Watch verdict below is set with that timing as an explicit input.

Reference table

Street consensus$355.20 (+32.9%) · median $364 (+36.2%) · high $441 (+65.0%) · low $296, 10.8% ABOVE spot · 0 strong buy / 14 buy / 6 hold / 0 sell across 20 analysts. We are 15.5% BELOW consensus and below its lowest published target
Valuation26.8x trailing adjusted EPS ($9.99) · 22.8x FY2026E $11.734 · 20.0x FY2027E $13.335 · 15.6x FY2028E $17.085 · 14.2x EV/EBITDA
The Calpine acquisition — closed 2026-01-07, absent from the payload50 million newly issued shares plus ~$4.5 billion cash; total consideration $17,503M per the cash-flow statement · ~$12.6 billion of Calpine debt assumed ($7.6B corporate, $5.0B project financing) · adds ~23 GW of gas, geothermal, battery and solar and a retail platform serving ~62 TWh annually · March-quarter revenue +63.9% to $11,122M
PJM capacity — auctioned, dated, contractedAll plants cleared the 2028-2029 auction at $325/MW-day, effective 2028-06-01: COMED nuclear 9,800 MW, EMAAC nuclear 4,325 MW + fossil 2,225 MW, MAAC nuclear 1,575 MW + fossil 150 MW, BGE fossil 375 MW. Nuclear capacity revenue counts toward the production-tax-credit gross-receipts calculation
The DOJ deadline — 31 days awayDefinitive agreements to divest five PJM generating assets, one ERCOT asset and the Gregory Power minority interest required by 2026-09-04. Gregory closed January 2026; ~4.4 GW of PJM assets under agreement with LS Power since March 2026; $5,735M held for sale
Capital returnDividend $1.6286, 0.61% yield · FY2025 buyback $400M; $558.0M repurchase on 2026-06-02 absorbing 2,000,000 shares from a secondary by former Calpine holders; ~$3.5 billion of authority remaining
ConvictionLow-positive, and the sweep was 90% noise — 29 raw KB hits, 3 used, 26 DISCARDED as Constellation SOFTWARE homographs. Concentration test failed at 67% on the surviving three
Technicals−33.8% from the 52-week high of $403.95; only +13.0% above the low of $236.50; +1.7% above the 50-DMA ($262.68) but −12.9% BELOW the 200-DMA ($306.75); RSI 56.4; MACD +1.98; max drawdown −33.8%; 12-month −21.6% vs SPY +24.3%

What the experts actually said 1 traceable claims on CEG · showing the highest-conviction voices

“Merchant nuclear plants have been revalued sharply higher: IRA price floor (~$43.75/MWh starting next year), green premiums for data centers, and tight short-power markets — 'put that in your DCF, worth a lot more'.”
Business Breakdownsbullishconviction 752024-06-01business_breakdowns-31hAkx6w2RU:6ff55fb3e3

Every claim reconciles to a real claim_id in the Synthos knowledge base — this is the evidence the verdict is built on, not vibes. Management (the company itself) is shown but half-weighted; one cautionary voice is included on purpose.

Price & moving averages 12 months · 50 & 200-day averages · 52-week range

223272320369417Aug '25Nov '25Jan '26Apr '26Jun '26Aug '2652w hi $404200-DMA 298Price 27750-DMA 26552w lo $236

Solid line = price · dashed line = 50-day average · dotted line = 200-day average · the two thin horizontal lines mark the 52-week high and low. Price above both averages is an uptrend.

Data summary: last close $276.75, 4% above the 50-day average ($265), 7% below the 200-day average ($298) — a mixed trend. 31% below the 52-week high of $404, 17% above the 52-week low of $236.

Bollinger Bands 20-day average ± 2 standard deviations

217269320372423Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26Price 27720-day avg 274

The shaded band widens when the stock gets more volatile. Riding the upper edge = strong momentum (sometimes stretched); the lower edge = weak / potentially oversold.

Data summary: price $276.75 is currently inside the band (band $263–$286).

RSI (14) momentum gauge · 0–100

705030Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26RSI 54.4

Above 70 (overbought zone, shaded) = overbought, below 30 (oversold zone, shaded) = oversold. Currently 54.

MACD 12 / 26 / 9 · trend & momentum

0Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26MACD 3.6signal 3.5

The MACD line crossing above the signal line (bars flip to the up color) = momentum turning up; crossing below (bars flip to the down color) = turning down. Bar height = the size of that gap.

Data summary: MACD is currently above its signal line by 0.14, positive momentum.

Relative performance vs S&P 500 & its sector (XLU (sector)), set to 100 a year ago

7085100115131Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26S&P 500 119XLU (sector) 101CEG 87

Solid = CEG · dashed = S&P 500 · dotted = XLU (sector). A rising line means it is beating that benchmark — the sector line shows whether it is a leader or laggard within its own group.

Forward revenue & earnings actual → estimate · "FY" = fiscal year, "E" = estimate

012233546$24BFY23EPS $6$24BFY24EPS $8$24BFY25EPS $9$33BFY26EEPS $12$34BFY27EEPS $13$36BFY28EEPS $17$40BFY29EEPS $20$41BFY30EEPS $21

Darker bars = actual results, brighter = analyst estimates. Taller bars to the right = expected growth.

Key stats an RIA wants

Price$276.75
Market cap$99B
P/E trailing28×
P/E FY26E / FY27E23× / 21×
EV / Sales3.9×
EV / EBITDA14.5×
Gross margin94.9%
Net margin11.1%
Dividend yield0.60%
Beta1.118
52-wk range$236 – $404
RSI(14)56
50 / 200-DMA$265 / $298
12-mo return+-12% (SPY +19%)
Street target$354 ($296–$441)
Analyst grades15 Buy · 6 Hold · 0 Sell
FMP ratingB+
Next earnings'2026-08-06 (Q2 2026 earnings, TWO DAYS AWAY; vendor consensus adjusted EPS $2.40 and revenue $7,789M, against $1.91 and $6,101M in the year-ago quarter — implying +25.7% EPS growth and +27.7% revenue growth, almost entirely Calpine). A second hard date follows: the DOJ resolution requires definitive divestiture agreements for five PJM generating assets by 2026-09-04, thirty-one days away.'

1. What the company is now — and it is not what the payload describes

Constellation Energy is a Pennsylvania-headquartered independent power producer, Nasdaq-listed, with Joseph Dominguez as chief executive and approximately 15,339 employees before Calpine. Before the acquisition it operated approximately 32,400 MW of generation across nuclear, wind, solar, natural gas and hydroelectric, organised by market region: Mid-Atlantic, Midwest, New York, ERCOT and Other.

Revenue by region, from seg_prod — and note this block does NOT reconcile:

FY2025 regionRevenueshare of block
Mid-Atlantic$6,487M29.3%
Midwest$5,804M26.2%
Other Regions$5,583M25.2%
New York$2,389M10.8%
ERCOT$1,904M8.6%
Sum$22,167M100%
inc_a FY2025 revenue$25,533M—
Sum as % of revenue86.8%—

The block accounts for only 86.8% of reported revenue — $3,366 million is missing — and seg_geo is byte-for-byte the same five regional lines, so the payload contains no product or geographic disclosure distinct from this one incomplete regional split. Detail in Section 5.

The Calpine acquisition — closed 7 January 2026, and it changes the company

From the 10-Q filed 2026-05-11:

> "On January 7, 2026 (the 'Acquisition Date'), we acquired all of the outstanding equity interests in Calpine in a cash and stock transaction... The merger consideration consisted of 50 million newly issued shares of our common stock, no par value, and approximately $4.5 billion in cash. In connection with the merger, certain of the newly issued shares will be subject to a lock-up period, which expires on June 30, 2026, for 50% of the shares and on June 30, 2027, for the remaining 50%."

> "Calpine operates a competitive retail electric supplier platform serving approximately 62 TWhs of load annually. Calpine also owns and operates a generation fleet of natural gas, oil, geothermal, battery storage, and solar assets with approximately 23 GWs of generation capacity, after considering divestitures required by certain regulatory approvals..."

And the debt assumed:

> "Upon completion of the acquisition of Calpine in January 2026, we assumed approximately $12.6 billion of debt inclusive of approximately $7.6 billion of corporate long-term debt... in addition to approximately $5.0 billion of various project financing arrangements. Pursuant to the Exchange Offers... we issued new notes in January 2026 effectively replacing $2.3 billion of Calpine's senior unsecured and secured notes with Constellation senior unsecured notes."

The cash-flow statement puts total consideration at $17,503 million ("Acquisition of Calpine $17,503") with $2,537 million of net cash outflow after cash and restricted cash acquired, and $17,507 million of common stock issued for the acquisition.

The strategic logic, in the company's own words: "The merger couples the largest producer of clean, emissions-free energy with the reliable, dispatchable natural gas assets of Calpine, and also creates the nation's leading competitive retail electric supplier." That is the pairing that matters for the data-centre thesis: nuclear supplies the firm carbon-free baseload, gas firms the rest, and the retail platform sells it.

The DOJ condition, and it has a deadline thirty-one days from this dive:

> "The final regulatory clearance for the merger was the DOJ resolution, which requires the divestiture of five generating assets located in PJM, one in ERCOT, and Calpine's minority interest in the Gregory Power Plant, also in ERCOT. The DOJ resolution requires us to enter into definitive agreement(s) to divest these assets within 240 days of closing the acquisition, by September 4, 2026."

Progress to date: the Gregory Power minority interest was divested in January 2026. In March 2026 the company agreed to sell five PJM generation assets, approximately 4.4 GW, to LS Power Equity Advisors. $5,735 million sits in assets held for sale at 2026-03-31 against $126 million a year earlier.

None of this appears in the vendor payload. bal_a ends 2025-12-31 — six days before the acquisition closed — and cf_a ends FY2025. The vendor's fiscal-2025 balance sheet shows total debt of $8,992 million and net debt of $5,244 million for a company that assumed $12.6 billion of additional debt six days later. Anyone screening this name on leverage is looking at a balance sheet that no longer exists.

The PJM capacity auction — a dated, auctioned earnings step

From the 8-K filed 2026-07-14:

> "On July 14, 2026, Constellation Energy Corporation (the Company) learned the results of the PJM capacity auction for the 2028-2029 planning year. Each of the Company's power plants located in the PJM market cleared in the auction. Capacity revenues for nuclear units are included in the gross receipts calculation for the Production Tax Credit. The auction results take effect June 1, 2028."

ZoneNuclear (MW)Fossil/Other (MW)Capacity Performance price
COMED9,800400$325/MW-day
EMAAC4,3252,225$325/MW-day
MAAC1,575150$325/MW-day
BGE—375$325/MW-day
RTO——$325/MW-day

Cleared capacity for assets held for sale is excluded from these figures, so this is the retained fleet.

Read the arithmetic on the largest line alone. 9,800 MW of COMED nuclear at $325 per megawatt-day for 365 days is approximately $1.16 billion of annual capacity revenue from one zone, before any energy sales. That is why the FY2028 consensus steps up 28.1% and it is the single most concrete forward-looking fact in this file. It is also three years out, and this dive's fair values are anchored on FY2027 — before it starts.

2. Earnings — a small-beat record and a very large acquisition

Report dateQuarterRevenue (inc_q)Adjusted EPSestimatevariance
2025-05-06Q1 2025$6,788M$2.14$2.18−1.8% MISS
2025-08-07Q2 2025$6,101M$1.91$1.84+3.8%
2025-11-07Q3 2025$7,184M$3.04$3.11−2.3% MISS
2026-02-24Q4 2025$5,460M$2.30$2.28+0.9%
2026-05-11Q1 2026$11,122M (+63.9%)$2.74$2.54+7.9%
2026-08-06Q2 2026—?$2.40—

Trailing four quarters of adjusted EPS: $1.91 + $3.04 + $2.30 + $2.74 = $9.99. At $267.25 that is 26.8x trailing.

Two misses in five quarters, and the beats other than the most recent are under 4%. This is not a company with a reliable positive-surprise record, and the one large beat came in the first quarter that contained Calpine — when the consensus revenue estimate of $8,458M was 31.5% below the actual $11,122M, because the street had not fully modelled the acquisition. We do not treat that beat as evidence of operating momentum.

The annual record is distorted by fair-value accounting on the generation hedge book and should be read with care:

Fiscal yearRevenueOperating incomeNet incomeGAAP EPSDiluted shares
FY2021$19,649M−$346M−$205M−$0.63326.7M
FY2022$24,440M$495M−$160M−$0.49328.0M
FY2023$24,918M$1,610M$1,623M$5.02324.0M
FY2024$23,568M$4,352M$3,749M$11.90315.0M
FY2025$25,533M$3,086M$2,319M$7.40314.0M

GAAP earnings swung from a $0.49 loss in FY2022 to $11.90 in FY2024 and back to $7.40 in FY2025 on revenue that moved 4%. That is mark-to-market on hedges and nuclear decommissioning trust returns, not operations. Both the company and the street run on adjusted operating earnings, and so does this dive.

3. Valuation — priced in or room?

At $267.25 (market capitalisation $95.97B, implied 359.1M shares):

TrailingFY2026EFY2027EFY2028E
Adjusted EPS$9.99$11.734 (11)$13.335 (13)$17.085 (11)
Estimate range—$11.305–$11.974 (5.9%)$11.866–$14.314 (20.6%)$14.532–$23.119 (59.1%)
Growth——+13.6%+28.1%
P/E26.8x22.8x20.0x15.6x
Revenue$25,533M (FY2025)$33,097M (10)$34,576M (10)$36,309M (13)
EV / EBITDA14.2x———
Net debt / EBITDA2.58x (stale — pre-Calpine)———
Price / book2.83x———
Return on equity20.1%———
Dividend yield0.61%———

Estimate coverage is adequate and the dispersion widens sharply with time. Eleven analysts on FY2026, thirteen on FY2027, eleven on FY2028. The FY2027 range of $11.866 to $14.314 is 20.6% wide; the FY2028 range of $14.532 to $23.119 is 59.1% wide. The street agrees there is a step in 2028 and disagrees by nearly a factor of two about its size — which is a fair reflection of an auctioned capacity price layered on top of merchant energy revenue. FY2029 and FY2030 rest on 4 analysts each and are excluded from every conclusion.

est.ebitdaAvg and est.ebitAvg fail two tests simultaneously and are rejected outright. ebitAvg EXCEEDS ebitdaAvg in all eight years in the file — FY2026 shows EBIT of $10,202M against EBITDA of $9,373M, FY2027 $10,658M against $9,792M, and so on. That is arithmetically impossible: earnings before interest and tax cannot exceed earnings before interest, tax, depreciation and amortisation. It is the defect measured at 8.9% of files across the programme, present here in every year rather than one. Both rows also carry the fixed-ratio signature: ebitdaAvg is exactly 28.319% of revenueAvg and ebitAvg exactly 30.825% in every forward year. All forward valuation uses epsAvg.

Enterprise value is directionally usable but stale. The vendor prints $117,265M, implying net debt of approximately $21.3 billion — which is consistent with the post-Calpine balance sheet rather than the fiscal-2025 one in bal_a (net debt $5,244M). So the EV field appears to be computed on a more recent balance sheet than the annual statements in the same payload, which is helpful but means netDebtToEBITDATTM of 2.58x and debtToEquityRatioTTM of 0.671 — both computed on the stale annual figures — understate leverage materially. Enterprise value also omits $336M of noncontrolling interests, a 0.3% understatement — the mildest in this batch.

Peer context is good. The vendor set contains AEP, Clearway, Duke, GE Vernova, NextEra, National Grid, Southern, Sempra and Vistra. Vistra ($48.3B) is the single most apposite comparable — the other large merchant generator with nuclear exposure — and GE Vernova ($271.2B) is the equipment counterpart to the same demand thesis. The regulated utilities (AEP, Duke, Southern, Sempra, National Grid) are a different business model entirely and are not comparable on multiple. No cross-multiple is drawn, because this file contains no forward-EPS data for the peers.

3a. What today's price assumes (the inversion)

At $267.25 — 22.8x FY2026 consensus, 20.0x FY2027, 15.6x FY2028 — the price embeds:

3b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: adjusted EPS growth (+13.6%, from FY2026E $11.734 to FY2027E $13.335) + multiple drift (roughly HELD, 22.8x on the forward year moving to about 22.5x on the then-forward year) + shareholder yield (+0.6% dividend, plus a buyback running near 1% at the June pace) ≈ +13% to +15%.

Our base of $300 applies 22.5x to the FY2027 number, which is essentially the multiple the market currently pays on FY2026 — so the base assumes the multiple HOLDS rather than expands or compresses. That is a deliberate refusal to underwrite either side of a stock that has already de-rated by roughly 40% in twelve months. The bull argument is that a scarce, non-replicable asset in front of a structural demand shock deserves more than 20x; the bear argument is that a merchant generator with a 59%-wide FY2028 estimate range deserves less. We are declining to adjudicate that and are taking the earnings growth instead.

Note the composition: essentially all of the return is earnings growth, with almost no yield. A 0.61% dividend and a buyback of roughly $1 billion a year against a $96 billion market capitalisation gives a total shareholder yield near 1.5% — there is very little to collect while waiting.

If the multiple compressed to 17x FY2027E the price is $227 (−15.1%). If it expanded to 24x, $320 (+19.7%).

3c. Variant perception (where we differ, what would surprise)

Synthos fair values

All three anchors are multiples of the FY2027 consensus adjusted EPS distribution (mean $13.335, low $11.866, high $14.314, 13 analysts).

Base is 12.3% above spot; asymmetry roughly 1.59:1 (21.4% down, 34.0% up), plus a shareholder yield near 1.5%. That is genuinely interesting and it is short of the Tactical bar, which in this batch has required roughly 12%+ WITH a payoff ratio of 2:1 or better. Combined with results in two days and a regulatory deadline in five weeks, the correct tier is Watch, and the trigger is named in Section 6.

4. Knowledge base — twenty-nine hits, twenty-six of them about a Canadian software company

Raw hits: 29. Entity matches: 3. Free-text hits: 26. Used: 3. Discarded: 26. Distinct channels in the used lane: 2.

The sweep ran case-sensitive entity tokens CEG, Constellation Energy and the bare word Constellation, plus case-sensitive free-text patterns for the same, across all 51,928 distilled claims.

THE DISCARD IS THE FINDING. All twenty-six free-text hits concern Constellation Software (CSU.TO), the Canadian vertical-market-software serial acquirer, or are generic references to it. They arrive overwhelmingly from business_breakdowns and we_study_billionaires and cover: Mark Leonard taking no salary and travelling on his own dime; strict 20-30% acquisition hurdle rates described as "magnetic"; decentralised capital-allocation authority enabling "100+ acquisitions a year"; approximately 1,200 disciplined deals; a claim that Constellation "trades ~28x EV/EBITDA yet acquires at ~5x"; the Topicus and Lumine spin-offs; and a 2026 software selloff described as "momentum-driven and forced-selling."

Not one of them concerns a power producer.

This is the most dangerous homograph this programme has encountered, and the reason is that the wrong company is a good one. The PH/petahash, Vertex/cloud-platform and USB/hardware-standard collisions catalogued elsewhere are obviously noise on inspection. Here the noise reads as substantive investment analysis, from credible channels, about a genuinely high-quality business — and a careless pipeline would have credited Constellation Energy with a 20%-ROIC serial-acquirer moat, mid-90s revenue retention and Mark Leonard's compensation discipline. The entity check is what prevented it, and the entire free-text lane is discarded.

The three surviving entity claims, verbatim:

> 2024-06-01 · bullish · conviction 75 · horizon: thesis · entities: VST, CEG · channel: business_breakdowns · speaker: John Dullis · speaker_role: independent

> "Merchant nuclear plants have been revalued sharply higher: IRA price floor (~$43.75/MWh starting next year), green premiums for data centers, and tight short-power markets — 'put that in your DCF, worth a lot more'."

> 2024-09-29 · bullish · conviction 70 · horizon: thesis · entities: Constellation Energy · channel: raoul_pal · speaker: null

> "Deep symbiotic AI-energy relationship; nuclear restarts (Three Mile Island supplying Microsoft) prove nuclear powers intense compute demand."

> 2025-09-12 · bullish · conviction 65 · horizon: fact · entities: Constellation Energy, Three Mile Island, Microsoft · channel: business_breakdowns · speaker: null

> "AI power demand is reviving nuclear — Microsoft contracted 3 Mile Island output ahead of its 2027 reopening to feed data centers."

Concentration test — RUN and FAILED. business_breakdowns supplies two of three used claims — 67%, twice the one-third threshold. Removing that channel leaves a single claim, from raoul_pal. The lane does not survive the removal of its dominant voice, and conviction is scored low-positive accordingly, with breadth 2.

Attribution note. One claim carries a named speaker (John Dullis on business_breakdowns, speaker_role: independent); two carry speaker: null and are channel-attributed. No management voice appears. On the raoul_pal claim we apply the standing house treatment for that channel: the framework is credited, the claim is taken at moderate weight, and no additional weight is given for the channel's own market positioning.

GRADING THE LANE, which is the substantive part. The thesis was right about the asset and has been wrong about the last twelve months. These claims date from June 2024, September 2024 and September 2025. Merchant nuclear was indeed revalued sharply higher over that period. But the stock is DOWN 21.6% over the trailing twelve months against SPY's +24.3%, sits 33.8% below its 52-week high, and fell another 2.36% today. A structural thesis that is correct and a position that has lost a third are compatible, and reporting only the first would be dishonest. We report the lane as thesis-correct, timing-poor, and currently unresolved.

Conclusion. Breadth 2, claim count 3, net conviction positive-low. Twenty-six of twenty-nine hits were the wrong Constellation. Three survived, all bullish on nuclear and AI, two from one channel, and the stock has fallen a third since the most recent of them.

5. Data integrity — what we rejected and why

Nine findings. This is the second-most-damaged vendor file in this batch after KKR's, and two of the errors are arithmetically impossible on their face.

1. est.ebitAvg EXCEEDS est.ebitdaAvg in ALL EIGHT YEARS — REJECTED. FY2026: EBIT $10,202,049,527 against EBITDA $9,372,851,711. FY2027: $10,658,030,466 against $9,791,771,627. FY2028: $11,191,988,057 against $10,282,330,441. FY2029, FY2030 and the three historical rows show the same inversion. Earnings before interest and tax cannot exceed earnings before interest, tax, depreciation and amortisation for a company with $2.6 billion of annual depreciation. Both rows additionally carry the fixed-ratio fabrication signature: ebitdaAvg is exactly 28.319% of revenueAvg and ebitAvg exactly 30.825% in every forward year. The measured base rate for the impossible-ordering defect across 3,249 files is 8.9%; this file exhibits it in 8 of 8 years. All forward valuation uses epsAvg.

2. inc_q reports December-quarter gross profit of $15,700M on revenue of $5,460M — 288% of revenue. The quarter's operating income is $145M and net income $432M, so the gross-profit line is not merely mislabelled but internally inconsistent with the rest of the same row. grossProfit for this name is unusable in inc_q and is not used anywhere.

3. inc_a changes its cost basis between FY2024 and FY2025, producing a 75.8% gross margin for a power generator. FY2024 reports costOfRevenue of $17,578M and grossProfit of $5,990M — a 25.4% gross margin. FY2025 reports costOfRevenue of $6,159M and grossProfit of $19,361M — a 75.8% gross margin — on revenue that rose only 8.3%. A three-fold change in gross margin without a corporate event is a change of presentation, not of economics. ratios_ttm.grossProfitMarginTTM of 77.9% inherits it. This is the MO/APP two-basis defect appearing in the ANNUAL rather than the quarterly series, and every gross-margin figure in the file is rejected.

4. inc_a FY2025 depreciation of $985M contradicts cf_a FY2025 depreciation of $2,601M — a factor of 2.6. The two blocks agree in FY2024 (both $2,700M) and disagree in FY2025. The cash-flow figure is the credible one for a company with $57 billion of assets; the income-statement figure is not, and inc_a FY2025 ebitda of $5,007M is built on it. All EBITDA figures derived from inc_a FY2025 are rejected.

5. cf_a reports NEGATIVE operating cash flow in three of the last four fiscal years — treated as unreliable. FY2022 −$2,353M, FY2023 −$5,301M, FY2024 −$2,464M, then FY2025 +$4,237M. A generator with $23-25 billion of annual revenue and $2.5 billion of annual depreciation does not consume $5.3 billion of operating cash in a year in which it earned $1.6 billion. The most likely explanation is collateral and margin movements on the hedge book being classified as operating rather than being netted, but the payload does not disclose the composition. The consequence is that freeCashFlow for FY2022 through FY2024 (−$4,042M, −$7,723M, −$5,029M) is not a usable series, and incomeQualityTTM of 1.199 rests on one good year. We use FY2025's $4,237M of operating cash flow and $1,288M of free cash flow and state that the prior three years are not trusted.

6. seg_prod accounts for only 86.8% of revenue, and seg_geo is a byte-for-byte DUPLICATE of it. The FY2025 regional block sums to $22,167M against $25,533M of revenue — $3,366 million, 13.2%, unaccounted for. And the seg_geo block contains exactly the same five lines with exactly the same values. So the payload contains ONE incomplete regional split presented twice under two different keys, and no product-level or true geographic disclosure at all. The FY2022 and FY2021 entries additionally contain a "Constellation Natural Gas" line ($4,967M and $3,379M) that disappears thereafter, and the FY2020 and FY2019 entries use an entirely different two-line structure ("Energy Commodities and Service", "Public Utilities") summing to $33.0 billion and $34.6 billion — more than the company's revenue in those years, because they are pre-spin Exelon figures. No segment figure in this dive comes from the payload.

7. bal_a predates the Calpine acquisition by six days and therefore understates leverage by roughly $12.6 billion. The most recent balance sheet is dated 2025-12-31; Calpine closed 2026-01-07. totalDebt of $8,992M, netDebt of $5,244M, netDebtToEBITDATTM of 2.58x, debtToEquityRatioTTM of 0.671, debtToAssetsRatioTTM of 0.232 and financialLeverageRatioTTM of 2.894 are all struck on a balance sheet that no longer describes this company. Note that enterpriseValueTTM of $117,265M implies net debt near $21.3 billion and therefore appears to use a MORE RECENT balance sheet than the annual block in the same payload — an internal inconsistency that happens to favour accuracy in the EV field and not in the ratios.

8. priceToEarningsRatioTTM and netIncomePerShareTTM are mutually inconsistent. priceToEarningsRatioTTM of 23.2189 implies trailing EPS of $11.510 at $267.25; netIncomePerShareTTM reads $10.746. The four quarterly inc_q net income figures sum to $3,804M, which on the ~354M share base implied by revenuePerShareTTM gives $10.75 — so the per-share figure is right and the ratio is computed on something else. Neither is used: this dive runs on the adjusted-EPS basis ($9.99 trailing) that the company and the street use.

9. Enterprise value omits $336M of noncontrolling interests — a 0.3% understatement. The mildest instance of this defect class in the batch (KKR 35.6%, FCX 11.5%, SPGI 3.9%, DUK 1.1%). Recorded and immaterial.

Also checked and recorded. quote.yearHigh/yearLow ($412.70 / $228.63) against tech.hi52/lo52 ($403.95 / $236.50) — a 2.1% and 3.4% discrepancy, the widest low-end gap in this batch; tech is used throughout, which is the more conservative choice since it places the stock nearer its low. earn_cal.revenueActual disagrees with inc_q revenue for two of the four completed quarters — the calendar reports $6,570M for Q3 2025 against inc_q's $7,184M, and $6,074M for Q4 2025 against inc_q's $5,460M — while agreeing exactly on Q1 2026 ($11,122M). Two revenue bases coexist in the file; only the Q1 2026 figure is corroborated and only it is quoted. The EPS column of earn_cal is internally consistent on the adjusted basis and is used. dividendYieldTTM of 0.609% is consistent with dividendPerShareTTM of $1.6286 at $267.25. effectiveTaxRateTTM of 31.0% is consistent with FY2025's incomeTaxExpense of $1,187M on incomeBeforeTax of $3,511M (33.8%).

returnOnEquityTTM — source stated. Present in km_ttm at 20.07% and absent from ratios_ttm. We report it and note it is struck on a pre-Calpine equity base of $14,517M, so the post-acquisition figure on a materially larger equity base will be lower.

Vendor composite rating — noted, low weight. B+ / 3 overall, with 4 out of 5 on discounted cash flow, return on equity, return on assets and debt-to-equity, and 1 out of 5 on price-to-book. The debt-to-equity score of 4 is struck on the pre-Calpine balance sheet described in finding 7 and is therefore flattering by roughly $12.6 billion of assumed debt. The discounted-cash-flow score rests on a free-cash-flow series that is negative in three of four years per finding 5. Neither carries weight.

Non-equity tripwire — checked and passed. CEG is common stock, no par value, Nasdaq-listed. Price of $267.25 is not par-like; beta is 1.121; the dividend is variable and small ($1.6286, 0.61%); volume was 3.70M shares (~$989M of turnover); the 52-week band of $236.50 to $403.95 is a 71% range. This is common equity.

6. Verdict, kill-criteria and flip conditions

Watch.

The case for interest. Constellation owns the largest fleet of merchant nuclear generation in the United States, an asset that cannot be replicated on any relevant timescale, in front of a demand shock our own knowledge base identified in June 2024 and that has since produced a contracted restart of Three Mile Island for Microsoft. On 2026-01-07 it added Calpine — roughly 23 GW of dispatchable gas and a retail platform serving about 62 TWh — for 50 million shares and $4.5 billion. On 2026-07-14 every one of its PJM plants cleared the 2028-2029 capacity auction at $325 per megawatt-day effective 2028-06-01, with 9,800 MW of COMED nuclear alone worth roughly $1.16 billion of annual capacity revenue. Consensus expects 13.6% earnings growth in FY2027 and 28.1% in FY2028. And the stock is 33.8% below its high.

The case for waiting, which is the operative one. Our base fair value of $300 is 12.3% above spot with 1.59:1 asymmetry — short of the roughly 2:1 that has earned the Tactical tier elsewhere in this batch. Results are released in two days, and this company has missed consensus in two of its last five quarters with beats otherwise under 4%. The DOJ requires definitive divestiture agreements for five PJM assets by 2026-09-04, thirty-one days away, with $5,735 million held for sale — a seller against a statutory deadline is a price-taker. Purchase accounting for a $17.5 billion acquisition was still provisional at the March quarter. Leverage is roughly $12.6 billion higher than the payload's most recent balance sheet shows. The knowledge-base lane collapses to one claim if its dominant channel is removed, and the surviving thesis, though correct about the asset, has coincided with a 21.6% twelve-month loss. And the vendor data is the second-most-damaged in this batch: an impossible EBIT-above-EBITDA relationship in all eight estimate years, a quarterly gross profit 288% of revenue, an annual gross margin that triples without a corporate event, and negative operating cash flow in three of four years.

Watch rather than Hold, because there is a genuine catalyst path and a genuine price at which the arithmetic works. Watch rather than Buy, because 12.3% with 1.59:1 asymmetry two days before a print and a month before a forced-sale deadline is not a position.

The named trigger: $240. At $240 the same FY2027 consensus of $13.335 is 18.0x, the base case becomes +25%, asymmetry improves past 2.5:1, and the entry sits just above the 52-week low of $236.50. That is where this becomes Buy — Tactical on arithmetic alone, absent an adverse development below.

Pre-registered conditions that would move CEG to Buy — Tactical:

Pre-registered conditions that would move CEG to Avoid:

Where CEG fits in the Synthos Framework Portfolio. The utilities/power sleeve — on the watchlist at zero weight today, with a $240 limit and a review scheduled for the 2026-08-06 print and again after the 2026-09-04 divestiture deadline. Sizing note: this is the highest-exponential name in this batch and also one of the two most volatile, and the correct response to a 34% drawdown that has not demonstrably ended is a named entry price rather than a partial position. Logged as a tracked Synthos call (Watch) as of 2026-08-04 at $267.25, with the fair-value anchors, the $240 trigger and both condition sets gradeable.

Single biggest risk: merchant power prices between capacity auctions. The $325 per megawatt-day cleared for 2028-2029 and the production-tax-credit floor near $43.75 per megawatt-hour are floors, not the whole revenue line. Between them, Constellation sells energy into competitive markets at prices set by gas, weather and demand. The street's FY2028 estimate range of $14.532 to $23.119 — a 59% spread — is the honest expression of how wide that outcome is, and it sits on top of an integration seven months old, a balance sheet carrying $12.6 billion of newly assumed debt, and a forced divestiture of roughly 4.4 GW against a 4 September deadline. The asset is genuinely scarce and the demand thesis is genuinely good. Neither of those facts sets the price of a megawatt-hour in 2027, and the kill conditions above are written accordingly.

Provenance & disclosures