⚛️

CEG

💡
Thesis. Owns the single most irreplaceable asset in AI power — existing nuclear with interconnection, which cannot be built quickly at any price — and is converting it into 18.5-year contracts with investment-grade counterparties.
Reviewed 19 Sep 2026, after Q2 2026 results on 6 Aug. Next review after Q3 2026 earnings on 3 Nov 2026.

What it does

Constellation is the largest owner of nuclear generation in the United States, with a fleet of roughly 60 GW after the $16.4B Calpine acquisition added dispatchable gas. It sells power wholesale and under long-term contracts. CEO Joe Dominguez.
A PPA — power purchase agreement — is a contract to sell a fixed quantity of power over a fixed term at a fixed price. It converts merchant exposure, where the price moves with the market every hour, into a bond-like revenue stream. The counterparty's credit is therefore as important as the megawatts.
Co-location is the arrangement at issue in front of PJM: a datacentre built next to a power plant, drawing power directly rather than through the grid. Whether and how that is allowed determines what an existing nuclear plant's output is worth to a hyperscaler.
An existing nuclear plant with an interconnection is the one thing in this file that money cannot conjure. Everything else in the power stack can be built faster with more capital.

Bull case

  • Q2 2026 adjusted operating EPS $2.55, up $0.64 YoY. FY26 guidance raised to $11.50–12.50 — midpoint up $0.50 with half the year still to run
  • ~920 MW of new long-term nuclear PPAs signed, averaging 18.5 years, every counterparty investment-grade. That brings ~30% of expected clean baseload under contract by 2032
  • Includes Walmart's first-ever nuclear PPA (176 MW), enabling a 30 MW uprate at Dresden
  • Nuclear fleet ran at a 93% capacity factor, generating 40 TWh through six planned refuelling outages averaging 23 days
  • Crane Clean Energy Center restart on track for H2 2027 — the NRC approved the fuel licence amendment and FERC granted the waiver transferring capacity injection rights from the retiring Eddystone units
  • Filed to extend the Ginna and Nine Mile Point 1 licences to 2049
  • Selling Brazos Valley to LS Power for $860M (~$1,420/kW), satisfying the last DOJ condition from Calpine
  • $2.2B of buybacks year to date, $2.8B remaining. Management outlined a path to 20% earnings growth through 2029

Bear case

  • Revenue of $7.5B missed the $7.94B estimate. The guidance raise came on margin and contracting, not on top line
  • Co-location rules remain unresolved. PJM files its response around November 2026 and clarity is not expected until Q1–Q2 2027. This directly determines how nuclear plants can sell power to adjacent datacentres — the biggest single regulatory variable on this page
  • ~70% of expected clean baseload remains merchant and uncontracted. The 30%-by-2032 figure is the achievement; the other 70% is the exposure
  • The stock sits ~36% below its 52-week high of $412.70, reflecting valuation caution rather than operational disappointment
  • Nuclear outage timing swings quarterly output — six refuelling outages in a single quarter shows how much the calendar matters
  • A weaker ERCOT market was absorbed in the quarter
  • Regulatory and political exposure on licence renewals

Major customers

  • Microsoft — the anchor relationship, underwriting the Three Mile Island / Crane Clean Energy Center restart through a 20-year PPA
  • 🏛️
    META     — 20-year agreement for ~1.1 GW from the Clinton nuclear plant from 2027; Meta's largest energy agreement
  • Walmart — 176 MW, the first nuclear PPA signed by a major retailer
  • ~920 MW of newly contracted investment-grade customers, 15–20 year durations, beginning 2029–2032
  • Commercial and industrial retail customers through the Constellation retail business
🔗
Read-through. Constellation's customer list is a direct index of hyperscaler power procurement —
🏛️
META's capex decisions flow straight into this contracting pipeline, which makes the Meta page the place to sanity-check whether these PPAs get honoured at scale. The Walmart deal matters beyond its size: it signals nuclear PPAs spreading beyond tech into general corporate procurement, which broadens the demand base past a handful of buyers — the single most bullish structural datapoint here. It competes with
⚛️
VST and
⚛️
TLN for the same hyperscaler counterparties, and its new gas build queues behind
🔌
GEV turbines like everyone else's.

What would change the view

  1. The PJM co-location ruling — filed ~November 2026, clarity Q1–Q2 2027. The biggest single regulatory catalyst in the Power Producers group
  2. Additional PPA signings, and the percentage of baseload under contract moving above 30%
  3. Crane restart staying on schedule for H2 2027
  4. Nuclear capacity factor and outage duration
  5. Buyback pace against the remaining $2.8B
  6. Whether non-tech corporate buyers follow Walmart

Update log

19 Sep 2026 — Converted to the standard template. Current through Q2; no new disclosures since. Promoted the uncontracted ~70% of baseload into the bear case, since the 30%-contracted figure only means something alongside it.
6 Aug 2026 — Q2 2026 reported. Adjusted operating EPS $2.55 (+$0.64); FY26 guidance raised to $11.50–12.50. Revenue $7.5B missed $7.94B. ~920 MW of new nuclear PPAs at 18.5-year average duration, all investment-grade, including Walmart's 176 MW. Nuclear capacity factor 93%, 40 TWh generated. Crane restart on track for H2 2027. Brazos Valley sale to LS Power for $860M. $2.2B of buybacks YTD.
Probabilities: strengthen ~60% / weaken ~25%.
Research and education only — not investment advice.