Constellation’s November report tests the cost of contracted power

Constellation has buyers for output it has yet to deliver

Constellation Energy goes into the report scheduled for 2026-11-09 with signed power contracts but no disclosed spending schedule for delivering the latest deals. The date comes from the DeadRisk market-data earnings calendar captured in August 2026, not from a filing; scheduled dates can move. For this preview, the Power Wall is the constraint behind the deals. The question for the report is how much spending Constellation must carry before contracted output becomes delivered power.

The sequence starts with Microsoft. Bloomberg reported in September 2024 that Constellation planned to spend $1.6 billion restarting Three Mile Island and sell the reactor’s output under a 20-year agreement. Meta later signed a 20-year Illinois power deal, but that contract starts in mid-2027, when an existing state subsidy expires. Neither announcement makes the future delivery date disappear.

The two latest agreements do not buy the same thing. Amazon agreed to purchase 690 megawatts for 20 years at Calvert Cliffs. Bloomberg says the deal supports more than $3 billion in Maryland investment. Google contracted for 3,590 megawatts of power in the PJM grid; Reuters says about one-quarter is new nuclear output. That is roughly 900 megawatts of new nuclear supply by calculation, compared with 3,590 megawatts of contracted supply—not 3,590 megawatts of new reactor capacity. These agreements put Constellation inside the AI capex supply chain. They do not establish when each project will earn cash.

A project plan is not an annual cash baseline

Bloomberg’s Maryland figure is a project plan, not one year of spending. Its more-than-$3-billion Maryland investment figure is more than $1.4 billion above the $1.6 billion Constellation planned for the Three Mile Island restart, using Bloomberg’s Maryland report and its September 2024 restart report. That difference is derived from two announced project amounts. The Maryland figure covers investment supported by the Amazon deal; the earlier figure is Constellation’s planned restart cost. Neither is a filed annual cash-flow line.

The DeadRisk market layer labels an FY2025 field “capex” but does not define its cash-flow components. It cannot serve as a verified annual capital-expenditure baseline here. Bloomberg’s Maryland report also gives no annual schedule for that investment. The street cannot stack a multiyear announcement against one annual figure and call the difference cash due.

The Microsoft restart offers a separate financing check. Bloomberg put the planned Three Mile Island investment at $1.6 billion in September 2024. In November 2025, the Wall Street Journal reported that the Department of Energy would provide a $1 billion loan for the restart. The reported loan equals 62.5% of the earlier announced investment, by calculation. Those reports do not establish that the loan was drawn, that project costs stayed fixed, or how any remainder was funded. They do show why a contract count is not a financing schedule.

TEXXR’s quiet quarter concealed a naming problem, not a business verdict

TEXXR coverage of Constellation Energy ran at 0 articles in 2026Q3, 0% of the trailing eight-quarter average of 1 article per quarter. Those are articles in this coverage record, not a count of contracts, reactor work or power demand. The October Amazon and Google reports arrived after that measured quarter.

TEXXR also shows why an edge count needs inspection. Its Constellation Energy company-name graph contains six edges tied to the same Microsoft announcement as it appeared across three September 2024 dates. One announcement did not become six contracts. The graph records the later federal financing under the shorter name “Constellation,” so a full-name search misses the loan reported by the Journal. A reader counting only full-name edges would double-count the original story while overlooking a financing event. That is a property of TEXXR’s extraction, not evidence that Constellation changed its plans.

New coverage must connect the contracts to delivery

The scheduled 2026-11-09 report can narrow the gap between what Constellation has signed and what it must build or restart. The next checks are dates, spending definitions and delivered power:

• Maryland spending: Does Constellation assign dates, a responsible payer and a cash-flow definition to the Amazon-linked investment plan? Without them, the project figure cannot become an annual Constellation spending forecast.
• Google’s supply mix: Does subsequent coverage distinguish new nuclear output from the rest of Google’s 3,590-megawatt contract?
• Contract start dates: Does reporting put delivered output against Meta’s mid-2027 start and the Microsoft restart, rather than counting signatures as current production?

Constellation has supplied the street with contracts. The report’s harder job is to show the cost and timing of supplying the power.

Sources

The record:

• Constellation Energy plans to invest $1.6B to revive the Three Mile Island nuclear plant in Pennsylvania and sell all the output to Microsoft data centers — Bloomberg, 2024-09-20 — original · TEXXR record
• Meta agrees to buy power from Constellation Energy's Illinois nuclear plant, signing a 20-year deal that starts in mid-2027 when a 10-year state subsidy expires — Bloomberg, 2025-06-03 — original · TEXXR record
• The US DOE will provide Constellation with a $1B loan to restart the Three Mile Island nuclear plant; the output will be sold to Microsoft under a 20-year deal — Wall Street Journal, Wed, 19 Nov 2025 00:00:00 GMT — original · TEXXR record
• Amazon signs a 20-year deal to buy 690 MW from Constellation Energy, helping the nuclear operator boost its capacity in Maryland, supporting $3B+ in investments — Bloomberg, 2026-10-01 — original · TEXXR record
• Google contracts for 3,590 MW of power from Constellation Energy in the US' largest power grid, with new nuclear energy accounting for ~25% of that supply — Reuters, 2026-10-07 — original · TEXXR record