Amazon’s 7.65 GW Shortcut Carries Permit Risk • The Weighted Average

Amazon’s planned Pecos County campus is tied to 7.65 gigawatts of gross gas nameplate carrying a 30.13-million-metric-tonne annual CO2e permit ceiling. Divide those sourced inputs and the envelope is about 3.94 million metric tonnes per permitted nameplate GW-year—a ceiling, emphatically not a forecast—so buyers should contract for permit and delay risk before treating private power as delivered compute.

The shortcut moves the queue onto the campus

Cleanview’s investigation says Amazon confirmed its campus will buy power from GW Ranch, a private-grid project initially outside ERCOT. Pacifico’s permit announcement describes 7.65 GW of gross gas capacity and a guaranteed path to more than 5 GW. That structure promises a seductive shortcut around a congested grid queue. It does not remove the state, local, fuel, construction or emissions constraints that determine when turbines can run.

The permit arithmetic supplies a due-diligence benchmark. The project application reports 33,212,284.72 U.S. short tons of CO2e a year at maximum potential across 35 gas turbines. Multiply by the standard 0.90718474 metric conversion to get about 30,129,678 metric tonnes, then divide by the independently published 7.65 GW gross nameplate: 3.94 million metric tonnes per permitted nameplate GW-year. This two-source ratio describes the outer permit envelope normalized by proposed capacity. Plants rarely operate at their maximum permitted pollution level, and nameplate is not dispatch. It must never be presented as expected emissions, Amazon-attributable emissions or a prediction of utilization.

The broader record makes the ceiling consequential even with that caveat. A Texas Tribune and Floodlight permit investigation identified GW Ranch in Pecos County and reported that at least nine Texas data-center gas plants could emit more than 130 million U.S. tons annually at their permitted limits. The project’s individual ceiling therefore belongs to a portfolio-scale permitting question, not an isolated generator filing.

Official scrutiny is already visible. A Senate Environment and Public Works letter to Pacifico cites the project’s scale and permit concerns. Pacifico says the TCEQ permit authorizes construction and operation under air-quality requirements, but counterparties should reconcile the direct project application, approval and commercial milestones before assuming all proposed capacity shares one settled delivery date.

The operator decision extends the logic of the state’s 474-GW large-load audit. Grid-connected projects now face a verification gate; behind-the-meter generation exchanges that queue exposure for plant-level permitting, fuel and operating exposure. “Power secured” is not a binary field. It should be decomposed into permitted, financed, under construction, mechanically complete, commissioned and contractually deliverable.

Put the permit into the commercial terms

Infrastructure buyers should allocate four risks explicitly. First, who pays if a permit appeal, control requirement or construction condition delays commercial operation? Second, may the customer terminate, reduce its take-or-pay commitment or reset price? Third, who bears carbon costs and reporting duties if actual dispatch rises? Fourth, what substitute power can the provider deliver without weakening reliability or emissions commitments?

Those clauses matter because the public evidence does not disclose actual hourly dispatch, final Amazon ownership or offtake terms, capacity factor, fuel price, carbon-control performance or the share operating as backup. Each missing input can move the outcome. Lower utilization would make annual emissions far below the ceiling. More efficient generation, renewables, storage or carbon controls could alter the operating profile. A narrower Amazon role could reduce attributable exposure. None of those possibilities erases the need to allocate risk; they are the reasons the contract needs measurable conditions.

The right switch is phased capacity. Developers contemplating a single multi-gigawatt gas dependency should combine staged turbine commissioning with grid, storage or renewable options where feasible. Enterprise customers with climate covenants should delay a long-tenor commitment until they can inspect final permit scope, reporting, control guarantees and contractual responsibility. They should require hourly energy and emissions data rather than an annual sustainability narrative.

A sensible diligence file should reconcile the named turbines, gross and delivered capacity, expected commissioning sequence, permit identifiers, control equipment and reporting owner. Finance should then model at least three cases without calling any one of them a forecast: delayed energization, partial capacity and full contracted delivery. The purpose is to price remedies, not to guess dispatch from a permit maximum. Sustainability teams should separately define which emissions sit in the customer inventory and which remain the generator’s responsibility.

The commercial trigger should be measurable. Release a take-or-pay obligation only after the relevant phase has its final authorization, passes commissioning and demonstrates the contracted output. Tie later phases to the same evidence rather than letting one completed block validate an entire 7.65-GW plan. If the provider will not accept milestone-backed remedies, the buyer is being asked to finance uncertainty while the developer retains the upside.

The conclusion would change with evidence: final turbine and permit schedules; executed Amazon purchase or offtake documents; hourly dispatch assumptions; guaranteed controls; and proof that most generation is emergency rather than baseload. Until then, the 3.94 Mt/GW-year ratio remains a normalized permit ceiling useful for comparing contractual exposure, not a forecast useful for declaring environmental outcome.

This is the infrastructure version of today’s Astra release-gate argument. A powerful asset can exist and still be unavailable until evidence clears. The same discipline shaped yesterday’s Firmus capital-stack analysis: financing and announced capacity do not equal an energized accelerator-hour. GW Ranch may eventually deliver enormous power. Buyers should pay for delivered output and defined remedies, not for the adjective “private.”