Planned Amazon Data Center Could Become the Biggest Climate Polluter in the U.S.

A natural gas power plant permitted to release up to 33 million tons of carbon dioxide per year — more than any single facility operating in the United States today — is being built to power a new Amazon data center campus on a West Texas ranch. The permit was issued for a site in Pecos County called GW Ranch, and the scale of what Amazon is constructing there reframes the AI infrastructure buildout as a story not just about compute, but about emissions. The project is not a proposal. Land clearing has already begun.

The Claim

Amazon has acquired the GW Ranch site in Pecos County, Texas, where it plans to build a large-scale AI data center campus powered by an on-site natural gas generation facility rather than drawing from the state's electrical grid. According to multiple reports citing the New York Times and data center tracking firm Cleanview, the planned power plant would consist of 35 natural gas turbines with a combined capacity of 7.65 gigawatts. The plant is being developed by Pacifico Energy. Amazon confirmed the Pecos County project separately to the Houston Chronicle.

Amazon's stated rationale for keeping the facility off-grid centers on cost and civic responsibility. "Amazon believes in paying the full costs of powering our operations," a spokesperson told Chron. "Our new planned data center campus in Pecos County does just that: it's powered by new on-site generation that won't raise electricity costs for Texas families and is designed to transition to grid-connected service as interconnection timelines allow." The implication is that Amazon isn't exploiting the Texas grid — it's building around it, at least until ERCOT interconnection can accommodate demand at this scale.

The climate permit that has drawn attention allows the plant to emit up to 33 million tons of CO₂ annually. That figure, if reached, would place this single facility above every existing U.S. power plant in terms of greenhouse gas output. Multiple reports — from The Verge, TechCrunch, and the Houston Chronicle — confirm this reading, all citing the Cleanview analysis as the underlying source for the comparison.

What We See

The number that matters here isn't 7.65 gigawatts. It's the 33 million ton emissions ceiling — and the fact that Amazon's overall carbon output was already up 16 percent year-over-year before this plant enters operation, with company leadership explicitly attributing that increase to AI workload growth. This project doesn't represent a future risk. It accelerates a trend already in motion.

What makes this harder to wave away as an overblown environmental scare is the language coming from Amazon itself. A spokesperson acknowledged publicly that the company's original net-zero commitment is under pressure: "The world looks different now than when we co-founded the climate pledge." Our read is that sentence carries more weight than the permit number. Amazon is conceding that the 2040 target it helped establish under the Climate Pledge framework — co-founded alongside Global Optimism — no longer reflects its operational direction. That is not a defense of this project. It is an admission of a gap between stated values and actual infrastructure choices.

The off-grid decision deserves scrutiny on its own terms. Amazon says the plant "won't raise electricity costs for Texas families," and on that narrow claim it is probably accurate — a captive generation facility doesn't draw from ERCOT's public grid and doesn't compete with residential demand. But framing that as a public benefit conflates grid neutrality with environmental neutrality. Burning natural gas at 7.65 gigawatt scale affects the atmosphere regardless of whether the electrons route through a utility. The more candid version of Amazon's argument is that ERCOT doesn't have spare interconnection capacity for a campus this large anyway, so captive generation is a workaround for a grid bottleneck, not a principled climate choice. Satellite imagery reviewed by Cleanview confirmed that land clearing at GW Ranch is underway, and Amazon has already filed three construction permits with the state for data center buildings at the site.

A separate source worth noting: multiple outlets confirm that Amazon is not the only hyperscaler taking this path. Meta and Google are pursuing their own off-grid captive power strategies for AI infrastructure expansion. This is less an Amazon-specific story than a sector-wide story about what AI compute actually costs in carbon terms — and the answer, given current grid constraints and the pace of capacity buildout, appears to be considerably more than anyone publicly committed to when they signed the pledges.

Where It Falls Short

The most important caveat in current coverage is that 33 million tons is a permit ceiling, not an operational forecast. Emissions permits routinely authorize far more than facilities actually release — they set a legal maximum rather than a utilization target. Amazon has not published projected annual output for the plant, and neither Pacifico Energy nor Amazon has disclosed a turbine utilization schedule. It is plausible that the facility operates at a fraction of permitted capacity for years; the gap between what a permit allows and what steady-state operations produce can be substantial, and none of the reporting to date addresses that distinction with any precision.

What is also missing from the current wave of coverage is how this project interacts with Amazon's existing renewable energy portfolio. Amazon has historically claimed the title of the world's largest corporate purchaser of renewable energy, and it presumably intends to apply renewable energy certificates against its data center emissions in some form. Whether those offsets apply at the facility level or only at the corporate aggregate level — and how the accounting framework works for an off-grid captive plant — goes unaddressed in every source that covered this story. That omission matters because the "biggest climate polluter in the U.S." headline depends entirely on where you draw the accounting boundary.

Sources