An Amazon Data Center Could Have the Worst Polluting Power Plant in the Country

Amazon has filed construction permits in Pecos County, Texas for a data center campus that will draw power from a dedicated 35-turbine natural gas plant — one that holds a state permit to emit up to 33 million tons of carbon dioxide per year, a figure that would exceed the output of the largest coal-fired power plant currently operating in the United States.

What Changed

The project, known as GW Ranch, is a 7.65-gigawatt natural gas facility being developed by Pacifico Energy on land Amazon recently acquired in West Texas. The data center campus it will feed is designed specifically for AI workloads, joining a wave of hyperscale AI infrastructure investments across the American South and Southwest. Satellite imagery reviewed by Cleanview — the infrastructure-tracking outlet that first reported the project's scope — shows land clearing already underway at the site, and Amazon separately confirmed the Pecos County project to the Houston Chronicle.

That 33 million ton CO₂ ceiling is a permit maximum, not a guaranteed annual output. Plants routinely operate well below their legal limit. But the number still matters: it establishes the legal upper bound for how much the facility could emit, and at that ceiling it would outpace any single coal plant in the country. For context, the largest US coal plant — a category of facility that has been at the center of climate policy fights for decades — doesn't reach that threshold. A natural gas plant attached to one company's private AI infrastructure would.

Multiple reports, including coverage from The Verge and the Chronicle, confirm that Amazon plans to operate this plant off-grid initially, meaning none of the power will flow through ERCOT, Texas's main electricity grid. Amazon spokesperson language, quoted by the Chronicle, framed this as a feature rather than a liability: "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 company also invoked its stated commitment to paying "the full costs of powering our operations" — a framing that sidesteps questions about carbon costs entirely.

How It Works

The off-grid architecture here is deliberate and increasingly common. When a hyperscaler connects a data center directly to a captive power plant rather than drawing from the public grid, it bypasses the renewable energy certificate market that cloud providers have historically used to claim carbon neutrality. RECs — credits representing one megawatt-hour of clean electricity fed into the grid — only work when the grid is in the loop. A private wire from a gas plant to a server farm generates no REC, creates no offset, and produces emissions that don't appear in grid-level reporting unless regulators require separate disclosure.

The 7.65 GW figure is itself worth pausing on. For reference, the average US nuclear plant generates roughly 1 GW. GW Ranch, if built to full capacity, would exceed the output of most large power stations in the country — dedicated to a single tenant's AI compute. The Cleanview tracker, which maps data center construction against associated power projects across the US, flagged this plant as notable precisely because of that scale: it's not a backup generator or a renewable supplement, it's the primary and only power source for a facility large enough to run frontier model training.

The Trump administration's regulatory posture adds a layer of context here. Multiple reports indicate that federal restrictions on new fossil-fuel generation are being actively relaxed, which shortens the permitting timeline for projects like GW Ranch. That regulatory tailwind matters because projects of this scale have historically faced environmental review delays of years — and the permit ceiling of 33 million tons suggests Texas and federal reviewers approved this ceiling without requiring the kind of mitigation that would bring emissions exposure down to, say, the range of a mid-sized coal plant.

What It Means for Developers

Amazon isn't alone in this pattern. Meta and Google have both pursued or announced captive fossil-fuel generation for AI data centers in recent years. Our read is that this is the moment the "cloud is green" assumption breaks definitively for the AI era. The marketing around cloud infrastructure has leaned heavily on renewable energy commitments and carbon neutrality pledges — Amazon's own Climate Pledge, which the company announced in 2019 with a target of net-zero carbon by 2040, is directly at odds with a gas plant operating off-grid with no offset mechanism. When the same company holds a permit to emit more CO₂ from a single facility than any US coal plant, the pledge becomes a talking point rather than a constraint.

For developers building products on top of AI APIs and cloud compute, this creates a new category of upstream liability. ESG reporting requirements — particularly in Europe, where the Corporate Sustainability Reporting Directive is already in effect for large companies — increasingly require disclosure of Scope 3 emissions, which include emissions from purchased goods and services. A startup using Amazon Bedrock or AWS GPU instances as part of an AI product is, on paper, a downstream consumer of GW Ranch's output. The accounting is indirect and the standards for how to handle it are still evolving, but the direction of travel is clear: regulators and investors will eventually demand transparency about where compute-intensive AI workloads actually get their power.

The Hacker News discussion around this story surfaced a tension worth taking seriously: both New York and Texas have data center construction moratoriums reflecting genuine public opposition, yet neither state has a coherent clean-energy-first policy that would give tech companies a viable alternative to building their own gas plants. If green energy interconnection timelines run to a decade and AI training deadlines run to quarters, companies will keep choosing gas. That's not an excuse — it's a policy failure with a concrete beneficiary.

The Cleanview tracker is now probably the most useful single tool for developers who want to understand the actual carbon footprint of the infrastructure they're building on. It maps which data centers are tied to which power projects, which is a level of specificity that provider sustainability reports have never offered.

Sources