The numbers coming out of Pecos County, Texas, read like a misprint. Amazon is financing a natural gas power plant permitted to release 33 million tons of CO2 per year, more than any single operating source in the United States. The plant, called GW Ranch, would generate 7.65 gigawatts from 35 gas turbines to feed an AI data center campus that Amazon confirmed it acquired from developer Pacifico Energy. At full capacity, it would be the largest power plant in the country by output, surpassing the Grand Coulee hydroelectric dam in Washington State.
AI data center power demand has outgrown the grid, and the bridge being built runs on fossil fuel.
What exactly is Amazon building in West Texas?
The project came to light through Cleanview, a market intelligence firm that tracks data center power projects and reviews satellite imagery of construction sites. Cleanview found that land clearing had already begun at the location described in GW Ranch's air permit records. The Amazon connection surfaced when the company filed three construction permits with the state of Texas for data center buildings on the site.
Pacifico Energy, the developer, announced in January that it had secured permits from Texas regulators for 35 "simple cycle" gas turbines. Simple cycle turbines are less efficient than the "combined cycle" units used in baseload plants. Combined cycle plants capture waste heat from the first turbine to drive a second generator, improving efficiency by 30 percent or more. Simple cycle units vent that heat. They are typically reserved for peaker plants that run a few hours a year. GW Ranch plans to run them as primary generation for a data center.
The site is also permitted for 750 megawatts of solar and 1.8 gigawatts of battery storage, according to the same Heatmap News report. But the gas turbines dominate the planned capacity by a factor of nearly three.
Amazon confirmed to Cleanview that it acquired the site and plans to buy power from the plant. A company spokesperson told the New York Times, as reported by The Verge, that the project is "powered by new on-site generation that won't raise electricity costs for Texas families and designed to transition to grid-connected service as interconnection timelines allow."
The key phrase is "as interconnection timelines allow." Grid connection for a project of this scale can take years. In the meantime, the data center runs on gas.
How does 7.65 gigawatts compare to every other US power plant?
By generation capacity, GW Ranch would be the largest power plant in the country. The current leader is Grand Coulee at 7 gigawatts, followed by Arizona's Palo Verde nuclear plant at 4 gigawatts, and Florida's West County Energy Center, a combined cycle gas plant, at 3.7 gigawatts. GW Ranch at 7.65 GW would top all of them.
The chart above shows the gap. GW Ranch is the largest gas plant planned in US history, and it would burn the least efficient type of gas turbine available at industrial scale.
The emissions picture is even starker. The permit allows up to 33 million tons of CO2 per year. According to Heatmap's analysis, that is twice the 2023 emissions of the James H. Miller Jr. coal plant in Alabama, currently the most polluting power plant in the country. Facilities rarely emit at their permitted ceiling, but even a fraction of 33 million tons would make GW Ranch one of the largest single greenhouse gas sources in the US.
The technology choice compounds the problem. Combined cycle turbines are on backorder for years, according to the same report, and developers racing to bring data centers online are turning to simpler, dirtier alternatives as a quick fix. The result is a plant that produces more electricity per turbine but emits more CO2 per megawatt-hour than a modern gas plant would.
Why go off-grid instead of waiting for grid connection?
The Texas power grid, operated by ERCOT, has interconnection queues that can stretch for years. A 7.65-gigawatt connection request would be unprecedented. For context, that is more than the rated output of many utility-scale solar farms combined. Waiting for grid approval means waiting years to turn on the compute.
Going off-grid lets Amazon bypass the queue. The gas plant and data center are co-located, so power flows directly from turbines to servers with no grid in between. This is the same strategy other hyperscalers are pursuing. AFP reported that Meta, Microsoft, and Oracle have all explored behind-the-meter power arrangements to avoid interconnection delays. The Rhodium Group developed a scoring system to help investors differentiate between data center power projects that accelerate the energy transition and those that slow it down. Off-grid natural gas plants scored at the bottom.
The tradeoff is carbon. An off-grid gas plant burning simple cycle turbines produces electricity at roughly double the carbon intensity of grid-connected combined cycle power. When the alternative is waiting two to four years for grid connection, the hyperscalers are choosing speed.
What does this mean for AI builders tracking their carbon footprint?
If you are running inference or training workloads on AWS, the carbon accounting gets harder. Amazon's Climate Pledge commits the company to net-zero carbon by 2040, and the company says it has enabled 10 gigawatts of carbon-free energy across 40 projects in Texas. But the company's emissions have climbed for several consecutive years, driven by AI demand. Adding a 33-million-ton permitted source to the portfolio does not reverse that trend.
For builders who report Scope 3 emissions or who have made net-zero commitments of their own, the implications are direct:
- Cloud carbon dashboards may understate the real footprint. AWS reports carbon data per region, but behind-the-meter generation may not flow through standard grid emission factors. If your sustainability team uses grid averages to estimate cloud emissions, the actual carbon intensity of your Pecos County workloads could be far higher.
- Off-grid power creates accounting blind spots. When a data center is not connected to the grid, it does not appear in grid emission datasets. Your cloud provider may report it separately, or may not report it at all until regulatory disclosure rules catch up.
- The gap between pledged and actual emissions is widening. Amazon spokesperson Margaret Callahan told the New York Times that "the world looks different now than when we co-founded the climate pledge," but insisted "our commitment hasn't changed." Builders relying on cloud provider net-zero pledges for their own compliance should treat those pledges as directional, not settled.
The chart below puts the emissions gap in perspective: 33 million tons for GW Ranch versus roughly 16.5 million for the worst coal plant and zero for a hydroelectric dam of comparable output.
This matters beyond Amazon. New York has already passed a data center moratorium that redraws where AI infrastructure can be built, and other states are watching. If Pecos County becomes the template for AI power, the regulatory backlash will follow.
Does Amazon's 2040 climate pledge survive this?
Amazon's Climate Pledge, co-founded in 2019, commits the company to reaching net-zero carbon by 2040. The company has made progress on renewable energy procurement, but its absolute emissions have risen as AI infrastructure expands. The GW Ranch plant, if it operates at even half its permitted capacity, would add roughly 16.5 million tons of CO2 per year to Amazon's footprint. That is the equivalent of adding the worst coal plant in the country to the balance sheet.
Amazon points to its 10 GW of carbon-free energy projects in Texas as evidence of parallel investment. But those projects serve the broader grid, while GW Ranch is dedicated to a single data center campus. The renewable energy offsets the gas plant on a portfolio basis, an accounting convention rather than a physical reality.
The same AFP report noted that if all planned Amazon gas projects reach their potential, total emissions could top 200 million tons of CO2 per year, roughly equivalent to the annual emissions of more than 45 million cars, or about the entire carbon output of Pakistan. Amazon says it plans to transition GW Ranch to grid-connected service over time and add solar and battery storage on site. But the gas turbines are the first thing being built, and there is no binding date for the transition.
What should builders watch next?
The GW Ranch story has three threads to follow:
- Whether the plant actually gets built at full scale. Permits allow more than facilities emit, and construction timelines for 35 turbines are long. Watch for turbine procurement announcements and construction milestones from Pacifico Energy.
- Whether other states push back. New York's data center moratorium set a precedent. If Texas regulators or environmental groups challenge the permit, the timeline slips. The permit was issued under Texas's own air quality program, and the Trump administration has signaled it will not obstruct fossil fuel projects.
- Whether cloud customers start asking about behind-the-meter carbon. Enterprise customers with their own net-zero targets are the likeliest force for change. If enough large AWS customers demand site-level carbon data, Amazon may need to disclose more than regional averages.
The bet Amazon is making is that speed to compute matters more than carbon optics, and that by the time 2040 arrives, offsets and grid transition will close the gap. The bet is defensible only if the grid transition happens fast enough. Right now, the evidence points the other way.
The cost of the shortcut
Every gigawatt of AI compute has to come from somewhere. Amazon chose 7.65 gigawatts of simple cycle gas in West Texas because it was the fastest path from dirt to electrons. The 33-million-ton permit is the price of that speed. Builders who care about their carbon footprint, or who have regulators who care for them, should start asking their cloud providers where the power comes from and what kind of turbine is burning the fuel.
Sources
- The Verge: An Amazon data center could have the worst polluting power plant in the country
- Heatmap News: Amazon Is Building America's Biggest Fossil Fuel Plant to Power AI
- AFP: Amazon behind massive private gas plant for new data centers
- Distilled/Cleanview: Amazon Is Behind One of the Largest Planned Gas Power Plants in the US
