Amazon's Pecos County Gas Plant for AI Data Centers

Amazon has confirmed it is the financial backer and power off-taker behind GW Ranch, a 7.65-gigawatt, 35-turbine natural gas plant under construction on more than 8,000 acres in the Permian Basin of Pecos County, Texas ^[1]. The company says the plant, developed by Pacifico Energy, will run "new on-site generation that won't raise electricity costs for Texas families" ^[1], and pairs the gas turbines with 1.8 GW of battery storage and up to 750 MWac of solar on the same site ^[2]. Pacifico first announced the project a year earlier, in August 2025, describing a smaller 5-gigawatt scope before it grew to today's 7.65 GW across 35 turbines ^[3].

To put that number in perspective: an AP wire story reports that GW Ranch's permitted 7.65 GW capacity would be larger than any natural gas power plant currently operating in the United States ^[4]. That figure would also exceed the roughly 6,809-megawatt (6.8 GW) capacity of Grand Coulee Dam, currently the largest power station of any kind operating in the U.S. ^[9]- a comparison between two independently sourced numbers, not a claim either source makes directly. Developer Pacifico Energy itself has called GW Ranch the "largest power project" in the United States, as reported by Inside Climate News ^[8], though that superlative is the developer's own characterization rather than an independently verified finding. Amazon itself had not been named as the plant's financial backer until early August 2026, when the market-intelligence outlet Cleanview/Distilled Earth matched satellite imagery of active land-clearing at the site to three newly filed Amazon data center construction permits and got the company to confirm the connection on the record ^[1], a scoop that was quickly picked up by wire outlets and tech press ^[4]. Amazon is targeting first power delivery by the first quarter of 2027 ^[2].

The defining design choice at GW Ranch is not the plant's size but its isolation: it will run entirely disconnected from the Texas grid, at least in its early phase, with Amazon describing it as "designed to transition to grid-connected service as interconnection timelines allow" ^[1]. That framing is deliberate. Amazon's public position is that behind-the-meter generation lets it "pay the full costs of powering our operations" rather than relying on new transmission and generation infrastructure that would ultimately be billed to ordinary Texas ratepayers, whose electricity costs have already climbed alongside AI-driven demand growth ^[1]. The plant will also draw on non-potable, brackish groundwater rather than water sources used for drinking or irrigation, an apparent attempt to head off a second front of local opposition around water use in an already water-stressed region ^[6].

That doesn't mean the project avoids local friction. Reporting on the deal notes a broader, increasingly bipartisan wariness toward large data centers landing in rural and suburban Texas communities, driven by concerns that these facilities bring relatively few permanent local jobs while carrying a track record - at grid-connected sites elsewhere - of utility rate increases and reliability problems ^[5]. Amazon counters that framing by pointing to its existing Texas footprint: roughly 40 carbon-free energy projects totaling near 10 GW of capacity across the state, positioned as evidence the company is a net contributor to Texas's energy mix even as GW Ranch itself burns gas ^[6].

In January 2026, the Texas Commission on Environmental Quality quietly issued GW Ranch an air permit allowing it to emit up to 33 million tons of carbon dioxide annually ^[5]. Run at that ceiling, GW Ranch would exceed the emissions of the largest coal plant in the United States and become, by a wide margin, the single largest greenhouse-gas source anywhere in the country ^[5]- a claim that requires the caveat that power plants rarely run at their full permitted maximum, but one that still sits uneasily next to Amazon's own Climate Pledge commitment to reach net-zero emissions by 2040 ^[6].

The criticism isn't limited to the raw tonnage. John Rogers, an energy analyst at the Union of Concerned Scientists, argues the deeper problem is a missed opportunity: "There are more efficient ways to use gas. Virtually none of these plants are proposing to use those, and that matters for on-site pollution, both direct and climate" ^[4]. His point is that GW Ranch and its peers are choosing conventional gas turbine technology over more efficient alternatives at exactly the moment scale would make efficiency gains matter most. Not every reading of the story agrees the plant is uniquely dirty, though - some pushback has argued that GW Ranch isn't dirtier per megawatt than a typical gas plant, it's simply several times larger, and that the "most polluting plant in America" framing conflates scale with intensity. Either way, if the roughly 97 GW of behind-the-meter gas capacity now planned across hyperscalers is ever run at full output, estimates put the collective toll north of 200 million tons of CO2 a year - comparable to the annual emissions of more than 45 million cars ^[4].

GW Ranch is not an isolated bet - it's the leading edge of an industry-wide pattern. Roughly 30 miles away, Microsoft and Chevron announced their own 2-gigawatt off-grid natural gas data center campus near Pecos in June 2026 ^[1]. Meta, meanwhile, has struck a reported $1.6 billion power-and-pipeline infrastructure deal with Williams to build a natural gas-powered data center campus in Ohio ^[7]. Data center developers have announced nearly 60 behind-the-meter gas power projects industry-wide since the start of 2025, adding up to roughly 90 gigawatts of combined planned capacity, with Amazon, Microsoft, Google, and Meta all committing significant capital to natural gas power in 2026 ^[7].

The pattern reveals the actual constraint driving these deals: it isn't a gas preference, it's grid queue length. Every hyperscaler racing to bring AI compute online faster than interconnection timelines allow is reaching for the same playbook - buy land, buy turbines, skip the queue. That convergence is also what turns an individual company's power decision into a collective climate liability; no single hyperscaler's behind-the-meter plant would reshape the national emissions picture on its own, but a dozen of them running simultaneously plausibly could ^[4]. Whether regulators, competitors, or the companies' own sustainability pledges catch up to that arithmetic before 2028 - the year Pacifico projects GW Ranch's first gigawatt comes online - remains the open question ^[2].