Amazon Plans a 7.65 GW Gas Plant to Power Texas Data Centers
A gas plant built mainly to feed data centers, initially off the state grid, gets a permit capping annual CO2 at up to 33 million tons.
Amazon is planning a 7.65-gigawatt natural gas power station in Texas, built primarily to supply its data center operations — and, in an unusual arrangement, initially disconnected from the state grid.
The numbers
Seven point six five gigawatts is not a data center backup generator; it is a medium-sized power plant. The project, called GW Ranch and located in Pecos County, lines up 35 gas turbines, 1.8 GW of battery storage and 750 MW of solar, off the ERCOT grid. The permit allows annual carbon emissions of up to 33 million tons — a figure TechRadar notes could make it one of the biggest CO2 polluters in the US. That number is a legal ceiling rather than an expected operating level, but even the possibility says something about the scale of the AI buildout.
Why dedicated power
AI data centers cannot wait for the grid. They need enormous, reliable, always-on electricity, and utilities cannot build transmission fast enough. The industry's answer has been to build generation next to the load: gas plants, and increasingly nuclear deals, attached directly to data center campuses. The GW Ranch complex — developed by Pacifico Energy Group, with Amazon disclosing ownership after permit filings and satellite images surfaced — is one of the largest examples yet.
The trade-off nobody has resolved
The environmental math is uncomfortable. Natural gas is the fastest way to get gigawatts online, and it is a fossil fuel. The 33-million-ton ceiling is the permit's way of saying "this scale has consequences". Meanwhile the demand only grows — every frontier model training run and every agent deployment consumes power that someone has to generate.
There is no clean answer in this story, only a clear trend: the AI era's energy question is being answered with whatever can be built fastest, and the climate bill arrives later.
Every conversation about AI's future eventually runs into electricity. This is what that conversation costs.