Amazon Acquires Texas Site for 7.65 GW Gas-Powered AI Campus

Amazon confirmed on August 7 that it acquired a Pecos County, Texas, site for an AI data center campus initially powered by Pacifico Energy's proposed 7.65 GW GW Ranch gas project. The project's air permit allows up to 33 million metric tons of annual greenhouse-gas emissions, a regulatory ceiling rather than a forecast of actual emissions.
Amazon confirmed on August 7 that it acquired a site in Pecos County, Texas, for a planned AI data center campus initially supplied by on-site natural-gas generation. The associated GW Ranch project, developed by Pacifico Energy, is permitted for up to 7.65 GW of generation.
Cleanview's Michael Thomas reported the connection after reviewing three Amazon construction permits and satellite imagery, then obtaining confirmation from Amazon. AFP separately reported that Amazon is financing the private power project. The proposed plant would use 35 gas turbines and initially operate apart from the Texas grid.
Amazon said the campus is designed to transition to grid-connected service when interconnection timelines allow. The company also said it is exploring on-site solar and battery storage and plans to use non-potable brackish groundwater rather than water suitable for drinking or irrigation. Those statements describe Amazon's plans; the retrieved sources do not establish that the optional solar or storage components have been contracted.
The permit sets a ceiling, not a forecast
Texas Commission on Environmental Quality records show Pacifico GW received its state air permits in January 2026. Pacifico announced the milestone on January 26, describing authorization for 7.65 GW of gas-fired generation.
Cleanview and AFP report that the permit allows up to 33 million metric tons of greenhouse-gas emissions per year. That figure is a permitted maximum, not an estimate of likely annual output. Plants commonly emit less than their permit ceilings, so the number should not be read as a prediction. If fully used, however, the authorized level would make the project an unusually large single source of US greenhouse-gas emissions.
Amazon told AFP it remains committed to its Climate Pledge goal of net-zero carbon emissions by 2040 and cited nearly 10 GW of carbon-free energy across 40 Texas projects supporting existing operations. The new campus's reliance on dedicated gas generation nevertheless creates a direct tension between faster compute deployment and the emissions profile of its power supply.
Power availability shapes AI deployment
GW Ranch shows how power procurement and interconnection timing are becoming first-order constraints for large AI clusters. Behind-the-meter generation can let a campus start before a full utility interconnection is available, but it also concentrates dependence on fuel supply, turbine delivery, site permits, water planning, and the reliability of local generation assets.
For infrastructure teams, the relevant comparison is not simply grid versus off-grid. It is the complete operating design: generation redundancy, fuel contracts, emissions compliance, later grid integration, cooling-water availability, and whether planned lower-carbon additions materially change the facility's delivered power mix.
Key Points
- 1Amazon confirmed a Pecos County AI campus tied initially to Pacifico Energy's permitted 7.65 GW GW Ranch gas-generation project.
- 2The reported 33 million metric tons of annual greenhouse gases is a permit ceiling, not a forecast of actual emissions.
- 3The project illustrates how interconnection queues, fuel, permitting, and on-site generation now shape the timing and risk of large AI deployments.
Scoring Rationale
The project is an unusually large example of dedicated power infrastructure tied to AI data center demand. It is relevant to infrastructure practitioners because grid access, generation procurement, emissions permitting, water sourcing, and operational power reliability are increasingly material constraints on large-scale compute deployment.
Sources
Public references used for this report.
Practice with real Retail & eCommerce data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Retail & eCommerce problems
