Amazon Explores Grid Connection for 8,000-Acre AI Campus in Texas

Amazon is developing its own power generation system for an 8,000-acre artificial intelligence campus in West Texas and has confirmed it is evaluating a potential shift to grid-connected service that could supply electricity to other customers. The move signals a pivotal evolution in how hyperscale cloud providers are approaching energy infrastructure amid surging AI-driven demand.

In response to questions from Data Center Knowledge, an Amazon spokesperson said the company is "actively exploring" a transition to front-of-the-meter grid service and sees an opportunity for its energy infrastructure to eventually "add supply to benefit other customers." Front-of-the-meter refers to power generation connected directly to the grid, as opposed to behind-the-meter systems that supply electricity exclusively to an on-site facility. Amazon did not disclose a timeline for the transition, nor did it confirm whether it has submitted a large-load interconnection request to the Electric Reliability Council of Texas (ERCOT) or specify the grid capacity the campus might seek.

The project, known as GW Ranch, combines a planned private natural gas-fired power plant with a massive AI data center in Pecos County, approximately 17 miles north of Fort Stockton. Cleanview first reported Amazon's involvement on Aug. 7 after linking construction filings to the site, and Amazon subsequently confirmed it had acquired the property and planned to purchase power from Pacifico Energy. The Texas Commission on Environmental Quality (TCEQ) preliminary determination for Pacifico describes a plant with a nominal output of 5,000 megawatts (MW) powered by 35 simple-cycle combustion turbines. The facility will supply electricity exclusively to the on-site AI data center and will not initially sell or receive power through the local utility distribution system.

The project has drawn national scrutiny over its potential environmental impact. Reports from The New York Times, The Verge, Futurism, and other outlets have cited TCEQ documents estimating allowable annual emissions of approximately 33 million tons of carbon dioxide equivalent (CO₂e). That figure reflects maximum operating assumptions—including 8,760 operating hours per turbine annually—rather than a forecast of actual emissions, which would depend on real-world operating conditions and demand levels.

Pacifico proposed the plant as a greenfield development, with the utility describing GW Ranch as a broader project encompassing up to 7.65 gigawatts (GW) of gas generation, 1.8 GW of battery storage, and up to 750 MW of solar capacity. First power is targeted for the first quarter of 2027, with 1 GW expected online in 2028 and more than 5 GW by 2031. Pacifico's public materials do not explain the relationship between the 7.65 GW figure and the 5,000 MW, 35-turbine configuration in the TCEQ filing, and Amazon has not disclosed the campus's planned IT load.

Industry experts note that behind-the-meter generation at this scale remains highly unusual. Joshua Rhodes, a University of Texas researcher who studies electricity markets and infrastructure, said that while many companies have discussed such arrangements, few have followed through. "Many have talked about doing behind-the-meter generation, but didn't proceed when they realized how hard it is," Rhodes said. He also highlighted a broader market constraint: "It will still compete with the electric grid in that turbines are scarce right now. If 5 GW of turbines go to this behind-the-meter data center, they won't be available for other uses."

Pacifico has framed the private-grid design as a strategy to bypass lengthy interconnection queues and deliver power directly to hyperscale data centers. Amazon, however, said it is intentionally structuring the arrangement so its energy investment can eventually contribute to a more reliable and affordable grid. The company said it sees an opportunity to transition the campus to front-of-the-meter service, adding supply for other customers, though it did not explain how that would work—including whether the generation would become an ERCOT resource, how much capacity could be made available, or how the campus's own load would be treated.

Rhodes said it is too early to assess how GW Ranch's existing generation might affect a future ERCOT connection. "This part is still unclear," he said, adding that clarity should emerge as ERCOT's Batch Zero process and subsequent large-load batches move forward. ERCOT's Batch Zero is a systemwide study process for qualifying large-load requests of at least 75 MW, grouping eligible projects for reliability assessments, capacity allocation, and identification of required transmission upgrades. The framework also includes a process for large loads seeking to connect with co-located generation under a Withdrawal-Limited Private Use Network (WLPUN), also referred to as bring-your-own-generation.

For GW Ranch, the unresolved question is how that framework would apply if Amazon transitions from its initial private power system to a grid-connected configuration while retaining some or all of its on-site generation. Rhodes does not expect hyperscalers to embrace power generation as a permanent core business. "I don't think that hyperscalers want to be in the power business, the margins are too low," he said. "I think this is just a necessity for some right now."

Amazon's response to Data Center Knowledge also provided new details on the campus's planned energy mix. The company said it is exploring solar and battery storage and is "actively exploring how to maximize deployment of carbon-free resources on site." The initial phase requires no water for power-generation cooling, while later phases are designed to rely primarily on brackish groundwater, which Amazon said does not compete with community drinking-water supplies. The company is also exploring produced water as another non-potable source. Amazon said the campus is expected to create thousands of jobs but did not disclose its planned investment in the data center.

The potential shift to front-of-the-meter service leaves open several critical questions: how Amazon's generation assets and data-center load would be treated in ERCOT's interconnection process, what transmission upgrades would be required, and how much of the campus's eventual demand would rely on the grid. As hyperscale AI campuses continue to strain regional power systems, the GW Ranch project represents a test case for whether private generation can eventually integrate into public grids without compromising reliability or affordability.