Amazon is investing in a massive natural gas power facility for a planned AI data center in West Texas that is permitted to emit up to 33 million tons of carbon dioxide annually, potentially making it the largest single source of carbon pollution in the US.

The Pecos County project is designed to generate as much as 7.65 gigawatts of electricity using 35 natural gas turbines. Actual emissions could be significantly below the permitted maximum because power plants typically do not operate at their regulatory limits. Still, the scale of the authorization demonstrates the extraordinary power requirements created by today’s rapid AI buildout. 

Amazon recently acquired the Pecos County property where a developer is building the power plant. The facility is expected to operate initially without a connection to the Texas electrical grid, with its power primarily intended for Amazon’s data center. This on-site power generation approach is as an important strategy for hyperscalers as they attempt to expand AI capacity faster than utilities can accommodate them. Connecting these deployments to the existing utility grid involve lengthy interconnection timelines.

Natural gas offers a comparatively fast route to adding large generation capacity. But it also creates a significant environmental problem for tech companies that have made aggressive carbon-reduction commitments.

The tech sector now faces the challenge of rapidly building the physical capacity required for AI while reconciling that expansion with carbon-reduction goals established before the current AI boom.

The Climate Pledge

Amazon co-founded The Climate Pledge in 2019 and committed to achieving net-zero carbon emissions across its operations by 2040. More than 700 companies and organizations have joined the initiative. Yet Amazon’s emissions have risen for several consecutive years as the company expands infrastructure to support cloud computing and AI.

“The world looks different now than when we co-founded the climate pledge,” Amazon spokeswoman Margaret Callahan told The New York Times, while adding that the company’s commitment has not changed.

Amazon says the Pecos County project is designed to prevent its growing power requirements from increasing electricity costs for Texas consumers. The company is also considering solar generation and battery storage at the site, which would not replace but supplement the natural gas turbines.

To reduce its use of natural resources, Amazon plans to use water unsuitable for drinking or irrigation, deploy custom cooling technology to reduce consumption and investigate the use of produced water (water that surfaces during oil and natural gas production).

The issue here is far larger than the Pecos County project. AI data centers are placing new strains on electrical grids just as hyperscalers are investing heavily in building computing capacity. The resulting bottleneck is prompting data center operators to consider dedicated power generation rather than relying exclusively on utilities.

This shift is likely to reshape data center architecture. Historically, cloud providers purchased electricity from utilities that supplied power generated from a mixture of natural gas, coal, nuclear and renewable sources. With today’s growing demands, the power plant itself could become an integrated element of an AI data center campus. Hyperscalers are exploring alternatives that include nuclear power, solar generation and energy storage.

However, natural gas has advantages in availability and deployment speed, particularly in regions with abundant gas supplies like West Texas.