New tracker scores AI data centers on carbon emissions

A new emissions tracking tool is giving corporations visibility into the carbon footprint of their AI infrastructure choices, revealing stark differences among the industry's biggest players.

Climate Power, a climate communications nonprofit, has launched the Data Center Air Pollution Tracker to rank eight leading cloud and AI services companies on their environmental impact. The tool draws on facility data from research firm Cleanview and applies carbon intensity metrics from the EPA, United Nations, Financial Times, and Guardian to assess both planned and operational data centers.

According to the tracker, Amazon earned the highest score at 68 out of 100, with an average carbon intensity of 716 pounds of CO2 per megawatt-hour. The company achieved this rating partly through the lowest ratio of behind-the-meter natural gas projects among the eight firms evaluated.

Microsoft followed at 64, while Google and Meta both scored 61. The two most prominent AI model providers, Anthropic and OpenAI, received scores of 59 and 56 respectively. Oracle landed at 51.

At the bottom sits Elon Musk's xAI, with a score of just 6. The company's Colossus data center in Memphis relies on largely unregulated gas turbines that produce smog-forming emissions.

How the scoring works

The tracker evaluates companies on two equally weighted factors: their use of behind-the-meter natural gas generation and the carbon intensity of local electricity grids serving their facilities.

A perfect score of 100 would indicate exclusive reliance on grid power from the world's cleanest electricity sources. The tool considers grids producing 200 pounds of CO2 per megawatt-hour as cleanest, while the dirtiest grids generate 1,000 pounds per megawatt-hour.

The metrics update daily, providing corporations with current data as they make procurement decisions that could affect their own emissions reduction commitments.

Why it matters

The AI data center expansion is projected to consume 194 gigawatts by 2035—roughly 20 percent of total U.S. power production. This buildout has triggered community resistance not only because of electricity demand but also due to the Trump administration's push to power these facilities with coal and natural gas rather than renewable sources.

That policy direction contradicts public sentiment. Climate Power polling from June found voters strongly favor data centers committed to clean energy, with projects tied to coal or natural gas receiving significantly less support than those connected to solar or wind generation.

For enterprises expanding AI capabilities, the tracker offers a concrete tool to evaluate whether their infrastructure partners align with sustainability goals—or undermine them.

The details were first reported by Trellis.