AI is driving rapid growth of energy-intensive data centers across the United States. Today, all of the hyperscaler tech companies’ operating data centers in the country are powered through the local energy grid or via gas-fired power plants on-site. Where tech companies build and how they power their facilities can have drastically different impacts on local air pollution and global greenhouse gas emissions.
This scorecard evaluates the eight hyperscaler tech / AI companies with U.S.-based data centers on one simple question: are their data centers running on clean energy, or fossil fuels like gas, oil, and coal? Each company is scored on a 0–100 scale, with 100 representing entirely clean power, based on two factors: whether they burn gas behind the meter (BTM) on-site, and how clean the electrical grid is where they build.
The analysis covers all planned and operating U.S. data centers tracked by Cleanview, with grid pollution data from the EPA's eGrid database, the location-based methodology used by the Financial Times, The Guardian, and the United Nations to evaluate tech company pollution.
Company rankings
| Company | Score | Key Factors |
|---|
How the score works
Every data center in a company's portfolio is classified as either behind-the-meter, or BTM, (burning gas on-site, bypassing the grid entirely) or grid-connected (drawing power from the local electrical grid).
The company score is the simple average of two equally-weighted factors:
Behind-the-meter gas
What share of the company's data center projects burn gas on-site? This is the most-polluting possible practice and means that 100% of the data center's power comes from gas. 0% BTM = score 100. 100% BTM = score 0.
Grid carbon intensity
For grid-connected projects, how clean is the local grid? Measured in lbs of CO₂ per MWh using EPA eGrid data. Building in coal-heavy regions can delay coal plant retirements, leading to extreme amounts of pollution from carbon dioxide and other highly toxic pollutants like mercury, sulfur dioxide, and particulate matter. The scoring system creates a range based on the minimum and maximum carbon intensity on the US grid today – 200 lbs/MWh is the cleanest grid you can find and 1,000 lbs/MWh is the least clean. 200 lbs/MWh = score 100. 1,000 lbs/MWh = score 0.
Why these two factors?
Coal, oil, and natural gas release toxic pollutants and greenhouse gases into the air that harm public health and the environment. Polling repeatedly finds that Americans strongly prefer to see AI data centers powered with clean energy vs. gas or coal.
The behind-the-meter (BTM) score captures the most-polluting behavior, namely building on-site, natural gas burning power plants that wouldn’t need to be built otherwise, increasing air pollution in the community surrounding a data center.
The grid intensity score captures whether companies are choosing to build their data centers where the electric grid is powered with cleaner energy sources like wind and solar, or where the grid’s power is heavily dominated by coal and gas.
Additional Resources
- “Clean Energy Is The Cheapest Way To Meet Rising Demand”
- “Behind-the-meter data center gas plants will raise US energy bills”
- “American Voter Priorities for Data Center Policy: Clean Energy, Accountability, and Protecting Communities from Pollution”
- “Opposition to Data Centers Is Growing. On-Site Renewable Energy May Be the Answer”
- “Google Is Proving You Don’t Need On-Site Gas to Build Large Data Centers Quickly”
- “A.I. Is on the Rise, and So Is the Environmental Impact of the Data Centers That Drive It”
This tracker is not intended to be a comprehensive analysis of tech / AI companies’ environmental impacts; it is solely focused on the hyperscalers’ data center power generation.
Citations
- Data sources:
- Cleanview data center tracking database (project locations, capacity, behind-the-meter status, tenant relationships). EPA eGrid subregional emissions data (CO₂ intensity, fuel mix, renewable share). Grid carbon intensity is the location-based methodology required by the GHG Protocol.
- Scope:
- All planned and operating US data centers tracked by Cleanview. Stargate projects are attributed to both OpenAI and Oracle as joint partners. Anthropic projects include facilities where Anthropic is a known tenant.
- Scoring:
- BTM score = (1 − share of projects burning gas, oil and coal on-site) × 100. Grid intensity score = linear scale from 200 lbs CO₂/MWh (score 100) to 1,000 lbs CO₂/MWh (score 0), averaged across all grid-connected projects. Overall = simple average of both factors.