Saad Ullah
⬤ Epoch AI's latest estimates show the frontier AI data-center landscape in the U.S. as of July 2025, measured in megawatts of power capacity. Google dominates with 927 MW, while Anthropic follows at 819 MW and OpenAI trails at 684 MW. The data zeroes in on the biggest U.S. facilities, assigning capacity based on who's expected to use them. These numbers paint a clear picture of the massive infrastructure buildout happening across the AI sector.
⬤ Anthropic's total gets a significant boost from 478 MW connected to the Anthropic-Amazon Project Rainier partnership, showing how Amazon (AMZN) is playing the infrastructure game through strategic partnerships rather than going solo. OpenAI's capacity includes 147 MW from their Stargate Abilene project with Oracle, highlighting how AI companies are teaming up with cloud providers to build these massive data centers. Further down the list, xAI sits at 368 MW, Meta at 319 MW, while Microsoft registers 0 MW in this particular dataset focused on select large-scale sites.
⬤ Looking ahead, projections suggest OpenAI could more than double any competitor's data-center capacity by early 2028. But right now, Google's clearly out in front, with Anthropic and OpenAI battling for second tier. Anthropic's expansion solidifies Amazon's (AMZN) position in the AI infrastructure race through partnerships rather than direct development.
⬤ These power capacity numbers are becoming a critical market indicator—they signal who'll have the compute muscle to train bigger models and scale commercial AI operations. With Amazon (AMZN) tied closely to Anthropic's growth and overall demand for hyperscale capacity climbing, how this infrastructure gets distributed among AI leaders offers real clues about where the competition's headed in the years ahead.
Saad Ullah
