Photo: panumas nikhomkhai / Pexels
The independent research organization now tracks 86 large AI facilities with a combined 13.3 gigawatts of IT power and nearly 14 million GPUs.
If you wanted to understand the sheer scale of AI infrastructure being built around the world, you used to be stuck stitching together press releases, earnings calls, and educated guesses. Epoch AI just made that job significantly easier. The independent research organization’s AI Data Centers database now tracks 86 large facilities globally, covering an estimated 46% of all deployed AI compute as of June 2026. That’s roughly double the coverage from early 2024, when the figure sat around 22%. The tracked sites carry a combined IT power capacity of approximately 13.3 gigawatts and house an estimated 13.9 million H100-equivalent GPUs.
What’s inside the database #
Epoch’s database focuses on high-capacity, AI-optimized sites rather than traditional enterprise data centers. The methodology combines satellite imagery with public documents to build detailed timelines covering power consumption, computing capabilities, and construction progress for each facility.
The biggest entries read like a leaderboard for the AI arms race. SpaceXAI’s Colossus 2 tops the list with roughly 1.112 million H100-equivalent GPUs. Microsoft’s Fairwater Atlanta facility follows at approximately 769,000 H100-equivalents. Amazon’s Anthropic-dedicated site in New Carlisle, Indiana, rounds out the top three with around 686,000 H100-equivalents.
The geographic distribution skews heavily toward the United States, though the database also captures international sites in countries including China and Malaysia.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
From niche tracker to essential infrastructure data #
Epoch AI has been tracking compute trends in the AI ecosystem for years, but the data center database represents a significant expansion of scope. The organization previously maintained a GPU Clusters database that offered more limited coverage of where compute was actually deployed.
The amplified project kicked off in earnest around mid-2025, and the pace of additions has been aggressive. Going from 22% coverage to 46% in roughly two years suggests Epoch’s researchers are adding facilities faster than new ones come online, closing the gap between what exists and what’s documented.
The database as of September 16, 2026 encompasses those 86 sites, with the organization noting that coverage continues to increase as it adds detailed data for more facilities worldwide. Infrastructure improvements and expansions projected through 2028 are also captured in the construction timeline data.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our