Global AI Datacenter Database
Track the world's largest GPU clusters powering frontier AI: 786 datacenters from xAI, Meta, Microsoft, Google and Tesla, with live GPU counts, H100-equivalent compute, and megawatts online and planned.
Filters
Datacenters (10)
| Name | H100 Equiv |
|---|---|
| Abu Dhabi UAE/USA 5GW Campus Phase 2 | 20.3M |
| Stargate UAE Phase 2 | 2.5M |
| Stargate UAE Phase 1 | 252.7K |
| G42 Microsoft 100Mw UAE Cluster | 55.0K |
| G42 Microsoft 30Mw UAE Cluster A | 15.0K |
| G42 Microsoft 30Mw UAE Cluster B | 15.0K |
| Core42 SuperPOD | 1.3K |
| Core42 AI-03 | 395 |
| G42 Artemis | 81 |
| Alibaba Cloud in UAE | — |
What Is an AI Datacenter?
An AI datacenter is a facility purpose-built to house and power dense clusters of GPUs for training and serving artificial intelligence. Unlike a traditional datacenter tuned for storage and web traffic, these sites are engineered around high-bandwidth GPU racks, liquid cooling, and the enormous electrical supply that modern accelerators demand — often tens or hundreds of megawatts for a single cluster. We track them all in one place, normalised to H100-equivalent compute so different hardware can be compared on a single scale.
The AI Compute Race
Frontier AI now runs on the largest computers ever built. From xAI's Colossus in Memphis to Meta's 100K-GPU clusters, labs are racing to deploy compute faster than the grid can supply power. Tracking who is building where — and how much capacity is online versus planned — reveals where the industry is heading. Explore the systems that fill these racks, or compare the GPUs inside them.
Explore More
GPU Systems
Explore pre-configured multi-GPU systems like DGX, HGX, and SuperPOD that fill these datacenter racks.
Frequently Asked Questions
Expert answers on GPU architectures, Tensor Cores, datacenter hardware, and performance.
GPU & Architecture Glossary
Understand the terminology — from Tensor Cores and NVLink to HBM memory, FLOPS, and precision formats.
The Largest Cluster
xAI's Colossus in Memphis leads the race at roughly 275K H100-equivalent GPUs.
View ColossusFrequently Asked Questions
What is an AI datacenter?
A facility built to house and power large clusters of GPUs for training and serving AI models. Unlike a traditional datacenter, it is engineered around dense GPU racks, high-bandwidth networking, and the enormous power and cooling those accelerators demand, often tens or hundreds of megawatts for a single site.
What does "H100 equivalent" mean?
A normalised measure of compute that expresses a cluster's total AI performance in units of NVIDIA H100 GPUs. Because clusters mix different chips, converting everything to H100 equivalents lets you compare the real compute capacity of datacenters built on different hardware on a single scale.
Which companies operate the largest AI datacenters?
The largest clusters are operated by well-funded labs and hyperscalers including xAI, Meta, Microsoft, Google, Amazon, and Tesla, ranging from tens of thousands to hundreds of thousands of H100-equivalent GPUs.
How much power does an AI datacenter use?
Power draw scales with GPU count. A modern flagship cluster can require anywhere from tens of megawatts to over a gigawatt at full build-out, comparable to a small city. Power capacity is now one of the primary constraints on how fast new AI compute can be deployed.
Datacenter data from Epoch AI, licensed under CC-BY 4.0.