SpaceXAI’s Colossus 2 has the largest deployed entry, with 550,000 GB200/GB300 GPUs. Its additional 440,000 announced GPUs are excluded from that deployed count.
The count is based on the operator’s reported figures, not an independent hardware audit. The index is a selected inventory; its rows include both individual clusters and combined fleets, and chip counts do not measure usable computing performance.
This index does not establish a complete GPU total for either lab. Anthropic has disclosed access to more than 220,000 NVIDIA GPUs through its Colossus agreement, but access to a partner’s cluster is not ownership or a complete inventory.
A gigawatt measures power capacity and cannot be converted into a reliable chip count without knowing the hardware and facility assumptions. The lab chart mixes existing access, future commitments, “up to” ceilings, and OpenAI’s NVIDIA letter of intent; it is not a measure of capacity operating today.
The index records 23,536 deployed GH200 GPUs in JUPITER’s Booster module at Jülich. The operator specifies 5,884 compute nodes, each with four Grace-Hopper Superchips, matching that count.
This counts the GPUs in the Booster’s standard compute nodes, not every processor or login node in the wider system. A GPU inventory is also different from a supercomputer performance ranking such as TOP500.
A GPU cluster connects GPU-equipped servers so workloads can use multiple machines together. A GPU fleet is an inventory that can span several clusters or locations; it does not necessarily form one interconnected system.
The index includes both individual clusters and aggregate fleet entries, as identified in the row labels and sources. Compare the same scope and deployment status. GPU generation, memory, networking, and workload also affect performance, so twice as many GPUs does not automatically mean twice as much useful compute.
Citation estimates use inputs through September 1, 2026; fleet counts have an October 1, 2026 cutoff. The research-topic sample covers January 1 to June 1, 2025. See each answer for its period and limitations.
