Anthropic is ramping Claude inference on Colossus 2 using NVIDIA GB200 GPUs throughout June via its expanded SpaceX partnership. The May 2026 post from @nottombrown states the company will scale GB200 capacity and thanks Elon Musk for hosting Claudes. Primary source: https://xcancel.com/nottombrown/status/2057194829986300375.

GB200 deployments follow xAI's Colossus 1 build and NVIDIA's Blackwell architecture specs showing 4x training and 30x inference gains over Hopper. Related coverage from NVIDIA confirms GB200 NVL72 rack-level liquid cooling for 120 kW density. Source: NVIDIA Blackwell platform announcements.

OpenAI and Google have similarly secured custom clusters from CoreWeave and TPU v5 pods to address inference demand spikes, per 2025 earnings filings and cluster utilization reports. This pattern shows frontier labs prioritizing atom-moving speed and power density over traditional data center timelines.