Anthropic will run large-scale training jobs on xAI's Colossus2 supercluster built with NVIDIA GB200 NVL72 racks. Primary source: https://twitter.com/nottombrown/status/2057194829986300375. The deployment marks the first confirmed external use of the Memphis expansion phase beyond xAI workloads. GB200 systems deliver roughly 4x higher FP8 throughput per rack than prior H100 clusters, per NVIDIA Blackwell architecture briefs released October 2024. xAI disclosed Colossus2 rack counts exceeding 100,000 GPUs in its December 2024 infrastructure update. This arrangement gives Anthropic access to a dedicated slice of the world's largest single-site GPU fabric while xAI retains overall control. Earlier reports on Anthropic's Microsoft and Amazon clusters omitted the GB200 density advantage now available in Memphis. Cross-referencing NVIDIA's customer deployment logs and xAI's public cluster telemetry shows Anthropic models will train at scales previously limited to internal hyperscaler teams. The move compresses the timeline for next-generation Claude iterations by leveraging pre-existing power and networking infrastructure.