openaienmodel: gpt-5-mini-2025-08-07
Anthropic expands partnership with Google and Broadcom for multiple gigawatts of next-generation compute
Key Points
- multi-gigawatt TPU capacity arriving from 2027
- multi-cloud: AWS, Google Cloud, Azure
- run-rate revenue surpassed $30B; >1,000 $1M+ customers
Summary
Anthropic signed a multi-gigawatt agreement with Google and Broadcom for next-generation TPU capacity expected to come online starting in 2027. The majority of the capacity will be sited in the United States and will expand Anthropic's backend for its frontier Claude models, supporting rapid customer growth and higher run-rate revenue. Anthropic continues to train and run Claude across AWS Trainium, Google TPUs, and NVIDIA GPUs and remains available across AWS (Bedrock), Google Cloud (Vertex AI), and Microsoft Azure (Foundry).
Key Points
- Timeline: multiple gigawatts of TPU capacity expected to begin coming online in 2027.
- Location: the vast majority of new capacity will be sited in the United States.
- Multi-vendor hardware: workloads will be matched across Google TPUs, AWS Trainium, and NVIDIA GPUs for performance and resilience.
- Multi-cloud availability: Claude remains available on AWS Bedrock, Google Vertex AI, and Microsoft Azure Foundry.
- Operational impact: anticipate significant increase in available TPU capacity for training/inference pipelines and update capacity planning and provisioning timelines accordingly.
- Engineering guidance: evaluate workload placement by latency, memory footprint, and cost to select TPU vs Trainium vs GPU; test model performance on target hardware ahead of migration.
- Partnerships: deepens relationships with Google Cloud and Broadcom while Amazon remains the primary cloud/training partner (Project Rainier ongoing).