Anthropic just locked in a major infrastructure boost. The company signed a new deal with Google and Broadcom for multiple gigawatts of next generation TPU capacity set to come online starting in 2027. This expansion powers frontier Claude models and meets surging customer demand. 🤖
The scale of the new compute agreement
Broadcom will supply roughly 3.5 gigawatts of TPU based computing capacity as part of the pact. The hardware comes from Google‘s custom tensor processing units which Broadcom helps design and manufacture. Anthropic already uses a mix of AWS Trainium Google TPUs and NVIDIA GPUs to match workloads to the best chips. The Wall Street Journal reported on the expanded collaboration details.
Krishna Rao Anthropic‘s chief financial officer described the move as a continuation of disciplined scaling. He noted it builds capacity for exponential customer growth while keeping Claude at the frontier of AI development. The vast majority of this new infrastructure will locate in the United States. That aligns with Anthropic‘s November 2025 pledge to invest 50 billion dollars in American computing. Reuters covered the long term chip supply agreement through 2031.
Growth driving the need for more power
Customer demand for Claude accelerated sharply in 2026. Anthropic‘s run rate revenue now tops 30 billion dollars up from about 9 billion dollars at the end of 2025. The number of business customers spending over 1 million dollars annualized jumped from 500 in February to more than 1000 today. Bloomberg highlighted the revenue surge and compute expansion.
This deal deepens Anthropic‘s existing ties with Google Cloud including last October’s TPU capacity increase. It also strengthens the relationship with Broadcom. Anthropic remains available on all three major clouds AWS Bedrock Google Cloud Vertex AI and Microsoft Azure Foundry with Amazon as the primary training partner. TechCrunch detailed how the partnership addresses skyrocketing demand.
Readers interested in AI infrastructure can explore related resources on BestAITools.com. The AI compute infrastructure overview breaks down hardware options. Check the Google Cloud TPU solutions directory for practical tools. For broader context see AI hardware comparisons and enterprise AI scaling strategies.
The agreement shows how quickly leading AI labs must secure power to keep up with real world usage. It leaves open questions about how this diversified approach will affect performance and costs over time.
