Anthropic Secures Multi‑Gigawatt TPU Deal with Google and Broadcom in LLM Push

Anthropic will begin deploying the new TPU capacity in early 2027, with most of the hardware slated for U.S. data centers, the company said in a statement to The‑Decoder. The multi‑gigawatt allocation expands the firm’s existing Google Cloud partnership, which was first announced in October 2025, and adds Broadcom as a silicon supplier for the next‑generation TPU pods.

The compute boost comes as Anthropic’s annualized revenue run‑rate has surged past $30 billion, up from roughly $9 billion at the end of 2025, according to the same source. The revenue jump reflects a rapid influx of enterprise contracts: customers generating more than $1 million in annual spend have doubled since February, now exceeding 1,000 accounts.

Anthropic continues to train its Claude models on a heterogeneous mix of silicon. It uses Amazon’s AWS Trainium chips, Google’s TPUs, and Nvidia GPUs, making Claude the only major foundation model available across all three leading cloud platforms—AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure AI, The‑Decoder notes. Despite the new Google‑Broadcom deal, Amazon remains Anthropic’s primary cloud partner for both training and inference workloads.

The agreement also deepens Anthropic’s strategic alignment with Broadcom, which will supply the custom interconnects and networking ASICs needed to stitch together the massive TPU clusters. Broadcom’s involvement is expected to improve power efficiency and latency, key factors for scaling Claude’s next‑generation models, the report adds.

Analysts see the multi‑gigawatt commitment as a signal that Anthropic is positioning itself to compete head‑to‑head with OpenAI and Google’s own Gemini models on both performance and reach. By securing hardware across all three cloud ecosystems, Anthropic aims to lock in a distribution moat that could translate into further enterprise growth, The‑Decoder writes.