Anthropic hits $30 billion run rate, clinches Broadcom partnership

Anthropic’s disclosed revenue run rate now exceeds $30 billion, a three‑fold increase from the $9 billion figure reported at the end of 2025, according to Bloomberg. The company attributes this acceleration to a surge in enterprise‑grade deployments of its Claude‑series models, which have been integrated into a growing number of large‑scale SaaS platforms and internal business workflows. Bloomberg notes that Anthropic has formalized a partnership with Broadcom Inc. and Google to “power its burgeoning operations,” suggesting a joint effort to scale the compute infrastructure required for its next‑generation language models.

The Broadcom component of the deal is expected to involve the semiconductor maker’s high‑performance ASICs and networking silicon, which are engineered for low‑latency, high‑throughput AI workloads. While Bloomberg does not detail the specific product lines, Broadcom’s recent portfolio includes the Trident and Tomahawk series of Ethernet switches, as well as custom AI accelerators that can offload transformer inference from general‑purpose CPUs. By leveraging these chips, Anthropic can reduce the per‑token inference cost and improve the latency profile of its API services, a critical factor for enterprise customers with strict service‑level agreements.

Google’s role, as reported by Bloomberg, centers on providing cloud compute capacity and the underlying software stack that orchestrates large‑scale model training and serving. Google Cloud’s Tensor Processing Units (TPUs) and its proprietary Kubernetes‑based AI orchestration platform are likely to be the backbone of Anthropic’s expanded deployment pipeline. The integration of Google’s cloud services with Broadcom’s silicon could enable Anthropic to achieve higher model parallelism and more efficient data movement across distributed training clusters, thereby shortening the time required to iterate on model improvements and to roll out new capabilities to customers.