- Anthropic said its annualized revenue run rate has exceeded $30 billion, up from $9 billion at the end of 2025 — a more than threefold increase in roughly one quarter.
- Broadcom confirmed on April 6, 2026 that it has reached an agreement to supply Anthropic with Google-designed Tensor Processing Unit (TPU) chips.
- The three-party arrangement has Google designing the TPU architecture, Broadcom manufacturing and routing the chips, and Anthropic deploying them for AI workloads.
- Financial terms and TPU generation were not disclosed; Bloomberg’s reporting does not specify whether the run rate figure is gross or net revenue.
What Happened
Anthropic PBC disclosed that its annualized revenue run rate has surpassed $30 billion, according to Bloomberg reporting published April 6, 2026. That figure is more than triple the $9 billion run rate Anthropic reported at the close of 2025. Simultaneously, Broadcom Inc. confirmed it has struck a deal to procure and ship Google-designed Tensor Processing Units to Anthropic, adding a major infrastructure development to the same disclosure.
Why It Matters
Moving from a $9 billion to a $30 billion annualized run rate in approximately one quarter reflects sustained growth in enterprise API adoption and signals that Anthropic’s Claude model family is reaching significant commercial scale. The company has historically sourced compute capacity primarily through Nvidia GPU hardware and cloud agreements with Amazon Web Services and Google Cloud.
Google committed more than $3 billion to Anthropic by early 2025 as part of a broader strategic investment. The Broadcom-mediated TPU agreement extends that relationship from cloud credits to custom silicon supply, a structurally different form of infrastructure dependency.
Technical Details
Google’s Tensor Processing Units are application-specific integrated circuits (ASICs) purpose-built to accelerate the matrix multiplications central to transformer-based model training and inference. Unlike general-purpose GPUs — including Nvidia’s H100 and H200 — TPUs are not available on the open market; Google controls both the architecture and production pipeline. Broadcom manufactures TPUs under contract for Google and, under the newly confirmed arrangement, will route a portion of that supply directly to Anthropic.
The specific TPU generation involved was not identified in Bloomberg’s reporting. Google’s current production line includes the Cloud TPU v5e, rated at up to 459 teraflops of bfloat16 performance per chip, and the v5p, targeted at large-scale training runs. The choice of generation would have material implications for whether Anthropic is using TPUs primarily for inference, training, or both.
Who’s Affected
The deal most directly affects Anthropic’s infrastructure capacity and the enterprise customers — including AWS, which resells Claude models through its Bedrock platform — that depend on Anthropic’s inference throughput. If TPU deployment expands available capacity or reduces per-token costs, API pricing for downstream developers could shift. Nvidia, which has served as the default chip supplier for frontier AI labs, faces incremental pressure from this arrangement, though the volume of TPUs involved remains undisclosed.
What’s Next
Broadcom has not released the financial terms, volume, or duration of the TPU supply agreement. Anthropic has not confirmed a production timeline for deploying TPU-based infrastructure. The company is also expected to face continued capital expenditure pressure as run rate growth typically requires proportional increases in compute provisioning.
