Summary: Anthropic will access around 3.5 gigawatts of Google TPU compute capacity via Broadcom from 2027, marking its largest infrastructure move yet, as its revenue surpasses $30bn, tripling from $9bn at 2025’s end.
Anthropic announced securing significant gigawatts of compute power through a Google-Broadcom agreement while reporting substantial revenue growth, reflecting their now monumental infrastructure needs. The agreement, unveiled on 6 April 2026, allows Anthropic to use approximately 3.5 gigawatts of Google’s TPU capacity via Broadcom starting in 2027, building on an existing 1 gigawatt supply in 2026.
Krishna Rao, CFO of Anthropic, described it as their “most significant compute commitment to date,” reflecting a disciplined scaling strategy. Most of this additional capacity will be U.S.-based, extending Anthropic’s November 2025 $50bn American AI infrastructure investment commitment.
Three-Party Infrastructure
The announcement heavily involves Broadcom in this infrastructure setup. Broadcom serves as a middle layer between Google’s silicon and Anthropic’s needs while holding a separate long-term agreement with Google for future custom TPU chip design and supply, and components for AI data racks through 2031.
This positions Broadcom as an essential player in AI infrastructure. Without engaging in AI model creation, Broadcom constructs the silicon and interconnects essential for AI. Consequently, Broadcom shares rose about 3% following the news, indicative of investor interest in foundational AI technology providers. Mizuho, led by Vijay Rakesh, projected $21bn from Anthropic in 2026 for Broadcom, surging to $42bn in 2027.
Broadcom hinted at the scale of its relationship with Anthropic in September 2025 when Hock Tan revealed a $10bn order from a mystery customer for custom TPU racks, which he later confirmed was Anthropic alongside an additional $11bn order.
Revenue and Customer Growth
The compute agreement aligns with Anthropic’s commercial expansion, with its ongoing revenue now exceeding $30bn, rising from about $9bn by end of 2025. This impressive growth, more than tripling in three months, follows accelerated enterprise sales and Anthropic securing $30bn in its Series G funding round closed on 12 February 2026, achieving a $380bn valuation.
Over 1,000 customers now spend over $1m annually, doubling in less than two months. This surging enterprise adoption necessitates more compute, which in turn requires increased infrastructure investment.
Multi-Cloud Architecture
Anthropic’s infrastructure plan includes a multi-vendor strategy, deploying Claude across three platforms: Amazon’s Trainium, Google’s TPUs, and Nvidia GPUs. Anthropic highlights Claude as the sole model available on AWS, Google Cloud, and Microsoft Azure.
This strategy offers resilience and bargaining power. Workloads can be moved if capacity or supply on a platform is challenged. Microsoft mirrors this approach with its AI models, hedging against single-vendor reliance.
Amazon remains Anthropic’s primary cloud partner since 2024, with $8bn invested. The Rainier project will increase Amazon Trainium 2 chips in use by the end of 2025. The Google partnership extends with Broadcom’s involvement.
U.S. Infrastructure Focus
The April agreement is part of Anthropic’s November 2025 $50bn domestic infrastructure commitment, initially in cooperation with Fluidstack to deploy U.S.-based data centers. The enhanced Broadcom capacity will expand this commitment.
This focus aligns with the AI Action Plan prioritizing U.S. compute capacity, signaling both strategic intent and regulatory alignment.
The Compute Arms Race
The Anthropic-Google-Broadcom deal is part of a trend requiring expansive financial efforts for infrastructure. For Broadcom, involvement with Anthropic illustrates its growing significance in AI hardware development alongside other major tech players like Nvidia.