Anthropic secures 3.5GW Google TPU deal as annualized revenue hits $30 billion

Anthropic has disclosed that its annualized revenue run rate has surged past $30 billion, more than tripling from $9 billion at the end of 2025, while announcing its largest-ever compute deal: access to approximately 3.5 gigawatts of next-generation Google TPU capacity starting in 2027, supplied through Broadcom.^[1][2][3][4] The agreement, revealed in a Broadcom regulatory filing on April 6, 2026, expands Anthropic's existing partnerships with Google and Broadcom to fuel the explosive demand for its Claude AI models.^[1][2] According to Anthropic CFO Krishna Rao, this "groundbreaking partnership" represents the company's most significant infrastructure commitment yet, aimed at sustaining "exponential growth" in its customer base.^[2][3][4]

The deal builds on Anthropic's prior commitments, including a 1 gigawatt supply of TPUs already ramping up in 2026 and an October 2025 expansion with Google Cloud for up to one million TPUs worth tens of billions of dollars.^[1][5] Broadcom plays a central role as the intermediary, designing and supplying custom Tensor Processing Units (TPUs) to Google under a separate long-term agreement, along with networking gear and components for Google's AI data racks through 2031.^[1][2] As reported by Bloomberg, this positions Broadcom as a key enabler for major AI systems like Google's Gemini and Anthropic's Claude, transforming it from a lesser-known player in AI hardware two years ago into a critical infrastructure provider.^[1]

Skyrocketing demand is driving these moves, with Anthropic doubling its business customers to over 1,000 in just two months and imposing caps on Claude Code tokens due to overwhelming usage.^[3] The company emphasized that the vast majority of the new 3.5 gigawatts will be located in the United States, expanding a November 2025 pledge to invest $50 billion in American computing infrastructure.^[4] This multi-gigawatt scale underscores the immense power needs of frontier AI training and inference, equivalent to the output of several large nuclear plants.

Anthropic maintains a diversified hardware strategy to optimize performance and resilience, training and running Claude on AWS Trainium chips, Google TPUs, and Nvidia GPUs.^[3][4][5] Amazon remains its primary cloud provider and training partner via Project Rainier, a massive cluster with hundreds of thousands of AI chips across U.S. data centers.^[4][5] Claude is uniquely available on all three major cloud platforms: AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry, serving everyone from Fortune 500 firms to AI startups.^[3][4]

For Broadcom and Google, the partnerships lock in revenue streams and technological leadership amid fierce competition in AI infrastructure.^[1][2] Anthropic's growth trajectory—fueled by Claude's capabilities—highlights how AI demand is reshaping global compute markets, with companies racing to secure capacity years in advance.^[1][3] Next steps include the new TPUs coming online in 2027, further TPU generations from Broadcom through 2031, and ongoing expansions to match usage spikes, as TechCrunch and Slashdot reported on the surging run-rate revenue.^[3] This deal signals sustained investment in U.S.-based AI infrastructure, potentially influencing energy demands, job creation in tech hubs, and the pace of AI innovation worldwide.^[4]