Broadcom reveals its mystery $10 billion customer is Anthropic

In a recent earnings call, Broadcom disclosed that it has secured a monumental $10 billion order from Anthropic, an AI research lab. This significant deal involves the supply of the latest Google tensor processing units (TPUs), a pivotal component in the booming field of artificial intelligence. CEO Hock Tan revealed the identity of the mystery customer during Broadcom's fourth-quarter earnings call. He confirmed that the order was specifically for the advanced TPU Ironwood racks, and additionally, Anthropic has placed another order worth $11 billion in the latest quarter. This revelation has captured the interest of investors, particularly as the AI infrastructure market continues to flourish. Typically, Broadcom remains tight-lipped about its major clients, but the mention of the $10 billion order earlier in September had already piqued curiosity among stakeholders. Notably, a Broadcom representative clarified in October that Anthropic is not affiliated with OpenAI, which has a separate agreement with Broadcom for chip procurement. Broadcom specializes in custom chips, known as ASICs, which some experts suggest may provide greater efficiency for specific AI algorithms compared to the widely used Nvidia chips. The company collaborates with Google in the production of TPUs, and recently, Google highlighted that its cutting-edge Gemini 3 AI model was exclusively trained using its TPUs. On the earnings call, Tan emphasized that the delivery to Anthropic includes entire server racks, marking Anthropic as Broadcom's fourth customer for its XPUs, the designation for its custom AI chips. Furthermore, Tan announced that a fifth customer has also placed a $1 billion order, though their identity remains undisclosed. In a significant move for the AI sector, Anthropic and Google recently forged a comprehensive cloud partnership valued in the tens of billions. This collaboration grants Anthropic access to approximately 1 million TPUs, expected to generate over a gigawatt of new AI computing power by 2026. Adopting a multi-cloud strategy, Anthropic utilizes a diverse range of computing resources, balancing workloads across Google’s TPUs, Amazon’s Trainium chips, and Nvidia’s GPUs. This flexible approach allows the startup to optimize its models for training, inference, or research based on the most suitable platform. For Google, Anthropic's commitment to TPUs underscores the strong performance and efficiency advantages that the startup has experienced over the years. Google Cloud CEO Thomas Kurian noted that Anthropic's significant expansion in TPU utilization is a testament to the value these chips offer. As Google evolves its TPU offerings into a cloud service rather than mere hardware sales, analysts suggest that these specialized chips could play a crucial role in driving growth within Google's cloud business. With power limitations emerging as a key issue for AI development, Broadcom's tailored ASICs may provide a much-needed alternative to Nvidia's GPUs, enhancing Google's competitive edge in the fast-paced AI landscape.