Broadcom announced a long-term arrangement with Google to design and supply successive generations of custom artificial intelligence chips and associated components for Google’s forthcoming AI rack hardware through 2031. The agreement covers development and supply for Google’s next-generation AI infrastructure, but the companies did not disclose financial terms.
In a separate deal, Broadcom also agreed to provide Anthropic - the AI startup - with access to about 3.5 gigawatts of AI computing capacity that will draw on Google’s AI processors, with provision of that capacity slated to begin in 2027. Terms of the Anthropic arrangement were likewise not disclosed.
Broadcom’s stock reacted positively to the announcements, rising about 3% in extended trading.
Demand for specialised chips used for AI workloads, such as Google’s tensor processing units (TPUs), has risen sharply as businesses look for alternatives to high-cost graphics processors from Nvidia. Google has been working to position its TPUs as a viable alternative to Nvidia’s market-leading GPUs, and TPU sales have grown into an important driver of Google Cloud revenue as the company seeks to demonstrate to investors that its AI investments are producing returns.
Anthropic said the new agreement with Broadcom builds on the startup’s broader commitment to invest $50 billion in strengthening U.S. computing infrastructure. The company reported that demand for its AI model Claude accelerated in 2026, with run-rate revenue now exceeding $30 billion, up from roughly $9 billion at the end of 2025.
Anthropic also noted that it trains and operates Claude across a mix of AI hardware, including Amazon Web Services’ Trainium, Google TPUs, and Nvidia GPUs. The company said that Amazon remains its primary cloud provider and training partner.
Key context and implications: The Broadcom-Google engineering and supply deal runs through 2031 and focuses on custom processors and components for next-generation AI racks. The separate Broadcom-Anthropic agreement provides about 3.5 gigawatts of Google-processor-based compute capacity beginning in 2027. Financial terms were not disclosed.
The announcements underscore growing market activity around custom AI hardware and the multi-party relationships between chip designers, cloud infrastructure providers, and AI model developers. Companies across cloud services, chip manufacturing, and AI model providers are implicated by shifts in demand for specialised AI compute.