Broadcom tightens grip on AI infrastructure with Anthropic, Google deal
Broadcom’s latest agreement with Anthropic and Google strengthens the chipmaker’s position as one of the key firms sitting underneath the AI boom, as AI firms race to lock in even larger amounts of compute.
Under the new arrangement, Anthropic will have access to multiple gigawatts of Google tensor processing unit (TPU) capacity from 2027, while Broadcom will supply the infrastructure that enables it.
Anthropic said roughly 3.5 gigawatts will be involved, making this its biggest compute commitment yet.
Broadcom is already a major partner to Google on custom TPUs, and it has now extended that role through a long-term agreement covering future TPU generations and networking components for Google’s next AI racks.
This places Broadcom deeper into the physical layer of the AI market, where demand is now rising fastest, and customers are prepared to commit years in advance.
Anthropic’s revenue run-rate has now passed $30bn, up from about $9bn at the end of 2025, while the number of business customers spending more than $1m a year has doubled in under two months to more than 1,000.
Its chief financial officer, Krishna Roa, said the company was making its “most significant compute commitment to date” in order to keep pace with growth and support future Claude models.
Most of the additional compute capacity secured through the agreement is expected to be located in the United States, extending Anthropic’s earlier commitment to invest $50bn (£38bn) in domestic AI infrastructure.
Infrastructure demand drives long-term deals
The deal reflects an industry trend in which firms are increasingly entering into longer-term arrangements to gain access to the latest computing infrastructure.
By working with hyperscale providers like Google, Broadcom has positioned itself comfortably within that supply chain.
Its involvement in TPU production and deployment links it directly to the expansion plans of AI developers like Anthropic.
The firm had previously indicated that demand from Anthropic could exceed 3 gigawatts of compute capacity over time, highlighting the sheer scale of AI workload growth.
Partnerships of this type have been seen as a significant revenue opportunity for chipmakers, as spending on AI infrastructure shows no signs of slowing.
Anthropic’s multi-cloud approach
Anthropic is continuing to deploy its models across multiple hardware platforms, including Google TPUs, Nvidia GPUs, and Amazon Web Services (AWS) infrastructure.
Big Tech said this approach allows it to match workloads to different types of hardware and maintain flexibility as demand evolves.
Amazon remains its primary cloud and training partner, including through large-scale projects such as its AI supercomputing cluster.
The expanded agreement with Google and Broadcom adds further capacity alongside these existing tie-ups, as Anthropic scales its infrastructure to meet growing enterprise usage