Broadcom is helping finance Anthropic’s access to tens of billions of dollars worth of Google AI silicon as custom accelerators become industrial-scale infrastructure.
According to the Financial Times, the deal centres on Google’s Tensor Processing Units, with Anthropic committing heavily to TPU capacity as it expands the computing estate behind Claude.
Broadcom is deeply embedded in that hardware chain because it works with Google on custom AI silicon and is expected to become one of Anthropic’s largest chip design suppliers from 2027.
Anthropic has committed $125.2 billion over five years to leasing TPU computing capacity, showing how far frontier AI has moved from buying racks of servers to building infrastructure on the scale of a small national utility.
The current financing package reaches $60 billion. Of that, $42 billion is senior secured debt partly supported by Broadcom, while another $18 billion is planned as junior financing.
Anthropic needs access to vast quantities of advanced accelerators, networking, memory and data centre infrastructure to train and run increasingly large AI models.
Google’s TPUs are central to that expansion. Unlike general-purpose GPUs, TPUs are custom accelerators designed specifically around machine-learning workloads and deployed in tightly integrated Google infrastructure.
Broadcom’s involvement gives it exposure well beyond ordinary merchant silicon sales. It is helping design the hardware, supporting the financing and positioning itself for a large slice of Anthropic’s future compute requirements.
Anthropic has said it runs Claude across Google TPUs, AWS Trainium and Nvidia GPUs, giving it a deliberately mixed hardware strategy rather than chaining the entire operation to a single accelerator supplier.
The company said most of the new compute capacity from its expanded Google and Broadcom relationship would be located in the US.
That deployment builds on Anthropic’s earlier commitment to invest $50 billion in US computing infrastructure, with the latest arrangement pushing its hardware requirements considerably further.
Broadcom’s willingness to support up to $42 billion of Anthropic’s financing shows how peculiar the AI hardware market has become, with chip suppliers no longer simply waiting for customers to turn up with purchase orders. They are increasingly helping arrange the money required to deploy the accelerators they helped design.
The arrangement resembles Nvidia’s broader strategy, where financing and strategic investment can help customers build infrastructure that eventually consumes more silicon.
Anthropic is expected to become Broadcom’s largest custom chip design customer in 2027, while demand for Google-designed TPUs gives Broadcom another route into a market dominated by Nvidia.
The company has projected AI semiconductor revenue of $115 billion in 2027 and $230 billion in 2028, making custom accelerators rather more than an interesting sideline. The hardware bill is becoming so enormous that financing the chips is turning into part of the semiconductor business itself.
