Broadcom is assembling a financing package worth roughly $60 billion to fund the development and supply of custom AI chips for Anthropic, according to a report from The American Bazaar.

The figure, if confirmed, would rank among the largest financing commitments ever tied to a single custom-silicon customer relationship, and it signals how far Broadcom is willing to go to lock in the next generation of AI accelerator programs beyond its existing hyperscaler base.

Broadcom has built its AI franchise on custom accelerators designed with large cloud providers, using its IP portfolio in networking, serdes and memory interconnect alongside its long-standing partnership with TSMC on advanced process nodes. The reported Anthropic deal would extend that model to a frontier AI lab whose compute demands have grown steeply as it scales its Claude model family and API business.

The structure described in the report — a financing package rather than a straightforward design contract — suggests Broadcom would help underwrite the enormous cost of capacity and components needed to bring a custom chip program to volume production. Cutting-edge AI accelerators are typically fabricated on leading-edge nodes and packaged with high-bandwidth memory, and securing both foundry allocation and HBM supply routinely requires multi-year, multi-billion-dollar commitments.

Neither Broadcom nor Anthropic has publicly confirmed the size or terms of the arrangement, and the $60 billion figure should be treated as a report rather than a disclosed company figure. Both companies have declined or not yet responded to requests for comment on the matter, according to the report.

For Anthropic, a dedicated custom accelerator program would reduce dependence on merchant GPUs, which have dominated AI training and inference but remain supply-constrained and expensive. For Broadcom, the deal would deepen its exposure to AI silicon at a moment when investors increasingly value the company on the strength of its AI revenue rather than its legacy semiconductor and infrastructure-software lines.

The reported arrangement also fits a broader industry pattern: AI labs and cloud providers are shifting toward custom silicon to control cost per token, and suppliers able to combine chip design, networking and access to advanced manufacturing capacity are capturing disproportionate value from that shift. Broadcom's existing custom-chip relationships with major cloud operators have already made it one of the largest beneficiaries of AI infrastructure spending outside the GPU vendors themselves.

Financing at this scale would also carry supply-chain implications. It would effectively pre-commit capital toward foundry wafers, advanced packaging capacity and high-bandwidth memory years ahead of delivery, tightening an already constrained allocation environment for leading-edge AI silicon.

Whether the package closes at the reported size, and how the capital translates into shipped accelerators, will shape the competitive balance between custom silicon and merchant GPUs in the next phase of AI infrastructure build-out.