Broadcom has put a $42 billion loan proposal on the table for Anthropic, structuring the financing around AI chip leases, according to a Yahoo Finance report.

The offer ties the loan's deployment directly to lease lines covering AI accelerators. Broadcom's custom-silicon business has become a central node of the AI infrastructure build-out, shipping ASICs and the high-speed networking fabric that connects accelerator clusters at hyperscale customers.

What does the deal look like?

The headline figure is $42 billion, framed as a loan rather than equity or a direct chip purchase. The lease-based structure echoes financing models already circulating through the data-center build-out.

Under those arrangements, operators secure long-term access to capacity without taking immediate title to the underlying silicon. Broadcom would operate as both chip supplier and credit provider, capturing value across multiple layers of the stack.

The structure also lets Anthropic secure compute capacity that is otherwise constrained by allocation queues at the major accelerator vendors.

Why is Anthropic the counterparty?

Anthropic, the AI model developer behind the Claude family of frontier models, has run an aggressive compute-expansion program. The company competes for share against OpenAI, Google DeepMind and a small set of other frontier labs.

Anthropic has used a mix of Google TPUs, Amazon Trainium chips and Nvidia GPUs to train and serve its models. A loan of this magnitude, tied to chip leases, would fund additional accelerator deployments without forcing further equity dilution or large cash purchase orders at current accelerator prices.

What is Broadcom's role in custom AI silicon?

Broadcom has emerged as one of two principal partners — alongside Marvell Technology — for hyperscalers building custom accelerators. The company co-develops ASICs with customers including Google, Meta and ByteDance.

Broadcom also supplies the high-speed Ethernet switching and routing silicon that connects thousands of accelerators into a single training fabric. That dual exposure — silicon plus networking — gives the firm an entrenched position in the AI capex cycle.

Recent earnings commentary has highlighted custom-AI silicon as a multibillion-dollar revenue line, with multi-year backlog commitments from multiple hyperscaler customers.

Why does the size matter?

A $42 billion commitment ranks among the largest single-vendor financing arrangements yet proposed to a model lab. The figure exceeds the cumulative quarterly AI-chip revenue reported by many smaller merchant accelerator vendors.

The scale also approaches full-year AI infrastructure budgets at some hyperscalers. Training runs for next-generation models now require tens of billions of dollars in accelerator spend, pushing labs beyond traditional equity raises into structured debt.

Vendor financing of this size blurs the line between chip supplier and capital provider — a pattern more familiar from the aircraft and heavy-equipment industries than from semiconductors.

What remains unclear?

The reported terms — including interest rate, tenor, collateral and which accelerator families the lease lines cover — were not disclosed. It is also unclear whether Anthropic has accepted the proposal or whether competing offers are on the table.

The structure may require co-financing from Anthropic's existing cloud partners, though no such arrangement was confirmed in the report.