Broadcom is providing a $42 billion loan facility to Anthropic to finance AI training capex and custom AI chip development.

Anthropic's recent IPO prospectus leak has revealed the true cost of building frontier artificial intelligence. Despite strong revenues, the company faces substantial cash burn to secure the computing infrastructure necessary to compete with rivals like OpenAI. To address this challenge, Broadcom has agreed to lend Anthropic $42 billion through a convertible debt arrangement—an unprecedented financing structure designed to accelerate the startup's infrastructure buildout.

The deal reflects a strategic partnership between the two companies. Anthropic has committed to a $125.2 billion five-year lease for Tensor Processing Unit capacity, with Broadcom's $42 billion loan covering approximately one-third of that obligation. The arrangement secures Anthropic the raw computing power it requires while guaranteeing Broadcom a substantial and reliable revenue stream. Broadcom and Google are supplying next-generation TPUs beginning in 2027.

This financing strategy mirrors NVIDIA's approach over the past several years. NVIDIA has used its balance sheet to fund ambitious AI startups with substantial cash infusions, enabling them to purchase NVIDIA GPUs in return. Broadcom is executing the same playbook: by extending credit to Anthropic, the company locks the startup into its hardware ecosystem for years. Anthropic is projected to become Broadcom's largest compute customer by 2027.

Anthropic's upcoming public offering could value the company at $2 trillion, and the arrangement carries regulatory scrutiny. The company disclosed in its filing that Broadcom's dual role—as both hardware supplier and primary debt financier—creates material conflicts of interest. If Anthropic fails to meet performance metrics, substantial portions of its lease obligations could become immediately due, creating significant financial risk.

The AI infrastructure race is consolidating around a handful of massive institutional players. Building frontier AI models requires multiple gigawatts of power and thousands of advanced chips. Broadcom's willingness to commit $42 billion demonstrates how lucrative the custom AI chip market has become. This scale of infrastructure investment is historically significant, reflecting the foundational buildout of the next era of computing.