Asset management giant Blackstone and three Wall Street banks—Bank of America, Citi, and Morgan Stanley—announced on October 26 a specialized financial vehicle (SPV) valued at $63 billion to finance the procurement of Broadcom-designed artificial intelligence accelerators and networking hardware, which will be leased back to Anthropic to support frontier model research.

The arrangement marks a significant shift in how technology firms fund hardware expenses, expanding earlier discussions regarding the TPU financing package between Anthropic and Blackstone to a substantially larger infrastructure scale.

Special Purpose Vehicle Mechanics and Broadcom’s Commitment

Under the unveiled structure, an independent special purpose vehicle (SPV) will be created to hold ownership of all AI servers and networking equipment supplied by Broadcom. The entity will raise debt financing from the four financial institutions and subsequently enter into a long-term lease agreement with Anthropic to deploy and operate the hardware across data centers.

A pivotal element mitigating lender risk is Broadcom’s agreement to provide a residual value guarantee on the batches of AI chips. This guarantee covers an amount equivalent to the entire first-lien financing tranche, ensuring that if compute demand subsides or the lease experiences disruptions, the banks can still recover their capital through direct backing from the semiconductor manufacturer.

Shift from Outright Purchases to Infrastructure Leasing

This arrangement does not constitute an entirely new hardware order, but rather serves as the financing framework for an existing computing capacity supply agreement signed between Anthropic and Broadcom. Deliveries of the custom AI chips are slated to officially commence in 2027.

The structure enables Anthropic to curb direct capital expenditures on its balance sheet ahead of an initial public offering, complementing similar long-term commitments such as the previously disclosed $45 billion infrastructure deal with Nscale. Rather than deploying tens of billions of dollars in cash for rapidly depreciating hardware, AI companies are pivoting toward recurring operating expense models.

Nevertheless, financial markets channeling tens of billions of dollars in debt backed by specialized chips as collateral have heightened concerns regarding the risk of a hardware credit crisis should the AI investment cycle run into bottlenecks in generating realized revenue.