Reuters reporting on Anthropic's IPO prospectus has surfaced a detail that reframes how deeply intertwined the AI buildout has become with its own financiers: Broadcom has agreed to lend Anthropic up to $42 billion to help fund infrastructure spending. The arrangement sits alongside existing chip supply and equipment leasing agreements between the two companies, and it positions Anthropic to become Broadcom's largest compute customer by 2027. A separate, five-year commitment tied to Google-designed TPUs, supplied through Broadcom, is valued at $125 billion.
What the Prospectus Actually Shows
IPO prospectuses are meant to give prospective investors a clear-eyed view of a company's finances, obligations, and risks. What this one reveals is a lender-customer relationship that runs in multiple directions at once. Broadcom isn't simply selling chips to Anthropic - it's financing the infrastructure spending that allows Anthropic to buy more chips, lease more equipment, and expand its compute footprint. That layered structure, where a supplier also acts as a creditor, is the kind of detail that typically draws close scrutiny from analysts trying to understand who holds the risk if revenue projections fall short.
The Circular Financing Problem
Commentators have increasingly flagged a pattern across the AI sector: large technology firms extending credit, committing to long-term purchases, or taking equity stakes in the very companies that spend that money back on their own products and services. Broadcom's loan to Anthropic fits this description closely. The companies acknowledge on air that it isn't clear whether the arrangement sits on or off balance sheet, and that ambiguity matters. Off-balance-sheet financing can understate the real leverage embedded in a company's accounts, making it harder for outside observers to judge how exposed a lender truly is if a borrower underperforms.
Why the Repayment Question Looms Large
The arrangement only works if Anthropic's revenue eventually catches up to its spending. That is far from settled. The same prospectus reporting has drawn attention to the scale of Anthropic's recent losses, and separate research from firms including Bain and Company, along with commentary from economist Torsten Slok, has pointed to a widening gap between what AI companies are spending on infrastructure and what they currently have the revenue to support. For Broadcom, the deal looks attractive on paper: guaranteed chip demand, a locked-in leasing relationship, and interest income on the loan. But all of that depends on Anthropic generating enough commercial revenue to service $42 billion in financing commitments, not just to in the near future, but over a sustained multi-year period.
Implications Beyond the Two Companies
This deal is instructive for anyone trying to assess the health of the broader AI infrastructure boom. When suppliers become lenders, and lenders become customers, the usual market signals - pricing, demand, creditworthiness - get harder to read independently. Investors in Broadcom, and eventually in Anthropic once it goes public, will need visibility into how these obligations are structured, how much risk sits where, and what happens if growth disappoints. None of this suggests imminent failure. It does suggest that the AI capital structure has become unusually interconnected, and that a disruption at one node - a missed payment, a revenue shortfall, a shift in chip demand - could ripple further than a simple vendor relationship would.