- Broadcom is in talks to raise more than $60 billion in debt to help Anthropic and others secure chips and compute, according to Bloomberg.
- The financing highlights how much of the AI buildout is now funded with debt rather than equity.
- Analysts flagged concern about “circular financing,” where chipmakers, clouds, and labs fund one another.
- The report landed the week before Nvidia’s next earnings, a barometer for AI hardware demand.
What Happened
Broadcom is in talks to raise more than $60 billion in debt to help Anthropic and other AI companies secure chips and computing power, Bloomberg reported on August 21, 2026. The scale of the prospective raise underscores a shift in how the AI infrastructure boom is being financed, moving from venture equity toward large debt facilities carried on suppliers’ balance sheets.
Why It Matters
The cost of funding the AI buildout has become the central concern hanging over the trade. On Bloomberg Tech, Science and Technology Partners founder and chief investment officer Erica Klauer discussed the risk of “circular financing,” in which chipmakers, cloud providers, and model labs increasingly fund one another’s purchases. When a supplier borrows to finance a customer’s ability to buy its own products, reported revenue can look stronger than the underlying demand justifies — a dynamic that amplifies both upside and downside across the sector.
The pattern is now widespread. Oracle took on tens of billions in borrowing to build data centers tied to OpenAI’s Stargate program; Nvidia has invested in customers such as CoreWeave and xAI that in turn buy its chips; and specialist clouds have financed GPU fleets with debt secured against the hardware itself. A $60 billion Broadcom facility aimed at customer compute would be among the most direct examples yet of a chip supplier underwriting demand for its own products.
Technical Details
A debt raise above $60 billion would rank among the largest corporate financings tied to AI infrastructure to date. The capital is aimed at securing accelerators and the data-center capacity to run them — the two bottlenecks for scaling frontier models — where a single large training cluster can cost well into the billions once power, networking, and memory are included. Because debt carries fixed repayment obligations regardless of how model revenue develops, the structure raises the stakes on utilization: the chips and facilities must generate enough return to service the borrowing, on a schedule that does not flex if demand slows.
Who’s Affected
Anthropic and other labs gain access to compute they could not fund from cash flow alone, at the cost of new obligations in the supply chain they depend on. Broadcom deepens its position as a financier as well as a supplier of custom silicon. Bondholders and the banks arranging the facilities take on exposure to AI demand, and Nvidia — reporting earnings next week — remains the clearest public gauge of whether that demand is holding.
What’s Next
The talks had not closed as of the report, and terms were not finalized. The signals to watch are whether the facility is completed at the reported size, the interest rate it carries, and whether other suppliers follow Broadcom into large debt raises to fund customer purchases.
