“Circular financing” describes a pattern that emerged in the 2025 AI buildout: a supplier of chips or cloud capacity invests in an AI lab, and the lab then uses that money to buy the supplier’s own products. Cash loops among a small set of interconnected firms, which can make demand look organic and revenue look robust even when much of it is the same dollars circulating through the loop.
The pattern is visible in the labs’ own announcements. On September 22, 2025 NVIDIA said it intended to invest up to 100 billion dollars in OpenAI to deploy 10 gigawatts of NVIDIA systems - that is, NVIDIA putting money into a company whose largest expense is buying NVIDIA chips. Days earlier, on October 6, 2025, AMD announced a 6-gigawatt GPU deal with OpenAI that included a warrant for up to 160 million AMD shares, again tying a supplier’s equity to a customer’s purchases. Layered on top were OpenAI’s multi-hundred-billion-dollar cloud commitments to Microsoft and Oracle, themselves major buyers of the same chips.
Critics argue this resembles vendor financing from past bubbles, where a supplier effectively lends customers the money to buy its goods, inflating reported demand. Defenders argue these are rational long-term commitments in a capacity-constrained market, where suppliers want to guarantee that the customers who anchor their growth can actually pay and build. The truth depends on whether the underlying AI revenue eventually materializes - the same open question raised by the “$600B” essay.
Why a business reader should care: circular financing makes it harder to read AI demand at face value. When the same handful of companies are simultaneously each other’s investors, suppliers, and customers, headline deal sizes and revenue figures can overstate how much genuinely independent demand exists - a key reason the bubble debate is so hard to settle.