Circular AI Financing Draws Scrutiny as Broadcom Lends Anthropic Up to $42 Billion and Nvidia Invests Over $40 Billion in AI Customers
The AI boom is increasingly running on circular money, and it’s reminding some investors of 2008, per Bloomberg. Chipmakers and cloud giants invest in or lend to AI companies, which then use that money to buy chips and computing power from the very same companies. Recent examples: - Broadcom, $AVGO, agreed to lend Anthropic up to $42 billion to buy chips and computing hardware - Nvidia, $NVDA, has put more than $40 billion into AI companies this year, including $30 billion in OpenAI.
Context
Vendor financed compute deals allow AI labs to scale multi gigawatt clusters ahead of cash flow, but they tie chipmaker revenues and equity valuations directly to the solvency of their own borrowers.
Throughout 2026, semiconductor suppliers and cloud hyperscalers have increasingly combined equity investments and debt facilities with hardware supply contracts for frontier AI labs. According to reporting by Bloomberg and Reuters, these arrangements create a capital loop in which chipmakers and cloud providers fund AI startups that immediately spend the proceeds leasing or purchasing chips and compute from the same backers. Analysts have compared the structure to vendor financing during the late 1990s telecom buildout and capital concentration patterns prior to the 2008 financial crisis.
Deep dive
Citing Bloomberg reporting, financial account unusual_whales highlighted growing concern among Wall Street analysts and investors that the AI infrastructure boom is increasingly powered by circular financing. Under this model, chipmakers and cloud providers invest in or extend credit to AI model developers, which then use that capital to purchase or lease semiconductors and cloud capacity from the same suppliers.
Two major 2026 disclosures anchor the discussion. First, according to Anthropic IPO prospectus details reported by Reuters and Bloomberg Law, chip designer Broadcom agreed to lend Anthropic up to $42 billion to lease AI chips and equipment, with the debt instruments structured to be convertible into Anthropic shares. That financing supports an expanded partnership announced in April 2026 among Anthropic, Broadcom, and Google for roughly 3.5 gigawatts of Tensor Processing Unit capacity coming online in 2027, which is projected to make Anthropic Broadcom's largest compute customer. Second, Nvidia has deployed more than $40 billion into AI companies during 2026, including a $30 billion investment in OpenAI alongside backing for dozens of neocloud and application startups that spend heavily on Nvidia GPUs.
According to Bloomberg's analysis by Shawn Donnan, this self reinforcing capital loop attracts global institutional cash and inflates reported hardware demand, but it also concentrates systemic risk. Skeptics warn that if end user enterprise revenue fails to keep pace with capital commitments, a pullback by any major lab could trigger cascading write downs across chipmakers, cloud providers, and private credit lenders.
What to expect next
As Anthropic prepares for its reported public listing in November and public chipmakers report quarterly results, investors and regulators will scrutinize how much data center revenue is backed by vendor loans, convertible notes, or reciprocal equity stakes rather than independent customer cash flows. Analysts will also track whether credit rating agencies adjust risk weights on debt facilities tied to circular hardware purchase commitments.