AI Circular Deals: $1.4 Trillion of IOUs and a Meta Stock Sale

AI circular deals are arrangements in which a supplier invests in, lends to or guarantees a customer that then spends the money on the supplier's own chips or cloud services. The International Monetary Fund counted 14 publicly disclosed deals since 2023 that link about $300 billion of equity and $650 billion of debt to $1.4 trillion of future revenue commitments. The short answer to whether they matter: they do not prove a bubble on their own, but they tie the fortunes of the biggest AI companies together, so a disappointment at one would show up in several balance sheets at once.

The question got sharper in the last week of October. Nvidia and AMD bought $4 billion of Meta stock in Meta's $25 billion share sale, OpenAI's latest funding round fell well short of its target, and the Nasdaq Composite closed out its worst month since March 2025. This analysis lays out what the deals are, how large they have become and what would turn them from a growth engine into a risk.

The IMF has put a size on the loop

In its Global Financial Stability Report published on October 13, the IMF looked at 14 publicly disclosed transactions since 2023. In them, investors led by the hyperscalers put about $300 billion of equity into financially weaker AI companies, backed by $650 billion of debt, in exchange for $1.4 trillion of future revenue commitments. About $320 billion of that debt sits with the companies receiving the money, including debt backed by investor guarantees, contingent financing and other off-balance-sheet borrowing.

The ratio is the point. Every dollar of equity in these deals is attached to almost five dollars of promised future spending. The investors are paid back only if the companies they funded grow into those commitments. If AI revenue arrives more slowly, the equity stakes fall in value at the same moment the revenue they were meant to secure shrinks, which is why the IMF treats the structure as a market stability issue rather than a single-company one. The same report notes that hyperscalers now issue about 10% of investment-grade bonds, up from under 1% in 2024.

Chipmakers are now buying their customer's stock

The newest example is Meta. On October 26, Meta priced 40 million Class A shares at $625 each, raising $25 billion before fees, according to its prospectus filed with the SEC. The filing names Nvidia, AMD and Broadcom as cornerstone investors that indicated interest in buying up to $5 billion of the shares, and it states that those indications were not binding. Nvidia said it would buy $3 billion and AMD $1 billion, 24/7 Wall St. and TipRanks reported; no Broadcom purchase has been confirmed.

All three chipmakers sell hardware to Meta, which is spending heavily on AI data centers. In effect, part of the money Meta raises to buy chips comes from the companies selling the chips. Demand for the offering was strong, at about five times the shares on sale according to people familiar with the deal cited by 24/7 Wall St. and TipRanks. Separately, S&P Global has moved Meta's credit outlook to negative over its spending, which we covered in Meta credit rating outlook cut to negative.

Nvidia's guarantees reach into OpenAI's data centers

Equity stakes are the visible part. Guarantees are harder to see. In August, Nvidia agreed to back OpenAI's lease of a large data center campus being built by SoftBank's SB Energy in Ohio. Nvidia's payment obligation is capped at $105 billion for its initial commitment, Constellation Research reported, quoting Nvidia's SEC filing. Nvidia pays only if OpenAI becomes insolvent and defaults, or fails to make lease payments, and then only the shortfall after SB Energy tries to re-lease or sell the capacity. Nvidia is also investing $1.5 billion in SB Energy and will supply the chips for the first phase, which is expected to start operating in 2028.

The structure means Nvidia's exposure to OpenAI now runs through chip sales, an equity relationship and a lease guarantee at the same time. That is the kind of linked exposure the IMF describes. It is also why OpenAI's fundraising matters beyond OpenAI. The company has raised about $11 billion toward a $30 billion target at a $1.25 trillion valuation, below the $1.4 trillion it sought, The Decoder reported, citing the Wall Street Journal. We covered the earlier cut to the target in OpenAI funding round cut to $20 billion.

Why circular deals are not automatically a bubble

There is a serious case that the worry is overdone. Owen Lamont, a portfolio manager at Acadian Asset Management, argues that bubbles need outside money: new shares sold to the public, hidden leverage or distorted fundamentals. Money moving between a few companies in a closed loop is not enough on its own.

The economist Noah Smith has made a related point: much of this looks like ordinary vendor financing, the way carmakers lend to their buyers, and it may even spread each supplier's exposure across several AI customers. The weakness of that defense is timing. Lamont's test asks whether equity is being sold to outside shareholders. Meta's $25 billion sale is exactly that, and the chipmakers sat on the same side of the book as the public.

Where the risk actually sits

• Revenue that has not happened yet. The $1.4 trillion in the IMF count is commitments, not sales. It depends on AI labs growing revenue fast enough to pay for capacity they have already ordered.
• Correlated losses. If an AI lab misses its targets, its suppliers lose on the stake, the receivable and any guarantee at once. That linked exposure is why the IMF singled out circular financing in its October report.
• Debt off the balance sheet. About $320 billion of the debt sits with the companies receiving the investment, some of it backed by guarantees that do not appear as borrowing on the guarantor's books.
• Funding windows. These structures need open equity and credit markets. OpenAI's smaller round and Meta's share sale show both markets are still open, though OpenAI is raising at a lower valuation than it sought.

What the AI Bubble Index shows

The AI Bubble Index read 62.9 after the October 30 close, in the Normal band, down from 64.1 a week earlier and the lowest reading of the last 65 sessions. The Fragility axis, at 68, sits slightly above Euphoria at 66.3, and the systemic risk family reads 68.2. The sentiment and attention family is the highest at 91.8. The index does not track individual deals, so it cannot measure circular financing directly. What it does show is that market stress now weighs as much as enthusiasm in the reading. The methodology explains how the axes are built.

The telecom era is the closest precedent

Vendor financing has a history in tech booms. In the late 1990s, equipment makers lent to the telecom carriers that bought their gear, and those loans turned into losses when the carriers could not grow into their debts. Our history of tech bubbles and dot-com bubble timeline trace how that unfolded. The difference today is that the biggest buyers, Meta, Microsoft, Alphabet and Amazon, generate large operating cash flows. The weaker links are the AI labs and the neoclouds that rent out chips, which we examined in neocloud stocks and debt, and the cash squeeze at the top shows in our analysis of AI capex and cash flow.