AI Circular Financing: Nvidia, OpenAI, and Ecosystem Risk

Update, November 2026: In September 2025 Nvidia announced an intention to invest up to $100 billion in OpenAI, a proposed staged investment rather than a completed transfer. Anthropic’s S-1 now shows $657 billion of compute contracts, $498 billion of them noncancelable. For the critic’s view, see Ed Zitron and the AI bubble case.

The pattern

An ordinary supply chain flows one way: a chip company sells to a cloud provider, which sells to a lab, which sells to customers. In AI, money also flows backwards. A chip company invests in a lab, and the lab uses that money to buy the chip company’s products. A cloud provider invests in a lab, and the lab commits to spend on that cloud.

Economists call the result circular or round-trip financing. Each individual deal can be sensible. The question is what it does to the picture of demand.

What has actually been disclosed

Be careful here, because many headlines compress complicated agreements.

OpenAI is valued at $852 billion after a $122 billion round in March and has reportedly filed confidentially for an IPO. Both labs depend on very large compute purchases from a small set of suppliers.

Why circularity matters

1. Demand can look bigger than it is. If a supplier funds the customer who then buys from the supplier, some revenue is the supplier’s own money coming back.
2. Risk is correlated. If one party stumbles, the others feel it. On the day OpenAI’s revenue was restated, Nvidia fell about 3%, Oracle nearly 6% and CoreWeave nearly 8%.
3. Metrics lose comparability. When labs count revenue differently, you cannot tell which is healthier without the filing.
4. Capacity is committed ahead of revenue. Noncancelable contracts turn a revenue slowdown into a cash problem.

The fair counterpoints

Not every deal is smoke and mirrors.

• Suppliers have always financed customers when it helps them sell. Equipment makers in many industries do.
• Real end-customers exist: Anthropic reported 11,900 customers paying over $100,000 a year in its S-1 coverage, up from 950 a year earlier.
• A public filing carries legal liability, which is a stronger check than a press release.

So circularity is a reason for caution when reading growth numbers, not proof that demand is fake. The most vocal critic of the structure is profiled in Ed Zitron’s five messages, alongside where his claims have and have not held up.

How to read a deal

Ask five questions when you see a headline:

1. Is the money committed or only announced?
2. Who gets paid, and where does the cash come from?
3. Can the customer cancel? What happens if they do?
4. Is the revenue counted gross or net?
5. What share of the seller’s growth comes from this one customer?

Implications for builders

• Pricing risk. Subsidised pricing may not last. Model your costs at higher prices and read the token price war piece.
• Availability risk. Capacity decisions upstream affect rate limits and model deprecations.
• Concentration in your own stack. If most of your product runs through one model, you are part of the same loop.
• Funding climate. If AI financing tightens, startups feel it in fundraising terms and valuations.

Practical steps:

1. List every AI vendor and the share of your product that depends on each.
2. Add a second provider or a self-hosted option for critical paths. See open-weight models.
3. Negotiate contract terms that limit sudden price or policy changes.
4. Keep prompts and evaluation sets portable between models.

A reading checklist for filings

When a prospectus or investor deck appears, look for revenue by customer and by partner, the cancellation terms of purchase commitments, related-party transactions, how revenue is defined and how much of the cash comes from the same parties that receive it.

Your next step

Draw a one-page map of your own AI supply chain: model provider, cloud, chips, your customers. Mark where money and dependency flow both ways. Anything that sits in a loop with a single counterparty deserves a fallback plan and a named owner.

Bottom line

Circular financing is a normal tool used at unusual scale. Read the deals for what is committed, who can cancel and how revenue is counted. Treat headline figures with care, and design your own dependencies so you can survive a shock to any one vendor.