Nvidia has pumped about $30 billion into OpenAI and roughly $10 billion into Anthropic, with bills more going into xAI and CoreWeave. The detail that actually matters is what those companies do with the money: they spend it back on Nvidia chips. Profit leaves Nvidia's investments and returns to Nvidia's revenue, and the report calls that loop circular financing. Now that founder CEO Jensen Huang has walked away with no succession plan in place, that structure is suddenly the whole story. But the loop is not only a Nvidia problem, and it is not only about chips.
A bank, not just a supplier
The reason Nvidia reads like a bank is that it underwrites the very companies that buy from it. It invests in an AI firm and that firm then spends the money back on Nvidia chips. A supplier who also holds stock in its own customers is a supplier who can steer demand, back the winners before the crowd, and treat its own accounts book as an investment portfolio. That is a different kind of power from the vendor that simply ships a part, invoices, and moves on. It is the power of the funder, and the funder decides who gets repaid and on what terms.
The report frames the stakes in market terms, noting that about three-fourths of all US economic growth has been attributed to AI, with Nvidia one of the biggest spenders in that camp even outside the government. If pain shows up at Nvidia, the source says, it trickles into the companies it funds, and from there into the wider market that AI has been carrying. The risk is networked. It does not sit in one contract.
The same loop, in your own stack
Any operator can build a version of this loop without meaning to. Over the last two years you let one AI vendor become the rails: the data models, the workflow, the integration that your whole downstream depends on. You stopped treating it as a supplier you could leave and started treating it as infrastructure you bank on. That is the moment a supplier relationship turns into a financing relationship, because you are no longer buying a product. You are funding a bet that the vendor stays solvent, keeps shipping, and keeps pricing against you.
Then your own company gets folded into someone else's numbers. Your subscription spend shows up on the vendor's side of the book, sometimes as a material line. Their roadmap, their pricing, their next raise, their next acquisition can all be redirected by forces you cannot see, investors or acquirers or a single big customer down the line. Their survival stops being their problem and starts being yours, because your ability to run your business now leans on a company whose books are partly your money.
Who carries the loss
When a leg of the loop breaks, the loss does not fall on the funder or the equity investor. It falls on the buyer, and it arrives in four forms. First, replatforming cost, the real engineering time to move off the rails. Second, immobility, because your data, your tuned models, and your pipelines were built to live inside the vendor's world and cannot simply be packed up. Third, lost productivity, the months it takes a team to rebuild muscle memory. Fourth, the renegotiated terms, because the vendor now knows exactly how hard it will be to leave, and the price moves against you the moment you prove you cannot walk away.
Picture a hypothetical claims team that has built its whole weekend turnaround around one AI platform, betting the vendor would keep improving and keep the price stable. Nobody named, no real company, and the deal looked fine until the vendor was acquired and the new owners reset the contract. The team that had been counting on it carried every cost of the move. That is not a cautionary tale. It is the standard shape of a circular dependency when one leg snaps.
The dependencies hiding in your stack
The question in the title is not rhetorical, and the answer is that your own stack is usually hiding three loops:
- Capital-dependency risk. You are relying on a vendor's funding, valuation, or investor appetite to keep its product alive. A well-funded vendor is not the same thing as a reliable one, but a vendor whose next round depends on keeping its customers happy is a vendor whose incentives can override your needs.
- The reverse loop. Your spend is a material part of their numbers, so their pricing and support move around their own survival and margin targets rather than your business. You have become an account they manage, not a client they serve.
- Concentration dressed as partnership. A strategic alliance or co-development agreement is often just a dependency wearing a shirt and tie, where one side breaking takes the entire arrangement down with it.
The tell and the answer
The tell is simple, and it will surface almost any one of these on its own. If the health, pricing, or roadmap of a critical AI vendor could be redirected by someone other than your needs, then you are in a loop. The funder, a large customer, or an acquirer can move the dial, and you hold the short end. You are the back leg of someone else's circular financing.
The cost of breaking it is already on your balance sheet, waiting in the form of migration time, stranded work, and the next vendor's onboarding premium. So the practical move is to treat that dependence as a quantified risk, not a given. Map where your spend feeds a vendor's books, isolate the single rails nobody else can run, and hold the right to leave in writing and in practice before you need it.
That is the answer to the question when your leading AI vendor acts more like a bank than a supplier. The circular dependencies your own stack is hiding are the relationships where you have stopped being a customer you can fire and become infrastructure the vendor can lean on, while letting them define the terms. The moment you can name who pays when a leg breaks, you will find it is almost always you, and the amount is everything it costs to walk away.
Sources
- Nvidia News: Jensen Huang Is Out, and the 'Bank' Behind American AI Just Lost Its Founder (Briefs Finance, 2026-11-14)