When your chip supplier is also your lender, you don’t have a supply chain — you have a monetary system, and Nvidia is running it.
That’s the uncomfortable thing about the “central bank of AI” label that’s been stuck to Nvidia through 2026. It started as a clever line. It has hardened into something closer to a description. Nvidia sells the compute that AI companies need, and increasingly it also provides the financial support and loans that let those companies buy the compute. Both sides of the transaction run through the same building.
I review AI tools for a living. Most of what I test is downstream of this arrangement, whether the founders realize it or not. So let me back up the verdict.
What a central bank actually does
A central bank sets the price of money, decides who gets credit, and acts as the lender of last resort when the private market won’t step in. Nvidia now does a recognizable version of all three for AI infrastructure. It sets the effective price of intelligence by pricing its chips. It decides which projects get financial backing. And when AI infrastructure deals need a backstop, Nvidia’s willingness to provide one has become part of the calculation.
The difference is that real central banks have mandates, oversight, and a public interest they’re supposed to serve. Nvidia has shareholders. Those are not the same thing, and pretending otherwise is how people get surprised.
The circularity problem
Here is what makes critics nervous, and they’re right to be. If a vendor helps finance the purchase of its own products, revenue starts to look less like demand and more like an accounting loop. Money goes out as credit, comes back as sales, and gets reported as growth. That works fine as long as the borrowers can eventually pay from real business results.
Which brings us to the actual weak spot analysts keep circling: the poor financial shape of the AI model companies themselves. A lot of them burn enormous amounts of cash and have not demonstrated they can service large debts. Nvidia’s backstop is what closes that gap. Remove it and several deals look considerably less solid.
The debt market concern follows directly. When one company becomes the credit engine for an entire sector’s buildout, its risk appetite becomes everyone’s risk appetite. Lenders price off it. Suppliers plan capacity around it. Startups build business models assuming compute stays available on current terms.
The numbers people cite, and what they hide
Nvidia took thirty years to reach a $1 trillion valuation and nine more months to reach $2 trillion. Analysts now project revenue could reach $1 trillion by 2029. Those figures get quoted as proof of a durable engine. I read them as proof of speed, which is a different property. Speed tells you how fast something moves, not how well it stops.
Record revenue also doesn’t tell you where the money came from. A dollar of revenue from a profitable enterprise customer paying cash is not the same asset as a dollar of revenue from a project Nvidia helped finance. Both look identical on the top line. They behave very differently in a downturn.
What this means if you actually build things
Most readers here aren’t trading the stock. You’re picking tools, choosing model providers, and deciding what to build on. This structure matters to you in a few concrete ways:
- Pricing you rely on may be subsidized. Cheap inference is partly a product of financed capacity. Financing terms change.
- Vendor concentration is worse than it looks. Switching model providers feels like diversification. If every provider depends on the same hardware and the same lender, it isn’t.
- Some AI companies are alive because of credit, not customers. Ask how a provider funds its compute before you make it a dependency.
- The bottleneck can move fast. A credit tightening cycle hits capacity availability before it hits headlines.
The honest read
I’m not predicting collapse. Nvidia makes genuinely excellent hardware, and demand for it is real. There is no conspiracy here, just a company doing the rational thing when it holds a position this strong: extending credit to keep the buildout moving and capturing the growth that follows.
What bothers me is the absence of anyone checking the work. A private company has quietly taken on a systemic role without any of the accountability that role normally carries. The crown may well stay where it is. But a financial system with one participant setting prices, granting credit, and backstopping failures is not a market. It’s a monopoly with a treasury department.
Build accordingly. Assume today’s compute costs are a promotional rate, keep your architecture portable, and treat any AI vendor’s runway as a question you’re allowed to ask out loud. The best defense against someone else’s balance sheet risk is not needing them as badly as everyone else does.
🕒 Published: