Thirty years. That’s how long it took Nvidia to reach a $1 trillion valuation. Getting to $2 trillion took nine more months.
I keep coming back to that gap, because it explains something that has nothing to do with chips. When a company grows that fast, it ends up with more money than it can spend on itself. And when the rest of your industry is short on cash, having too much of it changes what you are. Nvidia sells the hardware that AI runs on. It has also started financing the people who buy that hardware. That combination has earned it a nickname that keeps showing up in the financial press this year: the central bank of AI.
What a central bank actually does
If you don’t follow monetary policy, the comparison might sound like flattery. It isn’t, exactly.
A central bank is the institution that decides how much money flows through an economy. It lends to banks when they’re short. It sets the cost of borrowing. In a crisis, it becomes the buyer or lender of last resort, the entity that steps in when nobody else will. That role comes with enormous power and enormous responsibility, which is why central banks are usually public institutions with mandates and oversight.
Nvidia has none of that oversight. It’s a company with shareholders. But in the specific economy of AI infrastructure, it has started performing a similar function: providing significant financial support and loans to the projects building out AI capacity. If you want to construct a data centre full of AI chips, Nvidia may not just sell you the chips. It may help fund the purchase.
Why this matters for anyone using AI agents
Most readers here care about AI agents as tools, not as an asset class. So let me connect it.
The agents you use, the ones that draft your emails, summarise documents, or handle customer queries, run on models that live in data centres. Those data centres are extraordinarily expensive. The companies building them are, in many cases, not yet profitable. Reporting on Nvidia’s financing role has pointed directly at this weak spot: the poor financial shape of the AI model companies, and the real possibility that some of them can’t pay their way.
So the chain looks roughly like this:
- You use an AI agent, probably for a modest monthly fee or for free
- That agent runs on infrastructure that costs enormous sums to build and operate
- The companies building that infrastructure are borrowing heavily to do it
- Some of that borrowing is backstopped by the company selling them the equipment
The tools feel cheap and abundant at your end. The financing underneath them is doing a lot of quiet work to keep it that way.
What the critics are worried about
The concern isn’t that Nvidia is behaving badly. It’s structural. When a supplier finances its own customers, the line between a genuine sale and a circular arrangement gets harder to read from the outside. Demand looks strong. Some of that demand is being funded by the same company recording the revenue.
Critics also worry about the debt market more broadly. AI infrastructure borrowing has become a meaningful chunk of corporate credit, and a lot of it rests on the assumption that AI revenue arrives on schedule. Analysts predict Nvidia’s revenue could reach $1 trillion by 2029, which is a staggering forecast and also a demanding one. Forecasts like that get built into lending decisions. If the number slips, the loans made on the strength of it look different.
There’s a reason the coverage this year has used phrases like Nvidia’s crown potentially slipping. Concentration is the risk. Real central banks are backed by governments and can print money. Nvidia is backed by chip sales.
How I’d think about it
I’m not in the business of predicting crashes, and I’d be suspicious of anyone who claims to know how this resolves. What I’d suggest instead is a small mental adjustment.
When you evaluate an AI agent for your work, you’re used to asking about accuracy, privacy, and price. Add a question about durability. Is the company behind this tool funded by revenue from customers, or by capital that assumes future revenue? You often can’t answer precisely, but the question changes how you plan. It nudges you toward tools with export options, toward avoiding deep integrations you can’t unwind, toward keeping your own copies of anything important.
The AI agent space is being built at remarkable speed, partly because one company has been willing to act as its lender. That’s a genuine accelerant. It’s also a single point of dependency, and understanding that is just part of being a clear-eyed user rather than a passenger.
🕒 Published: