For three years the neat way to describe Nvidia was that it sold shovels in a gold rush: whoever won the AI race, the company supplying the picks got paid either way. That story is quietly being rewritten. Nvidia is no longer just selling the tools. It is increasingly helping its customers find the money to buy them, and in doing so it is starting to behave less like a hardware vendor and more like a bank. The Economist went a step further this week, calling it "the central bank of AI," the one institution that decides how much of the industry's scarcest resource, compute, gets allocated and to whom.
The clearest sign is the size of the cheques Nvidia is now helping to arrange. The company says it is working with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than 500 billion dollars in outside capital for AI infrastructure, pools of money its clients can draw on at reasonable rates. Bank of America has taken to describing chip suppliers like Nvidia and Broadcom as "credit intermediaries," firms that smooth over the financial risks that make giant data centres hard to fund. The bank estimates AI spending could top 5 trillion dollars between 2026 and 2030, with roughly 1.2 trillion of that needing to come from outside lenders. Somebody has to stand behind those loans, and increasingly it is the chipmaker.
Why would Nvidia want the job? Because the alternative is a traffic jam of its own making. Demand for the chips is not the problem: Oracle reported this month that its graphics processors were running at 97.9 percent utilisation, meaning new hardware is absorbed almost the instant it is plugged in, and its chief financial officer said Nvidia compute was fully used across every cloud Oracle serves. The bottleneck is money and power. AI data centres cost a fortune upfront and earn their keep slowly, a mismatch that is fine for cash-rich giants like Microsoft and Google but brutal for the smaller "neoclouds" doing much of the building. By guaranteeing revenue, backstopping leases and taking equity stakes, Nvidia keeps those buyers solvent enough to keep ordering.
Jensen Huang, Nvidia's chief executive, is candid about the shift. "We began by building chips," he said this month at a Goldman Sachs conference. "Today, we are helping create a new class of productive, investable infrastructure: AI factories." At the same event he repeated his forecast that global AI infrastructure spending will reach 3 to 4 trillion dollars by 2030, a number he first floated a year ago and has never walked back. Coming from the man now helping to finance that spending, it is both a prediction and a sales pitch.
The trouble with a bank is that it holds the risk it used to sell. When Nvidia simply shipped chips, a customer who overbuilt was the customer's problem. Now that Nvidia guarantees the loans, underwrites the demand and owns slices of its own buyers, a downturn flows back to its own balance sheet. The arrangement also looks uncomfortably circular: Nvidia invests in a cloud provider, the provider uses that money to buy Nvidia chips, and the resulting revenue helps justify the next investment. It works beautifully while demand climbs. Should it stall, the same wiring that pumped money outward would carry the losses home.
None of this means the boom is about to break. Utilisation is real, the order backlogs are real, and Goldman's analysts described visibility across the chip supply chain as historically unprecedented. But the character of the boom has changed. The safest bet in technology, the toll-booth that got paid no matter who won, has started writing the tickets on credit. Nvidia has made itself indispensable twice over, as the supplier and the financier. The question worth watching is what happens the first time one of its borrowers cannot pay.