Nvidia’s investment in OpenAI has a whiff of dot-com about it

SvD Näringsliv

Originally published in Svenska Dagbladet by Björn Jeffery, September 23, 2025

The world’s highest valued listed company — Nvidia — has an unusual problem. Its biggest customers are buying too much. But the company’s solution is making alarm bells ring.

When Jensen Huang meets his shareholders, it is hard to imagine anything other than nothing but happy faces. With a stock market performance of over 1,200 percent in the last five years, Nvidia has gone from a maker of graphics cards to the centrepiece of the AI boom.

The details of where the revenues actually come from have therefore been a detail that has somewhat slipped into the shadows. A closer look reveals that 85 percent of Nvidia’s revenues come from just six customers. So what do you do with this rather pleasant problem?

Well, Nvidia is helping new customers get up and running. But the method being used is somewhat reminiscent of similar arrangements during the dot-com crash.

Let us look at a deal Nvidia made recently. Data centre company CoreWeave received an order from Nvidia worth 6.3 billion dollars — around 59 billion kronor. CoreWeave is to deliver computing capacity for AI services that Nvidia can then sell on.

That might not sound so strange at first glance. But Nvidia is in effect buying back capacity in the chips it just sold to CoreWeave. Moreover, it is guaranteeing to continue purchasing that capacity until 2032. This means that CoreWeave’s large chip purchases from Nvidia become essentially risk-free. They know they will be paid for the next seven years.

Let us take another similar example. A week or so ago, Nvidia signed another deal to pay around 14 billion kronor to rent capacity from cloud company Lambda. Nvidia wanted to use 18,000 chips from Lambda’s data centres. But all of those chips had recently been purchased from Nvidia in the first place.

There is an additional interesting detail about these two deals. Nvidia is a major shareholder in both CoreWeave and Lambda. Through both arrangements, Nvidia is driving up revenues in companies whose profits it benefits from as an owner.

On Monday came a third variant, this time with a more familiar customer: OpenAI. A full 100 billion dollars is what Nvidia intends to invest in the creator of ChatGPT. But here too the details are somewhat vague — what exactly are they investing? Is it chips or cash? Or some form of credit for data services? Nvidia is also already a part-owner of OpenAI from before. The difference here is primarily the enormous scale of the investment.

The arrangements create a kind of feedback loop. Nvidia is customer, supplier, and major owner — simultaneously.

The reason for the deals is likely the customer mix described above. With such a large share of revenues coming from very few customers, the dependence on them is extreme. If even one of them were to decide to scale back their AI ambitions, it would hit Nvidia directly. It would therefore be beneficial for them if the number of successful chip customers were substantially greater. And a simple way to ensure that a customer grows is to buy services from them yourself.

The deals are not illegal, but they do resemble a couple of controversial business arrangements from the dot-com era.

In 2002, AOL purchased capacity from telecoms company WorldCom. WorldCom in turn bought advertising from AOL for roughly the same amount of money. The result was that both companies looked better on the revenue side without anything of substance actually having happened.

The same year it emerged that Enron — later to become one of the world’s biggest bankruptcies and financial scandals — had done a similar deal with telecoms company Global Crossing. What was really a loan between the companies was restructured to appear as revenue on both sides.

After Nvidia’s almost incredible run of successes, the question has been raised of what could possibly stop the company. The dependence on a handful of individual giant customers is one such thing. That Nvidia has an interest in building up new competitors is therefore logical.

But being on all sides of a transaction — buyer, seller, and owner — is messy and can lead to conflicts of interest. Which role does Nvidia represent when these deals are being negotiated? That is difficult to determine.

Right now everything points to a near-infinite demand for Nvidia’s products and services. The stock has surged over 32 percent — in this year alone. Perhaps Nvidia’s attempts to build up new customers are simply a smart strategy. And not a pitfall of strange circular business arrangements with built-in conflicts of interest.

The world’s stock exchanges — deeply dependent on Nvidia — are at least hoping that is the case.

The Author

Björn Jeffery is a Swedish technology columnist, advisor, and independent analyst based in Malmö, Sweden. He is the technology columnist for Svenska Dagbladet and co-hosts a podcast for the newspaper. He was previously CEO and co-founder of Toca Boca, the kids’ media company that grew to over one billion downloads. Through his advisory practice, Outer Sunset AB, he works with companies on digital strategy, consumer culture, governance, growth, and international expansion.