Nvidia’s Growing Dependence On a Few Big Customers

Nvidia CEO Jensen Huang has very good reason to try to develop new customers by investing in a variety of neocloud and AI firms. For all its success, Nvidia’s sales have become increasingly dependent on a handful of customers—some of which may cut back their purchases over time.

That becomes clear when you look at its own disclosures going back several years, which specify how many customers account for more than 10% of sales. In the first half of this fiscal year, ending in July, three such customers were responsible for 44% of total sales. Last fiscal year, two customers accounted for 36% of total sales. Going back to fiscal 2023, Nvidia had no customers accounting for 10% or more of sales (see the chart below).

Nvidia’s reliance on a handful of big customers has increased as its business has taken off. Since fiscal 2023, its data center business—where it records sales of the AI chips going into data centers—has ballooned to $193.7 billion last year from just $15 billion. That number is on track to double this year.

The issue has weighed on the minds of investors, such as Michael Burry, famed for his role in predicting the 2008 mortgage crisis, who in recent months has cited Nvidia’s increased customer concentration as a major risk.

To be sure, analyzing Nvidia’s customer concentration is complicated. The three biggest customers it cites in its most recent filing likely include either Dell or Hon Hai Technology, also known as Foxconn, both of which sell servers incorporating its chips to a range of other companies. Dell, for instance, recently reported that revenue from selling “AI-optimized servers” doubled in the second quarter to $16.4 billion, helping drive overall revenue growth of 58%.

Hon Hai reported 35% higher first-half revenue, also driven by surging AI equipment sales.

But Nvidia is heavily exposed to some cloud firms. In February, its chief financial officer, Colette Kress, said the five biggest cloud firms “and hyperscalers”—a phrasing that likely includes Meta Platforms and SpaceX—collectively accounted for more than 50% of data center revenues.

It’s unclear how many of those are in the top three, although the trio could include SpaceX, Meta or Microsoft, all of which have been investing heavily in new AI data centers stocked with Nvidia chips. Both Google and Amazon are major buyers of Nvidia chips for the same reason, but they’re also heavy users of AI chips they’ve designed themselves. That means they’re less likely to be among Nvidia’s biggest customers.

Microsoft, Meta and SpaceX are also developing their own chips, which means they could eventually reduce their purchases of Nvidia chips. That trend helps explain why Nvidia has been so focused on seeding new customers by investing in buyers of its chips, such as neoclouds like CoreWeave and Nebius.

Some of its efforts are aimed overseas. On Wednesday, for instance, Nvidia added several Australian companies to a program it calls AI Factory, under which it supplies chips and networking gear to companies building and running data centers. Nvidia promoted a similar build-out in India and Armenia this year, part of its push to seed more sovereign and neocloud customers beyond its biggest U.S. buyers.

Payment Due

Despite Nvidia’s efforts to diversify its business, the risk of overconcentration may be increasing. Nvidia disclosed in July that 70% of its accounts receivable balance—money owed for products due to be delivered—came from five customers. In contrast, three customers accounted for 56% of the accounts receivable balance in January, compared with two customers responsible for 33% of it at the end of fiscal 2025.

Notably, Dell reported recently that its accounts payable—money it owes suppliers—had jumped 48% in the first half of the year to $49.7 billion at the end of July. It likely owes much of that to Nvidia.

At the same time, the increase in Nvidia’s accounts receivable balance appears to represent, at least in part, the company giving customers more time to pay. Its accounts receivable balance jumped 64% in the first half of the year to $63 billion, “due to extended payment terms on large, multiquarter agreements with certain investment-grade customers,” Nvidia disclosed in late August.

Nvidia said it had extended payment terms to 60 days from 45 in the previous quarter. But in a securities filing at the same time, it said it may extend payment terms even longer.

“Payment from customers is generally due shortly after delivery of our products. In certain cases, for investment-grade customer purchases, we have and may in the future provide longer payment terms ranging from 90 days up to one year to assist customers with large data center builds depending on size,” the company said.

Extending payment terms holds a risk for Nvidia by reducing the amount of cash it generates in any one quarter. It also makes the chipmaker more dependent on the health of the customer who is getting extra time to pay.