A vulnerability was found in Nvidia’s business – a third of the company’s revenue depends on just three clients

Nvidia is heavily dependent on a handful of its largest customers, which actively buy computing accelerators for AI workloads and together account for more than a third of the company’s revenue. This puts Nvidia in a vulnerable position, although the company and its investors hardly need to worry in the near future – the demand for AI accelerators is only growing.

Image source: NVIDIA

In its quarterly Form 10-Q report, which the companies filed with the US Securities and Exchange Commission, Nvidia once again said that it has key customers that are so important that orders from each of them account for more than 10% of global revenue. Nvidia. However, the company does not disclose the names of these customers, which is logical, since it is unlikely that they would want their investors, employees, critics, activists and competitors to know exactly how much money they spend on Nvidia chips.

In its second-quarter report, Nvidia listed four of its largest customers, and in its most recent quarter, it cited three of them as one of them cut purchases. Although it is not known for sure who these clients are, Mandeep Singh, global head of technology research at Bloomberg Intelligence, believes that we are talking about Microsoft, Meta✴ and possibly Super Micro.

Nvidia itself simply calls them “Client A,” “Client B,” and “Client C.” Collectively, they purchased $12.6 billion worth of goods and services in the fiscal third quarter ended at the end of October. That’s more than a third of Nvidia’s total revenue, which was $35.1 billion. It is also noted that each of the “whales” purchased Nvidia goods and services worth from 10 to 11 billion dollars in the first nine months of the current fiscal year.

It is noteworthy that the contribution of the “whales” to the revenue turned out to be equal: each accounted for 12%, which suggests that they most likely purchased the maximum number of chips allocated to them, but not as many as they would ideally like. This is consistent with comments from CEO Jensen Huang that Nvidia is supply constrained. The company cannot simply produce more chips, since it does not produce them itself, but orders production from TSMC, whose capacity is scheduled for years in advance.

Because the names of Nvidia’s biggest chip buyers are classified, it’s hard to say whether they are “middlemen” like Super Micro Computer, which makes data center servers, or end users like Microsoft, Meta✴ or Elon Musk’s xAI. The latter, for example, built a powerful AI supercomputer almost out of nowhere in just three months.

However, relying on a handful of large customers is very risky – if any of them, or even worse, all of them at once, stop purchasing AI chips, Nvidia’s revenue will plummet. Luckily for Nvidia investors, that’s unlikely to happen any time soon.

Bloomberg Intelligence analyst Mandeep Singh sees few long-term risks for Nvidia. First, some large customers are likely to cut orders over time in favor of their own chips, which will lead to a decrease in the company’s market share. One such customer is Alphabet, which has its own TPU family of AI chips.

Secondly, Nvidia dominates in the field of accelerators for training AI, but cannot boast the same in the field of chips for inference – launching already trained neural networks. Inference doesn’t require such powerful chips, which means Nvidia faces much more competition not only from AMD and other direct rivals, but also from companies with their own chips, such as Tesla.

Ultimately, the launch of trained neural networks will become a much more significant business as more and more enterprises use AI, the analyst believes. “A lot of companies are trying to focus on inference capabilities because it doesn’t require the most powerful GPU accelerator,” Singh said. He also noted that in the long term, the transition to inference chips is “certainly” a bigger risk for Nvidia than losing market share in AI training chips.

Still, Singh says he believes Jensen Huang’s prediction that spending by major customers on AI chips will continue. Even if Nvidia’s share of the AI ​​chip market falls from its current 90%, the company could still make hundreds of billions of dollars annually.