Three unidentified hyperscaler customers have each purchased over 10 billion dollars in Nvidia AI chips in 2024.

Nvidia, the world's most valuable company by market cap, remains heavily dependent on a few anonymous customers that collectively contribute tens of billions of dollars in revenue. The AI chip supplier has disclosed in its quarterly 10-Q filing to the SEC that it has key accounts so crucial that their orders each crossed the threshold of 10% of Nvidia's global consolidated turnover. An elite trio of particularly deep-pocketed customers individually purchased between $10 billion and $11 billion worth of goods and services across the first nine months ending in late October. According to Mandeep Singh, global head of technology research at Bloomberg Intelligence, founder and CEO Jensen Huang predicts that spending will not stop: "The data-center training market could hit $1 trillion without any real pullback." By that point, Nvidia's share will almost certainly drop markedly from its current 90%, but it could still be in the hundreds of billions of dollars in revenue annually.

Looking at Nvidia's accounts on a quarterly basis, there were four anonymous whales that in the second fiscal quarter comprised nearly every second dollar of sales; only three still meet the 10% threshold criteria currently. In the fiscal third quarter through late October, the three designated Customers A, B, and C purchased a collective $12.6 billion in goods and services—more than a third of Nvidia's overall $35.1 billion recorded for that period. Their share was divided up equally with each accounting for 12%, suggesting they were receiving a maximum amount of chips allocated to them rather than as many as they might have ideally wanted. Singh speculated the anonymous whales likely include Microsoft, Meta, and possibly Super Micro, though Nvidia declined to comment on this speculation.

The customer designations are not fixed from one fiscal period to the next—they can and do change places, with Nvidia keeping their identity a trade secret for competitive reasons. For instance, "Customer A" bought around $4.2 billion in goods and services over the past quarterly fiscal period, yet appears to have accounted for less in the past since it does not exceed the 10% mark across the first nine months in total. Meanwhile, "Customer D" appears to have done the exact opposite, reducing purchases in the past fiscal quarter yet nevertheless representing 12% of turnover year to date. Whether they are middlemen like Super Micro Computer or end users like Elon Musk's xAI remains difficult to determine. Nvidia is supply constrained because it has outsourced wholesale fabrication of its AI microchips to Taiwan's TSMC and has no production facilities of its own.

There are only a handful of companies with the capital to compete in the AI race, as training large language models is exorbitantly costly. These typically include cloud computing hyperscalers such as Microsoft and Oracle, which recently announced plans to build a zettascale data center with over 131,000 Nvidia state-of-the-art Blackwell AI training chips—more powerful than any individual site yet existing. It is estimated the electricity needed to run such a massive compute cluster would be equivalent to the output capacity of nearly two dozen nuclear power plants. Bloomberg Intelligence analyst Singh identified longer-term risks for Nvidia, including the potential that some hyperscalers will eventually reduce orders, diluting its market share, with Alphabet being a likely candidate given its own training chips called TPUs.

The dominance Nvidia enjoys in training is not matched by inference, which runs generative AI models after they have already been trained and where technical requirements are not nearly as state of the art. There is much more competition in inference, not just from rivals like AMD but also companies with their own custom silicon like Tesla. Singh noted that "There are a lot of companies trying to focus on that inferencing opportunity, because you don't need the highest-end GPU accelerator chip for that." When asked if this longer-term shift to inferencing was a bigger risk than eventually losing share in the market for training chips, he replied: "Absolutely."