This article was originally published on Fool.com. All figures quoted in US dollars unless otherwise stated.
Key Points
- Two customers combined to provide 39% of Nvidia’s revenue in its latest quarter.
- Nvidia’s supply chain is complex, depending on numerous suppliers, equipment designers, and manufacturers.
- Broadcom’s AI sales growth is heavily dependent on a handful of hyperscaler customers.
Perhaps no company benefited more from the artificial intelligence (AI) trend than Nvidia (NASDAQ: NVDA).
In just five years, Nvidia's most important end markets shifted from gaming and professional visualization to data centers, though its core products remained the same -- graphics processing units (GPUs) and associated software and infrastructure.
Companies across a wide array of industries now depend on Nvidia's solutions to power their AI workflows. But without a couple of key customers, Nvidia's growth in recent years wouldn't have been nearly as impressive as it was.
Here's why Nvidia is increasingly relying on two customers, and if the AI growth stock is a buy now.
Peeling back the curtain
In its quarterly reports, Nvidia regularly highlights the importance of a select group of key customers. It doesn't disclose their names, however. Instead, it refers to them as Customer A and Customer B.
In its fiscal 2026 first quarter, Customer A represented 16% of total revenue and Customer B was 14%. But in its latest earnings report, delivered on Aug. 27, Nvidia said that Customer A had contributed 23% of total revenue and Customer B amounted to 16%. In just three months, those two companies went from 30% of total revenue to 39%.
Nvidia's revenue was $44.1 billion in fiscal Q1 and $46.74 billion in fiscal Q2. Customer A and Customer B accounted for $13.23 billion in Q1 revenue and $18.23 billion in Q2 revenue. But Nvidia only grew revenue by $2.64 billion quarter-over-quarter. So without Customer A and B, Nvidia's revenue would have been down by $2.36 billion quarter-over-quarter.
Customer A and Customer B aren't just Nvidia's key accounts: They are single-handedly supporting its investment thesis right now.
The chipmaker's sales to four other direct customers contributed 14%, 11%, 11%, and 10% of revenue, respectively, in the fiscal second quarter. All of these unnamed customers' revenue was attributed to Nvidia's Compute and Networking segment for data centers -- which provided 87.9% of total revenue in its latest quarter.
Understanding Nvidia's supply chain
Nvidia's dependence on Customer A and Customer B is a risk worth monitoring. It's also a bit misleading due to the structure of Nvidia's supply chain.
Nvidia's key end users are hyperscalers like Amazon Web Services, Microsoft Azure, Alphabet's Google Cloud, Meta Platforms, and Oracle Cloud Infrastructure. But these companies often buy Nvidia's chips through contractors that assemble AI server racks with Nvidia's GPUs and associated networking, storage, cooling, etc.
For example, a hyperscaler may order one of Nvidia's GB300 Blackwell Ultra rack-scale servers from Hon Hai Precision Industry, commonly known as Foxconn, rather than buying it from Nvidia directly. On Aug. 26, Nvidia reported that Foxconn was among a list of companies adopting its new RTX Pro Servers for AI reasoning, physical AI, and business workloads.
In its latest quarterly report, Nvidia listed add-in board manufacturers (companies that build graphics cards using Nvidia's chips), distributors, original device manufacturers, original equipment manufacturers, and system integrators as examples of its direct customers. By contrast, it described cloud service providers, internet companies, enterprises, and public sector utilities, among others, as indirect customers that largely purchase products through the direct customers. In other words, it's not that AWS or Microsoft is likely Customer A or Customer B. Rather, those tags are probably being put on major distributors that are likely supplying multiple cloud infrastructure providers.
Concentration has become the norm for AI chip makers
Nvidia's jargon around customer identities can make it difficult to determine just how much of its business is tied to each specific hyperscaler. However, based on the capital expenditures of top cloud companies, it's clear that Nvidia will need its key end users to profit from their AI spending to justify them boosting their order volumes.
Nvidia isn't alone in the tech world in relying on just a few customers. Broadcom (NASDAQ: AVGO) management forecast that its AI semiconductor sales will make up over 32% of revenue in its fiscal third quarter. And a large fraction of that revenue is coming from a handful of customers.
On its fiscal Q4 2024 earnings call in December, Broadcom management discussed how the company is working to expand its core customer base beyond the three hyperscalers that have been major contributors to its top line. Now, it has two additional hyperscalers that are developing their own next-generation AI XPUs. Broadcom forecasts that it will convert these prospects into revenue-generating customers by 2027. While we don't know for sure who these customers are, Meta and Alphabet are almost certainly two of the three established customers, and Apple and TikTok parent company ByteDance could be the prospects, based on other news reports that they are both building custom chips with Broadcom.
Meta Platforms is building the world's first multigigawatt "supercluster" data center next year. And it's highly likely that this data center will be jam-packed with Broadcom AI clusters, given that Broadcom is working with Meta on its custom AI chips. Google Cloud partners with Broadcom to make its custom Tensor Processing Units, which power Google DeepMind's AI chatbot, Gemini, as well as Google Search and other applications.
By following Broadcom's more direct commentary on its key hyperscale customers, we can infer that Nvidia is depending on the same cast of characters to drive its growth.
Nvidia is still a buy
Nvidia's increasing reliance on Customer A and Customer B isn't necessarily a red flag, as these are direct customers (likely suppliers) that are processing orders for multiple end-users. However, what does pose a risk to Nvidia, Broadcom, and the AI investing theme in general is the possibility of a pullback in capital expenditures by hyperscalers.
Right now, capex as a percentage of revenue is at a five-year high for Amazon, Microsoft, Alphabet, Meta Platforms, and Oracle -- meaning that these companies are in expansion phases. However, eventually, the cycle may shift as these companies redirect their focus from accelerated growth to generating free cash flow. And when that happens, it will likely affect the primary beneficiaries of that spending -- companies like Nvidia and Broadcom.
Nvidia is still a foundational AI growth stock to buy and hold, but investors who buy it should understand that these cyclical gyrations could lead to choppy stock price movements. However, volatility like this is worth enduring over the long term when a business's underlying fundamentals are sound.
This article was originally published on Fool.com. All figures quoted in US dollars unless otherwise stated.
