Two Nvidia customers made up 39% of Nvidia’s revenue in its July quarter, the company revealed in a financial filing on Wednesday, raising concerns about the concentration of the chipmaker’s clientele.
“Customer A” made up 23% of total revenue, and “Customer B” comprised 16% of total revenue, according to the company’s second-quarter filing with the Securities and Exchange Commission.
That’s higher than the same quarter a year ago when Nvidia’s top two customers made up 14% and 11% of sales, according to the filing.
The company regularly publishes information on a quarterly basis about its top customers, but the disclosure this week is fueling a renewed debate about whether Nvidia’s explosive growth is being driven by a handful of large cloud providers such as Microsoft, Amazon, Google and Oracle.
Nvidia finance chief Colette Kress said in a Wednesday statement that “large cloud service providers” made up about 50% of the company’s data center revenue. That’s important as the data center business made up 88% of Nvidia’s overall revenue in the second quarter.
“We have experienced periods where we receive a significant amount of our revenue from a limited number of customers, and this trend may continue,” Nvidia wrote in the filing.
Increasingly, analysts are looking to those cloud capital expenditure spending commitments to model the future growth of Nvidia.
“We see limited room for further earnings upside revision or share price catalyst in the near-term unless we have increasing clarity over upside in 2026 [cloud service provider] capex expectations,” wrote HSBC analyst Frank Lee in a note on Thursday. He has a hold rating on the stock.
But Nvidia’s Customer A and Customer B are not necessarily cloud providers. It’s a bit of a mystery, and an Nvidia representative declined to share the identities of Customer A and Customer B.
In its filing, Nvidia says it has both “direct customers” and “indirect customers.” Customer A and Customer B are listed as “direct customers.”
Direct customers are not the end users of Nvidia’s chips. They’re companies that buy the chips to build into complete systems or circuit boards that they then sell to data centers, cloud providers and end-users. Some of these direct customers are original design manufacturers or original equipment manufacturers like Foxconn or Quanta. Others are distributors or system integrators like Dell.
Indirect customers, meanwhile, include cloud service providers, internet companies and enterprises, which typically buy systems from Nvidia’s direct customers. Nvidia says it can only estimate revenue to indirect customers based on purchase orders and internal sales data.
Deciphering if any of those cloud providers are Nvidia’s mystery customers is difficult, in part because the chipmaker has wiggle room in the definitions of its direct and indirect customers.
Nvidia, for example, wrote in the filing that some direct customers buy chips to build systems for their own use.
Additionally, Nvidia noted that two of its indirect customers each accounted for over 10% of its total revenue, primarily buying systems through Customers A and B.
Contributing further to the mystery of it all, Nvidia said that an “AI research and development company” contributed a “meaningful” amount of revenue through both direct and indirect customers.
Nvidia told investors on Wednesday that demand for the company’s AI systems remains high, not just among cloud providers, but among other kinds of customers, including enterprises buying systems for AI and “neoclouds,” which are companies that are taking on the biggest providers with services more tuned for AI. Nvidia also listed foreign governments, saying it would record $20 billion in revenue this year for “sovereign AI.” All of these product categories are contributing to Nvidia’s revenue growth, Kress told analysts on an earnings call.
Nvidia CEO Jensen Huang also said that the company has a new forecast of $3 to $4 trillion in AI infrastructure by the end of the decade. It said that it could take about 70% of the total cost of a $50 billion AI-focused data center, not just for its graphics processing units but for other chips it sells, too.
Huang told investors it was a sensible target for the next five years because of how much hyperscalers were spending and committing to spend — $600 billion this year, according to Huang. He also said new kinds of customers, such as enterprises or overseas cloud providers, were joining the build-out.
“As you know, the capex of just the top four hyperscalers has doubled in two years as the AI revolution went into full steam,” Huang said.
