Nvidia relies heavily on three key customers who actively purchase AI accelerators, collectively generating over one-third of the company’s revenue. While this creates a potential vulnerability, Nvidia and its investors are unlikely to worry in the short term, as demand for AI accelerators continues to grow.
In its quarterly 10-Q report submitted to the U.S. Securities and Exchange Commission, Nvidia reiterated that it has major clients whose orders each account for more than 10% of its global revenue. However, the company does not disclose their names, likely to avoid revealing financial specifics to investors, employees, competitors, or critics.
In its second-quarter report, Nvidia identified four major clients, but the most recent quarter mentions only three, as one reduced its purchases. While the identities remain uncertain, Mandeep Singh, Global Head of Technology Research at Bloomberg Intelligence, speculates they are Microsoft, Meta, and possibly Super Micro.
Nvidia refers to them as “Client A,” “Client B,” and “Client C.” Together, these clients purchased $12.6 billion worth of goods and services in the fiscal quarter ending October, accounting for more than one-third of Nvidia’s total revenue of $35.1 billion. Over the first nine months of the fiscal year, each of these clients reportedly spent $10–11 billion, contributing equally to Nvidia’s earnings, with 12% each.
This balanced contribution suggests that each client likely bought the maximum number of chips allocated to them but still fell short of their ideal demand. This aligns with CEO Jensen Huang’s comments that Nvidia’s supply is constrained, as the company depends on TSMC for manufacturing, whose capacity is booked years in advance.
Given the confidentiality of Nvidia’s top buyers, it is unclear whether they are intermediaries like Super Micro, which manufactures data center servers, or end-users such as Microsoft, Meta, or even Elon Musk’s xAI, which built a powerful AI supercomputer in just three months.
Relying on a small group of major customers poses risks—if one or more of them stops purchasing, Nvidia’s revenue could plummet. However, this seems unlikely shortly.
Bloomberg’s Mandeep Singh highlights long-term risks for Nvidia. First, some customers may reduce orders over time, favoring their proprietary chips, and potentially shrinking Nvidia’s market share. For instance, Alphabet (Google) uses its own TPU chips for AI tasks.
Second, while Nvidia dominates in AI training accelerators, its presence is less robust in inference chips—used to run trained neural networks. Inference requires less powerful chips, increasing competition from AMD and companies like Tesla that build their hardware.
In the long run, inference will likely become a more significant market as businesses increasingly adopt AI, Singh explains.
“Many companies are focusing on inference solutions since they don’t require the most powerful graphics accelerators.”, he said, adding that the shift to inference chips poses a greater risk to Nvidia than losing AI training chip market share.
Still, Singh believes in CEO Huang’s forecast that large-scale spending on AI chips will continue. Even if Nvidia’s current 90% market share shrinks, it could still earn hundreds of billions annually from the AI chip market.
