Nvidia's Top Three Customers Account for 44% of Revenue as Concentration Risk Continues to Rise
nashnova research
Three Nvidia customers each crossed the 10% revenue threshold in the first half of this fiscal year, together accounting for 44% of total sales — up from zero such customers just two years ago — while those same buyers are building their own AI chips.
From zero customers above 10% to three at 44% — what happened?
In the six months to July, three customers each contributed over 10% of Nvidia's revenue, totalling 44%.
Last fiscal year, two customers accounted for 36%. In FY2023, no single customer reached 10%.
This means → customer concentration jumped from zero to nearly half in two years — faster than revenue growth itself.
The engine behind it: data-centre revenue surged from roughly $15 billion in FY2023 to $193.7 billion last fiscal year, and is on track to double again. In plain terms = the bigger the pie, the larger the slice held by the top few buyers.
Who are these three customers?
Nvidia has not named them publicly. The Information's analysis suggests Dell or Hon Hai (Foxconn) — both integrate Nvidia chips into servers and resell them.
Dell's "AI-optimised server" revenue doubled year-on-year to $16.4 billion in Q2, driving overall revenue up 58%. Hon Hai's first-half revenue rose 35%, led by AI equipment.
CFO Colette Kress said in February that the top five cloud and hyperscale customers together contribute over 50% of data-centre revenue. Microsoft, Meta, and SpaceX are all considered major buyers.
Google and Amazon also buy heavily, but their extensive use of in-house AI chips makes it unlikely either ranks among Nvidia's largest customers. This reflects how custom silicon is already reshaping the customer mix.
In-house chips from top buyers — a second layer of risk on top of concentration?
Concentration alone is a risk. What makes it worse: Microsoft, Meta, and SpaceX are all developing their own AI chips.
This means → the customers Nvidia depends on most are the same ones with the strongest incentive to reduce that dependence. Once their chips mature, orders could taper.
Short-seller Michael Burry has publicly flagged rising customer concentration as a major risk for Nvidia in recent months.
What do the receivables reveal?
As of July, 70% of Nvidia's accounts-receivable balance came from five customers. In January it was three customers at 56%; at the end of FY2025, two at 33%.
First-half receivables jumped 64% to $63 billion, driven by extended payment terms on large multi-quarter deals with investment-grade buyers.
Payment terms have already stretched from 45 days to 60, and filings say they could extend to 90 days or up to one year.
In plain terms = the goods are shipped, but the cash arrives slower. Nvidia's cash flow increasingly hinges on a handful of big customers paying on time.
How is Nvidia responding — can it diversify fast enough?
This week Nvidia brought several Australian companies into its "AI Factory" programme, supplying chips and networking gear. Similar efforts are under way in India and Armenia.
CEO Jensen Huang has invested in emerging cloud providers such as CoreWeave and Nebius, which are also significant chip buyers.
This means → Nvidia is cultivating new customers through an "invest-for-orders" model, but their combined scale is far too small in the near term to offset top-heavy concentration.
The pivotal question: whether customer diversification can build a meaningful hedge before the top buyers' in-house chips reach maturity — that will determine whether data-centre growth is sustainable.
市场有风险，内容仅供研究参考，不构成投资建议。