Direct vs Indirect: Who Actually Uses the ChipsNarrow moat
Nvidia (NVDA) — moat facet
Nvidia knows less about its own demand than its filings suggest — orders arrive through channels, and so do corrections, usually late.
A subtlety worth understanding before drawing conclusions from Nvidia's customer disclosures: the companies in the concentration table are frequently not the companies operating the hardware. Nvidia's direct customers include OEMs, ODMs, distributors, system integrators and channel partners — the firms that build the servers and racks. Its indirect customers are the cloud providers, neocloud builders, AI model makers, enterprises and public-sector bodies who ultimately run the chips, and who mostly purchase through those intermediaries1.
This matters in two directions. It means a 22% direct customer may represent demand from many different end buyers, which is less alarming than it first appears — a system integrator's order book is diversified even if Nvidia's invoice is not. But it also means the reverse: Nvidia's filing explicitly warns that it estimates some indirect customers individually represent 10% or more of revenue, so the true end-customer concentration is not fully visible in the table at all.
The practical consequence is that Nvidia knows less about its own demand than a company with direct relationships would. Orders arrive through channels; cancellations and inventory corrections arrive the same way, usually late. That is a familiar semiconductor problem and it is the mechanism behind most chip-cycle surprises. Watch inventory and purchase commitments alongside revenue: in a channel-fed business, the gap between what is sold and what is deployed is where the next correction hides.
The channel structure is long-standing and unchanged — OEMs and integrators invoice, hyperscalers and labs deploy. It neither improves nor decays; it simply means both the concentration table and the demand signal are one step removed from the end customer, which is the ordinary condition of the semiconductor industry and the reason its cycles surprise.
The table names distributors and system builders, not the clouds and labs that run the chips, and Nvidia estimates some of those indirect buyers also exceed 10%. Fewer, larger direct customers means the channel is consolidating around the biggest end buyers.
Source: NVIDIA Form 10-K, FY2026 ↗- ReportedDirect customers are OEMs, ODMs, distributors and system integrators; indirect customers — CSPs, neocloud builders, AI model makers, enterprises and public sector — buy through them, and NVIDIA estimates some individually represent 10% or more of revenue.NVIDIA Form 10-K, FY2026 — "For fiscal year 2026, sales to one direct customer represented 22% of total revenue and sales to another direct customer represented 14% of total revenue"; FY2025: one at 12% and two at 11% each; FY2024: one at 13%. Direct customers include OEMs, ODMs, distributors and system integrators; indirect customers (CSPs, Neocloud builders, AI model makers, enterprises, public sector) buy through them, and NVIDIA "estimate[s] some individually representing 10% or more of our revenue". "Our revenue is concentrated among a limited number of direct and indirect customers and this trend may continue." — FY2026 (ended Jan 25, 2026) · publ. February 2026 · source ↗