Major ClientsThin moat
Nvidia (NVDA) — moat facet
Two unnamed buyers are 36% of revenue, up from 23% a year earlier — and the famous names everyone assumes are Nvidia's customers mostly aren't, technically.
Nvidia's customer concentration is the most extreme in this collection, and it got dramatically worse in a single year. In fiscal 2026 one direct customer accounted for 22% of total revenue and another for 14% — 36% from two buyers1. A year earlier the largest was 12% and two others 11% each. Nvidia's own filing states plainly that revenue is concentrated among a limited number of customers and that the trend may continue.
Then comes the quirk that makes this page worth reading. Nvidia does not name those customers, and they are probably not the companies you would guess. Its direct customers are largely the OEMs, ODMs, distributors and system integrators who assemble and resell the hardware. The famous names — the cloud providers, the neocloud builders, the model labs — are mostly indirect customers who buy through those channels, and Nvidia says it estimates some of them individually exceed 10% of revenue too.
Read through the channel and the real picture is this: hyperscalers supplied 55% of data-center revenue in the latest quarter, down from 59% a year earlier, with the rest spread across AI clouds, enterprises, sovereign projects and industry2. Those are the buyers who matter, and the uncomfortable fact — treated in full on this company's root threat page — is that the four largest of them are each building accelerators of their own.
Set against Apple's page of the same name, the contrast is close to total: Apple has no customer worth naming and sells to hundreds of millions of individuals who cannot negotiate. Nvidia sells to a few dozen organisations that can negotiate, are building substitutes, and in several cases are financed by Nvidia itself. The pages that follow take the concentration, the channel, the hyperscalers and the labs in turn.
Concentration intensified sharply in one year — the top two direct customers went from about 23% of revenue to 36% — while the largest end buyers are all funding substitutes and a growing slice of demand comes from labs and neoclouds Nvidia has itself financed. Extraordinary revenue, resting on fewer and more conflicted counterparties than a year ago.
The most extreme customer concentration in this collection, and it worsened sharply in one year — Nvidia's own filing says the trend may continue. Hyperscalers supply about half of data-center revenue through the channel. Watch the concentration lines each year: this is what converts an AI-demand wobble into an earnings shock.
Source: NVIDIA Form 10-K FY2026 ↗- ReportedFY2026: one direct customer at 22% of revenue and another at 14%, versus 12% and two at 11% a year earlier; NVIDIA states revenue is concentrated among a limited number of direct and indirect customers and estimates some indirect customers individually exceed 10%.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 ↗
- Moat Explorer calcHyperscale customers were 55% of Data Center revenue in Q2 FY2027, down from 59% a year earlier.Moat Explorer calc from NVIDIA's Q2 FY2027 CFO commentary - Hyperscale revenue $48,710M of Data Center revenue $89,023M (54.7%) in Q2 FY2027, against $24,168M of $41,096M (58.8%) in Q2 FY2026, on the recast basis after one customer moved from ACIE to Hyperscale — Q2 FY2027 and Q2 FY2026 · publ. 2026-08-26 · source ↗