Double Counting the Same Customer Demand in a Supply Chain Graph

Key findings

• Nvidia explicitly distinguishes direct customers such as ODMs, OEMs, distributors, cloud providers, and system integrators from indirect customers that can purchase through those direct customers.
• A flat customer list can therefore double count one economic demand stream unless channel roles and multi-step paths are classified before exposures are aggregated.

Nvidia's direct and indirect customer disclosures are two views of demand, and their percentages should not be added without resolving overlap. A system builder can be the legal purchaser while a cloud provider supplies the underlying demand. The same hardware purchase may therefore appear in both views. This article uses Nvidia's filing definitions and five named Altsets relationships to build an exposure ledger that counts each demand pool once.

The fiscal 2026 annual report reports direct-customer revenue shares of 22% and 14%. It separately estimates that some indirect customers individually represent at least 10% of revenue. The indirect estimates use information including orders, product specifications and internal sales records. They are an attempt to look through the purchasing channel, not a second block of revenue to append to direct sales.

A 35% sum can describe only 20% of distinct demand

Consider a hypothetical example, with every number expressed as a share of the same supplier's revenue over the same period. A system builder buys products representing 20% of that revenue. An end customer drives purchases representing 15%, all routed through that builder. Adding the two percentages produces 35%, even though the distinct demand represented by the two views is only 20%.

Let $d$ be the direct-buyer share, $e$ the indirect end-demand share and $o$ the revenue counted in both. The distinct share is:

$$u=d+e-o.$$

The graph can contain both companies in all three cases. The nodes alone do not tell the analyst which row applies. With these assumed shares but unresolved overlap, distinct demand lies between 20% and 35%; choosing the upper bound silently assumes no shared sales. These are explanatory scenarios, not estimates for any named Nvidia counterparty.

The arithmetic also requires a common unit. Revenue received by a component supplier cannot be added to a system builder's revenue including assembly and other components. Likewise, annual and quarterly shares cannot be combined into a single exposure. Resolve the supplier, period and demand definition before calculating overlap.

Separate named connections from confirmed channel paths

The supplied Altsets Nvidia view contains Quanta Computer, Samsung Electronics, Amazon, Microsoft and Super Micro Computer downstream. It does not establish that one displayed node purchases through another. Classifying a company's business helps frame a channel hypothesis, but proving a path requires evidence about the particular transaction or product.

The ledger preserves the network without inventing a Quanta-to-Microsoft or Super-Micro-to-Amazon path. It also avoids assigning Nvidia's anonymous 22% or 14% filing entries to these names. The anonymous-customer analysis addresses the evidence needed for identity research. The Quanta and Samsung article compares the quantified edges themselves.

Aggregate at one layer, retain the other as an explanation

For legal-counterparty concentration, group comparable direct purchases by buyer identity. For end-demand concentration, allocate those purchases to the ultimate demand source where evidence supports the allocation. Save both views, but do not add one view's total to the other. If a cloud provider buys through several builders, an end-demand analysis may combine the portions attributable to that provider; the corresponding direct sales must then remain channel explanations rather than extra exposure.

A practical record needs the supplier, direct buyer, end-demand source, period, measurement basis and supporting source for the allocation. An unresolved end customer stays unresolved. It should not be distributed across famous cloud names merely to make the table complete. A later source can resolve the allocation without changing the original direct-buyer observation.

This changes the interpretation of an event. A cloud customer's spending cut and a builder's order reduction may be evidence about the same demand contraction. Treating them as two independent shocks can duplicate the exposure as well as the revenue. The analyst should first determine whether the second observation confirms the first path or reveals a distinct one. Relationship definitions are in the Altsets methodology.

Sources

• Altsets supply chain dataset
• NVIDIA fiscal 2026 annual report
• NVIDIA fiscal Q2 2027 filing

Related research

• How to Research Unnamed Major Customers in a 10-K

• Nvidia Customer Concentration: Quanta vs. Samsung

• Missing Supply Chain Metrics Are Not Zero Exposure

• Indium Phosphide Capacity Is an AI Networking Bottleneck

Methodology

Read the methodology for this research.

Cite this research

Altsets Research. "Double Counting the Same Customer Demand in a Supply Chain Graph." Published September 6, 2026. https://www.altsets.com/research/avoid-double-counting-direct-indirect-customers

Copy citationCanonical URLMethodology