4 Buyers Took 61% of Nvidia's Sales. Four Filings Show Where It Moved

In the quarter to 26 October 2025, four direct customers accounted for 61% of Nvidia's revenue, about $34.8 billion of a $57.0 billion quarter. Nine months later, Nvidia's 10-Q named only one direct customer above the 10% line. Read alone, that looks like Nvidia customer concentration easing. Read across four filings and converted into dollars, it says something narrower: the buyers moved, the money owed did not.

The short answer, for anyone checking Nvidia customer concentration before the next 10-Q, comes from the last four filings. Across them, between one and four direct customers each crossed 10% of revenue, at combined shares of 16% to 61%. Those buyers are mostly server assemblers. The concentration that matters for durability sits one layer down, and Nvidia only estimates it.

Nvidia customer concentration in four filings, tabulated

Every Nvidia 10-Q and 10-K carries one paragraph on customer concentration. It lists direct customers above 10% of revenue as percentages, without names. Coverage usually quotes the latest paragraph on its own. That loses the only thing the paragraph is good for, which is direction.

The table below puts the last four filings side by side. It covers the third quarter of fiscal 2026, the fiscal 2026 annual report, and the first two quarters of fiscal 2027. Nvidia's fiscal year ends in late January, so fiscal 2027 is mostly calendar 2026.

Percentages and revenue are from each filing. Dollar columns are Zan Digital arithmetic, share multiplied by revenue. Nvidia rounds shares to whole numbers, so each dollar figure carries an error of up to half a point of revenue, about $0.5B in Q2 FY2027. The 10-K row is a full year and is not comparable to the quarterly rows.

Two patterns stand out. The number of disclosed direct customers fell from four to one in three quarters. And the combined share in the latest quarter is the lowest in the series, at 16%. Our view is that neither pattern means what it appears to mean, and the rest of this post explains why.

Nvidia's top buyers in dollars of data center revenue

A share of total revenue is the wrong unit for anyone asking how much of the AI build rests on a few buyers. Total revenue includes gaming cards and workstations. The direct customers above 10% are, in Nvidia's words, primarily attributable to the Compute and Networking segment. So the useful comparison is against data center revenue.

Why the denominator changes the answer

In Q3 fiscal 2026, data center revenue was $51.2 billion. The four disclosed direct customers took about $34.8 billion of total revenue. If nearly all of that was data center hardware, which the segment attribution suggests, those four buyers accounted for roughly 68% of data center revenue. That is the most concentrated reading in the series.

In Q1 fiscal 2027, the three disclosed customers took about $44.1 billion against $75.2 billion of data center revenue. That is roughly 59%. In Q2 fiscal 2027, the single disclosed customer took about $15.4 billion against $89.0 billion, or 17%. The rest of Q2's data center revenue is spread across customers each below $9.6 billion, which is 10% of total revenue.

The half-year figure is the better reading

Quarterly disclosures exaggerate swings, because a customer near the 10% line flips in and out. Nvidia also gives a six-month figure. For the first half of fiscal 2027, three direct customers held 16%, 15% and 13% of revenue. That is 44% of $177.8 billion, or about $78.2 billion. Against first-half data center revenue of $164.3 billion, it is roughly 48%.

So the fair headline is not "concentration collapsed to 16%". It is that about half of Nvidia's data center sales in the first half of 2026 went through three direct buyers. We would use that figure, not the single-quarter one, in any model of Nvidia's dependence on its channel.

The chart adds context the concentration paragraph cannot. Q3 fiscal 2026 was the quarter hyperscale revenue jumped from $24.2 billion to $35.1 billion, while the other data center line, AI Clouds, Industrial and Enterprise (ACIE), slipped from $16.9 billion to $16.1 billion. A handful of the largest clouds buying through a handful of server makers produces exactly the 61% reading. The later decline in both measures tracks the rise of a broader set of buyers. Who sits inside ACIE is covered in the analysis of which customers carry Nvidia's $108 billion Q3 guide.

The 21% customer that dropped below the 10% line

The most informative fact in these filings is one Nvidia never states. It falls out of the gap between the quarterly and half-year figures, and it is the original arithmetic in this post.

The arithmetic, step by step

In Q1 fiscal 2027, one direct customer took 21% of $81.6 billion, about $17.1 billion. Allowing for rounding, at least $16.7 billion. In Q2, only one direct customer passed 10%, at 16% of $96.2 billion, at least $14.9 billion after rounding.

Suppose the Q1 leader was that Q2 customer. Its first-half revenue would be at least $31.6 billion. That is 17.8% of first-half revenue of $177.8 billion, which Nvidia would report as 18%. The half-year disclosure lists 16%, 15% and 13%. No 18%. So the Q1 leader was not the Q2 customer above 10%.

Zan Digital arithmetic on figures from Nvidia's Q1 and Q2 fiscal 2027 10-Q filings and the Q2 CFO commentary. The test needs only the published percentages and revenue totals. It does not identify the customer.

The implication is a fall of at least 44% in one quarter, from about $17.1 billion to under $9.6 billion, at the customer Nvidia relied on most in Q1. Nvidia's total revenue rose 18% across the same quarter. Something moved volume away from that buyer and toward others.

We do not think this signals lost demand. The likelier reading is a shift in which contract manufacturer assembled which racks, perhaps tied to the transition from one system generation to the next. But we cannot prove that from the filing, and neither can anyone outside Nvidia and its channel. The point is that the 16% headline hides a $7.5 billion swing at a single buyer.

What the 10-K hides: an implied Q4 fiscal 2026

Annual reports give full-year percentages only. Nvidia's 10-K says one direct customer took 22% of fiscal 2026 revenue and another took 14%. The fourth quarter, which ended on 25 January 2026, has no concentration disclosure of its own. It can be estimated.

The Q3 10-Q gives a nine-month figure. Two direct customers took 21% and 13% of nine-month revenue of $147.8 billion. If those are the same two customers named in the 10-K at 22% and 14%, subtraction gives their fourth-quarter revenue. The larger one comes to about $16.5 billion, or 24% of Q4 revenue of $68.1 billion. The smaller comes to about $11.0 billion, or 16%.

Rounding widens both. The larger customer's Q4 share could sit anywhere from 21.5% to 26.8%. The smaller could sit from 13.5% to 18.8%. Both ranges stay above 10%, which means the fourth quarter had at least two direct customers above the line and probably a combined share near 40%. That bridges the 61% of Q3 and the 54% of Q1 more smoothly than the annual 36% suggests.

This estimate depends on an assumption we cannot test: that the 10-K's two customers are the nine-month pair. Q3 had four customers above 10%, so a different pairing is possible. We are confident about the direction, at least two above 10% in Q4, and much less confident about the exact shares.

Receivables concentration: five customers, 70% of $63.1B

Revenue concentration measures who bought. Receivables concentration measures who still owes. Over the same four filings the two moved in opposite directions, and this is the divergence that deserves the most attention.

Shares and balances from Nvidia's Q3 fiscal 2026, Q1 fiscal 2027 and Q2 fiscal 2027 10-Q filings. The January 2026 shares appear as the comparison period in the fiscal 2027 10-Qs. Dollar column is Zan Digital arithmetic.

In Q2 fiscal 2027, revenue concentration fell to one disclosed customer. In the same quarter, the number of customers holding 10% or more of receivables rose to five. The dollars owed by that group roughly doubled in a single quarter, from about $26.1 billion to about $44.1 billion.

Nvidia explains part of it. Its Q2 CFO commentary says receivables of $63.1 billion carried 60 days sales outstanding, up from 45, "due to extended payment terms on large, multi-quarter agreements with certain investment-grade customers". That is a reasonable commercial choice when buyers are creditworthy. It also means the revenue is booked before the cash arrives, and a few counterparties hold most of the gap.

The full working capital picture, including inventory and supply commitments, is in the breakdown of Nvidia's receivables, inventory and $279 billion supply bill. For this post, one comparison is enough. The largest receivable at 26 July, 22% of $63.1 billion, was about $13.9 billion. That is close to a full quarter of revenue from the largest disclosed buyer.

One reclassified company moved $5.2B of data center revenue

Since fiscal 2027, Nvidia has split data center revenue into hyperscale and ACIE. During the second quarter it changed the split. The CFO commentary says: "we reclassified a company from AI Clouds, Industrial, & Enterprise (ACIE) to Hyperscale due to a change in their business model and recast the prior period revenue associated with this company".

Nvidia did not name the company or size the change. Both published versions of Q1 fiscal 2027 exist, so the size can be computed. The Q1 10-Q, filed in May, reported hyperscale revenue of $37.9 billion. The recast table published in August reports $43.1 billion for the same quarter. The difference, $5.2 billion, is the revenue associated with that one company in Q1.

The same subtraction works a year earlier. For Q1 fiscal 2026, the Q1 fiscal 2027 10-Q showed hyperscale at $17.6 billion. The recast table shows $22.3 billion. So that company accounted for about $4.7 billion of revenue in a quarter when Nvidia's total revenue was $44.1 billion. That is roughly 10.6% of Nvidia's entire revenue, from one indirect customer, by our arithmetic.

That figure matters for two reasons. It gives a concrete size to the filing's statement that some indirect customers individually represent 10% or more of revenue. And it shows that the hyperscale and ACIE trend, which much commentary treats as a demand signal, can move several points on a classification decision. By May 2026 standards, hyperscale was about half of data center revenue. By August standards, the same quarter was 57%.

Direct customers are not Nvidia's end buyers

Every figure above describes direct customers. That word carries a specific meaning in Nvidia's filings, and it is not the meaning most readers assume.

Who the direct customers are

The Q2 fiscal 2027 10-Q defines direct customers as those purchasing from Nvidia directly. It gives examples: "such as add-in board manufacturers, or AIBs, distributors, original design manufacturers, or ODMs, original equipment manufacturers, or OEMs, cloud service providers, or CSPs, AI model makers, and system integrators". In practice the largest are the contract manufacturers and server makers that turn Nvidia parts into racks.

That is why a single direct customer can swing by $7.5 billion in a quarter. It can reflect which assembler won which rack programme, not a change in who wanted the compute. The server makers' own side of this, including how much of their AI backlog sits with a few buyers, is examined in the breakdown of Dell's AI server backlog and customer mix.

The indirect layer Nvidia estimates

The concentration that matters for durability is indirect. The fiscal 2026 10-K states: "We generate a significant amount of our revenue from a limited number of indirect customers, some individually representing 10% or more of our revenue". It adds that Nvidia estimates "one AI research and deployment company contributed to a meaningful amount of our revenue purchasing cloud services from our customers in fiscal year 2026".

Nvidia gives no percentage for that company and does not name it. The Q1 fiscal 2027 10-Q repeats the warning in plain terms: "Our revenue is concentrated among a limited number of direct and indirect customers and this trend may continue". We read that sentence as the most honest line in the filing. A model lab renting GPUs from a cloud provider never appears in the direct customer list, yet it may be the reason those GPUs were bought.

The financing links between Nvidia and those labs, including the equity stakes and the guarantees, are covered in the explainer on how the AI circular deals work. For concentration, the takeaway is simple. The direct list tells you who invoices. The indirect estimate tells you who pays, and it is the less precise of the two.

Where this analysis is weakest

Three limits apply, and the first is the most serious. Nvidia does not name direct customers and does not label them consistently across filings. "Customer A" in Q3 fiscal 2026 may or may not be the 22% customer in the 10-K. Every cross-period inference above, including the implied Q4 and the fallen 21% customer, rests on percentages without identities. The Q1 to Q2 test is the strongest, because it needs no matching assumption. The Q4 estimate is the weakest, because it needs one.

Second, our conversion to data center revenue assumes the disclosed direct-customer revenue is almost entirely data center hardware. Nvidia says it is primarily attributable to the Compute and Networking segment. That segment is close to, but not the same as, data center. A customer buying workstations or automotive parts in volume would skew the ratio. We think the error is small. We cannot measure it.

Third, the strongest counter-argument is that none of this matters. Direct customers are channels, and concentration among channels is normal for a component supplier. If one assembler loses volume, another gains it, and the end demand is unchanged. That view is largely right about revenue. It is weaker about receivables, where $44.1 billion is owed by five counterparties on longer terms. Nvidia says those customers are investment grade, and we have no evidence otherwise. Investment grade still describes probability, not certainty.

Last verified: every figure in this post was checked against the filings listed in the references on 2 November 2026. The Q3 fiscal 2027 10-Q, due in late November, will add a fifth column, and we will update the tables then.

Frequently asked questions

Where to start

Build the fifth column before the Q3 fiscal 2027 10-Q lands. Copy the two tables in this post into a sheet, add a row for the next filing, and compute the dollar values the same way. Then run the half-year test on the nine-month figure. If the Q2 leader at 16% drops out in Q3 too, the channel is rotating fast, and the single-quarter percentage should carry even less weight in your view of Nvidia.

If you only track one number from that filing, make it receivables held by customers above 10%, in dollars. Revenue concentration has been falling for three quarters. Receivables concentration has been rising. When those two lines diverge, the one that measures cash still owed is the one we would trust.

Related reading