What Is Customer Concentration? Why a Few Buyers Bind the Cash
Customer concentration is a few buyers accounting for a large share of revenue. In the September 2026 quarter CoreWeave's largest share was 30% and Nebius's was 33%. The letter in the filing is not the buyer's name.
EXPLAINED · Mechanism · Last reviewed: 18 Nov 2026 · Next review: CoreWeave's 10-Q for the quarter ending 31 December 2026, when the 10% customer table is reprinted
Customer concentration is a few buyers accounting for a large share of a company's revenue. On the filings below, the line is 10%, and each buyer above it is printed as a letter, not a name.
In short. A neocloud is paid by a short list of buyers. In the quarter ended 30 September 2026, CoreWeave's largest share was 30% and the next was 24%. Nebius's largest was 33%. Our read: the stuck step is whether those buyers perform. The letter is not the company.
Why an investor cares: a GPU hour can be contracted and still depend on a few signatures. Those signatures sit on the Neoclouds layer. Who the specialists are is who the neoclouds are.
What is customer concentration?
Customer concentration is the share of revenue, or of the unpaid bill, that comes from a few buyers. Lose one buyer out of a wide set and the quarter still stands. Lose one of two, and it does not.
CoreWeave's note draws the line at customers that accounted for 10% or more of revenue. Nebius uses the same line for the same quarter. Anyone under it is not given a share. The line is a disclosure cutoff, not a credit score.
CoreWeave has one reportable segment, so these shares are shares of the whole company. Revenue recognized on customer commitments, including capacity delivered before a commitment's start date, was 98% of total revenue for the three and nine months ended 30 September 2026, and 98% in the same periods of 2025. The hours are mostly reserved. Who bears an idle hour is take-or-pay GPU contracts. This page is how few buyers clear 10%.
What did CoreWeave print for the September 2026 quarter?
CoreWeave's Form 10-Q for the quarter ended 30 September 2026 was filed on 12 November 2026. Revenue was $3,661 million for the three months, against $1,365 million a year earlier. For the nine months, revenue was $8,314 million, against $3,559 million.
The note lists every customer at 10% or more. A star in the filing means that customer did not represent 10% or more of revenue. The table below spells that star as "under 10%".
| Three months ended 30 Sep 2026 | Three months ended 30 Sep 2025 | Nine months ended 30 Sep 2026 | Nine months ended 30 Sep 2025 | |
|---|---|---|---|---|
| Customer A | 24% | 67% | 33% | 70% |
| Customer B | 30% | under 10% | 26% | under 10% |
| Customer C | under 10% | under 10% | under 10% | under 10% |
| Customer D | under 10% | under 10% | under 10% | under 10% |
Our arithmetic on that quarter, which the filing does not print as dollars: 0.30 × $3,661 million = $1,098.30 million for Customer B, and 0.24 × $3,661 million = $878.64 million for Customer A. Together that is 54%, or $1,976.94 million. The other 46% is $1,684.06 million, everyone the note does not break out.
The 10-Q says the quarter's revenue rose by $2.3 billion, or 168%, and that approximately 90% of the increase came from existing customers. The income-statement difference is $2,296 million. Our arithmetic on the filing's "approximately": 0.90 × $2,296 million = $2,066.4 million. It does not name those customers.
Why is the letter not a name?
Under the table, CoreWeave writes that letters A through D may represent different customers than in a previous period. That is why 67% in September 2025 and 24% in September 2026 are not one buyer getting smaller. The row is a label. We do not read Customer A as the company that went from 67% to 24%.
The June quarter shows the label moving. The Form 10-Q filed 12 August 2026, for the quarter ended 30 June 2026, carries the same warning. Revenue was $2,575 million. A was 36%, B 26%, and C 10%, together 72%, or $1,854 million, our arithmetic. By September, C is under 10% and the largest share is B at 30%. The letters may have been reassigned. That is not proof one account shrank.
The September risk factor still does not join a name to a letter. It says the top two customers were approximately 30% and 24%, with no other customer at 10% or more: the same two shares as B and A, with no legal entity attached. It also says OpenAI committed to pay up to approximately $6.5 billion through 31 May 2031, and that Meta initially committed to pay up to approximately $21.0 billion, for capacity through 20 December 2032 and an option through 10 April 2032. A credit-risk factor says the company announced in April 2026 that Jane Street had committed approximately $6.0 billion, and that other significant customers include Microsoft and OpenAI. "Up to" is a contract ceiling, not this quarter's 30%. The filing does not say which name, if any, is Customer A or B. We do not assign them.
Does the unpaid bill match the quarter?
No. The September 10-Q prints a second set of shares, this time for accounts receivable, net.
As of 30 September 2026, A, B, C, and D were 18%, 33%, 12%, and 14% of accounts receivable, net. The balance sheet states that line at $4,280 million. The four shares sum to 77%, which is $3,295.6 million, our arithmetic. B is 30% of the quarter's revenue and 33% of what is still owed. C and D are under 10% of revenue and above 10% of receivables. At 31 December 2025 the note had A at 68% and D at 11% of receivables. The letters may have changed, so that pair is not a trail to September.
Two other balances are not this split. Deferred revenue was $10.0 billion at 30 September 2026: cash already in for work not yet delivered. Deferred revenue and a customer prepayment are different mechanisms. Unsatisfied remaining performance obligations were $132.3 billion, and 36% of that was expected in the 24 months ending 30 September 2028. The note does not split the $132.3 billion by customer. The quarter's 30% is not 30% of the backlog.
What does Nebius print on the same line?
Nebius Group's unaudited condensed financial statements for the three and nine months ended 30 September 2026 were furnished as Exhibit 99.2 to its Form 6-K, filed 12 November 2026. Group revenue was $980.8 million for the quarter, against $146.1 million a year earlier, and $1,962.1 million for the nine months, against $302.1 million.
The note again lists customers at 10% or more. A star means the customer did not represent 10% or more.
| Three months ended 30 Sep 2025 | Three months ended 30 Sep 2026 | Nine months ended 30 Sep 2025 | Nine months ended 30 Sep 2026 | |
|---|---|---|---|---|
| Customer A | 29% | under 10% | 31% | under 10% |
| Customer B | 20% | under 10% | 16% | under 10% |
| Customer C | under 10% | 15% | under 10% | 20% |
| Customer D | under 10% | 22% | under 10% | 21% |
| Customer E | under 10% | 33% | under 10% | 21% |
Our arithmetic on the quarter, which the filing does not print as dollars: 0.33 × $980.8 million = $323.664 million for E, 0.22 × $980.8 million = $215.776 million for D, and 0.15 × $980.8 million = $147.12 million for C. Together that is 70%, or $686.56 million. On the nine months, C, D, and E are 20%, 21%, and 21%. E is a third of the quarter and only 21% of the nine months.
Nebius does not say whether a letter is the same buyer from one column to the next. We do not treat 2025's A as 2026's E, and we do not treat a Nebius letter as a CoreWeave letter.
The unpaid bill diverges. D was 75% of accounts receivable, net, at 30 September 2026, and E was 10%. The balance sheet states that line at $4,039.9 million. Our arithmetic: 0.75 × $4,039.9 million = $3,029.925 million. D is 22% of the quarter's revenue and 75% of what is owed. The note does not bridge the two. At 31 December 2025, C was 83% of receivables on a $720.3 million balance. The note does not say that C is September's D. It also says gross capital expenditure sat with two suppliers at 31 December 2025 and three at 30 September 2026, and it gives no shares.
Who gets paid — and where is the stuck step?
Our read: the neocloud is paid by the buyers over the 10% line, and by everyone under it. The letter hides the name. It does not hide the dependence.
CoreWeave's credit-risk factor states the stuck step. If customers fail to perform, the company may have excess capacity and may remain responsible for components, infrastructure, data center leases and build-outs, and related financing, without the corresponding revenue. The same factor says prepayments, credit checks, letters of credit, and guarantees do not fully remove that exposure, and that the company does not currently maintain credit insurance.
Who is paid on the build is the supplier of the components and the counterparty on the hall. Who is paid when the hour runs is the neocloud. What a neocloud is, and how a hyperscaler can absorb one lost GPU deal inside other lines, is neocloud versus hyperscaler. A signature reserves the hour. It does not make the buyer pay.
What would change this view?
- A filing that ties a letter to a legal name. Until then the names in the risk factor are not rows in the table.
- A quarter in which no customer reaches 10%.
- Receivable shares that match the quarter's revenue shares. On these two filings, they do not.
- A customer split of CoreWeave's $132.3 billion of remaining performance obligations. The note is silent.
Common misconceptions
- People assume the letter is the buyer's name. Actually neither filing ties a letter to a legal name, and CoreWeave says its letters may change between periods.
- People assume 67% last year and 24% this year are one buyer shrinking. Actually CoreWeave says the letters may be different customers. The row is a label.
- People assume the unpaid bill matches the quarter. Actually CoreWeave's B is 30% of September revenue and 33% of receivables, and C and D are under 10% of revenue but 12% and 14% of receivables. Nebius's D is 22% of the quarter and 75% of receivables.
- People assume "up to $21.0 billion" is this quarter's revenue. Actually that figure, and OpenAI's "up to $6.5 billion," are contract ceilings. The quarter's shares are 30% and 24%, and the filing does not join a ceiling to a share.
FAQ
What is customer concentration? Customer concentration is a few buyers accounting for a large share of revenue, or of the unpaid bill. On these filings the line is 10% of revenue, and each buyer above it is a letter.
What does the 10% line mean? It is the cutoff CoreWeave and Nebius print. Both list customers at 10% or more of revenue and mark everyone else as below that line.
Is Customer A the same company every quarter? Not necessarily. CoreWeave says letters A through D may represent different customers than in a previous period. Nebius does not say whether its letters stay put, so we do not assume they do.
How concentrated was CoreWeave in the September 2026 quarter? Customer B was 30% and Customer A was 24% of $3,661 million. Our arithmetic is $1,098.30 million and $878.64 million, together $1,976.94 million, or 54%. No other customer was at 10% or more.
Does the unpaid bill match that quarter? No. At 30 September 2026, A, B, C, and D were 18%, 33%, 12%, and 14% of the $4,280 million of accounts receivable, net. C and D were under 10% of revenue.
Who gets paid when a few buyers dominate? The neocloud is paid when those buyers perform. If they do not, CoreWeave says it may still carry the components, the hall, and the financing. The suppliers of that build are on a different clock.
Does a take-or-pay contract remove concentration? No. A take-or-pay changes who bears an idle hour. It does not change that a few buyers are the revenue.
Is this advice to buy or sell any company? No. This page explains how a few buyers show up in a filing. It does not rate companies or give buy or sell calls. Not investment advice. Do your own research.
Related terms
Related: what a neocloud is; who the neoclouds are; neocloud versus hyperscaler; take-or-pay GPU contracts; what deferred revenue is; what a customer prepayment is; remaining performance obligation.
Markets and stack doors
Layer: Neoclouds. CoreWeave's print is on the US Markets door. Nebius files the same 10% line with the US Securities and Exchange Commission. The company is in Amsterdam, and this site has no Netherlands door.
Get the Wire, a free daily note on who pays for AI computing.
How we check this. Every share, total, ceiling, and warning above is taken from CoreWeave's Form 10-Q filed 12 November 2026, its Form 10-Q filed 12 August 2026, or Exhibit 99.2 to Nebius's Form 6-K filed 12 November 2026. Where a dollar is a share times a printed total, it is marked our arithmetic. Who gets paid, and the stuck step, are our read. Not investment advice. Do your own research.
Last reviewed: 18 Nov 2026 · Next review: CoreWeave's 10-Q for the quarter ending 31 December 2026, when the 10% customer table is reprinted
Spot an error? Tell us and we will correct it and note the change here.
Sources
- CoreWeave — Form 10-Q for the quarter ended 30 September 2026, filed 12 November 2026. Revenue, the 10% customer table, the letter warning, receivables, commitments, deferred revenue, remaining performance obligations, and the named contract ceilings.
- CoreWeave — Form 10-Q for the quarter ended 30 June 2026, filed 12 August 2026. Revenue of $2,575 million and the 36%, 26%, and 10% shares, with the same letter warning.
- Nebius Group — Exhibit 99.2 to Form 6-K, filed 12 November 2026. Group revenue, the 10% customer table, receivable shares, and the supplier count with no shares.
Writes Second Order by TKN. Focus: who gets paid, where money sticks, and which step is stuck.