Summary

  • Datadog's Q2 2026 revenue rose 36% to US$1.121454 billion. Of the US$294.694 million increase, the company attributed approximately 70% to existing customers and 30% to new customers.
  • The AI-native customer cohort, which includes Datadog's largest customer, contributed high-single-digit percentage points to company growth. Datadog then saw that largest customer reduce usage beginning in Q3.
  • The filing does not identify the customer, isolate its contribution, quantify the usage cut or say it cancelled. Q2 revenue, month-end ARR, trailing net retention and Q3 usage are four different clocks.
  • Broad customer growth, wider product adoption, US$279 million of quarterly free cash flow and the Q3 revenue guide argue against treating one customer signal as a company-wide collapse. They do not make the signal immaterial.

The quarter closed before the usage cut

Datadog's fiscal-Q2 Form 10-Q contains one sentence that changes how the preceding pages should be read. Revenue for the three months to 30 June 2026 rose 36% to US$1.121454 billion. The largest customer began reducing usage in the third quarter.

The order matters. The reduction did not cause the reported Q2 revenue, because it began after Q2 ended. Nor does the Q2 growth rate prove that the reduction has already been replaced, because the rate looks backward into a period before the change.

That simple boundary is easy to lose in an earnings narrative. A completed quarter arrives with recognized revenue, a quarter-end balance sheet and operating metrics measured through the closing date. A later operational fact arrives in the same filing because it is relevant to the next period. Placing the two on one line does not put them on one clock.

The correct reading is not “growth was false.” Datadog recognized the revenue, collected cash and expanded across a broad customer base. It is also not “the cut was already absorbed.” The filing says the reduction may decelerate revenue growth. The company gives neither the amount nor the duration.

This makes the disclosure unusually useful. It supplies a transition point without pretending to supply the whole bridge. Q2 is the last full quarter before the largest customer's lower usage. Q3 is the first test of how much other demand can refill that space.

Where the US$294.7 million increase came from

The reported arithmetic is exact. Q2 revenue increased from US$826.760 million to US$1.121454 billion, a gain of US$294.694 million. Datadog says approximately 70% of the increase came from existing customers and the remaining 30% from new customers.

Applied mechanically, those rounded shares correspond to about US$206 million of growth from existing customers and about US$88 million from new customers. These are calculations from an approximate attribution, not company-published revenue lines. They should not be given more precision than the disclosure carries.

Even with that limit, the mix explains why one usage change deserves attention. Existing-customer expansion supplied roughly seven dollars of every ten in the quarterly increase. Datadog's land-and-expand model is doing what it is designed to do: customers add monitored infrastructure, data, products and consumption after the initial purchase. The same mechanism also makes optimization economically visible. When an established customer produces fewer billable units, the change arrives inside the part of the model that created most of the prior growth.

The 70% does not belong to the largest customer. It covers the entire existing-customer base. It does not equal ARR growth, RPO growth, cash collected or the AI-native cohort's contribution. It is a company-wide allocation of the change in recognized revenue.

That separation protects both sides of the analysis. Assigning the whole US$206 million to AI would inflate one cohort. Ignoring the concentration signal because 33,400 customers exist would treat customer count as an equal-weight measure. Neither is supported.

The cohort is not the customer

Datadog offers a narrower, but still bounded, description in the risk section. Its AI-native cohort contributed high-single-digit percentage points to total company year-over-year revenue growth in Q2. The cohort includes the largest customer. It had rapidly increased usage; beginning in Q3, the largest customer reduced usage.

Three subjects appear in those sentences: the company, the cohort and one customer. They cannot be collapsed.

“High single digit” is not an exact percentage. It should not be rewritten as seven, eight or nine points. The contribution belongs to the cohort, not to the customer alone. The disclosure does not say how many customers are in the cohort, what portion came from the largest account, which products they used or whether the reduction changed committed, on-demand or both kinds of consumption.

It also does not name the customer. Public speculation would add a fact the filing withholds and could import unrelated assumptions about that company's finances or infrastructure. The economically relevant information is already available without the name: a high-growth cohort had enough scale to add a visible number of company growth points, and its largest member later lowered usage.

Reduced usage is not the same as cancellation. It can occur inside a live contract, against an unused commitment, in on-demand consumption, through telemetry filtering, workload change, price optimization, product removal or infrastructure consolidation. Those are possible mechanisms, not findings about this customer. Datadog does not disclose the reason.

The uncertainty is not a defect to be filled with a guess. It is the central control boundary. Investors can measure whether the aggregate business replaces the usage without knowing the customer's private operating record. They cannot honestly attribute cause, product or contract outcome until Datadog supplies more evidence.

Four clocks inside one usage business

Datadog's measures answer different questions.

Revenue asks what was recognized during the quarter. It includes performance delivered under subscriptions and usage arrangements during that period. Q2 revenue therefore belongs to the months ending 30 June.

Monthly run-rate revenue, or MRR, asks what one month's commercial run rate looked like. Datadog says it combines committed contractual amounts, additional usage, usage delivered as used against a commitment and monthly subscriptions. ARR multiplies that month's MRR by twelve. ARR is a point-in-time operating measure; Datadog expressly says it is not GAAP revenue or a revenue forecast.

Dollar-based net retention asks how a prior customer set changed. Datadog compares current ARR for the same customers with their ARR a year earlier, includes expansion and subtracts contraction or attrition, then reports a weighted average of the trailing twelve monthly observations. At 30 June, that trailing measure was in the low-120% range, up from about 120% a year earlier.

Usage asks what customers are consuming now. The largest-customer reduction began in Q3. A trailing average can remain strong while a current customer begins to contract because eleven earlier observations still carry weight. That is not a flaw in the metric; it is its design. The error is to use a trailing average as if it were a real-time meter.

RPO is a fifth witness with another boundary. Datadog reported US$3.4714 billion of remaining performance obligations at 30 June, only US$10.2 million above the 31 December balance. The company warns that timing for drawdown contracts is uncertain because future revenue can vary significantly from past revenue. RPO can preserve a contracted minimum or future instalment while actual consumption changes. It cannot reveal the undisclosed usage cut, and the six-month comparison is not a standalone demand verdict.

The result is a sequence, not one master number: contract rights set a perimeter; usage supplies units; MRR and ARR annualize a current state; retention averages the same-customer history; revenue recognizes delivered service; cash follows billing and collection. A large customer can move through those stages at different speeds.

Breadth is real counterevidence

The customer reduction matters because the customer is large. It does not follow that Datadog's demand base is narrow in every other dimension.

The company had approximately 33,400 customers at 30 June, up from 31,400 a year earlier. About 4,720 customers had ARR of at least US$100,000, up 23% from 3,850. That group represented 91% of ARR, compared with 89% a year earlier.

Product adoption also broadened. About 85% of customers used at least two products; 58% used four or more; 37% used six or more; 22% used eight or more; and 13% used ten or more. Each measure improved from the prior year. More products can deepen switching costs and create additional paths for expansion.

These figures are counterevidence, not an offsetting equation. The 4,720 large customers do not contribute equally, and the company does not publish the largest account's share. Product count says nothing by itself about dollars per product, margins or persistence. A customer can use many products and still optimize the volume inside them.

The relevant question is substitution capacity. Can thousands of existing customers expand enough, can new customers add enough, and can additional products contribute enough to replace the lower use of the largest account without expensive discounting? The disclosed breadth makes that plausible. Only later revenue, retention and margin can prove it.

Growth carries a delivery bill

Usage is valuable only after the cost of serving it. Datadog's Q2 cost of revenue rose 45%, faster than the 36% increase in revenue. Gross margin moved from 80% to 79%.

Of the US$74.135 million increase in cost of revenue, US$64.1 million—about 86.5%—came from third-party cloud infrastructure hosting and software. Datadog also spent more on infrastructure and software inside research and development. The platform is not a costless meter placed between a customer and its cloud.

A large usage reduction can therefore affect both sides of the income statement. Less consumption can reduce revenue, but some delivery cost may also fall. The net effect depends on committed capacity, provider terms, workload mix and timing. None is disclosed for the customer, so it would be wrong to apply the 79% gross margin mechanically to the missing usage.

The wider profit record remains strong. GAAP operating income was US$5.455 million, compared with a US$35.500 million loss a year earlier. The earnings release reported non-GAAP operating income of US$257 million. The large difference includes US$220.251 million of quarterly share-based compensation, US$27.643 million of employer payroll tax on employee equity transactions and other adjustments. Non-GAAP margin cannot erase those claims on value even when it is useful for comparing operating investment.

Cash adds another check. Q2 operating cash flow was US$316 million and free cash flow US$279 million. Cash and marketable securities totalled about US$5.0 billion. One usage cut is not evidence of liquidity stress. It is a question about the durability and mix of the next growth dollar.

The forward ruler is deliberately modest

Datadog guides Q3 revenue to US$1.135 billion–US$1.145 billion. Against Q2 actual revenue of US$1.121454 billion, that is a sequential increase of US$13.546 million–US$23.546 million, or about 1.21%–2.10%. The midpoint implies about 1.65% sequential growth.

This does not reveal the size of the customer cut. Guidance already combines the usage reduction with expansion elsewhere, new business, price, product mix and management's assumptions. It is a net range, not a lost-revenue estimate.

The full-year guide of US$4.45 billion–US$4.47 billion provides another constraint. First-half revenue was US$2.127880 billion, leaving US$2.322120 billion–US$2.342120 billion for the second half. Q3 and Q4 therefore must carry the usage change while preserving continued absolute growth.

The next quarter has three possible readings. If revenue reaches the range with stable retention and margin, broad expansion is replacing the cut. If revenue reaches the range but retention or gross margin falls sharply, replacement may be coming through a weaker mix or more costly demand. If revenue misses while the customer cut persists or spreads, the cohort signal becomes a wider growth problem.

The Q2 record makes the first scenario credible. The disclosure makes the other two impossible to dismiss.

Sources