Datadog's AI-Native Customers Hit 12% of Revenue, Then the Number Went Quiet

Datadog's AI-native customers made up 12% of its revenue in the third quarter of 2025. That was the last time the company published the share. Rebuilding it from the figures Datadog still discloses puts the cohort at roughly 16% of the $1.12 billion it booked in Q2 2026, and one unnamed customer inside that cohort has just cut its usage.

The short answer to how much Datadog revenue rides on AI labs: about one dollar in six, on a reasonable reconstruction, and a meaningful but undisclosed slice of that sits with a single customer. The cohort grew from 4% to 12% of revenue between Q2 2024 and Q3 2025. Datadog's largest customer, widely reported as OpenAI, reduced usage from Q3 2026.

This post is written for founders and directors whose own revenue leans on a handful of fast-growing customers. Datadog is the cleanest public case of a software vendor riding an AI lab's spend, and its filings let you practise the concentration arithmetic on someone else's numbers first.

The Datadog AI-native disclosure ledger, quarter by quarter

Datadog defines its AI-native cohort as companies whose core product is built on AI, from model labs to coding tools. For six quarters the CFO, David Obstler, reported the cohort's share of revenue or ARR in prepared remarks. Then the series stopped, and the company switched to other markers.

The table below collects every figure Datadog stated on its calls and in its annual report, in its own words where possible. Read the right-hand columns as a change of vocabulary, not only a change of numbers.

Sources: Datadog earnings call transcripts for each quarter; the Q4 2025 contribution figure is from the FY2025 Form 10-K risk factors; Q1 and Q2 2026 revenue from the Q2 2026 Form 10-Q. Transcripts are third-party transcriptions of the calls.

Why the switch in metric matters

On the Q3 2025 call, Obstler said he expected the share metric to "become less relevant as AI usage in production broadens beyond this group of customers". That is a reasonable argument. It is also the point at which the cohort was growing fastest and the largest customer was renegotiating.

The replacement disclosure, non-AI growth, is not a lesser number. It is arguably more useful, because it tells you what the rest of the business does without the cohort. But it does not tell you the cohort's size, and size is what a concentration question needs.

The definition also moved. By Q2 2026 the CFO described the group as about 750 "AI customers" that "now also includes hyperscalers using Datadog for in-house AI labs". A cohort that absorbs the AI divisions of the largest technology companies is a different object from the 2024 cohort of startups. Any series built across that change carries the drift with it.

Rebuilding Datadog's missing AI-native revenue share

Datadog stopped publishing the share, but it kept publishing two inputs that let you reconstruct it. It discloses total revenue in its 10-Q to the thousand dollars. And it discloses year-over-year growth for customers excluding the AI-native group.

If you know what the non-AI business was a year ago, and how fast it grew, the rest of this quarter's revenue belongs to the AI-native cohort. The arithmetic is simple enough to check by hand, which is the point.

The Q2 2026 working

Start from Q2 2025 revenue of $826.8 million. The CFO said AI-native customers were about 11% of it, so roughly $90.9 million was AI-native and $735.9 million was everything else.

On the Q2 2026 call, both executives said non-AI revenue grew in the "high 20%" range. Grow $735.9 million at 27% to 29% and you get $935 million to $949 million. Subtract that from Q2 2026 revenue of $1,121.5 million and the AI-native cohort comes to about $172 million to $187 million.

That is 15.3% to 16.7% of revenue. It also implies the cohort grew roughly 89% to 105% year over year, close to doubling, while its share of the company rose by about 5 points.

Checking the method against a quarter that was disclosed

A method you cannot test is a guess. So run it on Q3 2025, where Datadog did publish the answer. Q3 2025 revenue was $886 million, up 28%, which implies about $692 million a year earlier. At about 6% AI-native, the non-AI base was about $651 million.

Grow that base at the disclosed 20% and you get $781 million. The remainder, $105 million, is 11.9% of Q3 2025 revenue. Datadog's published figure was 12%. The method lands within rounding of the official number, which is as much validation as a reconstruction can get from public data.

I trust this reconstruction more than any analyst estimate I found, because every input is a number the company said itself. That does not make it exact. It makes it checkable.

The same method gives about 12% to 14% for Q1 2026. One caution: in the Q1 2026 transcript, the CEO is recorded saying non-AI growth was "20%" while the CFO says "mid-20s percent". The Q2 2026 call settles it, describing growth as "up from the mid-20s last quarter," so the estimate uses the CFO's figure.

How much of Datadog's growth the AI-native cohort supplies

Share of revenue is a stock. Share of growth is a flow, and for a company valued on growth, the flow is what moves the stock price. Datadog disclosed the flow directly for three quarters.

AI-native customers contributed about 2 points of year-over-year growth in Q1 2024, about 6 points in Q1 2025, and about 10 points in Q2 2025. In its FY2025 annual report, Datadog stated the cohort, "which cohort includes our largest customer," represented approximately 7 percentage points of Q4 2025 growth.

The Q2 2026 waterfall in dollars

Datadog's revenue rose by $294.7 million year over year in Q2 2026. On the reconstruction above, about $206 million of that came from non-AI customers and about $81 million to $96 million from the AI-native cohort.

So roughly 27% to 33% of the latest quarter's dollar growth came from a group that was 11% of revenue a year earlier. That ratio, a cohort supplying three times its weight in growth, is the signature of concentration risk. It is also the signature of a cohort that is working.

Rows three and four are estimates with the range driven entirely by how "high 20s" is read. Neither row is a Datadog disclosure.

The largest customer is the concentration that matters

Cohort share overstates the risk in one way and understates it in another. Most of the roughly 750 AI customers are small and diversified. One of them is not, and Datadog has spent six quarters managing investor attention around it.

In May 2025, CEO Olivier Pomel told analysts: "There's also a revenue concentration there. We now have our largest customer in that cohort, and they're growing very fast". In Q3 2025 he announced "a nine-figure annualized expansion with a leading AI company," an early renewal at "better economics for a higher commitment".

What changed in August 2026

On the Q2 2026 call, Obstler said: "Regarding our largest customer, we have seen a usage reduction, which is incorporated in our Q3 and full year 2026 guidance". Pomel added that the customer had renewed, uses 17 Datadog products, and that "there's not a lot more we can share".

The shares fell sharply on the day. Investing.com reported a premarket drop of more than 21% despite revenue and guidance both beating consensus. That reaction is the market pricing a concentration it cannot measure.

Who the customer is, and what is actually known

Datadog has never named the customer. In July 2025, Guggenheim analyst Howard Ma downgraded Datadog to Sell, citing risks related to its largest customer, which StockStory's report of the note identified as OpenAI. Ma estimated that optimisation by OpenAI could open a revenue gap of $150 million or more for Datadog in 2026.

Treat both the identity and the $150 million as analyst claims, not company facts. Still, the figure is useful as a ceiling for scale. Against the $4.46 billion midpoint of Datadog's 2026 revenue guidance, $150 million is about 3.4% of revenue. If Ma was right, the gap is real and survivable. The economics on the OpenAI side, and why its vendors should watch its commitments, are covered in the analysis of OpenAI's compute commitments against its revenue.

Pomel has described the same sequence for large customers since 2025: usage grows, the customer commits to more, it gets a lower unit price, revenue dips, then grows again. In Q3 2025 he put it plainly: "you see growth of usage, drops in revenue as customers renew and get a higher commit and a better price".

That is a standard volume-discount motion, and Obstler has said margins on large customers depend on size, not on whether they are AI-native. My read is that the August 2026 usage cut is different in kind, because the renewal came with less consumption rather than more. The pattern Datadog describes explains price drops. It does not explain usage drops.

What Datadog's guidance already assumes about AI-native risk

Datadog guides conservatively, and it says so. Obstler said the company bases guidance "on trends observed in recent months and apply conservatism on these growth trends". He said it treated the largest customer with "a higher degree of conservatism" than the rest of the base.

The numbers make that visible. Q3 2026 guidance of $1.135 billion to $1.145 billion implies 28% to 29% growth. Full-year guidance of $4.45 billion to $4.47 billion, minus the $2,127.9 million reported for the first half and the Q3 range, leaves an implied Q4 of about $1.187 billion to $1.197 billion.

Against Q4 2025 revenue of $953 million, that is about 25% to 26% growth, down from 36% in Q2. Some of the step-down is the largest customer. Some is conservatism. Datadog has beaten the top of its revenue range in every quarter in this ledger, so the implied deceleration is a floor, not a forecast. How Datadog's beat pattern compares with peers is tracked in the scorecard of SaaS earnings guidance against results.

The backlog line nobody is quoting

One more figure belongs in this picture. Datadog's remaining performance obligations, contracted revenue not yet recognised, were $3,471.4 million at 30 June 2026, against $3,461.2 million at 31 December 2025. That is flat across six months in which quarterly revenue grew about 11%.

RPO is lumpy, and a large early renewal in late 2025 can pull commitments forward. I would not read a trend into two data points. I would watch whether RPO resumes growth on 4 November, because a renewal with lower consumption should show up there first. Backlog disclosures across cloud vendors are compared in the breakdown of cloud backlog and RPO figures.

Where this Datadog analysis is weakest

Every reconstruction in this post depends on two inputs that are rounded speech, not audited figures. "High 20s" could be 27% or 29%, and that alone moves the AI-native estimate by more than $14 million. The ranges above are honest about that, but they are not small.

The cohort definition also drifts. Hyperscaler AI labs joined the "AI customers" group during 2026. If Datadog's non-AI growth figure now excludes a broader set than the 2025 cohort, the reconstruction compares two different baselines, and the 16% estimate would overstate the like-for-like share.

Third, the transcripts used here are third-party transcriptions, not Datadog's own documents. They contain visible errors, including the CEO's name transcribed as "Ali". Every figure above was checked across both executives' remarks where possible, but a transcription slip in a percentage would carry straight into the arithmetic.

Finally, the strongest counter-case deserves a fair hearing. Pomel argued on the Q2 2026 call that "if you back out a large customer from our growth, you get pretty much the same growth rate," and non-AI growth accelerated for five straight quarters. On that evidence Datadog's exposure looks more like an upside option than a dependency. I think that case is mostly right about the business and mostly wrong about the disclosure.

A concentration test you can run on your own revenue

The useful lesson from Datadog is not about Datadog. It is a reusable method for any vendor whose growth leans on a few fast-scaling customers, including your own company if you sell into AI labs.

Share of growth is the number to put in front of your board

A customer at 5% of revenue that supplies 25% of growth is a bigger risk than a customer at 10% supplying 10%. Datadog's AI-native cohort ran at roughly three times its revenue weight in growth contribution. If any account in your book runs above two times, the board should see it by name.

The way public companies report AI exposure varies widely in usefulness, and the ranking of AI revenue disclosure metrics sorts the useful ones from the decorative. The same concentration arithmetic, applied to a chip supplier rather than a software vendor, appears in the read of Nvidia's guidance and customer concentration.

What to read in Datadog's 4 November results

Datadog reports Q3 2026 results on 4 November 2026, two days after this post. Four lines will tell you more about AI-native exposure than the headline revenue number.

First, the non-AI growth rate. If it holds in the high 20s, the reconstruction method still works and the cohort share can be rebuilt again. Second, the count of AI customers above $10 million, which went from 5 to 8 in one quarter.

Third, any wording on the largest customer. "Stabilised" and "in line with our assumptions" mean different things. Fourth, RPO against the $3,471.4 million June figure. I would weight the RPO line more than any of the others, because it is the only one of the four that is audited rather than spoken.

Figures in this post were last verified on 2 November 2026 against the Q2 2026 Form 10-Q, the FY2025 Form 10-K and the earnings call transcripts listed below. They should be re-checked after the Q3 release.

Frequently asked questions

How to put this to work before 4 November

Open your own revenue by customer for the last four quarters. Compute one number for each top-ten account: its share of year-over-year growth, divided by its share of revenue. Any account above 2 goes on the next board deck with its renewal date beside it.

Then, if you hold or follow Datadog, write down your estimate of the Q3 AI-native share before the release, using the method above. Comparing your number with whatever Datadog says on Wednesday is a faster education in concentration risk than any analyst note.

Related reading