Datadog Grew 35% in Q3 2026 While Seat-Priced SaaS Grew 11%: The Metered Gap

Datadog's revenue grew 35% to $1,197.1 million in the third quarter of 2026, reported on 4 November. Across four usage-billed infrastructure companies, median growth in the latest reported quarter was 36%. Across five seat-priced SaaS companies it was 11%. That 25-point spread is the clearest evidence in the 8-Ks that AI agents are a tailwind for metered software, and the Datadog Q3 2026 earnings also show where the tailwind stops.

The short answer to whether AI agents help usage-based SaaS growth: yes, in the filings that exist so far, and more than they help seat-priced software. Every usage-billed company in this comparison accelerated across four quarters. Every seat-priced company either held flat or slowed. Datadog's own data shows agent workloads multiplying the spans and tool calls it meters.

This post is written for founders and directors deciding how to price a product that agents will increasingly use. The comparison below is the cheapest evidence you will get on that decision, because nine public companies have already run the experiment with real customers.

What Datadog's Q3 2026 earnings actually showed

Datadog reported revenue of $1,197.1 million, up 35% from $885.7 million a year earlier. Non-GAAP operating margin was 28% and free cash flow was $251 million. The headline beat the top of its own guidance range, which is the pattern Datadog has held for years.

The numbers underneath the headline matter more for the metered argument. They show whether customers are committing to more usage, or just consuming what they already bought.

Backlog grew faster than revenue

Remaining performance obligations, contracted revenue not yet recognised, were $3,847.7 million at 30 September 2026, against $3,471.4 million at 30 June. That is an 11% rise in one quarter. CFO David Obstler said on the call that RPO grew 38% year over year and current RPO in the mid 30s.

Billings were $1.16 billion, up 30%. Billings lag revenue here because many commitments are invoiced annually, so I would weight the RPO line more. A backlog growing faster than revenue means customers signed for more future usage than they burned this quarter.

Customers above $100,000 carry 91% of ARR

Datadog ended the quarter with about 4,860 customers paying $100,000 or more in annual recurring revenue, up 20% from about 4,060. CEO Olivier Pomel said those customers generate about 91% of ARR. Total customers were about 33,500, up from about 32,000.

The ratio is the story. Total customers grew under 5%, while large customers grew 20% and revenue grew 35%. Datadog is not winning on logo count. It is winning on how much each large account consumes, which is exactly what a metered model rewards.

Trailing 12-month net revenue retention was in the low 120s, similar to the prior quarter, with gross retention stable in the mid to high 90s. Why those two numbers diverge so sharply between metered and seat models is covered in the comparison of net revenue retention across 12 Q2 filings.

Usage-based SaaS growth against seat-priced SaaS, four quarters of 8-Ks

A single company proves nothing about a pricing model. So the test needs a basket, fixed by pricing model before the numbers were read. Four companies bill mostly on consumption: Datadog, Snowflake, MongoDB and Cloudflare. Five sell mostly per seat or per employee: Salesforce, Workday, HubSpot, Zoom and DocuSign.

Each figure is the year-over-year growth rate stated in the company's own earnings release, mostly read from 8-K exhibits on SEC EDGAR. Each column is that company's four most recently reported quarters as of 7 November 2026. Fiscal calendars differ, so the latest quarter ended between 31 July and 30 September 2026.

An earlier Zan Digital tracker, the four-quarter test of whether SaaS is dead, used a different basket and stopped at quarters ending in August. This version adds Zoom, DocuSign, MongoDB and Cloudflare, adds the calendar Q3 prints from Datadog, HubSpot and Cloudflare, and measures the gap rather than the median.

Sources: each company's earnings release for the quarter. Datadog's latest quarter ended 30 September 2026; Cloudflare's and HubSpot's also; the rest ended 31 July 2026. "About" guides are Zan Digital arithmetic from the guidance range against the year-earlier quarter. Snowflake is product revenue. Salesforce includes Informatica from its Q4 fiscal 2026. Reported growth, not constant currency.

Every metered company accelerated, no seat company did

All four metered companies grew faster in their latest quarter than four quarters earlier. Cloudflare went from 34% to 39%, Snowflake product revenue from 29% to 37%, MongoDB from 19% to 30%, and Datadog from 29% to 35%. None of them is a small company: the smallest, MongoDB, booked $771.8 million in its latest quarter.

Among seat-priced companies, the best four-quarter move was Salesforce from 9% to 11%. About 4 points of that came from buying Informatica, which Salesforce's own guidance spells out. HubSpot slowed from 20% to 17% and now guides 12%. Zoom held at about 5%. DocuSign and Workday barely moved.

The choice favours size and clean pricing models. Every company here books more than $600 million a quarter, which rules out small names whose growth swings on one deal. Each sits clearly on one side of the pricing line, although HubSpot and Salesforce now sell credits alongside seats.

The obvious objection is that infrastructure and applications are different markets. They are. That objection is addressed directly in the section on where this argument is weakest, because I think it is the strongest one available.

The metered growth gap widened to 25 points in four quarters

Take the median of each group in each quarter and subtract. The gap was 20 points in the oldest quarter, 17.5 in the next, 20 in the third and 25 in the latest. It is not a straight line, but the latest reading is the widest of the four.

What the gap is made of

The widening came almost entirely from the metered side. The seat median rose 2 points across the period, and the metered median rose 7. So this is not a story about seat software collapsing. It is a story about consumption software speeding up while seat software stayed where it was.

That distinction matters for how you read the seat compression argument. If agents were already shrinking headcount at customers, the seat median would be falling. It is not, yet. What the filings show so far is a missing upside, not a visible loss.

One more comparison makes the scale concrete. Datadog added $311.4 million of quarterly revenue year over year. Zoom, with a larger quarterly revenue base of $1,277.2 million, added roughly $60 million on its stated 4.9% growth. Two companies of similar size, one billed on what software does and one on who logs in, added revenue at a ratio of about 5 to 1.

How AI agents feed observability bills

The mechanism is simple once you look at what an observability vendor counts. Datadog bills on hosts, ingested and indexed log volume, traces and, for AI workloads, LLM spans. None of those units is a person. All of them scale with the amount of software running, and an agent is software that runs far more steps per task than a human user triggers.

A single agent task can call a model several times, call tools between each call, retrieve documents and retry on failure. Each step can emit a trace span and log lines. A seat vendor sees one user. An observability vendor sees every step.

The State of AI Engineering report draws on LLM telemetry from Datadog customers, which its own methodology note says biases the sample. It found 69% of input tokens in March 2026 traces were system prompts, and only 28% of LLM call spans on caching-capable models showed any cached reads. Waste on that scale is good for a vendor that meters volume.

Why spans growing 9x does not mean revenue growing 9x

Datadog's LLM product is billed per LLM span, according to its pricing page, with tool and retrieval spans not billed as LLM spans. So a ninefold rise in all agent spans overstates billable growth. And Datadog has not disclosed what Agent Observability earns. My read is that it is still a small line, because if it were large the company would say so.

The tailwind runs through the core products instead. An agent that calls your APIs generates host load, logs and traces in your existing stack, all billed through infrastructure monitoring and APM. That is harder to attribute and probably larger. Where specialist tools compete for the agent-tracing budget is covered in the comparison of LangSmith, Braintrust and Arize.

Obstler said growth from customers excluding the AI cohort accelerated to the high 20% range, the fifth straight quarter of acceleration. That is the most underrated line in the release. It means Datadog's speed-up is not only AI labs spending on monitoring. Ordinary enterprises are consuming more too, and some of that consumption is their own agent deployments.

What Datadog's Q4 guide says about one large metered customer

Datadog guided Q4 2026 revenue to $1.19 billion to $1.20 billion. That is 25% to 26% growth, and at the midpoint it is about 0.2% below Q3. For a company that just grew 35%, a flat quarter is a striking guide, and the market reacted: SiliconANGLE reported the stock fell about 14% in regular trading on 4 November.

Obstler gave the reason directly: "Regarding our largest customer, our Q4 guidance contemplates their revenue at their minimum commitment." That customer reduced usage in Q3 while staying above its contractual floor. Excluding it, he said, implied Q4 growth is in the low to mid 30s.

Metering cuts both ways

The same property that lets Datadog grow 35% lets one customer subtract several points in a quarter. A seat vendor losing its largest customer loses that customer's licences at renewal. A metered vendor loses consumption the moment the customer optimises, down to whatever floor the contract sets.

Pomel described the pressure behind it: "companies everywhere are grappling with a very fast growth in token and AI spend". When agent costs balloon, the monitoring bill attached to them gets reviewed too. The concentration side of this story, including who the largest customer is reported to be, is laid out in the analysis of Datadog's AI-native revenue share.

The gap on guidance is still about 20 points

Using each company's own next-quarter guidance, the metered median falls to 31.5% and the seat median edges up to 11.5%. The gap narrows to 20 points, back to where it started four quarters ago. Two of the four metered names, Datadog and MongoDB, guide well below their latest print. I would not read that as the trend reversing. Metered companies guide to recent usage and then beat, and Datadog has beaten the top of its range every quarter in this period.

Where the agent tailwind argument is weakest

The biggest weakness is that the comparison mixes markets. Every metered company here sells infrastructure, and every seat company sells applications. Infrastructure spending tracks cloud workloads, and cloud workloads were rising for reasons that predate agents. A pricing-model effect and a market effect produce the same chart. Nine companies cannot separate them.

The cleaner test would compare metered and seat products inside one company. Few disclose that split. HubSpot and Salesforce now sell credits beside seats, but neither reports credit revenue on its own line, so the within-company test is not yet possible from filings.

Agents are not the only cause of the acceleration

Snowflake's and MongoDB's acceleration has explanations unrelated to agents, including migrations and new products. Cloudflare sells security and networking whose demand is not mainly agent-driven. The CEO statements that tie Datadog's telemetry growth to agents are spoken, unaudited and come with no base. I think agents are a real contributor. I cannot show from these filings that they are the main one, and nobody else can either.

Five against four is a small sample, and the medians move a lot when one company moves. The quarters are not aligned: Datadog, HubSpot and Cloudflare reported September quarters, the rest July quarters. Salesforce's growth includes an acquisition. Calendar Q3 prints from Salesforce, Workday, Zoom, Snowflake and MongoDB land in late November and December, and any of them could change the latest medians by a point or two.

What usage-based SaaS growth means for your pricing model

If you sell software that agents will call, the filings above say the unit you bill on decides whether agent adoption shows up in your revenue. A seat model does not charge for the ten thousand extra API calls an agent makes. A metered model does, and absorbs the volatility that comes with it.

The hybrid most vendors are choosing

HubSpot, Salesforce, Workday and monday.com have all added credits on top of seats in the past year. Datadog moved the other way on its AI products, to a model of AI credits sold against commitments. Both are converging on the same hybrid: a committed floor, plus metered usage above it. Datadog's Q4 guide shows why the floor matters, because it is the only thing standing between one customer's optimisation and a much worse quarter.

My position is that a seat-only model is now the riskier default for any product that agents will use heavily, and that pure metering without commitments is too volatile for most companies below Datadog's scale. The evidence on how blended models perform on retention is in the analysis of hybrid pricing and the NRR gap.

Figures in this post were last verified on 7 November 2026 against the earnings releases and filings listed below. Re-check the seat medians after Salesforce, Workday and Zoom report in late November and early December.

Frequently asked questions

Where to start this week

Pull your last four quarters of revenue and split it by billing unit: seats, usage, or committed credits. Then find your five accounts with the most agent or API activity and check what each paid per unit of that activity. If the answer is close to zero, agents are using your product for free, and the Datadog filings show what that costs you in growth.

If you already meter, look at your largest customer the way Datadog's CFO had to. Write down its contractual floor and the growth rate you would report if it dropped to that floor next quarter. If that number would worry your board, negotiate higher commitments at the next renewal, before the customer starts optimising.

Related reading