Datadog clears $1 billion a quarter while its GAAP margin sits at 1%

---

On May 7, 2026, Datadog reported the kind of quarter that makes a CFO's job easy. Revenue of $1,006 million, up 32% year over year. The first time the observability company had ever cleared a billion dollars in a three-month span. Annual recurring revenue past $4 billion. Roughly 4,550 customers now spending $100,000 or more a year, up from about 3,770 a year earlier. Remaining performance obligations of $3.48 billion, up 51%. Free cash flow of $289 million. Full-year guidance raised. The stock surged. Every headline that crossed the wire told the same story, and that story was true.

This article is not about whether Datadog is a good company. It plainly is. It is about the distance between how good the company is and how good it is priced to be — and about a small set of facts that the celebratory coverage tends to mention in passing, if at all. The most important of those facts is this: in the same quarter Datadog generated $1,006 million of revenue and $223 million of non-GAAP operating income, its GAAP operating income was $7 million. Its GAAP operating margin was 1%. The 21-point gap between the margin the company emphasizes and the margin the accounting rules require is not noise. It is the cost of running this business, paid in a currency — its own shares — that does not appear on the cash flow statement as an outflow but appears on every existing shareholder's stake as quiet erosion.

The bridge no one wants to walk across

Start with the one number that does the most work in any software bull case: the adjusted operating margin. Datadog reported a non-GAAP operating margin of 22% for Q1 2026, a figure that sounds like a profitable, mature software franchise. The trouble is what gets removed to get there. The single largest reconciling item between GAAP and non-GAAP is stock-based compensation, which was $197 million in the quarter. On $1,006 million of revenue, that is 19.6% — nearly a fifth of every dollar of revenue paid out in equity to employees.

Stock-based comp is a real expense. It is not a paper abstraction or an accountant's invention. When a company pays its engineers in shares instead of cash, it is buying labor with ownership, and the bill is settled by every other shareholder whose slice of the company gets thinner. The standard defense — "it's non-cash, so back it out" — is half-true and wholly misleading. It is non-cash to the company in the moment the grant vests. But it is intensely cash-relevant over time, because the only way to prevent the share count from ballooning is to spend actual cash buying stock back. Datadog's GAAP net income per diluted share was $0.15 in Q1 2026. That is the number after the accountants force the equity expense back onto the page, and it is a fraction of the adjusted figure investors anchor on.

The honest way to read the 22% non-GAAP margin is to read it next to the 1% GAAP margin and ask which one you would underwrite if you owned the whole company and could not dilute yourself. The answer is uncomfortable, because at a 1% GAAP operating margin, this $79 billion company is, by the rules everyone uses to value every other asset, barely breaking even on operations.

A multiple that assumes the cycle never turns

Now layer on the price. Datadog's market capitalization sat near $79 billion as of mid-June 2026, and its forward price-to-earnings ratio was roughly 95. On a price-to-sales basis, with a revenue run rate around $4 billion, the stock trades at roughly 19 to 20 times sales. These are not the multiples of a company the market expects to grow 25%. They are the multiples of a company the market expects to compound at 30%-plus for years, expand margins meaningfully, and do it without the growth rate decaying.

The company's own guidance tells a quieter story. Full-year 2026 revenue is guided to $4.3 billion to $4.34 billion — growth of 25% to 27%. That is a deceleration from the 32% just printed. Non-GAAP operating income is guided to $940 million to $980 million, a 22% to 23% margin. The forward guidance, in other words, embeds slowing growth and flat-to-modest margins on the adjusted basis that already excludes the SBC. At 95 times forward earnings, the price is not paying for the guidance. It is paying for a world in which the guidance proves conservative for years running.

There is a name for this asymmetry: priced for perfection. When a stock trades at this multiple, beating expectations buys a little upside, because the beat was half-expected. Missing them — a single quarter where usage growth stalls, where a large AI customer optimizes its bill, where net retention slips a point — does not cost a little. It re-rates the multiple, and at 19 times sales a multiple re-rating is measured in tens of billions of dollars of market value. The math is brutally simple. The higher the multiple, the more the entire thesis rests on the denominator never disappointing.

Usage-based billing is a tailwind that can blow the other way

Datadog's revenue model is its great strength and its hidden fragility. The company bills largely on usage — hosts monitored, logs ingested, custom metrics, spans traced. When customers grow their infrastructure, Datadog's revenue grows automatically, without a single new sales contract. In a year when the entire industry is spinning up AI workloads, GPU fleets, and inference clusters, that usage-based meter spins fast. Management has leaned hard into this, describing AI as a secular growth driver and noting strong traction with AI-native customers and traditional enterprises adopting AI workloads.

But a meter that spins up when usage rises spins down when usage falls. Usage-based revenue is, by construction, the most cyclical revenue in software. It has no floor of committed seats the way a per-seat SaaS contract does. When a customer's CFO launches a cost-optimization drive — and in observability, cost optimization is a perennial enterprise sport, because monitoring bills have a way of surprising people — the revenue can shrink without anyone canceling anything. The customer simply ingests fewer logs, samples more aggressively, drops retention windows. Datadog has lived through exactly this: in prior cloud-spending pullbacks, its growth decelerated sharply as customers optimized, precisely because the revenue rides the customer's consumption rather than a locked contract.

This is the heart of the cyclical-priced-as-secular problem. The market is currently treating Datadog's AI-driven acceleration as a permanent regime shift. It may partly be. But the same usage-based mechanism that delivered 32% growth in an AI capex boom is the mechanism that delivered far slower growth the last time enterprises tightened. You cannot claim the upside of a consumption model without owning its downside. A stock at 19 times sales is owning only the upside.

The AI-native cohort: powerful, concentrated, and footloose

Look closely at where the acceleration comes from. Management has disclosed that AI-native customers are a fast-growing, fast-diversifying cohort, and that the quarter included new deals with two of the world's biggest AI research teams to optimize their training workflows. In prior disclosures the AI-native cohort contributed on the order of low-double-digit percentages of revenue. That is a meaningful chunk of the growth coming from a narrow, distinctive set of customers.

Here is why that matters. AI-native companies — the model labs, the inference providers, the well-funded startups burning capital to train frontier systems — are the most volatile customers a usage-based vendor can have. Their consumption is enormous and their loyalty is thin. They are sophisticated enough to build their own observability tooling, to negotiate aggressively, to multi-source, and to slam the brakes on spending the moment their own funding or strategy shifts. A cohort that drives outsized growth on the way up is the same cohort that drives outsized deceleration on the way down. Concentration in fast-growing customers feels like strength when they are scaling and reveals itself as fragility when they are not.

There is a second-order risk specific to this cohort. The AI labs are themselves building platforms, and observability of AI systems is an area many of them consider strategic. A vendor whose growth depends on selling shovels to the people most capable of forging their own shovels is in a structurally precarious spot, however good the relationship looks today.

Net retention: still strong, but watch the trajectory, not the level

Datadog's trailing-twelve-month net revenue retention was in the low 120% range in Q1 2026, up from around 120% the prior quarter, with gross retention holding in the mid-to-high 90s. By any normal standard this is excellent — customers are expanding their spend faster than any are leaving. The bulls will point to it, correctly, as evidence of a sticky, expanding platform.

But net retention is a backward-looking, trailing metric, and its level matters less than its direction and its composition. A low-120s net retention rate today is well below the rates Datadog posted in its hypergrowth years, when the figure ran meaningfully higher. The expansion is real, but it is expansion off a base that is now enormous, and the law of large numbers is relentless. More importantly, a usage-based net retention number is flattered in a consumption boom and punished in a consumption bust. The same metric that reads low-120s in an AI capex surge read very differently the last time enterprises optimized. Reading a trailing retention figure at the top of a spending cycle and extrapolating it forward is the denominator illusion in its purest form: the number looks durable precisely when it is most exposed to mean-reversion.

The dilution treadmill and the buyback that isn't

Return to the $197 million of quarterly stock-based comp and follow it forward. Annualized, that is close to $800 million of equity expense — against a company guiding to under $1 billion of non-GAAP operating income for the entire year. Put plainly: the adjusted profit the bulls celebrate is roughly the same size as the equity compensation the adjustment removes. The non-GAAP operating income and the SBC are nearly the same order of magnitude. That is not a rounding difference. That is the business model.

To keep the share count from drifting upward, a company has to buy back stock at roughly the pace it issues it. Buying back stock at 19 times sales and 95 times forward earnings is one of the most expensive uses of capital imaginable — you are retiring your own equity at a premium valuation to offset dilution you created by paying people in that same overpriced equity. Either the share count grows, diluting holders, or the company spends its hard-won free cash flow buying high to stand still. Both outcomes are costs. Neither appears in the 22% non-GAAP margin. This is the quality-of-earnings question that the headline numbers are engineered to obscure: how much of the reported profitability survives once you account for the true cost of the labor that produced it?

Free cash flow is real — and partly a timing story

To be scrupulously fair, Datadog's free cash flow is genuine. The company generated $289 million of it in the quarter, a robust number that reflects a business with favorable working capital dynamics. Subscription software collects cash up front and recognizes revenue over time, so deferred revenue and billings can make cash flow look stronger than GAAP earnings during periods of growth.

But that same dynamic means free cash flow is partly a forward bet. It is strong when billings are accelerating and customers are pre-paying for expansion. It compresses when growth slows and the deferred-revenue tailwind fades. The 51% growth in remaining performance obligations is encouraging on this front — it suggests committed future business — but RPO is a contracted figure, and in a usage-based model the contract is a floor, not a ceiling or even a reliable predictor of consumed revenue. The cash is real. Its durability is a function of the very growth rate the multiple already assumes will persist.

What the bulls genuinely get right

Here the case for the defense deserves a fair and specific hearing, because the bull thesis on Datadog is not a fantasy — it is grounded in some of the best operating metrics in software.

First, the platform is genuinely excellent and genuinely sticky. Gross retention in the mid-to-high 90s means customers almost never leave outright; they embed Datadog across infrastructure, logs, APM, security, and more, and ripping it out is painful. The land-and-expand motion works: those 4,550 customers spending over $100,000 a year, up from 3,770, are the proof. This is a product people choose and then buy more of.

Second, the AI tailwind, even if cyclical, is real and may be large for a long time. Every new AI workload generates telemetry that someone has to monitor, and Datadog is positioned squarely in the path of that data. Management's disclosure that roughly 20% of customers now use AI integrations, representing about 80% of ARR, suggests AI adoption is broadening across the base, not concentrated in a few labs. If the AI buildout runs for years — and it may — the usage meter spins in Datadog's favor for years.

Third, the company is GAAP-profitable, even if barely. A 1% GAAP operating margin is thin, but it is positive, and it is positive while the company invests heavily in R&D and sales. Many software companies at this growth rate are deeply GAAP-unprofitable. Datadog reached the black years ago and has stayed there. The SBC is high, but it is funding real engineering that builds real product breadth, and the company has a track record of converting that investment into new revenue lines — security, AI observability, and more — that genuinely expand the addressable market.

Fourth, free cash flow is strong and consistent. $289 million in a quarter is not a mirage; it is cash in the bank that the company can eventually deploy. A business throwing off this much cash has real optionality and real resilience.

None of these strengths is in dispute. The disagreement is entirely about price. A wonderful business and a wonderful stock are not the same thing, and the gap between them is the multiple.

Demonstration versus deployment, in the customer's budget

There is a subtler frame worth naming. Much of the current AI-driven usage represents experimentation — enterprises standing up pilots, labs training models, teams instrumenting new AI applications to see what works. Experimentation generates a burst of telemetry and a burst of Datadog revenue. But experimentation is not the same as durable production deployment with a stable, growing budget behind it.

If a meaningful share of today's usage is demonstration rather than deployment, then some of the 32% growth is pull-forward — spend that happens once as customers explore, not spend that compounds. When the experimentation phase matures, some workloads scale into production and keep paying; others get killed, optimized, or moved in-house. A company priced for secular hypergrowth needs the deployment story to dominate the demonstration story. We will not know the mix until the AI capex cycle cools enough to separate the two — and by then, at this multiple, the knowing will be expensive.

The moat is real, but so is the loophole

Bulls describe Datadog's switching costs as a moat, and they are right that ripping out an embedded observability platform is painful. But it is worth distinguishing a moat from a loophole. A moat is durable structural advantage; a loophole is a temporary gap a competitor can close. Observability is an intensely competitive field. The hyperscalers — the very cloud providers whose infrastructure Datadog monitors — each offer native monitoring tooling, often bundled and often cheaper. Open-source stacks built on Prometheus, Grafana, and OpenTelemetry give sophisticated engineering teams a credible, lower-cost alternative, and the largest, most cost-sensitive customers are exactly the ones with the talent to assemble those stacks. Datadog wins today on breadth, polish, and integration, and that is a genuine advantage. But it is an advantage measured in product velocity, not in structural lock-in. A moat that has to be re-dug every quarter through continued R&D spend — funded, recall, by that $197 million of quarterly stock comp — is a moat with a maintenance cost. The market is pricing permanence into a position that requires perpetual reinvestment to defend. That reinvestment is not free, and the bill for it is the same dilution the adjusted margin pretends away.

The kicker

Datadog is a superb company. Nothing here disputes the product, the retention, the engineering, or the cash generation. The quarrel is arithmetic. A business growing 32% with a 1% GAAP operating margin, paying nearly 20% of revenue in stock, and guiding to deceleration, is being valued at 95 times forward earnings and roughly 19 times sales — a price that requires the most cyclical revenue model in software to behave like the most secular. The usage meter that spun the company past a billion dollars a quarter is the same meter that can spin the other way the moment a customer's CFO decides the monitoring bill has gotten too large. You are not paying for what Datadog is. You are paying for a future in which it never has a bad cycle.

The non-GAAP margin tells you how good the business is; the GAAP margin tells you how much of that goodness is still there after you pay the people who built it — and at this price the market has decided to look only at the first number.