How does Datadog make money in 2027?

What the business model actually is and why it matters to RevOps

Datadog sells software-as-a-service monitoring. That sentence is accurate and almost useless, because the interesting part is not *what* it sells but *how the meter runs*. Datadog is a metered platform business wearing subscription clothing. A customer signs an annual commitment — say $400,000 — and then draws that commitment down through usage across whatever products they turn on. When they exceed the commitment, they either true up mid-term or roll into a larger commitment at renewal. The commitment is the contract; the meter is the revenue.

That structure has three consequences that matter to anyone modeling the business.

First, revenue is a function of the customer's infrastructure, not the customer's headcount. A seat-based tool like a CRM grows when the customer hires. Datadog grows when the customer's environment grows — more Kubernetes nodes, more services, more logs, more traffic. This is why Datadog's growth tracked cloud migration so tightly through the early 2020s and why cloud-cost-optimization initiatives at large customers show up directly in Datadog's numbers as decelerated expansion. When a customer's FinOps team turns off idle instances, Datadog's meter slows down. That is a real, structural exposure, not a hypothetical.

Second, the land is small and the expand is everything. A typical first Datadog contract is infrastructure monitoring on a few hundred hosts — call it $30,000-$60,000 annually. That is not a business. The business is what happens over the following three renewal cycles as the same account adds APM, then log management, then real user monitoring, then a security product, and the contract walks from $50K to $200K to $700K. This is why Datadog reports products-per-customer as a headline metric alongside revenue: it is the leading indicator of the compounding. As of the company's fiscal 2024 disclosures, the average customer used roughly 3.3 products, up from about 1 product a handful of years earlier. Pushing that average toward 4+ is the single highest-leverage lever in the model.

Third, net revenue retention is the whole game, and it is a physical quantity here, not a sales artifact. In a seat-based business, NRR above 100% requires the sales team to sell more seats. In a consumption business, a meaningful chunk of NRR happens automatically — the customer's environment grew, the meter ran, the bill went up, and no salesperson touched it. Datadog has reported NRR in roughly the 110-120% band. The RevOps-relevant read: some of that expansion is *earned* by the account team through cross-sell, and some is *inherited* from the customer's own growth. Separating those two is one of the hardest and most valuable analyses a RevOps team at a consumption company can run, because it tells you whether your quota model is paying commission for work the customer did themselves.

The reason this matters beyond Datadog: consumption pricing is the dominant model for cloud infrastructure software, and Datadog is the cleanest large-scale example of it executed well. Snowflake, Twilio, Confluent, MongoDB Atlas, and the hyperscalers themselves all run variants. Understanding how Datadog turns metered usage into predictable-looking subscription revenue is a template.

How a dollar moves from a customer's cluster to Datadog's revenue line

The mechanics are worth walking end to end, because each step is a place where revenue can leak or compound.

Step one — instrumentation. An engineer installs the Datadog Agent on a host, or deploys it as a DaemonSet on a Kubernetes cluster, or adds a tracing library to an application. Nothing is billed at install. What has happened is that a data pipe now exists. Every subsequent product Datadog sells to that account rides this pipe, which is why Datadog fights hard to be the agent on the box. The agent is the distribution channel.

Step two — the meter starts. Infrastructure monitoring bills per host per month, on published list pricing in roughly the $15-$23 range depending on tier and commitment, with Datadog billing on a high-water or averaged host count depending on plan. Containers get counted separately above an included allotment per host. APM bills per host on a separate, higher tier — list pricing lands in roughly the $31-$40 per host per month range — plus separate charges for ingested and retained spans. Log management splits into ingestion and indexing, with indexed logs billed per million log events and retention period changing the rate; list pricing for indexing sits in the low single dollars per million events. Real user monitoring bills per thousand sessions, with published pricing around $1.50 per thousand. Synthetic tests bill per ten thousand test runs. Custom metrics bill per hundred metric timeseries above the included allotment.

Step three — the allotment traps. This is where actual dollars diverge from the spreadsheet. Custom metrics are the classic example. Each host includes a fixed allotment of custom metrics; every high-cardinality tag a developer adds — a user ID, a request ID, a container hash — multiplies timeseries count. A single careless tag on a busy service can add tens of thousands of billable timeseries overnight. The same dynamic applies to indexed logs: a debug-level logger left on in production during an incident can index tens of millions of events in a day. Datadog's revenue and the customer's surprise bill are the same number viewed from two sides, which is why Datadog itself built cost-management tooling and log-exclusion filters — an unmanaged bill produces churn, and churn is more expensive than the overage.

Step four — commitment drawdown and true-up. Usage draws against the annual commitment. Datadog's enterprise motion sells a commitment with a volume discount ladder: larger commitments buy a lower effective per-unit rate. A customer committing to a higher tier gets meaningfully better unit economics than one on pay-as-you-go, which is deliberate — it converts variable usage into contracted, forecastable revenue and it makes the customer's cheapest path to lower unit cost *committing more*, not using less.

Step five — renewal and expansion. At renewal, the account team resets the commitment against the trailing usage curve and attaches the next product. This is the moment products-per-customer moves. A renewal that only resizes the infrastructure commitment is a failed renewal in Datadog's model even if the dollar value went up, because it did not add a product line and therefore did not add switching cost.

Step six — the marketplace path. A significant share of enterprise transactions route through AWS Marketplace, Azure Marketplace, and Google Cloud Marketplace. The customer's Datadog spend then draws down their committed cloud spend agreement with the hyperscaler. This is a procurement gift: the buyer spends money they already promised to spend, which shortens the deal cycle and often kills the "do we have budget" objection outright. Datadog pays a marketplace fee for the privilege, which is a real gross margin drag, but the conversion-rate improvement generally pays for it.

Costs, timelines, and the ranges that define the model

The unit economics of this business are unusually clean, and the published financials establish the ranges.

Gross margin sits above 80%. That is high even for infrastructure software, and it holds because the marginal cost of ingesting one more customer's telemetry is storage and compute on a multi-tenant platform Datadog has spent a decade optimizing. The pressure on that number comes from three places: log storage at long retention, marketplace fees on cloud-marketplace transactions, and the compute cost of AI features that run inference rather than just aggregation. Each of those is dilutive at the margin; none has been large enough to move the headline out of the 80s.

Sales and marketing runs roughly a quarter to a third of revenue. This is the number that separates Datadog from most enterprise software and it deserves explanation, because it is the most durable structural advantage in the model. A traditional enterprise seller — Salesforce is the standard comparison, running S&M in the mid-40s to mid-50s as a percentage of revenue — has to buy every dollar of growth with sales capacity. Datadog does not, for two reasons. First, product-led entry: engineers sign up for a free trial, instrument something, and the tool proves itself before a salesperson is involved. Second, the consumption meter means a meaningful slice of expansion arrives without a sales touch. Datadog gets paid when the customer's cluster grows whether or not an AE made a call that quarter. Structurally, roughly half the S&M ratio of a comparable seat-based vendor drops to operating margin.

R&D runs in the same rough band as S&M, a quarter to a third of revenue. For a company selling twenty-plus products, that is the cost of the strategy. The product-count expansion is not free; each new product is a multi-year engineering investment before it produces material revenue, and most of them spend their first two years as a rounding error in the revenue mix.

Non-GAAP operating margin lands in the low-to-mid twenties, with free cash flow margin running higher — mid-twenties to around thirty percent. FCF exceeds operating margin because of the annual-commitment structure: customers frequently pay annually up front, so cash arrives ahead of recognized revenue and deferred revenue funds working capital. This is a genuinely attractive cash profile and it is a direct consequence of the commitment model rather than the meter.

Customer concentration follows a steep pyramid. As of the fiscal 2024 disclosures, Datadog reported roughly 3,610 customers at $100,000+ ARR and about 510 at $1M+ ARR, against a total customer base in the high twenty-thousands. The arithmetic is stark: a few hundred accounts drive a very large share of ARR, several thousand mid-market accounts drive the next slice, and the long tail of tens of thousands of small accounts contributes proportionally little revenue while consuming platform capacity. That is not a flaw — the tail is the farm system, and today's $8,000 startup account is a plausible $1M account after two funding rounds — but it means the revenue line is far more sensitive to the top of the pyramid than the customer count suggests.

Timelines. Land-to-first-expansion typically runs a quarter or two, because the first expansion is usually just more hosts. Land-to-second-product runs longer — often two to four quarters — because it requires a different buyer or at least a different budget line. Land-to-$1M is a multi-year arc, generally three or more years, and it almost always involves the account crossing from a single team's tool into a platform standard with executive sponsorship. Enterprise sales cycles for a net-new six-figure commitment run roughly one to two quarters; regulated verticals run considerably longer.

Where 2027 revenue plausibly lands. Against fiscal 2024 revenue of approximately $2.7 billion, compounding at 25% annually for three years produces roughly $5.3 billion; at 30%, roughly $5.9 billion; at 20%, roughly $4.7 billion. Treat any point estimate as a model output rather than a fact — Datadog does not guide three years out, and the deceleration curve is the entire question. What is defensible is the shape: net new ARR in the high hundreds of millions to about a billion dollars per year, driven far more by expansion within existing accounts than by new logo acquisition.

The mix shift is the real 2027 story. Datadog started as infrastructure monitoring, added APM and logs to become an observability platform, and has been building outward into security (cloud SIEM, application security, posture management, workload security, sensitive data scanning), software delivery (CI visibility, code analysis, service catalog), cloud cost management, and AI-workload observability including LLM tracing. In 2024 the observability core was the overwhelming majority of revenue and the newer clusters were small. The directional bet through 2027 is that security and AI-workload products go from marginal to material — plausibly a combined quarter of revenue — while the observability core, still growing in absolute dollars, becomes a smaller share of a much larger number. That is the honest framing: the business model does not change, the product mix does.

Where teams get this wrong

Mistaking the meter for the moat. The pricing model is not the defensibility. The agent is. Once Datadog's agent is deployed across a thousand hosts and three hundred services are traced through it, ripping it out means re-instrumenting the estate — a multi-quarter engineering project with no user-visible benefit. Competitors quote a lower per-host price constantly; the price is rarely why they lose. Anyone modeling this business on pricing competitiveness is modeling the wrong variable.

Treating consumption NRR as sales performance. This is the RevOps error with the largest dollar consequence. If your comp plan pays expansion commission on total account growth, you are paying for the customer's Kubernetes autoscaler. Consumption businesses need to decompose expansion into organic meter growth versus seller-attributed growth — new product attach, commitment upsize beyond the trailing usage trend, migration of a new business unit onto the platform — and compensate the second category. Getting this wrong inflates apparent sales productivity in growth years and produces a brutal surprise when customer environments stop expanding.

Ignoring the FinOps counterparty. Every large Datadog customer eventually assigns someone to reduce the Datadog bill. That person is real, they have executive backing, and their tools are log exclusion filters, sampling rates, metric cardinality limits, and retention reduction. A revenue model that assumes usage grows monotonically with the customer's infrastructure ignores that a full-time employee is working against exactly that. Realistic modeling assumes a persistent optimization drag of several percentage points annually on mature accounts, offset by new product attach.

Underweighting OpenTelemetry. Open-source instrumentation standards decouple the act of instrumenting from the vendor consuming the telemetry. In principle that erodes the agent moat and turns the backend into a commodity bidding war. In practice Datadog has embraced OTel ingestion and competes on the analysis layer — correlation across metrics, traces, and logs, plus the workflow — rather than on data collection. The honest read is that OTel is a genuine long-run ARPU risk that has not yet shown up as one, and anyone who declares it either fatal or irrelevant is overclaiming.

Confusing the hyperscalers' native tooling with the actual competition. AWS CloudWatch, Azure Monitor, and Google Cloud Operations are all free-ish, adequate, and single-cloud. Datadog's structural answer is multi-cloud neutrality: the moment a customer runs two clouds plus on-prem, single-cloud native tooling stops being a platform. The customers most at risk of native-tool displacement are the single-cloud, cost-sensitive mid-market accounts — the middle of the pyramid, not the top.

Assuming new products convert linearly. Twenty-plus products does not mean twenty-plus revenue streams. Most new products spend their early years as attach motions that increase stickiness and NRR without contributing standalone revenue. Modeling a new product as an immediate revenue line rather than a retention mechanism overstates near-term revenue and understates the durability it buys.

A decision framework for reading the model

If you are evaluating Datadog — as an investor, a competitor, a partner, or a RevOps leader borrowing the playbook — the question you are actually asking determines which metric to watch.

If you want to know whether growth is durable, watch products per customer and the $1M+ account count. Revenue growth in a consumption business can be flattered for several quarters by customer infrastructure expansion that has nothing to do with Datadog's execution. Products per customer cannot be flattered that way — it moves only when an account team attaches something new. Similarly, the $1M+ cohort count is the cleanest measure of whether accounts are crossing from departmental tool to platform standard.

If you want to know whether the meter is under pressure, watch NRR against customer count. Rising customer count with falling NRR means the land motion works and the expand motion is being eaten by optimization. That is the specific failure pattern to look for, and it shows up in NRR a full year before it shows up in revenue growth.

If you want to know whether the mix shift is real, watch the disclosure language on security and AI products, not the revenue percentages. Companies disclose product-level revenue when it is flattering. The moment security or LLM observability is disclosed as a discrete number rather than described qualitatively, it has become material. Until then, treat any specific percentage — including the ones commonly modeled — as an estimate.

If you are borrowing the model for your own company, the transferable pieces are narrower than they look. Product-led entry into a technical buyer, a metered expansion path that grows without sales touch, and a multi-product attach motion at renewal are all portable. The 80%+ gross margin at consumption scale is not — it rests on a decade of platform engineering. And the low S&M ratio is a *consequence* of the first two, not an input you can decide on.

Related questions

Is Datadog's revenue recognized as subscription or usage revenue?

Datadog reports revenue predominantly as subscription revenue under a single reportable segment. Usage against annual commitments is recognized as the service is delivered, which makes the consumption meter appear on the income statement as ratable subscription revenue rather than transactional revenue.

Why is Datadog's sales and marketing spend so much lower than Salesforce's?

Product-led entry and metered expansion. Engineers adopt Datadog through free trials before a seller is involved, and a share of account growth arrives automatically as customer infrastructure scales. Salesforce must buy nearly every incremental dollar with sales capacity, so its S&M ratio runs roughly double.

What is the single biggest risk to Datadog's revenue model?

Sustained customer cost optimization. Because revenue tracks customer infrastructure usage, a broad enterprise push to reduce cloud spend, cut log retention, and limit metric cardinality directly compresses the meter — and unlike a seat-based business, no contract minimum forces the spend back.

How does the cloud marketplace channel change deal economics?

Selling through AWS, Azure, or Google Cloud marketplaces lets the buyer draw against an existing committed cloud spend agreement, which removes budget objections and shortens cycles. Datadog pays a marketplace fee, trading gross margin points for conversion rate and deal velocity.

Does adding more products actually raise revenue per customer?

Eventually, but indirectly. Most new products first function as retention mechanisms that raise switching cost and NRR, contributing little standalone revenue for a year or two. The revenue effect shows up at renewal, when a multi-product account commits at a materially higher tier.

FAQ

Does Datadog charge per user or per seat?

No. Datadog's pricing is tied to infrastructure and telemetry volume — hosts monitored, containers, APM hosts, indexed log events, RUM sessions, synthetic test runs, custom metric timeseries, and security events. Adding engineers to a Datadog account does not by itself increase the bill. This is why headcount reductions at customers barely dent Datadog's revenue while infrastructure consolidation does.

What does a typical Datadog contract look like?

An annual commitment with a negotiated volume discount, drawn down by usage across whichever products the customer enables. Larger commitments buy lower effective unit rates. Overages are either billed as they occur or resolved through a mid-term true-up, and the commitment resets at renewal against the trailing usage curve.

Why do Datadog bills surprise customers?

Almost always custom metric cardinality or log indexing. A single high-cardinality tag — a user ID or request ID attached to a busy service — can multiply billable timeseries overnight, and a debug logger left enabled in production can index tens of millions of events in a day. Both are customer configuration choices that the meter faithfully bills.

How much of Datadog's growth comes from new customers versus existing ones?

The large majority comes from existing customers. With net revenue retention historically in the 110-120% range, the installed base grows meaningfully every year before a single new logo is counted. New logos matter as the future top of the pyramid rather than as this year's revenue.

Is Datadog profitable?

Yes on a non-GAAP basis, with operating margin in roughly the low-to-mid twenties as a percentage of revenue and free cash flow margin running higher. GAAP results are materially lower because of stock-based compensation, which is substantial at Datadog as at most high-growth software companies. Always check which basis a figure uses.

What would make Datadog's revenue growth decelerate sharply?

A durable slowdown in enterprise cloud infrastructure growth, a successful industry-wide shift to open instrumentation that commoditizes the backend, or a coordinated enterprise push on observability cost reduction. Any of the three attacks the meter directly. Competitive price pressure alone historically has not, because switching cost rather than price drives retention.

Sources

• Datadog Investor Relations (filings, earnings releases, metrics): https://investors.datadoghq.com/
• Datadog pricing (list pricing by product): https://www.datadoghq.com/pricing/
• Datadog product documentation: https://docs.datadoghq.com/
• SEC EDGAR company filings for Datadog, Inc. (DDOG): https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=DDOG&type=10-K
• Salesforce Investor Relations (S&M ratio comparison): https://investor.salesforce.com/
• AWS Marketplace: https://aws.amazon.com/marketplace
• Microsoft Azure Marketplace: https://azuremarketplace.microsoft.com/
• Google Cloud Marketplace: https://console.cloud.google.com/marketplace
• OpenTelemetry project: https://opentelemetry.io/
• Bessemer Venture Partners, State of the Cloud: https://www.bvp.com/atlas/state-of-the-cloud

Related on PULSE

• How does Salesloft make money in 2027?
• How does the NFL make money and share revenue among teams in 2027?
• How do sports agencies make money and how does the business model work in 2027?
• How does the NCAA make and distribute its money in 2027?
• How do college athletic departments make and spend money in 2027?
• How does the IOC and the Olympics make money in 2027?