The most expensive sentence in a software outage is often: “It is not our side.” Datadog exists because Olivier Pomel and Alexis Lê-Quôc heard versions of it too often. At Wireless Generation, the New York education technology company where they worked, developers disliked operations and operators distrusted developers. Each camp had its own instruments. When something broke, evidence arrived already sorted into rival jurisdictions.
Their proposed fix was cultural as much as technical: one live view of the system, shared by the people who wrote the code and the people who kept it running. “The root of all of this collaboration is to make sure that everyone has the same understanding of the problem,” Pomel later said. In 2010, the pair incorporated Datadog. The product became generally available in 2012, just as Amazon Web Services and the DevOps movement were making static server maps look like artifacts from a calmer age.
The wedgeThey did not sell monitoring. They sold a shared reality.
Datadog installs software agents and connects to cloud services, applications, databases, networks, developer tools, and security systems. It gathers metrics, traces, logs, user sessions, security signals, costs, and other machine exhaust, then correlates the lot. A slow checkout can be followed from a shopper's browser to an application service, a database query, a crowded container, and the cloud bill attached to it. The point is not another graph. It is the short path between “customers are angry” and “this deployment made this query slow.”
The organizational trick, drawn plainly
That common data layer is the distinction. Dynatrace and New Relic contest application performance; Splunk and Elastic contest logs; Grafana and open-source stacks offer composable alternatives; AWS, Azure, and Google sell native monitoring. Datadog's answer is breadth without the integration project. Its 2025 annual report said a single agent and more than 1,000 integrations fed one platform processing trillions of events per hour. Customers can begin in minutes, the company says, then move sideways into adjacent modules without rebuilding the context.
The root of all of this collaboration is to make sure that everyone has the same understanding of the problem.Olivier Pomel, co-founder and CEO
The first failureInvestors said no, so customers got the meeting
The first thing to fail was fundraising. Pomel has recalled that West Coast investors would not listen to an infrastructure pitch from New York, while local investors did not understand the category well enough to take the risk. The rejection forced the founders to spend their time with people living the problem. It changed the source of truth for the young company: less venture theory, more customer complaint.
That did not turn Datadog into a bootstrapped fable. It raised $6.2 million in 2012, $15 million in 2014, $31 million in 2015, and $94.5 million in 2016, for just under $148 million before the public offering. But the sequence matters. Customer contact came before capital abundance. Datadog went public in September 2019 at $27 a share and raised about $648 million. The rejected New York infrastructure company had become a test case for how quickly cloud tooling could turn into enterprise software.
The expansion pattern is visible in the product history. Infrastructure monitoring was the beachhead. Application Performance Monitoring arrived in 2017. Log Management followed in 2018, joining what the industry calls the three pillars of observability: metrics, traces, and logs. Datadog then spread into real-user monitoring, synthetic tests, network performance, incident management, software delivery, cloud costs, security, data pipelines, and AI systems. Each product makes the others more useful because each carries context into the same console.
The meterA simple entry price can become complicated arithmetic
Datadog makes substantially all of its revenue from subscriptions, but “subscription” understates the number of dials. Infrastructure Monitoring Pro is listed at $15 per host per month when billed annually. APM Pro is $35 per APM host per month. Database Monitoring is $70 per database host. Then there are containers, ingested logs, indexed log events, spans, custom metrics, synthetic tests, sessions, devices, retention periods, and on-demand rates. A modern system is dynamic. So is the invoice.
Three public starting points, before volume and add-ons
This is the fair way to read the recurring complaint that Datadog is expensive: sometimes it is, especially when high-cardinality metrics, verbose logs, or short-lived infrastructure arrive without governance. But the comparison is not merely Datadog versus a free dashboard. It is Datadog versus several licenses, the engineers maintaining an internal stack, slower investigations, duplicated alerts, and outages that reach customers.
The customer evidence gives that trade-off some shape. Mondelēz International says it used Datadog to cut resolution time for lower-severity P3 incidents from 23 hours to nine, reduce incident volume by 76 percent, and save roughly $150,000 a year by rightsizing infrastructure. Seven.One Entertainment Group says adopting Datadog APM alone reduced costs by 78 percent. Those are vendor-published case studies, not universal promises. They show the condition under which the price works: the software must change operational decisions, not merely decorate a screen.
One customer, fewer hours in the incident room
The new operatorDatadog wants the dashboard to investigate itself
Datadog's current bet is that observability will move from search to delegated action. Bits AI began as assistance and now includes agents that investigate alerts, form and test hypotheses against live telemetry, draft incident summaries, answer questions in plain language, inspect security findings, and work with coding agents. At its June 2026 DASH conference, the company announced more than 100 capabilities, including autonomous detection and remediation, AI Guard, Bring Your Own Cloud, and a builder for custom agents.
The acquisition of Adaptive ML that month made the strategy less subtle. Adaptive ML develops methods for tuning specialized models with reinforcement learning, synthetic data, evaluations, and feedback from production. Datadog owns an unusually rich stream of real production signals. The ambition is to train agents on the evidence of what systems actually do, then let them act within guardrails. It is a logical extension of the original pitch: first give two human teams the same facts; next give software enough context to join the call.
Financially, the platform is still expanding quickly. Revenue reached $3.427 billion in 2025, up 28 percent, with $914.7 million in free cash flow. In the second quarter of 2026, revenue rose 36 percent to $1.12 billion. About 4,720 customers were spending at least $100,000 a year by June, up 23 percent from a year earlier. Large contracts do not prove that every module deserves to exist, but they show that enterprises prefer consolidation when the shared context is useful enough.
What another builder can copy
- Start with a handoff where two teams lose time arguing about reality.
- Make one shared data model the product, not an afterthought.
- Shorten time to value until a practitioner can adopt without a committee.
- Expand only into workflows made stronger by the data already collected.
- Expose usage early and assign an owner before scale turns convenience into surprise.
The practical boundaryVisibility does not repair bad instrumentation
Datadog works best in a complex cloud or hybrid environment where several teams need the same live context and where saved engineering time exceeds the platform cost. It works less well when an organization cannot control what it sends, has strict self-hosting or data-residency requirements the service cannot meet, or runs a small and stable application that needs one narrow monitor. Poor tags remain poor tags after ingestion. A flood of low-value logs remains expensive noise. Observability cannot manufacture operational discipline.
Three conditions the dashboard cannot fix
No ownership: If nobody owns tagging, sampling, retention, and alerts, the platform accumulates noise.
No decision: If a metric never changes an action, collecting it is storage with better typography.
No shared habit: A common screen only reduces blame when teams actually use it together.
That caveat is also why Datadog's origin remains the best explanation of the company. The founders did not begin with a taxonomy of monitoring categories. They began with colleagues pointing at each other. The cloud made that argument louder by multiplying services, containers, and owners. Datadog put the evidence in one place, made the first look easy, and charged as the evidence multiplied. The next chapter asks whether an AI agent can turn that evidence into judgment. The old lesson still applies: the answer will be in what customers do, not in what the dashboard promises.