← Back to blog
NewsAbout 6 min read

Cognition Doubled Its Revenue to a $1 Billion Run Rate in Four Months. The Question Is Whether It Is Real.

Published Oct 5, 2026
Cognition Doubled Its Revenue to a $1 Billion Run Rate in Four Months. The Question Is Whether It Is Real.

On September 25, 2026, Cognition announced that its annualized revenue run rate had passed $1 billion. Four months earlier, in May, that figure was about $492 million. By early September it had reached roughly $900 million, and then it crossed the line.

The company behind the coding agent Devin and the Windsurf IDE is now generating revenue at an annualized pace of $1 billion. It was bringing in low tens of millions as recently as mid-2025. Whatever you make of the run-rate metric, a company that scales from that starting point to this one in little over a year is doing something that most of its peers are not.

What Cognition actually sells

The core product is Devin, an autonomous AI software engineer. Devin became generally available in late 2024 and is designed to take a ticket, write code, run tests, and open a pull request with minimal human intervention. That framing drew plenty of skeptics when it launched. The revenue curve suggests enterprises stopped debating the framing and started paying for the output.

The second pillar is Windsurf, the AI-native integrated development environment Cognition acquired in July 2025. Where Devin runs autonomously, Windsurf keeps a human in the loop, working side by side with the developer. Buying Windsurf gave Cognition both ends of the spectrum: full automation for tasks that do not need a person, and augmented coding for everything else. Cognition said the acquisition more than doubled its annualized revenue and that combined enterprise revenue grew by more than 30 percent in the seven weeks after the deal closed, with less than 5 percent overlap between the two customer bases.

The customer list has grown correspondingly. Cognition names GE Aerospace, Rivian, Rohlik, and Exa in its own announcement, and press reports add Nvidia, Citigroup, Mercedes-Benz, Goldman Sachs, Dell, Santander, the US Army, and the US Navy, along with startups including Modal, Eight Sleep, and OpenRouter.

The pricing model is doing a lot of the work

Part of what let the run rate double so fast is pricing. Devin is billed through per-seat subscriptions and usage-based Agent Compute Units, which means revenue grows with the volume of work an agent handles rather than only with the number of seats sold. Expanding usage inside an existing customer becomes the primary growth lever, and enterprise usage reportedly grew about 50 percent month over month for six straight months heading into the company's raise in May.

Cognition has also pushed the product past the original model of assigning a single task to one agent. In March it introduced the ability for Devin to manage other Devins, breaking larger jobs into tasks and delegating them to multiple agents in isolated virtual machines. Scheduled Devins let recurring sessions run automatically while retaining state between tasks. In May came Devin Auto-Triage, which responds to reports and alerts from Slack, Linear, GitHub, Sentry, and Datadog, investigates problems, and opens a pull request when appropriate. Devin Security Swarm followed, able to find vulnerabilities, assess whether they are exploitable at runtime, and produce remediation pull requests. On the model side, Cognition shipped SWE-1.6, SWE-1.7 in July, and SWE-2 in September.

Enterprise distribution ran alongside the product work. Infosys partnered with Cognition in January 2026 to deploy Devin across its organization and client base, starting with financial services and expanding into retail, energy, and healthcare. Cognizant signed on to deploy Devin and Windsurf across its engineering organization, covering code migration, refactoring, testing, and maintenance.

The funding that followed

Money followed usage here rather than the other way around. In May 2026, Cognition raised more than $1 billion at a $26 billion valuation, led by Lux Capital, General Catalyst, and 8VC, with Founders Fund returning. In September it raised more than $2 billion at a $48 billion valuation, led by Andreessen Horowitz and Accel, alongside existing investors including Founders Fund, General Catalyst, and Avenir. Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, Altimeter, Bond Capital, Meritech, T. Rowe Price, and Bain Capital Ventures also participated.

That trajectory is unusual even by the standards of the current market. The company was valued at $10.2 billion eight months before the May round.

Why the number deserves a careful read

An annualized revenue run rate is not booked revenue. It takes recent performance, usually the last month or quarter, and projects it across a full year. During a rapid growth phase, a run rate tends to run ahead of actual annual revenue, and it can retreat as fast as it climbed if growth stalls. Cognition's figure comes from the company's own disclosure, and 89 percent of the code its engineers committed being committed by Devin is a company claim about its own product.

None of that makes the number meaningless. Investors price a young company on run-rate multiples while auditors recognize revenue only when it is earned, and in a year of rapid growth those two calendars rarely line up. The trajectory from tens of millions to hundreds of millions to a billion in roughly 15 months is hard to dismiss as a timing artifact, and the customer names are large, verifiable organizations with procurement processes. When a bank, an automaker, and a government agency all pay for a coding agent, the spend is real even if the annualization is a projection.

There is also the contrast the figure sets with the rest of the AI funding market. Plenty of startups are raising at high valuations on the promise of what agentic AI will eventually do to enterprise software budgets. Cognition is one of the few showing that some of that spend has already arrived. Strip away the annualization and a billion dollars in revenue is invoices going out.

The competitive backdrop

The market Cognition is winning in is also the one where the fight is most intense. It competes with OpenAI's Codex, Anthropic's Claude Code, and Cursor, whose parent company Anysphere was folded into a new SpaceXAI division in an all-stock deal valued at $60 billion that closed in August 2026. Large model providers are entering the coding market directly, which intensifies pressure on both price and performance.

Cognition's answer was a revenue line that kept climbing, paired with a two-pronged strategy that most competitors lack: a fully autonomous agent on one side and a collaborative IDE on the other. That combination covers buyers who want to remove humans from a task and buyers who want to keep them close, which is most large engineering organizations.

What to watch

The next question is conversion. A $1 billion run rate is a statement about momentum. Booked, durable revenue across a full year is a different test, and the answer depends on whether usage-based expansion holds as enterprise customers move past pilot deployments and into standard procurement. If the doubling time holds even at a slower pace, Cognition becomes the reference point every other agentic coding startup gets measured against. If it stalls, the number becomes a lesson in how fast a run rate can move in both directions.

Related articles