Investors just paid twice as much for Cognition as they did in May, and the company is not yet half as large as the competitor that sold out four months ago. The Devin maker closed a Series E of more than $2 billion at a $48 billion valuation on September 8, 2026, led by Andreessen Horowitz and Accel with Founders Fund, General Catalyst and Avenir returning, and more than thirty additional firms — Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, T. Rowe Price and Nvidia among them — filling out the round.
Key takeaways
- The valuation nearly doubled from $26 billion in May while annualized run-rate revenue moved from $492 million to about $900 million, pushing the multiple past 50x.
- Nvidia sits on both sides of the deal, backing the round while using Devin internally for chip design work.
- a16z, which profited from Cursor's $60 billion sale to SpaceX, returned to lead a round in Cursor's direct competitor — the round's clearest signal about market structure.
Why investors paid a higher multiple than Cursor commanded
The comparison that gives this round its meaning is arithmetic rather than narrative. Cursor was negotiating in April at a $50 billion valuation on annualized revenue already past $2 billion, a multiple in the mid-twenties, before agreeing later that month to sell to SpaceX for $60 billion; Cognition is being priced at more than double that ratio on roughly $900 million.
Paying more per revenue dollar for the smaller company is a bet purely on slope. The Information projects Cognition reaching $4 billion to $5 billion annualized by the end of 2026, which would close most of the gap in a single year. Taken together with a16z's willingness to fund a rival months after cashing out of Cursor, the round reads as a collective judgment that AI coding assistants will support several large independent businesses rather than consolidating around one.
What Devin actually does for enterprise customers
Scott Wu, a competitive-programming prodigy, founded Cognition in 2024 around a specific product thesis: engineers should behave like architects and hand execution to swarms of agents. Devin embodies that by accepting high-level requirements instead of line-by-line instructions, then planning, writing, testing, debugging and deploying inside a sandbox without a human in the loop for each step.
Where that lands commercially says something about how far agent tooling has travelled. Nvidia runs Devin on chip design, GE Aerospace on aviation work, Mercedes-Benz on automotive engineering, Modal Labs on AI infrastructure, and Citigroup and Goldman Sachs inside financial services — industries where code review is a regulated process, not a preference. The newest capabilities push further into operations rather than authoring: Auto-Triage investigates production incidents, Security Swarm sweeps a codebase for vulnerabilities, and Automations fires agent runs off Slack, GitHub and Linear events.
Cognition additionally owns Windsurf, the editor it picked up in July 2025 out of the wreckage of OpenAI's collapsed takeover and Google's licensing-plus-hiring deal for the founders. The company markets itself as model-agnostic, free to mix third-party and in-house models per task, which is a deliberate jab at rivals bound to a single provider.
The costs the valuation has to outrun
Against that growth sits an expensive infrastructure position. Cognition leases an Nvidia cluster costing hundreds of millions annually, and The Information estimates total cash burn could reach $800 million this year. That number is worth holding next to Cursor's exit: investors familiar with its financials say severe compute constraints were the main reason it sold rather than raised. Cognition's stated escape route is training its own models on open-source foundations, cutting dependence on OpenAI and Anthropic inference bills — the same play Cursor was attempting when it ran out of runway.
The company also carries reputational baggage that the valuation does not erase. Devin's March 2024 launch demo was picked apart in a widely circulated video arguing the agent had accomplished considerably less on an Upwork task than shown, and the resulting skepticism about benchmark transparency has followed the company since. In August 2025, weeks after the Windsurf acquisition, thirty staff were laid off and the remaining two hundred or so were offered a choice between nine months of pay to leave or six-day weeks of 80 hours and up, in a memo where Wu wrote that the company does not believe in work-life balance.
Outlook
Six new offices — Washington, Tokyo, Singapore, London, São Paulo and Madrid, on top of San Francisco, New York and Austin — is a distribution bet that only pays if enterprise demand keeps compounding. Meanwhile the competitive field keeps widening rather than narrowing, as Meta's Muse Code terminal agent for giant codebases illustrates. The question for the next twelve months is simple to state and hard to answer: does revenue reach the $4 billion forecast before the compute bill forces Cognition into the same corner that ended Cursor's independence?
FAQ
How much is Cognition worth after the Series E?
The September 8, 2026 round valued Cognition at $48 billion on more than $2 billion raised. That is close to double the $26 billion mark set in May, when the company took in over $1 billion.
Is Cognition profitable?
No. Roughly $900 million of annualized run-rate revenue sits against an estimated $800 million of cash burn this year, driven mainly by leased Nvidia compute. Training in-house models on open-source foundations is the company's declared path to narrowing that gap.
What is the difference between Devin and Windsurf?
Devin is the autonomous agent: give it a requirement and it plans, codes, tests and ships inside a sandbox. Windsurf, acquired in July 2025, is an editor for developers who want AI assistance while writing code themselves rather than delegating the whole task.
