Scott Wu's Cognition raises over $2B at a $48B valuation
Andreessen Horowitz and Accel led the Series E, nearly doubling Cognition's valuation four months after its previous round.
By RuntimeWire Staff · Published
Primary source: The Wall Street Journal
Why it matters
Cognition's $48B price is about 53 times its self-reported run-rate. The round gives Scott Wu capital to fund the compute, models and distribution behind Devin.

Scott Wu's Cognition raised more than $2 billion at a $48 billion valuation in a Series E announced on September 8th, giving Cognition another large pool of capital four months after its previous billion-dollar financing. The Wall Street Journal reported the round as $2 billion, while Cognition's announcement describes the amount as over $2 billion.
Andreessen Horowitz and Accel led the round. Existing investors Founders Fund, General Catalyst and Avenir also participated, alongside Benchmark, Bessemer Venture Partners, Kleiner Perkins, Greylock, Lightspeed, Altimeter, Bond Capital, Bain Capital Ventures and other investment firms. Peter Thiel (@peterthiel) is a partner at Founders Fund, which led Cognition's early financing and has continued backing Wu.
The financing nearly doubles down on a bet investors made in May. RuntimeWire reported then that Cognition said it had raised over $1 billion at a $26 billion valuation. The new price is about 85% higher, a rapid reset even by the standards of the AI coding market.
A founder built for the coding-agent race
Wu started Cognition with Steven Hao and Walden Yan, fellow competitive programmers he had known before the three began working together on AI. Wu won three International Olympiad in Informatics gold medals and placed first overall in 2014. He left Harvard after two years, became a founding engineer at Scale AI and later co-founded the professional-networking service Lunchclub.
In a June interview with David Senra, Wu described company-building as a strategy game in which he works through a decision tree toward victory. The same account says Wu, Hao and Yan began Cognition in a New York apartment in 2023.
Wu has since given a more personal account of that period. In an interview published by Colossus, he said his mother died on October 6th, 2023. The OpenAI board crisis on November 17th became a forcing function for the founders to commit to building an AI lab and move quickly. Wu called Cognition "the big one" and "the one for life."
That ambition shaped Devin, Cognition's AI software engineering agent. Wu and his co-founders designed Devin to receive a task, inspect a codebase, plan the work, write and test code, and return a result for human review. The aim was to give software agents their own working environment and let engineers delegate complete assignments instead of requesting isolated code snippets.
The $48B valuation rests on a run-rate
Cognition says its annualized run-rate revenue rose from $492 million in May to almost $900 million by September, an increase of roughly 83% in just over three months. Cognition attributes the growth to expanding use among customers including Nvidia, GE Aerospace, Citi and Mercedes-Benz.
The $48 billion valuation equals about 53 times Cognition's claimed $900 million run-rate. That metric annualizes a recent period rather than reporting audited revenue across a full year. It leaves contract duration, gross margin, retention and the revenue split between Devin and Windsurf outside the frame. TechCrunch noted that Cognition did not explain how it calculated the figure.
The revenue claim still marks a striking acceleration from Cognition's early commercial period. Devin became generally available in December 2024, initially at $500 per month for engineering teams. Cognition advised users at the time to give Devin well-scoped assignments, keep sessions under roughly three hours and expect to guide or manually clean up some of its work.
Cognition's September 2026 pitch describes a broader system. Devin can now begin incident investigations, triage security vulnerabilities and start work automatically when events occur in Slack, GitHub or Linear. Cognition is asking customers to move from assigning individual coding jobs to placing software maintenance and development loops under agent supervision.
The distance between those two product descriptions carries much of the investment case. A useful assistant can support a large software business. A reliable agent that continuously finds, scopes and completes engineering work could occupy a much larger share of corporate technology budgets.
What another $2B buys
Cognition has spent the past year assembling the pieces around Devin. In July 2025, Wu signed an agreement to acquire Windsurf, bringing an AI-focused development environment, intellectual property and an established enterprise sales operation into Cognition. Cognition has also opened offices outside its San Francisco headquarters and started training its own specialized models while continuing to use models from other labs.
Cognition is investing in model development and inference capacity while integrating Devin with Windsurf and serving enterprise customers. The supplied reporting does not establish its current compute costs or cash burn.
The Series E therefore finances several races at once: model development, inference capacity, enterprise distribution and the integration of Devin with Windsurf. Cognition is competing with products from Microsoft-owned GitHub, Anthropic, OpenAI, Google and Replit, while trying to establish a distinct category around long-running, delegated engineering work.
Wu's pitch has remained consistent through Cognition's fast valuation climb. He wants human engineers setting goals, designing systems and reviewing consequential decisions while agents handle a growing share of execution. Cognition's investors are pricing that operating model as a likely control point in how software gets made.
The next test is whether the reported $900 million run-rate develops into durable, high-margin revenue while Cognition pays for the compute and research required to keep Devin competitive. At $48 billion, Wu has the capital to keep searching the decision tree. Investors have also made the cost of choosing a wrong branch considerably higher.