Cognition's revenue run rate reaches $1B, four months after $492M

Why it matters

Cognition's reported run rate puts a large commercial number behind the bet that AI agents can take on real engineering work. The figure is a projection, so retention, customer expansion and recognized revenue will show whether demand lasts.

Cognition co-founder and CEO Scott Wu has spent years moving between competitive programming, engineering and startup building. Now his AI coding company is approaching a $1 billion annualized revenue run rate, according to a person familiar with the matter cited by Bloomberg. The figure is based on Cognition's performance in September, and describes projected annual sales based on a shorter period, not $1 billion in recognized revenue for a completed year.

A short burst of sales can make an annualized figure look enormous. Bloomberg reported that Cognition said earlier in September that its run-rate revenue had topped $900 million, after reporting $492 million in May. The latest $1 billion estimate comes from an anonymous source familiar with the matter; Cognition declined to comment on those latest financials.

A founder's path from code to company

Wu's route to Cognition ran through several very different jobs. He won three gold medals at the International Olympiad in Informatics, then worked in performance engineering at Addepar, where he said he helped make some financial calculations roughly 100 times faster. He later co-founded Lunchclub, a professional networking service that facilitated millions of meetings before he left in 2022. In an interview with Colossus, Wu described the years after Lunchclub as a period of exploring ideas and thinking about what he wanted to build next. The arrival of ChatGPT in late 2022 pulled his attention toward AI.

That background helps explain Cognition's product choice. Wu and his co-founders built Devin as an agent meant to handle multi-step engineering work, rather than stop at suggesting the next line of code. In Cognition's launch post, Wu described a system that can work in a sandbox with a shell, code editor and browser, while reporting progress and taking feedback from engineers. The commercial case depends on customers paying for completed engineering work and saved engineering time.

The customer list Bloomberg named - Nvidia, Citigroup and Mercedes-Benz - suggests Cognition is selling into large organizations with substantial software needs. It does not establish how many customers pay for Devin, how much they spend, or how much of the reported run rate comes from recurring contracts. Those details are important when evaluating whether the company can sustain the pace implied by its headline figure.

Revenue growth and investor expectations

Cognition's own May announcement put its run-rate revenue at $492 million. Its September financing announcement said the company had raised more than $2 billion at a $48 billion valuation, with Andreessen Horowitz and Accel leading the round alongside existing investors including Founders Fund, General Catalyst and Avenir. The Series E announcement frames the company's thesis around the amount of software work that remains undone because engineering time is limited.

Cognition's September financing valued it at $48 billion and brought its total funding above $2 billion, according to the company. Bloomberg reported that the valuation had risen from $26 billion in May. The current revenue report gives investors another operating metric to watch.

The comparison with May is striking: the reported run rate has roughly doubled in about four months. The figures have different levels of attribution. Cognition disclosed the $492 million figure in May; Cognition said in September that the run rate exceeded $900 million; Bloomberg's new $1 billion figure relies on a person familiar with the matter. None of those numbers, as presented, is a full-year recognized-revenue result.

For Cognition, the reported figures point to rapid sales growth, though the public data leave its durability unclear. Enterprise demand for AI coding tools has become a meaningful contest among products built around different workflows, from code editors and autocomplete to agents that take on larger jobs. Cognition is selling the ambition that Devin can absorb pieces of an engineering team's workload. The harder measure will be whether customers keep expanding usage after pilots, whether the work meets their standards, and whether revenue growth translates into durable, profitable contracts.

The new funding gives Cognition substantial backing for that effort, and the customer names give the sales pitch weight. As reported, the $1 billion run rate indicates demand; it does not show whether Devin can reliably take over large portions of software engineering. For Wu, the challenge is turning a product built to handle more of the job into a business whose revenue holds up beyond a fast-growing month's annualized projection.