Cognition Hit $1 Billion Run Rate Before Devin Turned Two

Cognition has joined the short list of AI application companies selling at a billion-dollar annual pace. On September 25, the maker of the Devin coding agent said on its blog that it had crossed $1 billion in annualized revenue run rate. The company says it started in January 2024 and reached the mark less than two years after Devin became generally available.

Bloomberg reported the milestone the same day, citing a person familiar with the matter: Cognition was on track for $1 billion in annualized revenue based on its September performance, roughly double its run rate from about four months earlier. In May, when it raised at a $26 billion valuation, the company had disclosed $492 million.

Four months, three numbers

The climb is easiest to read as a sequence. $492 million in May. "Nearly $900 million" when Cognition closed its Series E on September 8, raising more than $2 billion at a $48 billion valuation in a round led by Andreessen Horowitz and Accel. Then $1 billion seventeen days later.

That last step matters because it comes from Cognition's own blog, not only from an anonymous source. The post is short. It names GE Aerospace, Rivian, the European grocery delivery company Rohlik and the search startup Exa as customers, and restates the company's mission that "the world needs far more software than it can build." It does not break out revenue by product, customer count or margin.

Bloomberg's report lists larger names: Nvidia, Citigroup and Mercedes-Benz. Citi is the clearest signal of how Cognition sells. The bank told American Banker last year it was rolling out agentic AI tools, including Devin, to its 40,000 developers. A handful of deployments at that scale can move a run rate by hundreds of millions of dollars.

What a run rate is, and isn't

A run rate is one period's revenue, usually a month, multiplied to a full year. FourWeekMBA put it plainly in its analysis of the Bloomberg report: the figure is not money the company has received. For a company growing this fast, trailing twelve-month revenue sits far below the run rate, because the early months of the curve were so much smaller.

So the honest reading is narrow. Cognition's September billing pace, extended over a year, reaches $1 billion. It has not booked $1 billion. The run rate is still the right metric for momentum, and the momentum is real: a company does not double in four months on marketing alone.

At the $48 billion valuation set earlier this month, Cognition now trades around 48 times its stated run rate, down from roughly 53 times at both the May and September rounds. The revenue caught up with the price faster than the price moved again.

One product line, not two

The growth arrived while Cognition was collapsing its product lineup into a single brand. It bought the Windsurf editor in July 2025. By June 2026 the editor had been renamed Devin Desktop, and Windsurf's signature Cascade agent was retired on July 1 in favor of Devin Local. The September 8 release of Devin Desktop, version 3.9.19, removed Cascade entirely, with release notes stating that Devin Local is now the only agent in the app.

That consolidation is a bet on a specific kind of buyer. Devin's pitch has always been autonomy: assign a task, leave, return to a finished pull request. The editor is now positioned as the place where engineers supervise that work, not as a separate copilot competing with Cursor or GitHub Copilot on autocomplete. A single product family is simpler to sell to a procurement team buying seats for tens of thousands of developers.

Cognition has also been lowering its own costs. Earlier this month it shipped SWE-2, a coding model post-trained from Moonshot's open-weight Kimi K3 base, which Flux covered on September 11. Owning more of the model stack is the most direct lever an agent company has on gross margin, the number Cognition still has not published.

The field it is growing in

The competitive map shifted hard this summer. SpaceX closed its $60 billion acquisition of Cursor on August 14. Days later Bloomberg reported that SpaceX had also approached Cognition. Chief executive Scott Wu disputed the report on X, saying the company "is not for sale" and that the two had not been in talks, as TechCrunch reported on August 19.

That leaves Cognition as the largest independent company in autonomous coding, competing against frontier labs that sell their own coding agents and against an open-weight tier that keeps closing the model gap. Its customer list, heavy on banks, automakers and aerospace, suggests where it believes it wins: regulated enterprises that want an agent vendor that is neither a model lab nor part of a rocket company.

Why it matters

A billion-dollar run rate less than two years after general availability puts Devin among the fastest-scaling enterprise software products on record, and it answers the most common objection to autonomous coding agents: that companies would pilot them and never pay at scale. Enterprises are paying, and the bill is growing month over month.

The open questions are the ones a run rate cannot answer. Autonomous agents burn far more inference than autocomplete, so the share of that $1 billion that survives compute costs is still unknown. And much of the growth comes from large first deployments whose renewals have not yet come due. Cognition's next disclosure worth watching is not a higher run rate. It is a margin figure, or a renewal rate, that shows this revenue lasts.