Cognition's latest funding round is notable not just for the $26 billion valuation, but for the company's claim that 90% of its own codebase is now generated by its AI coding agent, Devin.
On May 27, Cognition announced it had raised over $1 billion in a funding round that valued the company at $26 billion post-money. Just eight months ago, the startup was valued at $10.2 billion. The round was led by Lux Capital, General Catalyst, and 8VC, with participation from existing investors including Founders Fund and Elad Gil. Cognition has now raised more than $2.5 billion in total.
The valuation jump is striking, but the revenue trajectory better explains investor urgency. According to TechCrunch, Cognition's annualized revenue run rate has reached $492 million, up from about $37 million a year ago — a 13-fold increase. The company also said enterprise usage has grown more than 10x since the start of the year, with month-over-month growth of 50% over the past six months.
Cognition's client list is also becoming more heavyweight. Disclosed customers include Goldman Sachs, Mercedes-Benz, NASA, Santander Bank, and multiple U.S. government agencies. For an AI coding company, landing such enterprise clients supports a high valuation far more than selling subscriptions to individual developers.
Devin is positioned as an "autonomous AI software engineer": given a task, it can plan, write code, debug, and submit results. By putting the "90% of our own code written by Devin" claim front and center, Cognition is essentially saying it is the product's first heavy user. This is more direct than marketing copy — and riskier, because it ties the company's own delivery quality to the claim.
The risks are clear. More code written does not mean better code; more pull requests do not necessarily mean real engineering efficiency gains. AI coding tools can significantly accelerate output, but review, testing, architectural decisions, and production stability still depend on humans and processes. Cognition's willingness to publicize the 90% figure suggests it wants to turn controversy into a sales pitch.
Another notable strategic choice: Cognition has so far not focused on building its own foundation model. Founder and CEO Scott Wu told Bloomberg TV that a multi-model approach often yields better results than relying on a single vendor. In other words, Cognition is betting on orchestration, execution, and engineering closure — not on training a flagship model from scratch.
This approach is lighter and faster, but it also leaves the company dependent on others. If underlying models like Claude, GPT, or Gemini raise prices, throttle usage, or experience service disruptions, the experience of the coding agent on top will suffer. Cognition needs to prove that its execution layer is thick enough that customers will pay a premium for it, rather than seeing it as just a wrapper around model APIs.
Between the $26 billion valuation and the $492 million annualized revenue run rate, the market is buying a bet: that AI coding will move from assisted completion to autonomous delivery. Cognition is using its own 90% code claim as a footnote. Whether it can approach $1 billion in annualized revenue by year-end will determine whether that footnote is a sign of confidence or a risk warning.
Sources: CocoLoop, AI Coding Startup Cognition Raises $1 Billion at $26 Billion Value (Bloomberg); AI coding startup Cognition raises $1B at $25B pre-money valuation (TechCrunch); Cognition just raised $1 billion at a $26 billion valuation (The Next Web)