Cognition's $2B Series E at $48B valuation

Cognition's Series E prices the company at $48 billion, nearly double the $26 billion valuation it carried just four months earlier ^[2], and ties that jump to run-rate revenue climbing from $492 million to almost $900 million over the same stretch ^[1]. Against roughly $900 million in revenue, a $48 billion price tag implies a rich multiple, and Forkast News flags real execution risk behind the number: Cognition reportedly spends close to $800 million a year on compute, mostly leasing Nvidia's server clusters, an amount that eats nearly as much cash as the company currently brings in ^[3]. A companion Forkast analysis frames the valuation as treating coding agents like infrastructure rather than a software subscription, pointing to large-scale deployments like Citi's as evidence the technology has moved past pilot status into infrastructure-grade adoption ^[4]. TechCrunch, meanwhile, estimates Cognition could exit 2026 with $4-5 billion in annualized revenue ^[2]- a projection that only justifies today's price if the growth curve of the last four months continues rather than reflecting a one-time wave of enterprise land-grabs.

Cognition's rise looks smoother in this week's press coverage than it has looked in practice. Devin's original 2024 launch demo was publicly picked apart by a YouTuber who showed the agent largely fixing errors in code it had written itself rather than completing the advertised task, and a later third-party evaluation from Answer.ai found Devin succeeding on only a handful of real-world coding tasks out of twenty attempted. What has visibly changed since is the deployment evidence behind this raise: Goldman Sachs has moved from a pilot to scaling thousands of Devin instances, targeting a three-to-four-times productivity gain over its prior AI tooling ^[3], and Citi is reported to run roughly 40,000 developers on Devin generating over 100,000 agentic coding hours a week - described as the largest measured enterprise agent deployment in financial services ^[4]. Cognition's customer list now spans NASA, Nvidia, Mercedes-Benz, GE Aerospace, Dell, Cisco, Palantir, and multiple global banks ^[5], and the company used the raise to ship Devin Auto-Triage for incident investigation and Devin Security Swarm for vulnerability discovery ^[8]- product moves aimed squarely at the kind of regulated, high-stakes engineering environments where a shaky 2024 demo would once have been disqualifying.

Much of Cognition's current product and revenue base traces back to a chaotic stretch in mid-2025. OpenAI had been negotiating to acquire the Windsurf coding-agent startup for roughly $3 billion, but the deal collapsed just as Google moved in with a reverse-acquihire, hiring away Windsurf's CEO and co-founder ^[6]. Cognition acquired what remained of Windsurf days later ^[5], folding its editor and proprietary models into a product rebranded Devin Desktop in mid-2026 ^[7]. Separate retrospective coverage of that saga adds color worth noting: Anthropic reportedly cut Windsurf's access to its Claude API in the middle of the acquisition chaos, and Google's side of the deal was structured as a technology license plus the hiring of a small team into DeepMind rather than a full acquisition, leaving most of Windsurf's staff at the company Cognition ultimately absorbed. That messy provenance is now packaged as a unified IDE-plus-agent platform, which Cognition is using to argue Devin is a full development environment rather than a thin wrapper around someone else's model.

Not everyone reading this week's headline number is convinced the valuation reflects durable differentiation. Public reaction skewed skeptical and bubble-cautious, with comparisons to dot-com-era excess and blunt questions about who actually uses Devin day to day, tempered by counter-anecdotes from engineers at both a small company running Devin as its primary coding tool with CI/CD integration and a larger public company with tens of thousands of employees. The core disagreement is whether Cognition and rivals like Cursor are meaningfully differentiated from the frontier models they're built on, or are simply resellers of Anthropic and OpenAI tokens whose margins may not survive real competition; one detailed rebuttal argued Devin's edge is dynamic model routing that shifts cheap models onto simple tasks and frontier models onto complex reasoning to control cost. A parallel subplot involving SpaceX's ownership of Cursor reportedly curbing Cursor's access to Anthropic and OpenAI models has some framing Devin as a more attractive alternative. TechCrunch's read on the raise leans toward the bullish side of that debate, framing it as a signal that venture capital sees room for several major AI coding companies to coexist rather than a single winner-take-all platform ^[2]- a framing that quietly concedes the wrapper question hasn't actually been settled either way.