Cognition Raises $2B at $48B Valuation as Devin Revenue Nears $900M

Four months ago, Cognition was a $26 billion company. On Tuesday it became a $48 billion one — and the only thing that changed in between was the revenue line, which nearly doubled. The maker of Devin, the autonomous software engineer that debuted to skepticism in March 2024, said it has raised more than $2 billion in a Series E co-led by new investors Andreessen Horowitz and Accel, with existing backers Founders Fund, General Catalyst and Avenir returning. Run-rate revenue, the company said, has climbed from $492 million in May to almost $900 million.

That is roughly a 53x multiple on annualized revenue — a number that would have been considered unserious in any prior software cycle, and one that is higher than what Cursor commanded in April when it was in talks to raise at a $50 billion valuation on more than $2 billion in annualized revenue, shortly before agreeing to sell itself to SpaceX for $60 billion.

The syndicate around the round is unusually crowded. Beyond the five co-leads, more than 30 firms participated, including Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, Bond, T. Rowe Price, Lux Capital, 8VC, Bain Capital Ventures and Nvidia — which now sits on both sides of the table as investor and customer, using Devin in chip design work.

The revenue story, and the burn behind it

Cognition's pitch to investors rests on enterprise stickiness rather than developer-seat volume. The company says Devin now powers engineering work at GE Aerospace, Citi, Mercedes-Benz and Modal, and its May disclosures added Goldman Sachs, Elevance, Dell, Santander, the U.S. Army and the U.S. Navy to the roster. Enterprise usage, the company said at the time, had grown more than tenfold since the start of 2026.

The product has also widened considerably. Recent additions include Devin Auto-Triage, which takes a first pass at production incident investigation; Devin Security Swarm, which hunts and triages vulnerabilities across a codebase; and Devin Automations, which lets teams trigger agent work from events in Slack, GitHub and Linear rather than opening a chat window for every task. Cognition has published a usage chart denominated in what it calls Agent Compute Units, showing steep growth between August 2025 and August 2026.

Growth of that shape is expensive. Cognition leases an Nvidia server cluster costing hundreds of millions of dollars annually, and total cash burn could reach roughly $800 million this year, according to The Information. The same report projects the company will finish 2026 somewhere between $4 billion and $5 billion in annualized revenue — impressive in isolation, though below the $6 billion-plus trajectory TechCrunch reported for Cursor in the spring. Like Cursor before it, Cognition has begun training its own models on open-source foundations; its SWE-1.6 model, launched in May, runs at up to 950 tokens per second and is the most-used model inside Devin Desktop. Reducing dependence on OpenAI and Anthropic inference bills is the clearest available path toward margin.

Avenir, which said the check it wrote to co-lead the round is the largest in the firm's history, laid out the bull case bluntly on X: "Cognition and [Cognition Co-Founder and CEO Scott Wu] saw the shift from autocomplete to autonomous agents before nearly everyone. Now it has what we believe is a generational team; 14 months ahead on the shift from autocomplete to agents; the stickiest AI coding product in our enterprise survey; [and] a [$26 billion] market with a credible path to [$800 billion]. And we think code is only the first act."

Why It Matters

The most important signal in this round is not the valuation — it is who wrote the check. Andreessen Horowitz was a major backer of Cursor and made a fortune when it sold to SpaceX. Four months later, a16z is back leading a round in Cursor's direct competitor. That is a fairly explicit bet that AI coding is not a winner-take-all market, and that the category can support at least two companies at nine-figure-plus revenue scale alongside GitHub Copilot, Anthropic's Claude Code and a long tail of assistants.

It is also a bet on a specific architectural thesis. Cognition positions itself as an independent agent lab that can, in its words, "choose and combine the models best suited to the work, including our own" — a pointed contrast to rivals structurally bound to a single model provider. If frontier model quality continues to converge, the durable moat shifts from model access to workflow ownership: incident triage, security sweeps, migrations, the long tail of maintenance nobody wants to do. Cognition is trying to own the orchestration layer above whichever model wins.

The company is not without baggage. Devin's launch demo drew a widely circulated rebuttal video arguing it overstated what the agent actually accomplished, seeding lasting developer skepticism about Cognition's benchmark claims. And three weeks after the Windsurf acquisition closed in August 2025, Cognition laid off 30 staff and offered the roughly 200 remaining employees a choice between nine months of severance or six-day, 80-hour weeks. "We don't believe in work-life balance," Wu wrote in the memo.

What to watch

Three things. First, whether the $900 million run rate converts into the $4 billion to $5 billion exit velocity investors are underwriting — a 53x multiple leaves no room for a growth stall. Second, compute: Cursor sold to SpaceX in large part because it was severely compute-constrained, and it is not yet clear whether $2 billion buys Cognition enough headroom to avoid the same squeeze. Third, consolidation. If a strategic acquirer decides an AI coding platform is worth owning outright, Cognition is now one of a very small number of assets that qualify — and at $48 billion, the list of buyers who could move is short.

Wu, asked in May about the recursive self-improvement future his own company keeps invoking, offered the shortest honest forecast in the industry: "I think we are in for a wild ride."