Cognition, the New York-based startup behind autonomous AI coding agent Devin, is reportedly already back at the fundraising table — and the numbers are striking. The company is in early talks with investors for a new funding round that may boost its valuation by more than 50% to at least $40 billion.

Prospective investors are circling the company less than three months after Cognition raised $1 billion at a $26 billion valuation.

Revenue Is the Real Story

The valuation jump is aggressive, but it's anchored in genuine commercial traction. Cognition's annualized revenue run rate is nearing $1 billion — around twice the figure the firm posted at its earlier financing.

Three months ago, when it announced its last raise, CEO Scott Wu confirmed to TechCrunch that Devin had already hit a $492 million annualized revenue run rate, with enterprises growing their usage 50% month-over-month for the prior six months.

At the new target valuation, that works out to roughly 40x revenue — down from approximately 52x at the $26 billion mark three months ago. That compression matters: Cognition is growing into its price tag faster than investors are inflating it, which is a healthier signal than the valuation jump alone suggests.

What Devin Actually Does

Cognition makes Devin, billed as an autonomous AI software engineer.

CEO Wu has been explicit that Devin is not positioned as a human replacement — the agent is typically tasked with long-tail grunt-work that many programmers dislike. That framing has helped Cognition land major enterprise partnerships: the company has struck deals with Cognizant, Endava, and Synechron, among others, signaling that it's finding traction inside large engineering organizations rather than just among individual developers.

A Funding Velocity That's Hard to Ignore

The pace of Cognition's re-rating is remarkable even by AI startup standards. The May 2026 round at $26 billion was itself a major leap from a $10.2 billion post-money valuation when the company closed a $400 million round just eight months earlier, in September 2025.

That May round was led by Lux Capital, General Catalyst, and 8VC, with existing investors including Elad Gil, Soma Capital, Founders Fund, and others, plus new entrants Ribbit Capital, Atreides, and Layer Global.

Sources told Bloomberg that Cognition would be expected to raise more than $1 billion in the new round. No lead investors for the current talks have been named publicly.

How It Stacks Up Against the Competition

The AI coding market is growing crowded — and expensive. For comparison, Cursor (built by Anysphere) was last valued at $29.3 billion against annualized revenue that had reached roughly $4 billion by June 2026 — a multiple closer to 7x. Cursor trades at a far tighter revenue multiple, suggesting the market is beginning to bifurcate between companies priced on agent-led enterprise bets (Cognition) and those priced on raw developer adoption volume (Cursor).

Both Anthropic and OpenAI are pushing further into coding agents, putting model developers in direct competition with dedicated startups like Cognition.

Meanwhile, SpaceX has agreed to acquire Cursor — a deal that illustrates just how strategically important AI-written code has become, even outside traditional software companies.

What This Means for Founders and Builders

For startup founders watching this space, the signal is clear: enterprise adoption of agentic coding tools is accelerating, and the window to build differentiated positioning is narrowing fast. Cognition's ability to double its revenue run rate in under a quarter suggests that large engineering teams are moving from pilots to production deployments — compressing the sales cycle for the entire category.

The open question remains whether a shrinking-but-still-extreme revenue multiple in a market this crowded means Cognition is genuinely earning its price, or just running out of room before someone has to prove the multiple was worth paying. That answer will likely come into focus once this round closes — and once Devin's enterprise retention numbers become harder to ignore either way.