AI coding startup Cognition reportedly already in talks to raise at $40B valuation

Three months. That is how long Cognition waited before floating a $40 billion price tag after closing a $26 billion round. The company behind Devin, the AI coding agent that writes software by absorbing entire repositories, is reportedly back in market testing whether investors will pay a 54 percent premium on the last markup. The justification: a $1 billion annualized revenue run rate, double the $492 million Scott Wu confirmed to TechCrunch in May.

The revenue trajectory is undeniably steep. Wu also disclosed that enterprise usage of Devin has grown 50 percent month-over-month for six consecutive months. That compounding curve is the kind that makes venture partners reach for term sheets before the dataroom opens. But a run rate is not a run rate is not a run rate. Annualized figures extrapolate from a snapshot. They assume no churn, no saturation, no macro shock. They also blur the distinction between new logos buying in and existing accounts buying more seats. In a land-grab phase, the latter often dominates. The distinction matters when the multiple hits 40 times revenue.

Cognition's positioning is deliberate. Wu has been explicit: Devin is not sold as a human replacement. It is sold as a specialist in the long tail of software drudgery — migrating applications off legacy stacks, updating decades-old codebases, executing the platform shifts that engineering teams defer until the risk of inaction exceeds the pain of action. This is honest product marketing. It also frames the addressable market differently. The total addressable market for "replacing developers" is a religious debate. The market for "automating the migrations nobody wants to touch" is a budget line item every CTO already owns.

The customer list — Mercedes-Benz, NASA, Goldman Sachs — signals that enterprises are writing checks for this specific flavor of automation. These are not logo trophies. They are organizations with deep technical debt, regulatory constraints, and migration mandates that cannot be hired away. Devin's ability to ingest a codebase, reason across files, and propose changes that pass existing test suites makes it a plausible accelerant for exactly that work. The 50 percent monthly usage growth likely reflects seat expansion inside these accounts as more teams point the agent at their own backlogs.

But the valuation conversation has detached from the product conversation. A $40 billion valuation on $1 billion run rate implies the market is pricing Cognition as a platform company, not a tool company. Platform companies compound value through ecosystem lock-in, data gravity, and network effects. Tool companies compound through seat expansion and feature depth. Devin today operates largely as a sophisticated tool — an agent that executes scoped tasks inside a repository. The platform play would require it to become the default execution layer for a widening span of the software lifecycle: not just migration, but feature development, testing, security remediation, operational automation. That transition has not been demonstrated at scale.

Investors chasing the round are not blind to this. They are betting on the trajectory of the underlying models — likely a blend of frontier foundation models and Cognition's proprietary orchestration — continuing to climb the capability curve fast enough that the platform transition becomes inevitable before a competitor forces it. They are also betting that the revenue quality is sticky: that Mercedes-Benz does not churn after the migration wave crests, that NASA expands into new mission-critical systems, that Goldman Sachs standardizes on Devin for regulatory remediation pipelines. These are plausible bets. They are not proven bets.

The speed of the markup itself is a signal. In a normal cycle, a company executes on the capital from the last round for 12 to 18 months before returning to market. Three months suggests either remarkable capital efficiency — unlikely at this burn rate — or a preemptive strike by investors terrified of missing the next AI infrastructure layer. The latter dynamic produces rounds that clear at prices the fundamentals cannot yet defend. It also produces cap tables that become fragile when the hype cycle turns.

Cognition's next challenge is not raising. It is proving that the revenue run rate converts to durable ARR with net revenue retention above 120 percent. It is showing that Devin's capability envelope expands beyond the migration use case into the daily rhythm of feature velocity. It is demonstrating that the agent can operate safely in regulated environments without human-in-the-loop bottlenecks that erode the economics. And it is doing all of this while the foundation model layer — OpenAI, Anthropic, Google, xAI — races to embed the same repository-level reasoning into their own first-party coding agents.

The $40 billion number will either age well or become a cautionary exhibit. The difference will not be decided in the term sheet negotiation. It will be decided in the migration pipelines at Mercedes-Benz, the mission systems at NASA, the regulatory engines at Goldman Sachs — and in whether the next wave of enterprises buys Devin for what it does today, or for what Cognition promises it will become.