Cognition, the company behind the AI coding agent Devin, is already back in front of investors. TechCrunch reported on August 12, citing Bloomberg, that the company is in talks for a new round at a valuation of at least $40 billion. In May it raised $1 billion at $25 billion pre-money — $26 billion post. That is a 54% markup in roughly twelve weeks.
The number attached to the talks is not the interesting one. The condition is. Per Bloomberg’s sources, the $40 billion figure is premised on Cognition reaching a $1 billion annualized revenue run rate. When the May round was announced, CEO Scott Wu confirmed to TechCrunch that the company was at $492 million ARR, with enterprise usage of Devin growing about 50% month-over-month for the preceding six months.
Do the division. At $26 billion on $492 million, buyers in May paid roughly 53 times revenue. At $40 billion on $1 billion, buyers today are being asked for 40 times. The headline valuation went up by more than half. The price per dollar of revenue went down by about a quarter.
Our take: This is what a maturing category looks like from the inside. Cognition is not being repriced on story any more — it is being repriced on receipts, and the receipts are arriving fast enough that the multiple can compress while the valuation still climbs. That is the healthiest version of an up round. It is also the version that stops working the moment revenue growth flattens, because there is no narrative premium left underneath to catch it.
Why the revenue is growing
Wu has been consistent that Devin is not sold as a headcount replacement. The agent is pointed at long-tail engineering grunt work — dragging legacy software up to current versions, migrating applications off one platform and onto another — the kind of backlog that never gets prioritised because no engineer wants to own it and no manager wants to fund it.
That framing matters commercially, not just rhetorically. A tool that replaces engineers has to win a political fight inside every account. A tool that clears work nobody wanted gets bought out of a line item that was already sitting there unspent. Cognition says its customers include Mercedes-Benz, NASA and Goldman Sachs — buyers with enormous legacy estates and no appetite for a headcount argument.
The number that has to hold
Roughly doubling from $492 million to $1 billion in a quarter and change is the entire investment case. Nothing in the reported terms suggests investors are paying ahead of it — if anything they are demanding it be true first. That makes this round a bet on a single operating metric rather than on a category.
Worth noting what is not confirmed: the round has not been announced, the valuation has not been struck, and every figure here traces to sources speaking to Bloomberg. Cognition has not commented publicly on the talks.
What to watch
- Whether the round is announced at $40 billion or higher. A print above $40 billion means the ARR milestone was cleared early; a print at or below means it was negotiated down to it.
- Net revenue retention, if it is ever disclosed. Month-over-month usage growth of 50% inside existing enterprise accounts is the metric doing the work here, and it is the one that decays first.
- What the rest of the coding-agent field prices at next. Multiple compression at the leader tends to set the ceiling for everyone behind it.
- Whether the “grunt work, not replacement” positioning survives contact with enterprise procurement. Buyers eventually ask what they can stop paying for.
The broader signal is the one to file away. When AI companies were priced on possibility, valuations and multiples rose together. Cognition’s reported terms are the first clean example of the two coming apart in the right direction — and the market now has a benchmark for what real revenue is worth in this category.