The numbers
| |
Series D (May 2026) |
Series E (8 Sept 2026) |
| Valuation |
$26B |
$48B |
| ARR |
$492M |
~$900M |
| Revenue multiple |
~53x |
~53x |
Round size $2 billion. New leads Andreessen Horowitz and Accel; Founders Fund, General Catalyst and Avenir returning.
Why the flat multiple is the story
"Valuation doubles in four months" reads as froth. It is the standard shape of a bubble headline and it is how most coverage framed this round.
But the multiple did not move. Investors paid the same ~53x times a revenue number that had itself nearly doubled. That is not a re-rating, it is the same price applied to a bigger business.
Which does not make 53x cheap. It makes it consistent, and consistency tells you something a single number cannot: two separate sets of investors, four months apart, independently landed on the same valuation method. The froth question is whether 53x was ever the right multiple, not whether something changed in September.
What Devin actually sells
Devin is an autonomous coding agent - it takes a ticket and produces a pull request, rather than completing lines in your editor. That is a different product from Copilot or Cursor, and it is priced and sold differently: per task completed, into engineering organisations, with procurement.
The customer list is the part that explains the revenue: Nvidia, GE Aerospace, Citi, Mercedes-Benz, Goldman Sachs, Dell, Santander, Itau, the US Army and the US Navy.
Two banks, two defence branches and an aerospace manufacturer is not an early-adopter list. Those are procurement cycles measured in quarters, security reviews, and contracts that do not churn on a whim. Whatever you think of the multiple, that revenue is stickier than a seat-based developer tool's.
The honest caveats
- ARR is annualised, not earned. Roughly $900m of run rate is a snapshot rate multiplied by twelve. It is not $900m in the bank.
- The multiple says nothing about burn. Agentic coding is inference-heavy by design - every task is many model calls. Gross margin is the number nobody in this round has published.
- 53x assumes the growth continues. Doubling from $492m to $900m is far easier than doubling from $900m to $1.8bn, and the multiple is priced for the second one.
What it means if you are choosing a coding agent
- You want tickets closed without supervision - Devin is the category-defining product and now has the balance sheet to keep being it.
- You want faster typing - this is not that product, and the price reflects it.
- You are in a regulated industry - the customer list is the signal. Banks and defence branches cleared it, which means the compliance paperwork exists.
- You are cost-sensitive per task - price it against Claude Code and Codex on your own repository before committing. Autonomous agents burn tokens proportional to problem difficulty, not to lines produced.
FAQ
How much did Cognition raise?
$2 billion in a Series E that closed 8 September 2026, at a $48 billion valuation.
What is Devin's revenue?
Roughly $900 million annualised run rate, up from $492 million in May 2026.
Is $48 billion justified?
It is the same ~53x revenue multiple the May round paid. Whether 53x is right is a separate question from whether anything got more expensive - it did not.
Who uses Devin?
Named enterprise customers include Nvidia, GE Aerospace, Citi, Mercedes-Benz, Goldman Sachs, Dell, Santander, Itau, the US Army and the US Navy.