Cognition has raised more than $2bn in a Series E at a $48bn valuation, almost doubling the $26bn valuation it received in May. The company also says run-rate revenue rose from $492m to almost $900m between the two rounds, a striking pace of expansion for an AI coding business. Andreessen Horowitz and Accel led the Series E as new investors. Founders Fund, General Catalyst and Avenir returned, and the round also includes T. Rowe Price, DST, Bain Capital Ventures and NVIDIA.
The valuation jump could be read as more of the same froth that has defined AI infrastructure financing over the past two years. But the more telling detail is that the multiple did not increase. In May, Cognition raised over $1bn at a $26bn valuation with reported run-rate revenue of $492m, implying a price of roughly 53 times revenue. The new $48bn valuation against almost $900m in run-rate revenue implies approximately the same multiple. Investors paid the same price per unit of revenue, but for a business that is roughly twice as large, at least as measured by the company's own figures.
The distinction matters because bubble-like dynamics are usually associated with multiple expansion. When investors pay more for each dollar of revenue even though growth is static, valuation theory begins to look detached. Here, the valuation re-rating tracked the reported business. That is not a sign of conservatism; 53 times revenue is still an extraordinary price, but it is a different risk profile from one in which the multiple itself is expanding.
Why the flat multiple is the real story
The same arithmetic could be described in simple terms: in May, investors valued Cognition at $26bn, and the company told them it had run-rate revenue of $492m. Now, the company says run-rate revenue has grown to almost $900m, and investors value the business at $48bn. If both prices are divided by their respective revenue claims, the implied multiple is roughly 53 times in each case.
That raises a question: did investors become more aggressive? On the available numbers, no. They simply paid the same multiple for a company with more reported revenue. That is not the profile of a market in which the same asset is being repriced upward solely because of optimism. It is also not proof that the valuation is sustainable; 53 times run-rate revenue is an enormous price for any company. But the expansion in the headline valuation is explained by the expansion in reported revenue, not by a change in how investors are pricing each dollar of that revenue.
Multiple expansion is often the fuel that turns high-growth financings into speculative bubbles. Companies can post flat revenue and still see rising valuations if investors decide the same unit of output deserves a higher price. That is not what happened with Cognition's move from $26bn to $48bn. The implied multiple stayed constant, and the valuation increase mirrors the company's claim that its run-rate revenue nearly doubled.
An important caveat about the numbers
None of that should obscure the fact that the underlying figures are unaudited and self-reported. Run-rate revenue is not the same as audited annual revenue. It annualises a recent period, which can turn a strong quarter or month into a much larger annual headline. Cognition is private and chooses what it discloses. No public filing requires the company to reconcile its run-rate revenue to a fuller set of financial statements.
No outside party has independently verified the revenue jump. The figures in the announcement come from the company, and potential investors in a private round would have access to additional materials that are not publicly available. For everyone else, the numbers must be taken on faith. That matters because the flat multiple is only meaningful if the revenue figures are comparable across rounds. If the earlier $492m figure and the new $900m figure were calculated under different assumptions, the comparison breaks down.
The unusual role of NVIDIA
NVIDIA appears on both sides of this deal. It is an investor in the round, and the company also names NVIDIA as a Devin customer, saying the product is used in chip design. That dual role has become common in the AI financing cycle. NVIDIA has committed more than $40bn to AI equity positions this year, giving it a growing web of strategic relationships across the industry.
There is nothing inherently improper about investing in customers. Many companies have done it over the years, and strategic investors may be better positioned than purely financial ones to evaluate a product they use. But it makes revenue quality harder to assess from the outside. If a major technology partner is also a launch customer, the line between product adoption and ecosystem support becomes blurry. Capital and customer relationships increasingly travel together in AI infrastructure, and Cognition is another example of that pattern.
A duopoly takes shape next door
The comparable situation is easy to spot. Cursor, another AI coding company, sits at almost exactly the same price. Cursor has been raising $2bn at a $50bn valuation. Two AI coding companies, both around $50bn, both raising $2bn, suggests the market is treating the category as an emerging duopoly rather than a broad collection of startups. The parallel pricing gives investors a benchmark and gives each company a reason to keep moving quickly.
Buyers have circled too. SpaceX approached Cognition days after the Cursor deal, a sign that the AI coding sector is attracting interest not only from financial investors but also from strategic acquirers. The details of that approach have not been disclosed, but the fact that it happened adds a layer of strategic tension to the pure fundraising story.
What Cognition says it sells now
Cognition's product, Devin, is no longer positioned only as an AI assistant that writes code on request. The company describes a broader platform: Auto-Triage takes a first pass at incidents, Security Swarm finds and triages vulnerabilities, and Automations trigger work from Slack, GitHub and Linear. In Cognition's telling, the engineering team becomes an architectural layer, setting goals and reviewing work while software agents execute.
The company calls this the dawn of the self-driving software era. That phrase captures both the opportunity and the hype risk. Autonomous coding tools could indeed change the economics of software development, but the evidence so far is mostly in controlled demonstrations and company-provided metrics. Named customers include GE Aerospace, Citi, Mercedes-Benz and Modal. The earlier Series D announcement cited Goldman Sachs, Dell, Santander, the US Army and the US Navy. The customer list is substantial, but it is short enough that customer concentration remains a live question.
The strongest proof point is also the hardest to verify
Cognition's most frequently cited claim is internal: the company said in May that around 90% of the code committed by its own engineers was committed by Devin. If true, that is powerful evidence that the product has real utility. A company selling engineering automation is the easiest possible customer for engineering automation. It can deploy its own product in-house, dogfood it, and use the results to persuade outside developers.
The claim is also self-referential and impossible to audit from outside. Nobody outside Cognition can see the commit history or verify how the 90% figure was measured. It may be the best evidence the company has that Devin works, but it is not evidence that an independent observer can test. The same limitation applies to the revenue figures and to the product descriptions: the public version of the company's story is controlled by the company.
What to watch between now and the next round
The first signal will be whether the valuation multiple holds at the next financing. A flat multiple through two doublings of reported revenue suggests discipline. An expanding multiple on slowing growth would be a more conventional late-cycle sign. The second signal is customer concentration. Run-rate revenue built on a handful of large enterprise deals behaves very differently from revenue spread across thousands of smaller accounts. One bad churn event can change the trajectory; a broad base is more resilient.
The third signal is the cost base. Cognition opened offices in Washington, Tokyo, Singapore, London, São Paulo and Madrid this year. That kind of global expansion does not come cheap. Hiring employees in new markets, building sales teams, and supporting enterprise customers around the world all cost money. The office footprint is a bet that the growth continues. If the growth decelerates, the fixed cost base becomes much harder to justify.
For now, Cognition has a valuation that is extraordinary but not increasingly extraordinary on a revenue-multiple basis. The next round will show whether investors continue to pay 53 times a company's self-reported run-rate revenue, and whether the revenue itself keeps doubling. Those two variables — the multiple and the reported growth — will determine