A company that was worth $10 billion last autumn is now negotiating to be worth four times that, on the strength of a revenue line that barely existed 18 months ago.

This week, Bloomberg reported that Cognition, the maker of the AI coding agent Devin, is in early talks with investors for a round that would value the company at $40 billion or more. The detail that should stop you is not the number. It is the calendar. Cognition raised $1 billion at a $26 billion valuation in May, roughly three months earlier. Before that, in September 2025, it was valued at $10.2 billion. Three financings, three seasons, a fourfold move.

For anyone trying to understand where AI value is actually accruing, Cognition is the cleanest signal on the board. Not because it is the biggest name in AI — it is dwarfed by the labs building the models it runs on — but because it sits at the exact point where model capability turns into revenue. And the market is pricing that conversion at a speed it has rarely paid for anything.

What actually happened

Cognition builds Devin, an autonomous software engineer that a company can assign work to the way it would assign a junior developer. The pitch is not autocomplete. It is delegation: hand Devin a ticket, and it plans, writes, tests, and opens a pull request.

The new valuation talk rests on one figure. According to the sources Bloomberg cites, the round would clear $40 billion on the back of an annualized revenue run rate approaching $1 billion. Put that against the last data point the company disclosed publicly, and the growth is stark. When it raised in May, founder Scott Wu told TechCrunch the company had reached a $492 million annualized run rate, with enterprise usage of Devin growing 50% month over month for six straight months. If the run rate is now near $1 billion, it has roughly doubled in a quarter.

That trajectory did not come from Devin alone. In July 2025, Cognition acquired Windsurf, the AI coding tool, in the wreckage of a collapsed Google deal — Google had paid $2.4 billion in a licensing-and-hiring arrangement that stripped Windsurf of its CEO and top researchers, leaving the rest of the company for Cognition to absorb. Buying a distressed asset for a fraction of its former price, then folding its users into your own funnel, is not a glamorous strategy. It worked. Two months later the combined company was worth $10.2 billion.

Why the number is not obviously crazy

It is tempting to file a $40 billion valuation on sub-$1 billion revenue under "bubble" and move on. The math deserves more care.

At $40 billion of valuation against a $1 billion run rate, Cognition would be priced at roughly 40 times annualized revenue. That is expensive by any normal software standard, where mature SaaS trades in the high single digits to low teens. But revenue multiples are a bet on the future, not a reading of the present, and the thing investors are underwriting is the derivative: how fast the line is bending. A company doubling revenue in a quarter is not valued like one growing 20% a year, because if the growth holds even briefly, today's multiple looks cheap in hindsight. The entire wager is that the growth holds.

There is also a real product reason usage is climbing. Wu has been careful to frame Devin not as a replacement for engineers but as a machine for the work engineers dislike — the long-tail grunt work of dragging old software up to date or migrating applications between platforms. That positioning matters commercially. Tedious, well-defined, low-glory tasks are exactly where an autonomous agent can be trusted without a human hovering over every line, and exactly the work enterprises are happy to hand off. Cognition says its customers already include Mercedes-Benz, NASA, and Goldman Sachs.

The stakeholders

For Cognition, the round, if it closes, is less about the cash than the signal. A $40 billion mark buys recruiting leverage against the labs and a war chest to keep acquiring. But it also resets expectations violently. At $10 billion you are a promising startup; at $40 billion you are expected to become an enduring platform, and the patience for a stumble shrinks with every zero.

For investors, Cognition is a concentrated bet on a specific thesis: that the application layer, not just the model layer, will capture durable value. For two years the anxiety in AI investing has been that foundation-model makers would commoditize everything built on top of them. Devin's revenue is the counter-argument — evidence that a company can wrap a model in enough workflow, trust, and enterprise plumbing to command real prices. The risk is symmetrical. If the next generation of frontier models makes autonomous coding a native feature rather than a product, the wrapper thins.

For the incumbents, the pressure is direct. GitHub Copilot, Anysphere's Cursor, and the coding features baked into every major chatbot are competing for the same budget. Cognition's differentiation is that Devin is sold as an agent that owns a task end to end, not an assistant that speeds up a human. Whether that distinction survives as the assistants themselves become more agentic is the open question the whole category turns on.

For engineers, the story is more nuanced than the headlines. The tools are aimed at the work most developers were never precious about. The near-term effect is less "replacement" than "reassignment" — fewer hours on migrations and maintenance, which is either liberation or the slow erosion of the entry-level rungs where junior engineers used to learn, depending on where you sit.

The velocity is the story

Cognition is not an outlier this week; it is the sharpest instance of a pattern. On the same day the $40 billion talk surfaced, Lovable confirmed a $13.3 billion valuation on a fresh $400 million raise, and the code-testing startup Blacksmith saw its valuation jump tenfold to $550 million. Capital is not drifting toward AI coding. It is stampeding.

That concentration tells you something about where the market believes the first durable AI profits will land. Software engineering is a near-perfect early market for autonomous agents: the work is text, the output is verifiable, the buyers are technical, and the value of a shipped feature is easy to price. If agents are going to prove their economics anywhere first, it is here.

The risk is the one every gold rush carries. Valuations that reprice fourfold in a year are pricing a future that has to arrive more or less on schedule. If enterprise adoption plateaus, or a frontier lab folds autonomous coding into its base model and gives it away, the multiples that look prescient today will look reckless in the same rearview mirror. The higher the promise, the lower the patience for breakage.

For now, Cognition has done the rare thing: it turned a research demo into a revenue line that doubles by the quarter, and it did it by aiming at the work nobody wanted to do by hand. That is not hype. That is a business. The only question left is whether the price the market is putting on it is a measure of the business or a measure of the mood. Watch the run rate, not the valuation. The valuation is a story about belief. The run rate is the only thing that can end the argument.

Frequently Asked Questions

What is Cognition's current valuation?

Cognition was last valued at $26 billion in a $1 billion round in May 2026. As of August 12, 2026, Bloomberg reported it is in early talks for a new round that would value it at $40 billion or more, though that round has not closed.

What does Cognition's Devin actually do?

Devin is an autonomous AI software engineer. Companies assign it tasks — often maintenance and migration work such as updating legacy code or moving applications between platforms — and it plans, writes, and tests the code, then submits it for review rather than acting as a real-time coding assistant.

How fast is Cognition growing?

In May 2026, founder Scott Wu said Cognition had reached a $492 million annualized revenue run rate with enterprise Devin usage growing 50% month over month. The August funding talks are reportedly based on a run rate approaching $1 billion, implying it roughly doubled in about a quarter.

Why are AI coding startups raising so much money?

Software engineering is an ideal proving ground for AI agents: the work is text-based, the output is verifiable, and buyers are technical. That has driven a surge of investment — alongside Cognition, Lovable confirmed a $13.3 billion valuation and Blacksmith's valuation jumped tenfold in the same week.

Editor's note — sources: Bloomberg (Aug 12, 2026, $40B talks); TechCrunch (Aug 12, 2026, run rate, customers, Devin framing); TechCrunch (May 2026, $1B/$26B round); TechCrunch (May 2026, $492M run rate, Scott Wu); CNBC (Sept 2025, $10.2B, Windsurf); TechCrunch (Lovable); TechCrunch (Blacksmith).

Share this article: