It was worth $26 billion in May. In August the number being discussed is $40 billion.

Here's the deal: Bloomberg reported on August 12 that Cognition is in talks for a new round targeting a valuation of at least $40 billion, with the raise itself expected to exceed $1 billion.

The number isn't the interesting part. The interval is. Cognition closed more than $1 billion at a $26 billion valuation in May — three months ago — in a Series D co-led by Lux Capital, General Catalyst, and 8VC. That round had itself more than doubled the company's valuation in under eight months.

So this business has printed three separate price points since late 2025, and the gaps between them keep shrinking. Standard venture cadence runs 12 to 18 months. Cognition is issuing a new price tag roughly every quarter.

It's still talk, to be clear. Bloomberg described the discussions as early, and noted the company might not raise at all or might take different terms. But the fact that this report exists says something about the state of the AI coding market right now.

The company: one product, then an acquisition that changed the math

Cognition was founded in 2023 by Scott Wu, and early coverage leaned heavily on the founding team's competitive-programming pedigree. The company entered public consciousness in March 2024 with Devin — pitched as an autonomous agent that takes a ticket, reads the codebase, plans, writes code, runs tests, and opens a PR.

Reactions split hard. The demo went viral, then a wave of independent testing argued it didn't hold up outside the demo, and there was a running fight over benchmark claims. As of September 2024, Devin's annualized revenue run rate was around $1 million. Enormous attention, negligible revenue.

Two things changed that. The first was organic growth through the first half of 2025, which took ARR to roughly $73 million by June. The widely credited reason was repositioning: Devin stopped being sold as a developer replacement and started being sold as something you run several of in parallel to clear grunt work.

The second was decisive. In July 2025, Cognition acquired Windsurf. Windsurf had pioneered the agentic IDE, and Cognition picked up the remaining organization immediately after an OpenAI acquisition collapsed and Google took the licensing deal and key personnel. In Cognition's own framing, the plan was to let teams plan a task in Windsurf with Devin's codebase understanding and delegate chunks of work to a fleet of Devins. The deal added roughly $82 million in ARR and, more importantly, an enterprise distribution channel.

That produced May 2026's headline number: $492 million ARR, up roughly 13x from $37 million a year earlier. That figure justified $26 billion. The figure now justifying $40 billion is an ARR that has doubled again, to near $1 billion.

Laying the numbers out in order

Date ARR Valuation Note
Sept 2024 ~$1M $4B Devin early commercialization
June 2025 ~$73M — organic growth phase
July 2025 +$82M — Windsurf acquisition added
May 2026 $492M $26B Series D, $1B+
Aug 2026 ~$1B $40B (in talks) new round under discussion

Run the revenue multiples and something counterintuitive shows up. May's round priced the company at roughly 53x ARR. The $40 billion now being discussed, against $1 billion ARR, is 40x. The absolute number jumped; the multiple actually came down.

Two readings. Optimistically, revenue is growing faster than valuation, so fundamentals are catching up to price. Skeptically, 53x and 40x are both far above what software businesses normally command — high-growth SaaS has historically traded at 10 to 20 times revenue.

Enterprise usage was reported growing 50% month over month across the six months preceding May. If that rate held, ARR doubling in a quarter is arithmetically coherent. The customer list includes Mercedes-Benz, NASA, and Goldman Sachs, and Cognition has a partnership with Cognizant to push Devin and Windsurf through that firm's engineering organization and its global client base.

Scott Wu also talks about the product differently than he did in 2024. He now says explicitly that Devin isn't sold as a human replacement, and frames it around long-tail grunt work — legacy software updates, platform migrations. That's a substantial retreat from "autonomous software engineer," and it appears to have helped rather than hurt enterprise sales.

Who benefits

Cognition buys runway. AI coding is currently a cash-burning contest on two fronts: model inference costs and go-to-market spend. Running agents consumes tokens in volume, and those tokens are paid to Anthropic, OpenAI, and other model providers. Costs scale with revenue, so growth itself increases the cash requirement. A billion dollars is ammunition for continuing that fight.

Existing investors book paper gains. Lux Capital, General Catalyst, and 8VC, in at $26 billion in May, would be up more than 50% in a quarter. Founders Fund, in far earlier, sees a much larger multiple. All of it unrealized — private marks hold only as long as the next round arrives, and when rounds stop, marks get rewritten.

New investors face harder math. Entering at $40 billion requires believing the company reaches roughly $80–100 billion, or exits near that level. That needs the AI coding market to grow accordingly and Cognition to stay at the front of it.

Customers get a stability signal. Deeply integrating an AI tool is a bet that the vendor still exists in three years, and a large raise reduces that anxiety. The counterweight: valuations this high eventually create pricing pressure that lands on customers.

Working developers gain little directly. Capital may translate into better products, but it also raises the question of how long today's price competition lasts. Several vendors are currently selling below cost.

What the precedents say

Rapid back-to-back rounds aren't new. The outcomes diverged.

The case cited by optimists is OpenAI. From 2023 through 2025, its valuation stepped from roughly $29 billion to the $80 billions, then $150 billions, then $300 billions. Each step drew a chorus of "this time it's too expensive," and each time revenue caught up. The lesson: at sufficiently high growth, time fixes high multiples. Only while the growth holds, though.

Databricks is another. It raised repeatedly in private markets, pushing valuation up while revenue followed. The distinction worth noting is layer — Databricks sits in infrastructure with high switching costs, while coding tools have comparatively low ones.

The failure case is the 2021 software bubble. SaaS companies that raised above 50x revenue took brutal down rounds when rates rose in 2022, with cuts of 70 to 80% common. Liquidation preference stacks meant employee options frequently ended up worthless in the process.

There's a fresher cautionary tale inside this exact market: Windsurf itself. A roughly $3 billion OpenAI acquisition collapsed at the finish line, Google took a licensing deal and the key people, and Cognition absorbed what remained — all within weeks. It demonstrated how fast valuations in this niche can reprice. The irony is that the beneficiary of that episode is the company now discussing $40 billion.

How competitors respond

Anysphere's Cursor leads the IDE layer, strong in individual developer subscriptions and pushing into enterprise. Overlap with Cognition is growing. Both have climbed fast on valuation and ARR. The difference is direction of travel: Cursor starts in an editor with a human in the chair and adds automation; Devin starts from unattended execution and adds collaboration.

Model providers moving downstream are the biggest structural threat. Anthropic's Claude Code, OpenAI's Codex, and Google's Gemini CLI are coding agents built by the companies that own the models. They internalize token costs and get model improvements first. An application-layer company buys its inference at retail and carries permanent margin pressure.

GitHub Copilot wins on distribution. Microsoft's enterprise sales motion plus an installed base of GitHub accounts is formidable, and Copilot has moved toward letting customers choose among multiple models. In procurement, "vendor we already have a contract with" is hard to beat.

Open-weight models push from below. Running DeepSeek or Qwen coding models on your own infrastructure keeps getting more viable. For cost-sensitive organizations, and especially where data can't leave the building, that erodes commercial agents' pricing power.

IT services firms are partners and future rivals at once. Deals like the Cognizant partnership expand reach, but these firms ultimately want their own agent layer. Today's channel becoming tomorrow's competitor is a well-worn pattern in that industry.

So what actually changes

For working developers, nothing immediate. But the capital flow is a signal that investors believe coding agents genuinely monetize. When choosing tools, think about how much vendor lock-in you're willing to absorb — this market is consolidating quickly.

For engineering leaders, this is procurement input. Cognition sits near the top on both funding and revenue, so vendor risk is comparatively low. Do check pricing-adjustment clauses at contract time: parts of this market are priced below cost today, and that can normalize.

For founders, it's a benchmark with an asterisk. Half a billion in ARR within a year is demonstrably achievable at the AI application layer — and roughly half of it arrived via acquisition. This was not purely organic.

For investors, the argument is entirely about the multiple. 40x ARR is defensible only while growth holds. Two things to watch: whether net new customer adds hold up month over month, and whether gross margin improves. The second matters more. In agent businesses, token costs scale with revenue, and without margin expansion there's no economy of scale.

For general readers, treat this as a window into the AI funding cycle. A market where a company's price rises 50% in three months is not a normal market. It may keep going or it may correct — what's true so far is that revenue has kept pace.

🥄 Three Things You're Probably Wondering

— Is the $40 billion a done deal? No. Bloomberg described the talks as early and noted the company might not raise, or might take different terms. Private valuation reporting is a target number until the round actually closes.

— Does $1 billion ARR mean they're profitable? Unrelated questions. ARR is a revenue run rate, not earnings, and Cognition doesn't disclose profitability. A large share of agent revenue flows straight back out as model API spend. Raising another billion is itself evidence that cash burn is substantial.

— Cursor or Devin — who wins? Too early to call, and they're not quite fighting for the same seat. The larger variable is agents built by the model companies themselves. With Anthropic, OpenAI, and Google entering from a cost-advantaged position, the question is what differentiation the application layer can hold.

Sources

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!