At a glance: Cognition AI. Product: Devin (autonomous AI software engineer). Round closed: May 27, 2026. Raised: $1B+. Post-money valuation: $26B. Pre-money: $25B. Lead investors: Lux Capital, General Catalyst, 8VC. New investors: Ribbit Capital, Atreides Management, Layer Global. Existing investors: Founders Fund (Peter Thiel), 8VC. ARR run-rate: $492M. Devin enterprise usage growth: 50% MoM for 6 months. Part of our AI Models & Companies reviews.
On May 24, 2026, we covered the rumors: Cognition AI in talks for hundreds of millions at a $25 billion valuation. The lead sentence of that article asked the question the market was asking.
Three days later, Bloomberg confirmed the answer: Cognition raised more than $1 billion at a $26 billion post-money valuation. The round closed May 27, 2026.
This is what changed.
What the Round Confirms
Cognition’s last confirmed valuation was $10.2 billion, set in September 2025 when the company raised $400 million led by Peter Thiel’s Founders Fund. That round closed eight months ago.
The new round is led by Lux Capital and General Catalyst, with 8VC co-leading. New investors include Ribbit Capital, Atreides Management LP, and Layer Global. Founders Fund participated again as an existing investor.
The jump from $10.2B to $26B in eight months represents a 2.5x increase. That is notable on its own. What’s more notable is the revenue trajectory behind it.
$492 Million. Growing 50% Every Month.
In June 2025, Devin’s annualized revenue run-rate was $73 million. That was already a 73x increase from $1M in September 2024 — nine months of growth that commentators treated as exceptional.
The figure announced with this round: $492 million in annualized revenue run-rate, with enterprise usage growing 50% month-over-month for six consecutive months.
50% month-over-month means compounding growth that most companies cannot sustain for a quarter. Six consecutive months at that rate transforms $1 of revenue into $11.39 in six months. Applied to an already substantial base, it explains the valuation math even without any projection assumptions: at $492M ARR and 50% MoM growth, the trajectory implies annualized revenue above $1 billion within the current quarter if the curve holds even partially.
Whether that curve holds is the central question for Cognition’s next chapter. For now, the number is the number.
The Enterprise Customer List, Expanded
When we wrote our May 24 article, the most prominent named enterprise customer was Goldman Sachs — 12,000 developers in a pilot targeting 20% efficiency gains.
The round announcement expanded that list significantly. Production customers now confirmed by Cognition include:
- Goldman Sachs — 12,000-developer hybrid workforce pilot
- Citi — enterprise deployment
- Dell — enterprise deployment
- Cisco — enterprise deployment
- Ramp — fintech infrastructure
- Palantir — AI/data infrastructure
- Nubank — Latin American fintech
- Mercado Libre — Latin American e-commerce
- Mercedes-Benz — automotive engineering
- NASA — government research
This is not a customer list assembled from one vertical. Financial services (Goldman, Citi, Nubank, Ramp), enterprise tech (Dell, Cisco, Palantir), automotive (Mercedes), government (NASA), and e-commerce (Mercado Libre) represent Devin being used in genuinely different engineering environments.
The diversity matters for a specific reason: Devin’s prior criticism centered on benchmark performance in abstract settings. A customer list spanning regulated banking (Goldman, Citi), government security protocols (NASA, Palantir), and global commerce infrastructure (Mercado Libre) is evidence of performance in constrained, production-grade environments — where the cost of a hallucinated PR or a missed security requirement is not a benchmark score but an incident.
What the May 24 Article Asked About
Our previous coverage raised four specific questions to watch. The round announcement addresses some of them directly.
The round closing. Answered: closed at $26B post-money on May 27.
SWE-Bench 2.0 figures. Unanswered: Cognition still has not published updated benchmark scores. The original 13.86% remains the last published figure. As of this writing, the company’s position appears to be that revenue is the more meaningful signal. That’s a defensible position with $492M ARR; it gets harder to hold as competing agents publish figures in the 50–70% range.
Goldman Sachs pilot results. Partially answered: the expanded customer list confirms Goldman remains a production customer, not just a pilot participant. Whether the 20% efficiency target has been disclosed is not confirmed as of this writing.
Windsurf 2.0 adoption. Addressed indirectly: the 50% MoM Devin usage growth presumably includes adoption via the Windsurf 2.0 IDE integration (Agent Command Center), but Cognition has not broken out Windsurf-sourced vs. API-sourced revenue in public communications.
The Competitive Context
The AI coding infrastructure space has continued to move since April.
Cursor (Anysphere) remains in the process of being acquired by SpaceX for $60 billion — the deal that triggered the valuation rerating across AI coding companies in late April. That acquisition has not formally closed as of this writing.
GitHub Copilot continues to expand under Microsoft’s enterprise distribution. The Gartner Magic Quadrant 2026 for enterprise AI coding agents listed GitHub Copilot and Cursor as the two most-recognized names in the category; Cognition’s differentiation — cloud autonomy vs. IDE assistance — is meaningful but requires continued explanation in enterprise procurement conversations.
OpenAI Codex and Anthropic’s computer-use agent infrastructure continue to develop as underlying capabilities that third parties can use to build competing agents. Cognition’s moat is not the model — it’s the product layer, the enterprise workflow tooling, and the distribution that Windsurf adds through IDE users.
At $492M ARR and a $26B valuation, Cognition’s implied revenue multiple is approximately 53x ARR. That is a significant premium to most enterprise software benchmarks, but consistent with the trajectory multiples being applied across the AI infrastructure category in Q2 2026.
What to Watch Next
Revenue deceleration. 50% MoM for six months implies either a continued acceleration toward $1B+ ARR or a deceleration as the installed base grows larger. The next quarter’s figures — if disclosed — will be the first signal of which dynamic is dominant.
SWE-Bench transparency. Cognition has let competitors define the benchmark narrative. Publishing Devin 2.0 scores would either challenge the narrative or confirm that the company has a legitimate reason not to.
Goldman Sachs 12,000-developer results. The pilot has been running long enough that measurable efficiency data exists. Disclosure — in any form — would be the most powerful third-party validation available to Cognition.
SpaceX/Cursor close. The $60B deal remains pending. If it closes, it resets the pricing anchor for the category upward and likely accelerates further consolidation.
The Series H. At the current growth trajectory, Cognition will be a potential candidate for a late-stage round approaching or exceeding $50B valuation within the next 12–18 months — or an IPO candidate alongside the broader AI infrastructure wave signaled by OpenAI’s confidential S-1 filing.
We asked three days ago whether Cognition’s round would close, and at what number. It closed. The number was $26 billion.
The question worth asking now is whether $492 million in annualized revenue — growing at 50% month-over-month, across Goldman Sachs, NASA, Palantir, Nubank, and eleven other named enterprise customers — is the beginning of the curve or the peak of it.
The $1 billion wired into Cognition’s accounts is an explicit bet on the former.
Previously: Our May 24 analysis covered Cognition’s funding talks and the SWE-Bench critique in depth. See also: our Windsurf 2.0 review and the OpenAI IPO guide for the broader AI IPO context.