Cognition AI in Talks to Raise Hundreds of Millions at $25 Billion Valuation

Cognition AI, the startup behind Devin — the first widely deployed autonomous AI software engineer — is in early-stage discussions to raise a new funding round that would value the company at approximately $25 billion, according to people familiar with the matter. The target represents a more than 2.4x increase over the $10.2 billion valuation Cognition achieved in September 2025 when it closed a $400 million round led by Founders Fund, with participation from Lux Capital, 8VC, Elad Gil, Definition Capital, and Swish Ventures. The new round is expected to total hundreds of millions of dollars, though terms remain in flux and could change before any deal is finalized.

The valuation jump reflects both Cognition’s extraordinary ARR growth and the extraordinary premium the market is placing on companies that have credibly solved the problem of autonomous software development at enterprise scale.

Devin’s Trajectory: From $1M to $150M ARR

When Devin launched in March 2024, it was widely reported on as a research curiosity — an AI agent that could resolve some GitHub issues autonomously, but whose real-world utility was still unproven. The benchmark numbers were striking: a 13.9% success rate on SWE-bench, versus 1.7% for GPT-4 and 4.8% for Claude 2 at the time. But benchmark success and enterprise adoption are different things.

Devin’s ARR tells the adoption story more clearly than any benchmark. The product went from $1 million ARR in September 2024 to $73 million ARR by June 2025 — a 7,200% growth rate over nine months, driven primarily by enterprise contracts with companies that found the autonomous agent useful for the specific workflows where Devin excels: long-horizon coding tasks, repository-wide refactors, integration of new services into existing codebases, and bug hunting across complex dependency trees.

At Linktree, Devin authored approximately 300 pull requests in a single month as of early 2025, with around 100 successfully merged. The tasks included end-to-end integration of new social platforms — RedNote, Lemon8 — that required backend modifications, URL handling updates, and UI component adjustments, handled without human intervention at the task level.

The growth story accelerated further in July 2025 with Cognition’s acquisition of Windsurf, an AI-powered IDE with its own enterprise customer base and approximately $82 million ARR at the time of acquisition. Combined, Cognition’s run-rate revenue reached roughly $150–155 million by mid-2025, making it one of the fastest-growing AI companies in history on a revenue basis.

The Competitive Landscape: Racing Cursor

The $25 billion target must be understood against the backdrop of Cursor’s $50 billion valuation, which it achieved in a $2 billion round earlier this month. Cursor, built by Anysphere, has positioned itself as an IDE-layer AI coding tool — deeply integrated into the developer’s existing workflow with autocomplete, chat, and code generation capabilities. Its $50 billion valuation reflects massive adoption among individual developers and the strength of its product-led growth model.

Cognition and Cursor are solving different problems from different angles. Cursor augments the human developer; Devin attempts to replace entire task-completion loops. This distinction matters enormously for enterprise buyers. A company evaluating AI coding tools must decide whether it wants a tool that makes engineers faster (Cursor’s model) or an agent that takes tasks off the engineer’s plate entirely (Devin’s model). Both markets are large; both are growing.

The tension is also playing out in Windsurf’s position. Cognition’s IDE, which serves a segment of developers who prefer a Cursor-style workflow, now sits inside a company whose flagship product is a full autonomous agent. The combined portfolio gives Cognition an unusual ability to compete across the spectrum of AI coding use cases — from individual developer productivity to enterprise-grade autonomous execution.

The broader field includes GitHub Copilot (Microsoft’s AI coding assistant with the largest installed base), Amazon CodeWhisperer (now integrated into Amazon Q), and a cohort of newer entrants like Augment Code, Magic.dev, and SWE-agent variants from academic labs. No single company has yet established monopoly-level lock-in, which is why investors are willing to fund multiple players at high valuations simultaneously.

Enterprise Adoption at Scale

Cognition’s customer list reads like a cross-section of the global financial and technology elite. Goldman Sachs, Citi, Dell, Cisco, Ramp, Palantir, Nubank, and Mercado Libre are among the companies using Devin in production. These are not pilot programs or proofs of concept — at the ARR figures Cognition has disclosed, these are sustained, recurring commitments from organizations whose engineering organizations have integrated Devin into their standard workflows.

For financial institutions in particular, Devin’s value proposition aligns with a decade-long push to reduce software development costs while maintaining or improving output quality. Goldman Sachs and Citi have both invested heavily in modernizing legacy codebases — a task that is expensive, tedious, and exactly the kind of long-horizon, repository-wide work that Devin is designed to handle.

Cognition has also been notably disciplined on spending. Despite its scale of revenue, net burn has remained under $20 million since the company’s founding approximately two years ago — an extraordinary capital efficiency ratio for a company generating $150 million in ARR. This efficiency partly reflects a product architecture where Devin’s marginal cost of serving additional tasks is primarily inference compute rather than engineering headcount.

What the $25 Billion Valuation Implies

At $25 billion on roughly $150 million ARR (extrapolating from mid-2025 figures; 2026 ARR is likely higher given the growth trajectory), Cognition would be trading at approximately 166x annualized revenue — a multiple that is high even by AI startup standards. Cursor’s $50 billion valuation implies similar or higher revenue multiples.

The market is not pricing these companies on current revenue; it is pricing them on the assumption that autonomous AI agents will compress or replace a significant fraction of software development labor over the next five to seven years, and that the companies establishing early enterprise relationships and platform lock-in will capture disproportionate value from that transition.

Whether that thesis proves correct is the central question of AI enterprise investing in 2026. The early evidence — from Cognition’s ARR trajectory, Cursor’s user growth, and the speed at which large enterprises are signing multi-million-dollar AI coding contracts — suggests the demand is real. The remaining uncertainty is around durability: will the advantage of early movers compound as their agents improve on customer-specific data, or will a better model from Anthropic, OpenAI, or an unannounced competitor commoditize the task-execution layer and leave the incumbents exposed?

For now, investors appear willing to pay the premium that uncertainty commands.