Cognition Raises More Than $2 Billion at a $48 Billion Valuation as Devin Expands
September 9, 2026
Cognition has raised more than $2 billion in a Series E financing that values the artificial intelligence coding company at $48 billion. The round, led by Andreessen Horowitz and Accel, arrives as enterprises move from testing coding assistants to assigning larger software projects to autonomous agents.
The scale of the investment makes Cognition one of the most highly valued private AI software companies. It also places a demanding price on the future of Devin, the company’s agent designed to plan, write, test and maintain software with less step-by-step supervision than a conventional coding copilot.
Cognition’s $2 billion Series E: key facts
Cognition raised more than $2 billion at a $48 billion valuation.
Andreessen Horowitz and Accel led the financing.
Founders Fund, General Catalyst and Avenir returned as investors.
The company says run-rate revenue increased from $492 million in May to almost $900 million.
Devin is used by customers including NVIDIA, GE Aerospace, Citi, Mercedes-Benz and Modal.
Why investors are paying a premium for AI coding
Software engineering has become one of the clearest commercial markets for generative AI. Companies already know how much they spend on developers, maintenance, security reviews and delayed product work. That creates a measurable comparison between the cost of an agent and the value of projects completed faster.
Cognition’s valuation assumes that AI coding will expand beyond autocomplete and chat. The company describes a future in which engineers act more like architects: people define systems, priorities and constraints, while agents execute substantial portions of the work. If that model succeeds, coding agents could become a new labor and infrastructure layer inside nearly every large organization.
Devin is moving from assistant to operating system for engineering work
Devin was introduced as an autonomous software engineer, but Cognition is broadening the product around continuous workflows. Devin Auto-Triage can take a first pass at production incidents. Security Swarm is designed to find and prioritize vulnerabilities. Automations can start work when events occur in Slack, GitHub, Linear and other systems.
These features matter because valuable engineering work rarely begins with a perfectly written prompt. A production alert, support ticket or code change creates context that must be gathered across multiple tools. The stronger commercial opportunity is an agent that can detect the work, assemble the evidence, attempt a solution and return a reviewable result.
Revenue growth provides the central investment argument
Cognition says annualized run-rate revenue climbed from $492 million in May to almost $900 million. Run-rate revenue is not the same as audited annual revenue, and rapid adoption can change as contracts renew. Even so, the company-reported figure gives investors a concrete explanation for why the valuation nearly doubled in a short period.
The customer list also points to demanding use cases. Chip design, aviation, banking, automotive engineering and AI infrastructure involve large codebases, security controls and expensive downtime. Expansion inside those organizations would indicate that agents are becoming part of real production processes rather than remaining experimental developer tools.
What the investor group signals
The round includes a long list of venture and growth investors, from Benchmark, Bessemer and Greylock to T. Rowe Price, DST, NVIDIA and others. A syndicate of that size gives Cognition capital for product development, computing, enterprise support and international expansion. It may also create strategic relationships across the technology ecosystem.
Cognition says it wants to remain an independent agent lab that can combine its own systems with models from multiple providers. That flexibility could help customers avoid dependence on one foundation-model vendor. It also means Cognition must keep improving its orchestration, evaluation and workflow technology as underlying models become more capable.
The competitive landscape is getting harder
AI coding is crowded. Model companies, cloud providers, developer-platform businesses and startups are all building agents that can edit repositories and use tools. Better foundation models can quickly narrow the performance gap between products, while enterprise buyers may prefer tools already integrated with their identity, cloud and source-control systems.
Cognition therefore needs durable advantages beyond access to intelligence. Those advantages could include proprietary interaction data, reliable long-running execution, enterprise deployment, security controls and a deep understanding of how engineering teams organize work. Customer trust will depend on whether Devin produces correct, explainable changes without creating more review work than it saves.
How coding agents change engineering teams
The near-term effect is more likely to be job redesign than total replacement. Engineers can delegate repetitive migrations, test creation, incident investigation and backlog tasks, then spend more time on architecture, product decisions and difficult debugging. Junior roles may change most because agents can perform work that traditionally helped new developers learn a codebase.
Organizations should preserve learning and accountability. Every agent-generated change needs an owner, test evidence and a clear approval path. Teams also need to prevent agents from accessing production secrets or deploying code without authorization. Faster execution raises the value of disciplined review.
What enterprise buyers should evaluate
Companies should measure completed tasks, defect rates, review time and total cost rather than counting generated lines of code. A useful pilot includes real repositories, realistic permissions and tasks of different difficulty. Buyers should also test how the agent recovers from missing context, failing tools and contradictory requirements.
Security questions are equally important. Leaders should ask where source code is processed, how long data is retained, which models receive it and how credentials are isolated. Logs should connect each action with a user, task and approval. Rollback must be easy when an autonomous change causes a problem.
What to watch next
The next proof points will be retention, expansion and reliability. Cognition must show that customers continue increasing usage after initial deployments and that revenue growth is not dependent on unusually expensive pilots. Independent evaluations of long-running tasks will matter more than benchmark scores on isolated coding questions.
The company’s global offices in Washington, Tokyo, Singapore, London, São Paulo and Madrid suggest that enterprise support is becoming a major part of the strategy. Local implementation teams may help Cognition understand customers’ systems, but they also increase operating costs and make scalable delivery more important.
Why this round matters for the broader startup market
The financing shows that venture capital remains available at extraordinary scale for companies that can pair frontier AI with visible enterprise demand. It does not mean funding conditions have improved equally for every startup. Capital is concentrating around a small number of businesses with rapid revenue growth, recognizable customers and a credible path to owning an important workflow. Founders elsewhere in the market will face pressure to demonstrate similar evidence rather than rely on model access alone.
The bottom line
Cognition’s financing is a landmark bet on autonomous software engineering. The $48 billion valuation reflects strong investor confidence that Devin can become a persistent layer of enterprise production, not simply another coding assistant. The opportunity is enormous, but so is the execution burden. Cognition must convert reported revenue momentum into dependable outcomes, strong renewals and secure operation across complex organizations.
Comments