Forty-eight billion dollars. That’s what investors think Cognition AI is worth after its September 2026 Series E, a round that raised more than $2 billion led by Andreessen Horowitz and Accel. The number is eye-catching. But for anyone running engineering teams or managing software delivery, the more useful data point is what’s happening inside the companies already using the product.
Cognition’s annualized run-rate revenue grew from $492 million in May to roughly $900 million by September, according to the company. The Information projects the company could reach $4 billion to $5 billion in ARR by year-end. Those are company-reported figures, not audited revenue, and the distinction matters. But the trajectory explains why a16z, which made a significant return when Cursor sold to SpaceX for $60 billion in April, is now leading a round in a direct competitor. This isn’t a winner-take-all market. Smart money is hedging.
The enterprise deployments tell the real story. Goldman Sachs brought Devin on as what it called its first AI software engineer in July 2025, CNBC reported at the time. The bank has since moved from pilot to scaling thousands of instances, targeting a three-to-four productivity boost over prior AI tools. Cognizant, meanwhile, struck a strategic partnership in January 2026 to integrate Devin and Windsurf into its Flowsource engineering platform. Thirty percent of Cognizant’s code is already AI-generated, with a target of 50%.
Those aren’t sandbox experiments. The customer list now reads like a cross-section of global enterprise: Citi, Dell, Cisco, Palantir, Mercedes-Benz, NASA, the US Army and Navy, Infosys, Nubank, Santander. The deployment gap that analysts spent 2025 warning about is narrowing, and these contracts are the evidence.
Here’s the tension underneath the headline: Cognition is spending roughly $800 million a year on compute, mostly leasing Nvidia server clusters, according to The Information. That’s nearly as much as its current ARR. The company is training its own models on open-source foundations to cut that dependency, but the math only works if the revenue keeps compounding at this pace. For enterprise buyers, the question isn’t whether Devin works. It’s whether the economics of the provider can sustain the infrastructure your team would depend on.
That’s the practical read on a $48 billion valuation. The technology is past the proof-of-concept stage. Goldman Sachs and Cognizant aren’t running pilots anymore-they’re running production. The next chapter isn’t about whether AI coding agents can work at scale. It’s about whether the companies building them can afford to keep the lights on while they do.