AI Is Writing Code Faster Than Ever. This Startup Just Hit a $1 Billion Valuation Testing All of It

Blacksmith, a startup that runs the continuous-integration infrastructure companies use to test code before it ships, has raised a $66 million Series C led by Sequoia Capital at a $1 billion valuation. The round comes just two months after a $45 million Series B valued the company at $550 million, led by Peak XV Partners, bringing its total funding to roughly $124.5 million. Founded in 2024 by CEO Aditya Jayaprakash and co-founder Aayush Shah, the company has grown quickly largely because of a problem AI coding tools created for it to solve.

Tools like Cursor, Codex, and Claude Code have sped up how fast code gets written, which has shifted the real bottleneck in software development from writing code to checking it. Anthropic has reported that validation workloads on its own systems grew 25-fold over six months, and Blacksmith has seen its own CI traffic jump 40% each time Anthropic released a new model, evidence that more AI-generated code means more testing demand, not less.

Unlike most CI providers, which rent compute from AWS, Google Cloud, or Azure, Blacksmith owns its own servers, which it says lets it run the automated builds and tests that code must pass before merging into a company's existing codebase more cheaply and reliably at the volumes AI-driven development now produces.

The growth numbers back up the valuation jump. Blacksmith's customer base has passed 10,000 companies, up from roughly 5,000 in August, and it's now running about 90 million validation checks a week, up from 5 million at the start of the year. Annual recurring revenue has grown 17-fold over that same stretch and is now in the tens of millions, with its largest customers each spending more than $1 million a year. Headcount has grown to around 40 employees, with a target of roughly 75 by the end of the year.

Jayaprakash argues the rise of AI coding assistants helps his business rather than threatens it. "Usage on us actually goes up," he said, describing how each more capable AI coding model tends to produce more code that needs validating, not less demand for the infrastructure that checks it.