Path Robotics has spent eight years on one of the hardest jobs to automate in a factory. Welding parts that never quite match their drawings. On November 10 the Columbus, Ohio company announced a $300 million Series E at a $2.45 billion post-money valuation to apply what it learned to the rest of the shop floor.
Georgian, an AI-focused growth equity firm, led the round. Existing backers Addition, Drive Capital, IronArc Ventures, Tiger Global and Yamaha Motor Ventures returned, and D. E. Shaw Ventures and NAVER Ventures joined. Bloomberg puts total funding at about $610 million. Georgian partner Emily Walsh takes a board seat.
Growth that justifies the jump
The valuation nearly quadrupled from the Series D in October 2024. Path says annual recurring revenue grew 17-fold in the last 18 months as it scaled deployments across shipbuilding and defense, heavy equipment and energy infrastructure. Those are industries with chronic welder shortages and long order books. Shipyards in particular can’t hire fast enough to meet Navy demand, and a production contract with HII has been reported alongside the round.
Path’s product is Obsidian, a welding model that lets a robot cell see the part in front of it, work out where the seam actually is and adapt the weld in real time. Traditional welding robots repeat a programmed path and fail when parts vary. That variation is why most welding is still done by hand.
From one task to the factory
The round is priced on a bigger claim. CEO Andy Lonsberry, who co-founded the company with his brother Alex, says the goal is to automate the entire factory, not one step. The money funds new form factors, more compute and more model training. The bet is that data from a hard task like welding transfers to the jobs around it: grinding, fitting, inspection, material handling.
If it works, Path becomes a physical AI platform for manufacturing rather than a welding equipment vendor. That’s the gap between a $600 million company and a $2.45 billion one. The ARR growth suggests customers are paying for the welding part already. Next comes proving the model generalises.