THE CRUNCH
San Francisco startup Niobia is building AI software for a step of chip and battery making that rarely gets attention: working out why production goes wrong. Its AI agents run on self-contained GPUs mounted directly on manufacturing equipment, much like a sensor, and can spot pinhole-sized flaws such as scratches. Engineers can then ask a chatbot what caused a defect, with the company aiming for answers in minutes rather than the days it can otherwise take before production restarts.
The stakes are high because chips and batteries are increasingly in short supply, and each flaw can cost millions of dollars and set back manufacturing timelines. Co-founders Gaurav Jha and Harveen Kaur, who both hold chemistry PhDs and have worked at chipmakers and battery manufacturers, are presenting the company at TechCrunch Disrupt 2026 in San Francisco, where it is a Startup Battlefield finalist.
The agents run on open source models that Niobia trains further with battery and chip research data. Jha said the company chose open source models largely because customers do not want their proprietary manufacturing data sent to a public AI cloud. Niobia also lets customers hold their own encryption keys, so the startup cannot decrypt their data.
For now the company is focusing on three specialties: film coating, electroplating (coating a surface with metal using an electric current) and fluidized bed reaction (a chemical process where solid particles are suspended in a gas or liquid). Jha said the longer-term aim is greater autonomy, eventually building a fully self-driving factory for customers.