Niobia says its AI software can help engineers trace battery and chip manufacturing defects to their causes in minutes instead of days. The San Francisco-based startup’s system flags problems on production lines and lets engineers query a chatbot about what went wrong.
Customers mount self-contained GPUs directly on their manufacturing equipment, much like a sensor. The AI agents run on those GPUs and scan for defects, including pinhole-sized flaws such as scratches. Engineers then investigate what may have caused a product to fail. Niobia’s software also pulls production data into one place, so they can ask a chatbot what went wrong and dig into possible causes.
Manufacturing issues can emerge during research and development as companies develop and improve their technologies. Finding a fix may take engineers days or longer, delaying the restart of production. A flaw can cost millions of dollars and push back manufacturing schedules. Chips and batteries are among the world’s most important products, and both are increasingly in short supply.
Niobia co-founders Gaurav Jha and Harveen Kaur have PhDs in chemistry and have worked for chipmakers and battery manufacturers. Their industry experience gave them firsthand knowledge of its challenges. In a call ahead of TechCrunch Disrupt 2026, where Niobia is a Battlefield finalist, Jha said the company serves a multibillion-dollar market that includes some of today’s largest battery and semiconductor makers, even though its work is specialized.
“It’s extremely niche; maybe people will never know about it,” Jha said of the battery and chip manufacturing market Niobia serves. “But it’s like behind the scenes, this is powering everything… we just don’t see it in the regular world.” He added: “Every product around you, like glasses or shoes, slippers, wallets, and headphones – these processes touch every single part of it.”
The agents use open-source models that Niobia trains further on battery and chip research and other data. Jha said this training allows them to spot pinhole-sized defects. He said Niobia chose open-source models largely because customers don’t want sensitive company information sent to a public AI cloud, where it could be used to train the model. “None of our customers want their proprietary data to go to the cloud, even if we tell them it’s safe,” he said.
Niobia outlines its cybersecurity measures against the backdrop of competition between adversarial nations to advance battery and chip innovation. For example, customers can hold their own encryption keys, which means Niobia cannot decrypt their data.
Jha said Niobia is currently concentrating on three areas: film coating, electroplating, and fluidized bed reaction. Electroplating uses an electric current to coat a surface with metal. Fluidized bed reaction suspends solid particles in a gas or liquid as part of a chemical process.
Niobia’s longer-term goal, Jha said, is to make its technology more autonomous and eventually build fully self-driving factories for customers.
