A single production flaw in chip or battery manufacturing can cost millions of dollars and idle a line for days. San Francisco startup Niobia says its AI software can shrink that diagnosis from days to minutes — and it's betting a multibillion-dollar market wants exactly that.

The company, co-founded by chemistry PhDs Gaurav Jha and Harveen Kaur, places AI agents directly on production lines to catch defects, then loops in human engineers to work out root causes. The agents run on self-contained GPUs mounted on manufacturing equipment itself, much like a sensor. They can spot pinhole-sized flaws such as scratches.

Niobia also pulls all production data into one place, so engineers can ask a chatbot what went wrong and dig into the underlying cause. That consolidation matters. During R&D, things inevitably go awry, and engineers routinely spend days or longer figuring out a fix before restarting production — a delay that grows more painful as chips and batteries stay in short supply worldwide.

Who's behind it?

Jha and Kaur both hold PhDs in chemistry and have worked inside chipmakers and battery manufacturers, so they've lived the failure-analysis problem firsthand. Speaking ahead of TechCrunch Disrupt 2026, where Niobia is a Battlefield finalist, Jha framed the market as specialized but enormous, serving some of the biggest battery and semiconductor makers operating today.

"It's extremely niche; maybe people will never know about it," Jha said of the market his company serves. "But it's like behind the scenes, this is powering everything… we just don't see it in the regular world."

"Every product around you, like glasses or shoes, slippers, wallets, and headphones — these processes touch every single part of it," he added.

Why open-source models, and why on-prem GPUs?

Niobia's agents run on open-source models that the company trains further with battery and chip research and other data. That extra training is what enables the pinhole-level detection. The open-source choice wasn't ideological. It was a customer demand.

"None of our customers want their proprietary data to go to the cloud, even if we tell them it's safe," Jha said.

The stakes go beyond commercial secrecy. Niobia points to the high-stakes race among adversarial nations to innovate in batteries and chips, and it has built its cybersecurity posture accordingly. The company lets customers hold their own encryption keys, so Niobia itself cannot decrypt customer data.

Three processes first, then autonomy

For now, Niobia is deliberately narrow. Jha said the company is focusing on three specialties:

• Film coating
• Electroplating — coating a surface with metal using an electric current
• Fluidized bed reaction — a chemical process where solid particles are suspended in a gas or liquid

Beyond that, Jha wants the technology to become progressively more autonomous, with the eventual aim of building out a fully self-driving factory for its customers.

Niobia will present on stage at TechCrunch Disrupt in San Francisco on Tuesday, competing against 19 other startups in the Startup Battlefield competition, which runs through Thursday, October 15. If its minutes-not-days diagnosis pitch lands with the fabs and battery gigafactories already strained by supply shortages, edge-deployed defect agents could move from niche tooling to standard line equipment.