San Francisco-based startup Niobia is betting big on a critical but often overlooked part of battery and chip making: helping engineers understand how and why things go wrong. Its AI software flags production defects and lets engineers ask a chatbot what caused them, aiming to deliver answers in minutes rather than days.

How it works broadly: The company’s customers place AI agents on their production lines to look for defects, then loop in human engineers to work out what might cause a product to fail. The agents run on self-contained GPUs mounted directly on manufacturing equipment, much like a sensor. They can spot pinhole-sized flaws such as scratches. Niobia’s AI also gathers all the data in one place so engineers can ask a chatbot what went wrong and dig deeper into the cause.

Speed is so critical because chips and batteries are some of the world’s most important products, while also increasingly in short supply. As manufacturers continue to develop and improve their technologies, things inevitably go awry during research and development. Engineers then spend days or longer figuring out a fix before they can restart production. Each flaw or mistake can prove costly — sometimes running into the millions of dollars — and set back manufacturing timelines.

Niobia co-founders Gaurav Jha and Harveen Kaur both have PhDs in chemistry and have worked at chipmakers and battery manufacturers, giving them firsthand experience with the challenges within the industry. In a call ahead of TechCrunch Disrupt 2026, where the company is a Battlefield finalist, Jha explained that while what Niobia does is specialized, it serves a multibillion-dollar market that includes some of the biggest battery and semiconductor makers today.

“It’s extremely niche; maybe people will never know about it,” said Jha, speaking of the battery and chip manufacturing market that his company aims to serve. “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,” added Jha.

Niobia’s agents run on open-source models that the company trains further with battery and chip research and other data. That training is what lets them identify pinhole-sized defects. Jha said that Niobia chose open-source models largely because its customers don’t want their highly sensitive company information going into a public AI cloud where it might 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,” said Jha. The startup also outlines how it approaches cybersecurity, given the high-stakes race among adversarial nations to innovate in the battery and chip space. Niobia for instance lets its customers hold their own encryption keys so that the startup cannot decrypt the customers’ data.

Jha said that the company’s priority for now is to focus on three specialties, such as film coating, electroplating (which is 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).

He said that the company also wants its technology to become more autonomous, with the eventual aim of building out a fully self-driving factory for its customers.

If you’re curious to learn more, Niobia will present on stage at TechCrunch Disrupt this afternoon in San Francisco. Join us to catch the company and the 19 startups against which it’s competing as part of our storied Startup Battlefield competition, happening today, Tuesday, through Thursday, October 15.

Doug Field has worked on EVs at Ford and Tesla, the not-so-secret car project at Apple, and was employee No. 1 at Segway. Now, the executive is turning his attention to Rivian CEO RJ Scaringe’s latest project: Mind Robotics.

Field has been hired at president of Mind Robotics, an industrial AI and robotics venture that spun out of Rivian last year. The news, shared on stage at Disrupt 2026, is another sign that Mind Robotics is getting ready to scale up its early operations just one year after its founding.

Scaringe will still be involved with Mind Robotics, but the day-to-day aspects of the business, including product, engineering, and operations will be handled by Field.

Mind Robotics, which has already amassed more than $1 billion in backing from Accel, Andreessen Horowitz, Eclipse, Kleiner Perkins and the venture arms of Volkswagen and Salesforce, is building a full stack for industrial manufacturing. The purpose-built robots will run on foundation models with an goal of automating industrial and manufacturing tasks at scale.

Unlike the humanoid-like robots being developed by Tesla and others, Scaringe says Mind Robotics won’t attempt to mimic all the biomechanics of a human. The startup is also using Rivian as “a live, high-volume manufacturing environment for model training and deployment,” with Rivian being Mind Robotics’ first customer.

“It’s incredibly exciting what’s happening in robotics and what we’re capable of doing today, when we think about the progress that’s happened on the hardware, modeling techniques, [and] bringing that together to develop robots that can actually be deployed at scale,” Scaringe said on the Disrupt stage.

Field left Ford in April after five years shaping the automaker’s electric vehicle and technology strategies. His departure was part of a wider reorganization of the automaker’s leadership. Prior to Ford, Field led Apple’s Project Titan, the electric, autonomous car project that was canceled in 2024. He also served as senior vice president of engineering at Tesla, where he oversaw development of the Model 3 and got the sedan through “production hell.”

“Integrating AI and robotics is one of the most important–and difficult–engineering challenges of our time,” Field said in a statement. “Mind has assembled an exceptional team and has a unique advantage: the ability to deploy and learn inside a real, high-volume factory from day one. I’m thrilled to join RJ in building a team, culture, and technology that will redefine what we can make in America.”

Mind Robotics is the second company that Scaringe spun out of Rivian, following Also, a micromobility play that’s starting with high-end e-bikes. Like with Field and Mind Robotics, Scaringe is involved as chairman at Also but has tapped someone to run the day-to-tay: Chris Yu, the former chief product officer of Specialized.

Mind Robotics and Also operate as standalone companies, but Rivian has minority stakes in both. The company reported owning about a third of each startup as of June 30, 2026.

Rockets are, infamously, a real pain in the butt. Our old friend physics forces rockets to be built with almost no room for error, and even then only a small fraction of their weight makes it to orbit.

There are, at least in theory, other ways to do the job—space elevators, centrifuges, or a big chain hanging from a satellite in space and scooping up cargo from Earth. (Blue Origin spent some time thinking about that one before the idea wound up in a Neal Stephenson novel.)

And then there’s an idea dating all the way back to Jules Verne: Use a big gun to shoot a projectile into space.

Longshot Space Technology, a startup based in Alameda, California, is working on exactly that. The cofounders, CEO Mike Grace and CTO Nato Saichek, see an opportunity to take a technology that has already proven its worth, and apply it to the pressing need for space launch. The startup has raised about $20 million from venture investors in service of its mission.

When it comes to frontier technology, Grace says, “you can build something that has a lot of novelty, or you can do something really hard. You don’t get to do both.” He points to SpaceX as an example of a company that has “taken classically de-risked technology and scaled it aggressively.” What SpaceX did with NASA rocket research, he says, Longshot will do with multi-injection guns — designs that give a projectile extra boosts as it moves down the barrel — built for World War II and during 1960s research programs.

The idea is to take all the work done by the first stage of a launch vehicle and collapse that into a reusable accelerator on the ground. Picture a steel pipe mounted horizontally over the ground, with tanks of high-pressure gas attached at ports spread across its length. When a projectile is fired and moves down the barrel, the tanks fire off at precise times to accelerate it — to speeds as high as Mach 4, thus far.

The company is currently firing an 70-foot gun with a 6-inch diameter at its test site near Barstow, California, and says it can accelerate projectiles up to near-hypersonic speeds. Soon, it expects to start firing a 140-foot-long gun with a 30-inch diameter at its Alameda test site. After that, the company is preparing to build out a 30-inch diameter gun that is 328-feet long, with plans to extend it to about 3,000 feet—making it the largest gun on the planet. It will be capable of launching 220-pound payloads up to speeds of Mach 8, the team says.

These guns will let Longshot keep proving its technology works while also generating revenue from performing hypersonic testing for the US military, which has awarded the company several contracts. Grace envisions building one of his guns on a US military installation like Fort Bliss, where it could be part of the infrastructure for testing the next generation of US interceptors. (And no, the guns don’t need to point up at an angle. Longshot’s projectile is a lifting body, meaning its shape generates lift, so it will aerodynamically gain altitude once it exits the barrel.)

“If you want to practice shooting down a Chinese or Russian hypersonic missile, wouldn’t it be nice not to pay $30 million for that?” Grace asks. “This is a way for half a million, a quarter of a million bucks, of doing hypersonic clay pigeons, throw them and shoot them down, and do it really fast and really cheap.”

The ultimate goal, however, is putting large payloads into space. The team is confident it can scale up its current design to accelerate payloads to as much as 5 kilometers per second (more than 11,000 mph). That’s fast enough to get into space if the payload, which effectively serves as the second stage of a rocket, has its own propulsion system. But doing that would require a gun 10 kilometers (about 6 miles) long and nearly 10 feet in diameter. As you might imagine, one problem the company faces is simply finding suitable worksites.

“The hard part is not building the thing, it’s finding a place where they will let you build the thing,” Saichek jokes.

But if Longshot can realize its vision, the math is very attractive: The system is cheap to construct and to fuel, and is highly reusable. Grace expects that if everything works out, they can send payloads to space for a hard-to-believe $50 per kilogram (about $22 per pound). Considering that the much-hyped orbital data center industry needs prices to fall to around $300 per kilogram to get off the ground, that leaves Longshot plenty of room to offer a competitive price.

It will require some modification to launch spacecraft in this form factor and conditions, but at that price point, satellite operators will be willing to try.

If you want to learn more about Longshot from the company itself — while also checking out dozens of startups, hearing their pitches, and listening to guest speakers on four different stages — join us this week at Disrupt, Tuesday to Thursday, in San Francisco. Learn more here.