The main Korean manufacturer is back Config, TSMC for robot data

Asia’s push into physical AI is fueled by the same innovation that made the region a global industrial hub. Across South Korea, Japan, China and Taiwan, manufacturing remains a central pillar of economic growth. Unlike economies that are heavily labor-intensive or software-intensive, these countries have been heavily dependent on large-scale manufacturing, export-driven industries, and supply chains. This stable foundation is now shaping how artificial intelligence is taken and where money flows.

Which makes it even more important Fix itthe original Seoul- and San Jose- based data layer architecture examples of robotics basics (RFMs), has received support from the business units of major South Korean companies.

Samsung Venture Investment led its $27 million seed round at a valuation of over $200 million, making Config worth $35 million. Hyundai Motor’s venture arm ZER01NE Ventures, LG Tech Ventures, and SKT America, the main VC arm of the South Korean telco giant, also joined as investors, along with angel investor Pieter Abbeel (founder of Covariant AI and UC Berkeley professor) and Korean financial backers, Mirao Development Bank, Korea Futures Ventures Ventures, and Z Ventures.

Config was founded in January 2025 by CEO Minjoon Seo, a former researcher at Meta and chief scientist at Twelve Labs, along with three co-founders from Waymo, Google and Naver. Instead of creating robots themselves, the team focuses on a simpler goal, providing robots with the data they need to learn and work. They believe that good data will be the key to making more useful robots.

Teaching examples of large languages ​​is expensive, because of the computer power needed to run them, but the resources, many articles from the Internet, are easy to find. Teaching robots to move is a different challenge, Seo said in an interview with TechCrunch. Any training information must be collected physically, such as what you want the robot to be, the place to drive it, and the people who will use it. This makes AI robotics more expensive to develop than software chatbots, according to Seo. As companies pursue intelligent robots, the cost of collecting and documenting can quickly rise.

Config wants to be the company that makes AI robots possible for everyone. The startup compares its position to TSMC, a Taiwanese chipmaker that makes products for Apple, Nvidia, and AMD without competing with any of them. Config aims to play a similar role in robotics by providing data. This trend is gaining momentum as major manufacturers are keen to develop their own robot AI systems instead of relying entirely on external suppliers. This is the market that Config is betting on.

Config is already making money, Config COO and co-founder Jack Bang said. Current customers include major manufacturers, machine integrators, and agricultural and defense companies, Bang told TechCrunch. Its partners in space include Physical Intelligence, Generalist AI and Skild AI.

Config records people performing physical activity in controlled studio environments and in the field. The startup operates out of Seoul and Hanoi, where about 300 employees work on data processing. To date, it has collected more than 100,000 hours of human traffic data, 30 times its size. AgiBot World, the largest parallel open source dataset about 3,000 hours.

Many robotics teams train AI models on human movement data and then adapt the robots. Config is taking a different approach, Seo said. The company focuses on transforming pre-learning data to fit the way robots move and interact with the world. Seo compared the process to translating a language. To train a model on one type of data and hope it will work seamlessly in another, Seo said, he’s trying to teach Korean using only English materials.

“The data should be changed, not the model. This changing technology is the main differentiator of Config,” said Seo.

The money will go to three important things: expanding its data service in Vietnam and Seoul to a million hours of collected data, expanding its business to $10 million in ARR by the end of 2027, and launching a cloud-based product Robot-as-a-Service that allows companies to run the Config model without needing on-board equipment.