Korea’s biggest manufacturers back Config, the TSMC of robot data

Asia’s propulsion into carnal AI is being fueled by the aforesaid manufacturing prowess that made the portion a planetary concern powerhouse. Across South Korea, Japan, China and Taiwan, manufacturing remains a cardinal pillar of economical growth. Unlike economies much heavy weighted toward services oregon software, these countries person agelong relied connected large-scale production, export-driven industries, and highly optimized proviso chains. That structural instauration is present shaping however artificial quality is adopted and wherever concern flows.

Which makes it peculiarly important that Config, a Seoul- and San Jose- based startup gathering the information furniture for robotic instauration models (RFMs), has secured backing from the task arms’ of South Korea’s biggest manufacturers.

Samsung Venture Investment led its oversubscribed $27 cardinal effect circular astatine a valuation of much than $200 million, bringing Config’s full raised to $34 million. Hyundai Motor’s task limb ZER01NE Ventures, LG Tech Ventures, and SKT America, a South Korean telco giant’s VC unit, besides joined arsenic strategical investors, alongside angel capitalist Pieter Abbeel (co-founder of Covariant AI and a UC Berkeley professor) and fiscal backers including Mirae Asset Ventures, Korea Development Bank, GS Futures, Kakao Ventures, and Z Ventures.

Config was founded successful January 2025 by CEO Minjoon Seo, a erstwhile researcher astatine Meta and main idiosyncratic astatine Twelve Labs, on with 4 co-founders with backgrounds astatine Waymo, Google and Naver. Instead of gathering robots themselves, the squad is focused connected a simpler goal, providing information robots request to larn and operate. They judge that amended information volition beryllium cardinal to making robots much useful.

Training ample connection models is expensive, due to the fact that of the computing powerfulness required to process them, but the earthy material, immense amounts of substance from crossed the internet, is casual to obtain. Teaching robots to determination is simply a wholly antithetic challenge, Seo said successful an exclusive interrogation with TechCrunch. Every portion of grooming information has to beryllium physically collected, similar you request the robot, the installation to tally it, and radical to run it. That makes robotics AI much costly to make than software-only chatbot, according to the Seo. As companies prosecute much susceptible robots, the outgo of gathering and labeling information tin emergence quickly.

Config wants to beryllium the institution that makes everyone else’s robot AI possible. The startup compares its relation to TSMC, a Taiwanese chipmaker that manufactures  for Apple, Nvidia, and AMD without competing with immoderate of them. Config aims to play a akin relation successful robotics by supplying the data. The attack is gaining traction arsenic ample manufacturers progressively question to physique their ain proprietary robot AI alternatively of relying wholly connected extracurricular vendors. That is the marketplace Config is betting on.

Config is already generating revenue, COO and co-founder of Config Jack Bang said. The startup’s existent customers see ample manufacturers, strategy integrators, and companies successful the agriculture and defence sectors, Bang told TechCrunch. Peers successful the abstraction see Physical Intelligence, Generalist AI and Skild AI.

Image Credits:Kate Park

Config records humans performing carnal tasks successful controlled workplace environments and successful the field. The startup operates retired of Seoul and Hanoi, wherever a workforce of astir 300 handles information production. To date, it has accumulated implicit 100,000 hours of quality question data, much than 30 times the size of AgiBot World, the largest comparable open-source dataset astatine astir 3,000 hours.

Most robotics teams bid AI models connected quality question information and past accommodate those models for a robot. Config is taking a antithetic approach, Seo said. The institution focuses connected transforming the information earlier grooming begins truthful it is amended suited to the mode robots determination and interact with the world. Seo compared the process to connection translation. Training a exemplary connected 1 benignant of information and expecting it to enactment seamlessly successful different setting, Seo said, is trying to thatch Korean utilizing lone English-language materials.

“The information indispensable beryllium converted, not the model. This conversion exertion is Config’s halfway method differentiator,” Seo said.

The backing volition spell toward 3 priorities: scaling its information cognition successful Vietnam and Seoul toward 1 cardinal hours of collected data, increasing its endeavor level concern to $10 cardinal successful ARR by the extremity of 2026, and launching a cloud-based Robot-as-a-Service merchandise that lets companies tally Config’s instauration exemplary without requiring onboard hardware.