Korea’s largest producers again Config, the TSMC of robotic information

Asia’s push into bodily AI is being fueled by the identical manufacturing prowess that made the area a world industrial powerhouse. Throughout South Korea, Japan, China and Taiwan, manufacturing stays a central pillar of financial progress. In contrast to economies extra closely weighted towards providers or software program, these international locations have lengthy relied on large-scale manufacturing, export-driven industries, and extremely optimized provide chains. That structural basis is now shaping how synthetic intelligence is adopted and the place funding flows.

Which makes it notably important that Config, a Seoul- and San Jose- based mostly startup constructing the information layer for robotic foundation models (RFMs), has secured backing from the enterprise arms’ of South Korea’s largest producers.

Samsung Enterprise Funding led its oversubscribed $27 million seed spherical at a valuation of greater than $200 million, bringing Config’s whole raised to $35 million. Hyundai Motor’s enterprise arm ZER01NE Ventures, LG Tech Ventures, and SKT America, a South Korean telco large’s VC unit, additionally joined as strategic traders, alongside angel investor Pieter Abbeel (co-founder of Covariant AI and a UC Berkeley professor) and monetary backers together with Mirae Asset Ventures, Korea Growth Financial institution, GS Futures, Kakao Ventures, and Z Ventures.

Config was based in January 2025 by CEO Minjoon Web optimization, a former researcher at Meta and chief scientist at Twelve Labs, together with three co-founders with backgrounds at Waymo, Google and Naver. As an alternative of constructing robots themselves, the workforce is targeted on a less complicated objective, offering information robots have to study and function. They imagine that higher information will likely be key to creating robots extra helpful.

Coaching giant language fashions is pricey, due to the computing energy required to course of them, however the uncooked materials, huge quantities of textual content from throughout the web, is simple to acquire. Instructing robots to maneuver is a totally completely different problem, Web optimization mentioned in an unique interview with TechCrunch. Every bit of coaching information needs to be bodily collected, such as you want the robotic, the ability to run it, and other people to function it. That makes robotics AI extra pricey to develop than software-only chatbot, based on the Web optimization. As corporations pursue extra succesful robots, the price of gathering and labeling information can rise rapidly.

Config desires to be the corporate that makes everybody else’s robotic AI doable. The startup compares its function to TSMC, a Taiwanese chipmaker that manufactures  for Apple, Nvidia, and AMD with out competing with any of them. Config goals to play an identical function in robotics by supplying the information. The method is gaining traction as giant producers more and more search to construct their very own proprietary robotic AI as a substitute of relying solely on outdoors distributors. That’s the market Config is betting on.

Config is already producing income, COO and co-founder of Config Jack Bang mentioned. The startup’s present prospects embrace giant producers, system integrators, and corporations within the agriculture and protection sectors, Bang informed TechCrunch. Friends within the area embrace Bodily Intelligence, Generalist AI and Skild AI.

Config information people performing bodily duties in managed studio environments and within the discipline. The startup operates out of Seoul and Hanoi, the place a workforce of almost 300 handles information manufacturing. To this point, it has gathered over 100,000 hours of human movement information, greater than 30 instances the dimensions of AgiBot World, the most important comparable open-source dataset at roughly 3,000 hours.

Most robotics groups prepare AI fashions on human movement information after which adapt these fashions for a robotic. Config is taking a special method, Web optimization mentioned. The corporate focuses on reworking the information earlier than coaching begins so it’s higher suited to the best way robots transfer and work together with the world. Web optimization in contrast the method to language translation. Coaching a mannequin on one kind of knowledge and anticipating it to work seamlessly in one other setting, Web optimization mentioned, is attempting to show Korean utilizing solely English-language supplies.

“The information have to be transformed, not the mannequin. This conversion know-how is Config’s core technical differentiator,” Web optimization mentioned.

The funding will go towards three priorities: scaling its information operation in Vietnam and Seoul towards a million hours of collected information, rising its enterprise platform enterprise to $10 million in ARR by the top of 2027, and launching a cloud-based Robotic-as-a-Service product that lets corporations run Config’s basis mannequin with out requiring onboard {hardware}.