Major Korean Manufacturers Support Config, the 'TSMC of Robot Data'

Asia's advancement in physical AI is driven by the same manufacturing expertise that established the region as a global industrial leader. In South Korea, Japan, China, and Taiwan, manufacturing continues to be a cornerstone of economic expansion. Unlike economies more focused on services or software, these nations have historically depended on mass production, export-oriented industries, and highly efficient supply chains. This structural base is now influencing the adoption of artificial intelligence and directing investment trends.

This context makes it especially noteworthy that Config, a startup based in Seoul and San Jose developing the data infrastructure for robotic foundation models (RFMs), has secured investment from the venture capital arms of South Korea's largest manufacturing conglomerates.

Samsung Venture Investment led its oversubscribed $27 million seed funding round, valuing the company at over $200 million and bringing its total funding to $35 million. Strategic investors in the round included Hyundai Motor's venture arm ZER01NE Ventures, LG Tech Ventures, and SKT America, the venture capital unit of the South Korean telecom giant. They were joined by angel investor Pieter Abbeel (co-founder of Covariant AI and a UC Berkeley professor) and financial backers such as Mirae Asset Ventures, Korea Development Bank, GS Futures, Kakao 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 who previously worked at Waymo, Google, and Naver. Rather than constructing robots, the team concentrates on a more fundamental objective: providing the essential data robots require to learn and function. They believe superior data will be crucial for enhancing robot utility.

Training large language models is costly due to the necessary computing power, but the raw material—vast amounts of text from the internet—is readily available. Teaching robots to move presents a distinct challenge, Seo explained in an exclusive interview with TechCrunch. Every piece of training data must be physically gathered, requiring the robot, a facility to operate it, and personnel. This makes developing robotics AI more expensive than software-only chatbots, according to Seo. As companies pursue more capable robots, the expenses for data collection and labeling can escalate rapidly.

Config aims to be the enabling force behind other companies' robot AI. The startup likens its role to that of TSMC, the Taiwanese chipmaker that manufactures for Apple, Nvidia, and AMD without competing with them. Config seeks to play a similar part in robotics by supplying foundational data. This strategy is gaining momentum as major manufacturers increasingly look to develop their own proprietary robot AI instead of relying solely on external vendors. This is the market Config is targeting.

Config is already generating revenue, according to COO and co-founder Jack Bang. The startup's current clientele includes large manufacturers, system integrators, and companies in the agriculture and defense sectors, Bang told TechCrunch. Its peers in the field include Physical Intelligence, Generalist AI, and Skild AI.

Image Credits:Kate Park

Config captures humans performing physical tasks in both controlled studio settings and real-world environments. The company operates from Seoul and Hanoi, where a workforce of nearly 300 manages data production. To date, it has amassed over 100,000 hours of human motion data, which is more than 30 times the size of AgiBot World, the largest comparable open-source dataset at approximately 3,000 hours.

Most robotics teams train AI models on human motion data and subsequently adapt those models for robots. Config is taking a different approach, Seo stated. The company focuses on transforming the data before the training process begins, making it better suited to how robots move and interact with their surroundings. Seo compared the process to language translation. Training a model on one type of data and expecting it to work flawlessly in another context, Seo said, is akin to trying to teach Korean using only English-language materials.

"The data must be converted, not the model. This conversion technology is Config's core technical differentiator," Seo emphasized.

The new funding will be allocated toward three key priorities: scaling its data operations in Vietnam and Seoul to reach one million hours of collected data, growing its enterprise platform business to achieve $10 million in Annual Recurring Revenue (ARR) by the end of 2027, and launching a cloud-based Robot-as-a-Service product that allows companies to utilize Config's foundation model without needing specialized onboard hardware.