Asia’s move into physical AI is driven by its manufacturing strength, making it a global leader. South Korea, Japan, China, and Taiwan, continue to focus on large-scale production and export-driven strategies. This foundation influences AI adoption and investment directions.
Significantly, Config, a Seoul and San Jose-based startup developing data layers for robotic foundation models (RFMs), has received investments from South Korea’s major manufacturers’ venture arms.
Samsung Venture Investment led Config’s $27 million seed round, valuing it at over $200 million and totaling $35 million in funds raised. Hyundai Motor’s ZER01NE Ventures, LG Tech Ventures, SKT America, angel investor Pieter Abbeel, and other financial backers participated.
Founded in January 2025 by CEO Minjoon Seo, with three co-founders from Waymo, Google, and Naver, Config doesn’t build robots but provides critical data for them to learn and function. Superior data is fundamental to making effective robots.
Training large language models is costly due to computing needs, but acquiring text data is easy. In contrast, robotics training requires physically collected data, making it more expensive. Seo explained that to train robots to move, you need robots, facilities, and operators. As robots become more capable, data gathering and labeling costs can escalate.
Config aims to facilitate robot AI development like TSMC does for chip manufacturing, assisting large manufacturers in creating proprietary robot AI. Config is already generating revenue, catering to manufacturers, system integrators, and sectors like agriculture and defense. Competitors include Physical Intelligence, Generalist AI, and Skild AI.
Config records human tasks in studios and the field from Seoul and Hanoi, with 300 staff managing data production. It has amassed over 100,000 hours of human motion data, surpassing AgiBot World’s 3,000-hour open-source dataset.
Most robotics teams train models on human motion data before adapting them for robots. Config transforms data pre-training for better robot adaptation, likening it to language translation. Seo emphasizes that data must be converted, not the model, marking Config’s technological edge.
Funding will expand data operations to one million hours, grow its platform to $10 million ARR by 2027, and launch a cloud-based Robot-as-a-Service, allowing the use of Config’s model without hardware.