Asia’s Manufacturing Backbone Drives the Rise of Physical AI
Asia’s push into physical AI innovation is deeply intertwined with the region’s established strength in manufacturing. Countries like South Korea, Japan, China, and Taiwan have long been recognized as industrial powerhouses, leveraging large-scale production, export-driven industries, and highly optimized supply chains to fuel economic growth. Unlike economies that rely heavily on services or software, these nations’ manufacturing foundations are now steering the direction of artificial intelligence adoption and investment flows, particularly in robotics and physical AI.
Amid this landscape, Config, a startup based in Seoul and San Jose, is emerging as a pivotal player. Focused on building the data layer for robotics foundation models (RFMs), Config recently secured significant backing from the venture arms of South Korea’s top manufacturers, signaling the strategic importance of their work within the region’s industrial ecosystem.
Config’s Strategic Backing and Vision
In a highly competitive $27 million seed round led by Samsung Venture Investment, Config achieved a valuation exceeding $200 million. The funding round also saw participation from Hyundai Motor’s ZER01NE Ventures, LG Technology Ventures, and SKT America, the venture capital arm of a major South Korean telecommunications company. Other notable investors include angel investor Pieter Abbeel—co-founder of Covariant and a professor at UC Berkeley—alongside Mirae Asset Ventures, Korea Development Bank, GS Futures, Kakao Ventures, and Z Ventures.
Founded in January 2025 by CEO Minjoon Seo, a former Meta researcher and chief scientist at TwelveLabs, Config’s leadership team also includes co-founders with experience at Waymo, Google, and Naver. Rather than building robots themselves, Config focuses on a critical but often overlooked component: providing the high-quality data robots need to learn and operate effectively. Seo emphasizes that superior data quality is essential to making robots more capable and useful in real-world applications.
The Unique Challenges of Robotics AI Data
Training large language models is resource-intensive but benefits from the abundance of text data available online. Robotics AI, however, faces a fundamentally different challenge. According to Seo, every piece of training data must be physically collected, requiring robots, specialized facilities, and human operators. This drastically increases development costs compared to software-only AI models like chatbots.
As robotics technology advances, the expense of gathering and labeling diverse, high-quality data continues to rise. Config positions itself as the crucial infrastructure provider in this space—akin to how Taiwan Semiconductor Manufacturing Company (TSMC) manufactures chips for giants like Apple and Nvidia without competing with them. Config’s ambition is to become the “TSMC of robot data,” supplying data that enables manufacturers to build proprietary robot AI rather than relying solely on external vendors.
COO and co-founder Jack Bang confirms that Config is already generating revenue, serving large manufacturers, system integrators, and clients in agriculture and defense sectors. Peers in the robotics data field include companies like Physical Intelligence, Generalist AI, and Skild AI.
Data Collection and Innovation in Robotics Training
Config’s data collection process involves recording humans performing physical tasks in both controlled studio environments and real-world field settings. Operating out of Seoul and Hanoi, the company employs nearly 300 people to handle data production. To date, Config has amassed over 100,000 hours of human motion data—more than 30 times larger than AgiBot World, the largest comparable open-source dataset, which contains roughly 3,000 hours.
Most robotics AI teams train models using human motion data and then adapt those models for robotic applications. Config, however, takes a novel approach by transforming the data itself before training. Seo likens this to language translation—expecting a model trained on one type of data to perform well in a completely different setting is like trying to learn Korean solely through English-language materials. Thus, Config’s core technical innovation lies in its data conversion technology, which adapts human motion data to better suit robot movement and interaction.
“The data must be converted, not the model. This conversion technology is Config’s core technical differentiator,” Seo explains.
Future Plans and Growth Trajectory
Config plans to use its newly raised funds to scale data operations in Vietnam and Seoul with an ambitious goal of reaching 1 million hours of collected data. The company also aims to grow its enterprise platform to $10 million in annual recurring revenue by the end of 2027. Additionally, Config intends to launch a cloud-based robot-as-a-service product that enables clients to run their robotics foundation models without the need for onboard hardware, further lowering barriers to adoption.
With the confluence of Asia’s manufacturing expertise and Config’s innovative approach to robotics data, the region is poised to become a global leader in physical AI development and deployment.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Source: Here
