The TSMC of Robot Data: How Config is Powering the Next Wave of Physical AI
While the world has been captivated by software-based chatbots, a new frontier is emerging in “Physical AI”—the intelligence required to make machines move and interact with the real world. Leading this charge is Config, a startup positioning itself as the foundational data layer for the robotics industry.
The company recently secured a $27 million seed round at a valuation exceeding $200 million. The oversubscribed round was led by Samsung Venture Investment, with strategic participation from Hyundai Motor’s ZER01NE Ventures, LG Technology Ventures, and SKT America. This heavy-hitting lineup reflects a broader shift: Asia’s manufacturing giants are betting that proprietary data, not just hardware, will define the future of industrial automation.
Solving the Robotics Data Bottleneck
Training a Large Language Model (LLM) is relatively straightforward because the raw material—text—is abundant on the internet. In contrast, teaching a robot to perform physical tasks requires high-quality, real-world motion data that is notoriously difficult and expensive to collect.
Founded in January 2025 by CEO Minjoon Seo—a veteran of Meta and TwelveLabs—alongside experts from Waymo and Google, Config aims to be the “TSMC of robot data.” Much like how TSMC manufactures chips for competitors like Apple and Nvidia without competing against them, Config provides the essential data infrastructure that allows manufacturers to build their own proprietary AI models.
A Unique Technical Edge
Config’s differentiator lies in how it processes human motion. Rather than simply recording humans and hoping a robot can mimic them, Config uses a proprietary “conversion” technology.
Why Conversion Matters
- Translation, Not Adaptation: CEO Minjoon Seo compares the process to language. Trying to train a robot on raw human data is like trying to teach Korean using only English materials; the data itself must be translated into a format the machine understands.
- Scale: The startup has already amassed over 100,000 hours of human motion data, which is roughly 30 times larger than the open-source AgiBot World dataset.
The Roadmap to 2027
Operating out of Seoul and Hanoi with a workforce of 300, Config is already generating revenue from sectors as diverse as defense and agriculture. The new capital will be used to scale their data library to one million hours and reach a target of $10 million in ARR by late 2027. Additionally, the company plans to launch a cloud-based “Robot-as-a-Service” platform, allowing enterprises to run advanced foundation models without specialized onboard hardware.
