Z Venture Capital invests in Config Intelligence's robot foundation models
What's the deal? Z Venture CapitalDealroom has a profile for this one. Try Dealroom → (ZVC), the venture arm of Japan's Z HoldingsDealroom has a profile for this one. Try Dealroom →, has invested in Config Intelligence, a South Korean startup building robot foundation models. Config operates a full-stack platform that spans data collection through model development for robotics AI.
The company's proprietary Action Estimator converts cheap human video footage into high-precision robot action data at roughly 500-micrometer accuracy — about one-tenth the width of a human hair. Its foundation model, CFG-1, has been trained on more than 100,000 hours of in-house data, works across different hardware, and is already generating revenue.
Config has secured strategic partnerships with Samsung, LG, and Hyundai, booking sales in its first year of operation. The founding team includes CEO Minjoon Seo (KAIST, Meta FAIR), CTO Hyungmok Son (Harvard PhD, Waymo), chief scientist Kimin Lee (KAIST, Google Brain), and COO Jack Bang (UBS, Glenwood).
Why now? ZVC describes the current moment in robotics as the "GPT-2.5 moment" — hardware is maturing and model architectures are stabilising, but the critical missing ingredient is data. Unlike text or images, robot action data can't be scraped from the internet; most of it has been collected through expensive teleoperation.
The market for robot foundation models and data infrastructure is projected to grow at over 50% annually, reaching more than $18B by 2030. The window to build a defensible data moat is narrowing fast.
South Korea, which has the world's highest robot density and a world-class manufacturing base, offers an ideal proving ground. Factory-floor deployment data collected there is difficult for overseas competitors to replicate.
What could go wrong? Config is entering a space with well-funded rivals. Physical Intelligence and Covariant have reached the same conclusion about end-to-end integration of data, models, and deployment. Competing on cost alone may not be enough if larger players accelerate their own data collection efforts.
The reliance on Korean industrial partners is a double-edged sword — it provides early revenue and proprietary data, but could limit global expansion if the technology proves too tailored to specific manufacturing environments.
The signal: ZVC's bet on Config Intelligence underscores a growing conviction among corporate venture arms that the robotics value chain is unbundling — and that data infrastructure, not hardware, will be the defensible layer. Config's "breakout" growth stage, combined with first-year revenue from Samsung, LG, and Hyundai, suggests the company is already moving beyond research-phase economics at a pace few physical AI startups have matched.
Read more: prtimes.jp
