The archive · Developer & Business Tools · Strategic decision · 2024–2026
Config Bets on 'TSMC of Robot Data': Samsung, Hyundai, LG, SK Invest $27M
Samsung, Hyundai, LG, SKT invest $27M in Config, a robotics data firm with 100k hours of motion data.
Config (컨피그인텔리전스)
What the business is
Supplies high-quality training data for robot foundation models (RFM): operates data factories of about 300 people in Seoul and Hanoi, collects human motion data, and converts it into formats robots can learn directly.
Starting capital：US$27 million seed round (led by Samsung Venture Investment, valuation exceeds US$200 million); cumulative funding US$35 million
How it started
CEO Minjoon Seo is a KAIST associate professor, former Meta researcher, and chief scientist at video AI company Twelve Labs; founded Config in 2024 (fully operating in Seoul and San Jose in early 2025); three co-founders came from Waymo, Google, and Naver.
What happened
In May 2026, completed an oversubscribed US$27 million seed round led by Samsung Venture Investment, with participation from Hyundai ZER01NE, LG Technology Ventures, SK Telecom America, Kakao Ventures, GS Futures, Z Ventures, Korea Development Bank, and UC Berkeley professor Pieter Abbeel, among others; cumulative funding US$35 million. The company has already run data collection teams of about 300 people in Seoul and Hanoi, accumulated more than 100,000 hours of human motion data, and generated revenue from customers in large manufacturing, system integration, agriculture, and other fields.
How it ended up
As of September 2026: valuation exceeds US$200 million; the goal is to scale data to 1 million hours, reach US$10 million annual recurring revenue (ARR) for the enterprise platform by the end of 2027, and launch a cloud Robot-as-a-Service product that lets customers run its RFM without local hardware; the decisive factor is whether data scale and conversion technology can stay ahead.
Background
Config is a company that supplies training data for robot foundation models (RFM), headquartered in Seoul and San Jose. It does not build robots itself; it operates data factories of about 300 people in Seoul and Hanoi, films humans performing physical tasks, and converts the data into formats robots can learn directly, calling itself 'the TSMC of robot AI'.
Founder Minjoon Seo is a KAIST associate professor, did research at Meta, served as chief scientist at video AI company Twelve Labs, founded the company in 2024 (fully operating in early 2025); three co-founders came from Waymo, Google, and Naver. The team's bet is clear: robot AI development costs are far higher than software because each training data point must be collected using equipment, locations, and people; whoever can scale high-quality data becomes a supplier the entire industry cannot avoid.
In May 2026, the company completed an oversubscribed US$27 million seed round led by Samsung Venture Investment, with participation from Hyundai ZER01NE, LG Technology Ventures, SK Telecom America, Kakao Ventures, GS Futures, Z Ventures, Korea Development Bank, and Covariant co-founder and UC Berkeley professor Pieter Abbeel; valuation exceeds US$200 million and cumulative funding is US$35 million. The four South Korean manufacturing giants putting competition aside and sitting at the same table has been interpreted as recognition of the prospect of a 'data monopoly supplier'.
Scale is currently its hardest data: it has accumulated more than 100,000 hours of human motion data, about 30 times or more the size of China's AgiBot open-source dataset AgiBot World (about 3,000 hours); it has generated revenue from customers in large manufacturing, system integration, agriculture, and other fields. The funding will be used to scale data to 1 million hours, achieve US$10 million ARR for the enterprise platform by the end of 2027, and launch a cloud Robot-as-a-Service product.
What has to be true
- Positioning avoids head-on competition: Config only does the data layer, supplying all RFM vendors—the TSMC model means more customers strengthen its position.
- Data is the real bottleneck: robot training data must be collected physically, expensive and scarce; scaling first builds hard-to-replicate barriers.
- Technical differentiation in 'conversion' not 'models': translating human data into robot-ready formats is less effortful than competing on algorithms.
- Joint endorsement by Samsung, Hyundai, LG, SKT CVCs: rare co-investment signals data infrastructure is key, giving Config orders and trust.
- Risks: open-source datasets and synthetic data advance rapidly; 100k-hour lead may not be permanent, and 1M hours and RaaS are the real test.
What can be applied
Treat data as an independent industry: Config supplies all competitors, avoiding head-on competition. But it bets the data gap lasts; if open-source or synthetic data catch up, the moat shrinks.
Aftermath
As of 2026-09-01: Config's valuation exceeds $200M. It is scaling data collection to 1M hours, targeting $10M ARR by end-2027, and launching a cloud RaaS product. COO Jack Bang said revenue comes from manufacturers, integrators, and agriculture. Rivals like Skild AI, Physical Intelligence, and Generalist AI compete; Config's position depends on keeping data scale and conversion cheapest and most usable.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card