Config Raises $27M to Become the 'TSMC of Robot Data'
Config, a Seoul- and San Jose-based startup building the data infrastructure layer for robotics foundation models, has raised a $27M seed round led by Samsung Venture Investment at a $200M+ valuation — backed by Hyundai, LG, and SKT in a sign that South Korea's manufacturing giants are betting heavily on AI-native robotics.
Original sourceConfig has raised $27 million in an oversubscribed seed round at a valuation exceeding $200 million, backed by Samsung Venture Investment, Hyundai's ZER01NE Ventures, LG Technology Ventures, and SKT America. The round also includes Berkeley robotics professor and Covariant co-founder Pieter Abbeel as an angel investor, along with Mirae Asset Ventures, Korea Development Bank, and GS Futures.
The company's pitch is to be the TSMC of robot data — not building robots itself, but providing the critical data infrastructure that makes robotic foundation models (RFMs) possible at scale. Founded in January 2025 by Minjoon Seo (former Meta researcher and chief scientist at TwelveLabs) alongside co-founders from Waymo, Google, and Naver, Config has assembled one of the more credible founding teams in the space.
The TSMC analogy is deliberate: just as the semiconductor giant doesn't design chips but manufactures them for everyone, Config aims to be the neutral, shared data infrastructure layer that model builders rely on regardless of their hardware or platform choices. The company currently operates data collection operations in Vietnam and Seoul, and plans to scale toward 1 million hours of collected robot data.
The funding will go toward three priorities: scaling data collection operations, growing its enterprise platform to $10M ARR by end of 2027, and launching a cloud-based Robot-as-a-Service product that lets companies access Config's foundation model without requiring onboard hardware. The Korean strategic investor base is particularly notable — Samsung, Hyundai, LG, and SKT collectively manufacture a significant share of the world's industrial robots and consumer electronics, giving Config potential distribution at scale.
Robotics foundation models are still early-stage, but the race to control the training data layer is accelerating. Config's bet is that whoever owns the canonical dataset for robot behavior will have an enduring structural advantage — similar to how ImageNet shaped a generation of computer vision research, but with 10x the commercial stakes.
Panel Takes
The Builder
Developer Perspective
“The founding team is exceptional — Waymo, Google, Naver, Meta pedigree building robot data infrastructure backed by the actual manufacturers. If you're building on top of robotics foundation models, this is the company you want to watch for the canonical training data layer.”
The Skeptic
Reality Check
“A $200M+ valuation on a 16-month-old seed-stage data collection company is aggressive. The TSMC analogy sounds compelling until you realize TSMC has irreplaceable physical infrastructure — Config's data moat could be disrupted by any large robot OEM that decides to collect its own training data.”
The Futurist
Big Picture
“Robotics foundation models will reshape manufacturing the way LLMs reshaped knowledge work — and the data layer is where the durable value accrues. Config's Korean strategic backers aren't just writing checks; they're securing preferred access to the training data infrastructure for their own robot fleets.”