TechCrunch reported on May 11 that Config, a robotics company founded only in January 2025, has closed a $27 million seed round at a valuation above $200 million. Its total funding now stands at $35 million.
The valuation is not the most interesting part. The investor list is.
Samsung Venture Investment led the round, joined by Hyundai Motor's ZER01NE Ventures, LG Technology Ventures and SKT America. Samsung, Hyundai, LG and SK Telecom rarely move in formation at the seed stage, yet all four are now backing the same four-month-old company.
It is not building robots
Config CEO Minjoon Seo previously worked as a researcher at Meta and, before that, as chief scientist at video AI company TwelveLabs. The three co-founders came from Waymo, Google and Naver.
But Config is not trying to make its own robot.
"What we want to build is the TSMC of the robotics industry: a company that does not compete with any robot brand directly, but supplies the basic material everyone needs."
That is Config's own positioning. TSMC manufactures chips for Apple, Nvidia and AMD without selling phones or graphics cards itself. Config wants to play a similar role by supplying training data for companies building robots.
Why robotics needs a data foundry
The core bottleneck was already obvious in 2024.
Large software models can scrape the web when they need more data. Robot models cannot crawl the internet for real-world manipulation.
To learn how to fold clothes, open a door or place a pan on a stove, a robot needs demonstrations that record joint angles, force feedback, visual input and action timing. Each useful example carries physical-world metadata, not just text or pixels.
The industry's usual options all have trade-offs:
| Approach | Problem |
|---|---|
| Build an in-house training site and pay people to demonstrate tasks with data gloves | Expensive, slow and duplicated company by company |
| Use simulated synthetic data | The sim-to-real gap remains hard to close |
| Let robots explore through reinforcement learning | Failures in the physical world are costly |
Config's bet is that instead of every robotics company building its own data-collection line, the work can be outsourced and supplied by volume.
The logic resembles the rise of cloud computing in the 2010s. Internet companies once built their own server rooms; many eventually found it more efficient to pay AWS. Robot training data may be approaching a similar platform shift.
Why Korean conglomerates are aligned
Samsung, Hyundai and LG all have robotics ambitions of their own.
- Samsung has consumer robotics lines such as Ballie.
- Hyundai owns Boston Dynamics, acquired in 2021.
- LG works on factory automation and service robots.
- SK is building positions in robotic sensors and AI chips.
Under normal circumstances, these companies would be expected to back different robotics startups because their product roadmaps overlap. This time, they have converged on a data supplier that does not compete with them downstream.
The message is clear: instead of each group building and guarding its own robot-data pipeline, they can jointly support a shared upstream company and buy data from Config when they need to train systems.
That is the same upstream-standardization logic that made TSMC central to chips. Apple and Qualcomm compete fiercely in end devices, but both rely on the same foundry layer.
By co-investing in Config, the Korean groups are effectively acknowledging that robot differentiation is more likely to sit in model design, mechanical engineering and applications than in raw data capture. The data layer can be shared.
Config already has revenue
Config says it is already generating revenue. Its customers include large manufacturers, system integrators, robotics companies in agriculture and robotics companies in defense.
The last two categories matter. Agricultural robots need demonstrations from outdoor, irregular terrain. Defense robots need tactical actions and decision samples in complex environments. If Config is serving both markets, its data-collection network is not limited to factory floors.
One read
Placed inside the global robotics supply chain, Config's role is easy to frame.
Nvidia is building the GPU layer for robots, from Jetson to the Cosmos world model. Google DeepMind is building the brain, with systems such as RT-2 and Gemini Robotics. Boston Dynamics, Figure and Unitree are building the body.
Config wants to supply the formula milk for robots: standardized input that many builders can use before they differentiate elsewhere.
That market has historically been mixed in-house, with uneven quality, high cost and poor compatibility. If Config can standardize robot training data across scale and scenarios, robotics could get its first shared base layer.
The open questions are practical. Can Config cut data-collection costs by an order of magnitude versus customers doing it themselves? Can its data format become a de facto standard? If either fails, the "TSMC for robots" label will remain a PR line.
What is already clear is that Korea's industrial giants are not trying to race the United States and China head-on in building every robot. They are quietly taking an upstream position instead. It is not the flashiest move, but it is a steady one.
Sources: Korea's biggest manufacturers back Config, the TSMC of robot data (TechCrunch), CocoLoop, Why Config Is the AI Robotics Data Story Nobody Tells (KoreaPlus)