The global race toward physical artificial intelligence has reached a critical turning point as venture capital increasingly floods into foundational infrastructure rather than consumer-facing applications. Across East Asia, an industrial stronghold anchored by advanced manufacturing ecosystems in South Korea, Japan, China, and Taiwan, economic growth has long depended on hyper-optimized supply chains, heavy manufacturing, and high-volume exports. Unlike Western economies that pivoted heavily toward software-as-a-service (SaaS) and consumer internet applications over the past two decades, Asia’s industrial backbone has positioned it as the natural proving ground for physical AI and intelligent robotics.
Capitalizing on this structural shift, Config, a specialized artificial intelligence startup operating out of Seoul and San Jose, has officially closed an oversubscribed $27 million seed funding round. The investment values the company at over $200 million and brings its total capital raised to $35 million since its inception. The funding round was led by Samsung Venture Investment, with strategic participation from the venture arms of several other South Korean industrial and technology giants, including Hyundai Motor’s ZER01NE Ventures, LG Technology Ventures, and SKT America, the corporate venture capital arm of telecommunications leader SK Telecom.
Additional financial and angel backers joined the round, reflecting high-profile confidence in the startup’s trajectory. These include prominent robotics pioneer Pieter Abbeel—a co-founder of Covariant and professor at the University of California, Berkeley—alongside institutional investors such as Mirae Asset Ventures, the Korea Development Bank, GS Futures, Kakao Ventures, and Z Ventures. This formidable coalition of tech giants and financial institutions underscores the strategic urgency manufacturers feel as they race to integrate autonomous systems into physical supply chains, assembly lines, and logistical networks.
Foundational Origins and the Leadership Team
Config was founded in January 2025 by a cohort of seasoned artificial intelligence researchers and engineers with deep roots in some of the world’s most advanced technology labs. Chief Executive Officer Minjoon Seo previously worked as a researcher at Meta and served as the chief scientist at video-understanding AI startup TwelveLabs. He is joined by three co-founders who bring extensive technical experience from Waymo, Google, and South Korean internet giant Naver.
Rather than attempting to build proprietary humanoid or industrial hardware—a capital-intensive and fiercely competitive arena occupied by countless robotics manufacturers—Config’s leadership team identified a far more critical bottleneck in the ecosystem: the scarcity of high-fidelity data required to train modern robotics foundation models (RFMs).
In the landscape of artificial intelligence, large language models (LLMs) like GPT-4 or Claude benefited from an abundance of freely accessible training data scraped from the open internet, including digitized books, academic papers, forums, and websites. While training LLMs demands immense computational power, the raw material itself is virtually limitless and inexpensive to acquire.
Teaching physical machines how to interact with the real world presents a vastly different economic and logistical dilemma. According to Seo, every single piece of training data required for physical AI must be painstakingly captured in the physical world. This requires specialized hardware, operational facilities, testing environments, and human labor to execute tasks while capturing teleoperation or sensor metrics. Consequently, the marginal cost of gathering, cleaning, and labeling robotics training data scales rapidly, making physical AI significantly more expensive to develop than text-based software systems.
The "TSMC of Robotics": Business Model and Market Positioning
Config aims to resolve this bottleneck by positioning itself as the foundational data provider for the entire robotics industry. The startup draws a direct parallel to Taiwan Semiconductor Manufacturing Company (TSMC), which transformed the global semiconductor landscape by manufacturing microchips for competitors like Apple, Nvidia, and AMD without ever releasing a competing consumer device of its own. By remaining a neutral supplier, TSMC enabled the fabless semiconductor boom.
Config intends to replicate this model in the realm of robotics AI. As major global manufacturers, logistics providers, and defense contractors increasingly look to develop proprietary robot intelligence tailored to their unique operational environments, they face a severe shortage of specialized training data. Instead of relying entirely on turnkey third-party robot vendors, these enterprises want custom intelligence. Config’s platform is designed to supply the underlying data and translation infrastructure that makes this independent development possible.
Despite being founded only months ago, Config is already generating commercial revenue. Chief Operating Officer and co-founder Jack Bang noted that the startup’s initial customer base spans large-scale manufacturers, industrial system integrators, and enterprises operating within the agriculture and defense sectors. In this burgeoning ecosystem, Config competes alongside well-funded peers such as Physical Intelligence, Generalist AI, and Skild AI, all of which are racing to solve the foundational software and data challenges of general-purpose physical robotics.
A Unique Approach to Data Collection and Model Translation
To power its enterprise platform, Config has established a massive data-gathering operation spanning Seoul and Hanoi. In these controlled studio environments and field testing grounds, a workforce of nearly 300 employees records human workers performing complex physical tasks, dexterous manipulations, and operational workflows.
To date, Config has successfully accumulated over 100,000 hours of high-resolution human motion and interaction data. This repository dwarfs existing open-source alternatives; for comparison, AgiBot World—previously holding the title for the largest publicly accessible robotics dataset—features approximately 3,000 hours of training material, making Config’s dataset over 30 times larger.
However, the startup’s true competitive moat extends beyond the sheer volume of data it collects. Most contemporary robotics teams train their artificial intelligence models on raw human movement data and subsequently attempt to force those models to adapt to the mechanical constraints, degrees of freedom, and sensor configurations of a robotic chassis.
Config rejects this conventional pipeline. Seo explains that the company focuses on fundamentally transforming and translating the data before the model training phase begins, re-engineering the information so it natively aligns with the physics and kinematics of robotic movement.
Drawing an analogy to linguistic education, Seo compared naive data adaptation to attempting to teach fluent Korean to students using exclusively English-language instructional materials without proper translation structures. "The data must be converted, not the model. This conversion technology is Config’s core technical differentiator," Seo emphasized during an exclusive interview. By translating human demonstrations into robot-optimized data structures prior to ingestion, Config reduces training friction and enhances the operational reliability of the resulting foundation models.
Strategic Roadmap and Future Implications
With its newly secured $27 million seed capital injection, Config has outlined three core strategic priorities to guide its expansion through 2027:
- Data Infrastructure Scaling: The company intends to aggressively scale its operational facilities in Seoul and Vietnam with the goal of expanding its proprietary motion and interaction dataset toward 1 million collected hours.
- Enterprise Revenue Growth: Config aims to scale its enterprise platform business to achieve $10 million in Annual Recurring Revenue (ARR) by the conclusion of fiscal year 2027.
- Robot-as-a-Service (RaaS) Deployment: The startup plans to launch a cloud-based robotics-as-a-service product, allowing enterprise clients to execute Config’s proprietary robotics foundation models remotely via the cloud without requiring expensive, high-powered computing hardware onboard the physical robots themselves.
The massive financial backing from industrial heavyweights like Samsung, Hyundai, and LG highlights a broader macroeconomic realization across East Asian markets: leadership in the next generation of industrial automation will not be determined solely by who builds the best hardware, but by who controls the underlying data layer that brings machines to life. As manufacturing industries face persistent labor shortages, rising operational costs, and an urgent imperative for factory automation, foundational data startups like Config are poised to serve as the critical architects of the physical AI economy.
