Domestic physical artificial intelligence (AI) companies identified "inter-company ecosystem cooperation" and "national-level infrastructure and standardization support" as key tasks to secure competitiveness in the global AI market. As competition for leadership in physical AI intensifies, particularly between the United States and China, there is an urgent need for private coalitions and tailored government policy support.
On September 3, the 2026 Global Good Growth Forum (GGGF), hosted by Aju Economy, featured a panel discussion among key representatives from domestic physical AI companies and startups on the topic of "The Success of Integrating Manufacturing and Physical AI." They unanimously agreed that comprehensive collaboration among data providers, simulation specialists, and AI model developers is essential to accelerate the commercialization of physical AI.
Instead of focusing on individual survival, they are seeking pathways through "data alliances" and the establishment of "global standards." For instance, BrainCommerce, which has infrastructure in over 900 manufacturing sites, provides on-site data that startups like Endlight, RealWorld, and CarbonSix process and simulate, creating a virtuous cycle in the private sector.
"We are creating a collaborative structure that provides real-world data collected from various manufacturing sites to robotics and AI companies, and then deploying the developed models back to the field for validation (Proof of Concept)," said Hwang Hee-seung, CEO of BrainCommerce, emphasizing the importance of a privately-led ecosystem.
Choi Soo-hwan, Executive Director of RealWorld, cited collaborations with companies like NVIDIA and Endlight, stating, "Global companies and domestic startups are jointly creating a global benchmark to evaluate action data, such as precise hand movements of robots. To be competitive in the global market, we need to establish a consortium-type cooperation system that allows excellent domestic technology companies to work together."
The panelists agreed that systematic government policy support is necessary to respond to rapidly growing overseas markets, including China.
"China is already quickly seizing the market by prioritizing standardization and modularization," Hwang noted, adding that he hopes for the establishment of a national-level "standardization policy" to enable the organic connection of outstanding domestic startup technologies. Kim Seon-tae, CTO of Endlight, also mentioned the need for "data standard guidelines" to systematically collect and utilize high-quality data, as data formats vary across sectors like manufacturing and logistics.
The panelists also focused on the essential role of data in physical AI. Kim Seon-tae emphasized the importance of converting static data, such as CAD files held by manufacturers, into a format that robots can learn from, incorporating material properties and joint information, and transitioning to a standard format for simulation. Although vast amounts of data already exist in manufacturing sites, utilizing it directly for physical AI training remains challenging.
Choi Soo-hwan pointed out that various regulations hinder the data acquisition process. He stated, "When collecting human work data, privacy issues arise, and in key industrial sites, we must also consider restrictions on industrial technology protection," highlighting the need for institutional improvements.
There were also opinions that simply accumulating large amounts of data is insufficient for advancing physical AI. Hwang Hee-seung remarked, "Robots must go beyond merely mimicking human actions; they need to understand the process itself and why a task needs to be performed." He noted that South Korea could leverage its strengths in precision manufacturing process data as a competitive advantage.
Additionally, building infrastructure to effectively utilize data for AI training was identified as a challenge. Seo Hyung-joo, CTO of CarbonSix, stated, "Since video and manufacturing process data are significantly larger in size than language data, developing physical AI foundation models requires substantial costs, and government support for high-performance computing infrastructure is essential."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.