Google, Nvidia Target Korea's Manufacturing Data for Physical AI Race
The battleground for artificial intelligence supremacy is shifting from language-based virtual spaces to the physical world. Physical AI—systems that understand real-world physics and apply them to robotics and autonomous driving—has emerged as the next frontier beyond large language models (LLMs) that generate responses from text learning. In this transition, Korea's manufacturing data is becoming a critical asset that could determine winners and losers, surpassing even semiconductors in strategic importance.
Global tech giants including Nvidia and Google are intensifying efforts to secure physical data as they advance their AI capabilities, according to industry sources on Sunday. Existing generative AI systems like ChatGPT and Gemini, trained on text and video data, excel at language tasks but have clear limitations in fully understanding real-world physics such as gravity and friction.
AI has already learned most of the data available online. Like a student advancing to the next grade, AI must now learn about the physical world, with subjects becoming more numerous and understanding deepening. Some recently unveiled video-generation AI systems produce results that violate physics laws precisely because their training data was limited to text and two-dimensional video, analysts say.
To address this gap, big tech companies are using digital twin technology that replicates reality in virtual spaces. However, simulations alone cannot capture all real-world variables. AI needs data accumulated from actual production sites—subtle vibrations during factory operations, temperature changes, and equipment tolerances—to accurately perceive and make decisions about the physical world.
This is why Google, Meta, and other tech giants are focusing on Korea. While these companies have amassed vast web data, they lack data from actual industrial sites. Korea, by contrast, possesses a complete portfolio of advanced manufacturing: semiconductors (Samsung Electronics (005930.KS), SK hynix (000660.KS)), automobiles (Hyundai Motor, Kia), shipbuilding (HD Hyundai, Hanwha Ocean, Samsung Heavy Industries), and batteries (LG Energy Solution, Samsung SDI). Korea is an optimal training ground for AI to learn physical intelligence.
Nvidia CEO Jensen Huang's recent efforts to strengthen cooperation with Korean business leaders—Samsung Electronics Executive Chairman Jay Y. Lee, SK Group Chairman Tae-won Chey, and Hyundai Motor Group Executive Chairman Euisun Chung—reflect this strategy. Industry observers see these moves as going beyond securing high-bandwidth memory (HBM) chips, positioning them as a data acquisition strategy to enhance Nvidia's digital twin platform Omniverse. Nvidia's AI competitiveness will be complete when combined with Hyundai's robotic production process data and precision manufacturing data from Samsung Electronics and SK hynix, they say.
In the coming physical AI era, Korean companies are expected to transcend their role as mere members of the semiconductor ecosystem. Just as Tesla dominated the autonomous driving market by monopolizing driving data, Korean companies holding manufacturing data could gain leverage in negotiations with big tech, analysts say.
The semiconductor market environment is also favorable to Korea. As Google's Tensor Processing Units (TPU) and Amazon's custom ASIC chips called Trainium emerge as viable alternatives to reduce Nvidia dependence, demand for Korean memory semiconductors and foundry services will structurally increase. Combined with spillover effects from TSMC's saturated production capacity, new opportunities are opening for Samsung Electronics' foundry business, industry observers say.
*Gap World is a column by reporter Seo Jong-gap that explores the gaps in news during the era of technology supremacy competition. Check Gap World for insights and outlooks on cutting-edge technology and semiconductor issues.*