Jensen Huang, CEO of Nvidia, previewed the emergence of physical artificial intelligence (AI) at CES, the world's largest electronics and IT trade show, last year. At Nvidia's annual developer conference GTC this year, he declared, "Physical AI has arrived," officially signaling that AI's stage has moved into the real world. This shows that the center of AI competition is rapidly shifting from text and image generation to physical AI — systems that perceive, judge, and directly act in real-world environments such as factories, logistics networks, hospitals, and everyday spaces.
Physical AI differs fundamentally from conventional industrial automation in its operating principles. Traditional automation moved only along pre-programmed, rule-based paths and had no choice but to halt when exceptions arose. Physical AI, by contrast, perceives unpredictable real-world variables in real time, makes autonomous judgments, and adjusts its actions accordingly, making industrial sites more flexible.
As the way these systems operate changes, the "unit of competition" is also shifting. In the past, the performance of individual machines mattered most. Now what matters more is "system completeness" — how organically the chain of semiconductors, software, robot hardware, and services works together. Physical AI competition goes beyond the superiority of any single technology; it is a contest among prepared ecosystems.
This shift presents an opportunity for Korea as well. Unlike the language model market dominated by Big Tech, the physical AI field has yet to produce a definitive leader. As AI that handles the physical world is just entering a full-scale growth phase, now is the time to concentrate capabilities on leading the race.
Moreover, Korea already possesses the core assets needed for physical AI competition. Industrial sites spanning manufacturing, logistics, shipbuilding, automotive, and semiconductors provide an optimal environment for accumulating and validating the real-world data required for physical AI training. Korea's AI chip capabilities — the heart of physical AI — are a decisive strength. If proprietary AI model technology is combined with these assets, Korea can compete as a frontrunner, not a follower.
To translate this opportunity into real results, three tasks must be pursued simultaneously. First is securing the foundation models that serve as the brain of physical AI. Physical AI requires Vision-Language-Action (VLA) models that integrate learning across the processes of seeing objects, understanding commands, and executing actions. Second, developing "world models" that replicate the real world in virtual spaces for pre-training and verification is critical. These models generate diverse synthetic data and test numerous exception scenarios, thereby enhancing physical AI's real-world responsiveness and stability. Third, AI semiconductor capabilities — the heart of the system — must be elevated further. Semiconductors for physical AI must make real-time decisions and exercise control within limited power and space constraints. What matters, therefore, is not merely increasing chip speed but realizing semiconductors optimized for low-power, real-time operation.
Ultimately, the outcome of physical AI — a comprehensive art — will be determined not by individual component technologies but by a nation's ability to weave them into an industrial ecosystem. Only when field data, R&D, semiconductors, software, industrial demonstration, and institutional frameworks mesh precisely can a country secure leadership over future industries. Globally, few nations simultaneously possess manufacturing prowess, field data, and AI semiconductor competitiveness the way Korea does. Now is the time to strategically consolidate these strengths and make the leap to become a leading nation in physical AI.