Data Emerges as Battleground for Physical AI as Korean Tech Firms Join Race

Mobile World Congress 2026, the world's largest mobile communications trade show, proved difficult to define as merely a "mobile" event. The exhibition, held March 2-5 in Barcelona, Spain, drew approximately 2,900 companies and organizations, attracting more than 100,000 visitors from over 100 countries. The theme was "The IQ Era," but the exhibition halls were dominated not by smartphones but by AI systems that move and make decisions in the physical world. GSMA organized a dedicated session titled "Physical AI: From Virtual to Real," with robots, autonomous systems, and industrial AI taking center stage throughout the venue.

Chinese companies mounted a particularly aggressive presence. Honor, ZTE, China Mobile, and robotics firms showcased physical AI and robot technologies prominently at their booths. China Mobile unveiled a "robot restaurant" where ordering, cooking, and serving processes are coordinated between robots. Industry observers interpreted this as evidence that generative AI is moving beyond virtual assistants on screens into real-world service environments. Samjong KPMG identified physical AI as one of MWC's key trends, noting that implementations where AI perceives, judges, and acts in physical environments have now taken center stage.

Physical AI Growth Accelerates; Data Becomes Critical Factor

Physical AI refers to technology that combines AI with physical devices such as robots, autonomous vehicles, and smart devices, enabling them to independently perceive, judge, and act in the real world. While generative AI represented by ChatGPT is "speaking AI" that handles text and images, physical AI is "moving AI" that navigates factory floors, handles logistics, and drives on roads. According to market research firm Statista, the global physical AI market is projected to grow from $22.5 billion in 2025 to $64.3 billion in 2030, representing 23% annual growth. Goldman Sachs forecasts the global humanoid robot market will reach $38 billion by 2035—six times its projection of $6 billion from just one year ago.

The expanding physical AI market signals that this technology is rapidly approaching everyday life. However, many global companies agree that full commercialization still requires more time. Current technology remains at the stage where robots demonstrate natural walking or repeat standardized tasks. Responding flexibly like humans in unpredictable real-world environments requires vast learning data from diverse situations. Ultimately, data is the key that determines physical AI's completeness.

This explains why Korean IT software companies are increasingly eyeing physical AI. Recently, Kakao Mobility, Krafton, Naver, and KT have successively announced physical AI market entries or made related investments. Their moves are notable because these are software and platform companies rather than manufacturing conglomerates like Samsung, Hyundai Motor, or LG.

Krafton signed a memorandum of understanding this month with Hanwha Aerospace to establish a joint venture. The company also plans to participate as an investor in a $1 billion global AI, robotics, and defense fund being organized by Hanwha Asset Management, expanding cooperation across the ecosystem. Kakao Mobility has officially announced its transformation into a physical AI technology company and is pursuing autonomous driving and robot delivery demonstrations. Naver's startup investment arm D2SF has officially added physical AI startups to its new investment targets. KT is launching a robot service platform business targeting public services, logistics, and manufacturing sites, centered on K-RaaS, which it showcased at MWC.

Government Also Building Data Pipeline to Foster Ecosystem

These companies share a common weapon: data. For physical AI to function properly in the field, it must continuously collect and learn from vast physical-world data generated by robots and sensors. Field data—including human movement patterns, vehicle traffic, indoor operations, logistics flows, and equipment operations—is far more difficult to obtain than text or images. However, once accumulated, it creates high barriers to entry that competitors cannot easily overcome. Kakao Mobility's mobility data and real-time road information accumulated over five years, Krafton's simulation and digital twin technology built through game engines, Naver's startup investment network, and KT's nationwide network and control capabilities are each calculated to become differentiating assets in this market. Kakao Mobility CEO Ryu Geung-sun said, "Physical AI technology will create a virtuous cycle that improves operational efficiency and innovates service quality when combined with existing services."

The government also views physical AI as a next-generation strategic industry and is developing concrete support systems. Rather than simply supporting robot or humanoid development, the plan aims to create a full-cycle ecosystem that secures data accumulated at actual manufacturing sites and converts it into learnable formats. The government has proposed three initiatives: building data pipelines, securing core physical AI technologies, and fostering an organic ecosystem. Among these, the data pipeline is considered the most urgent priority. Physical AI requires physical-world data far more complex than text or images because it must operate in actual environments such as factories, logistics centers, shipyards, and manufacturing robots. The government considers it essential to convert tacit knowledge—the experience and expertise that skilled workers and field practitioners have accumulated over time—into explicit data.

To achieve this, the government plans to establish a nationwide data collection system. Through proof-of-concept projects underway in North Jeolla and South Gyeongsang provinces, it will pursue field data collection, acquisition of human behavior processes and factory operation data, equipment input-output data collection, and systematic accumulation of integrated data. The accumulated data is expected to serve as the foundation for AI models for robots, as well as core technologies such as "world models" that learn various variables and environments in the physical world and "Vision-Language-Action models (VLA)" that connect perception, judgment, and action.