The training stage for physical artificial intelligence (AI) is expanding from the real world to virtual space. Because it is hard to repeatedly recreate unexpected situations from Autonomous Driving, robots, and manufacturing sites in reality, Digital Twin technology that reproduces real environments in virtual space to build a "practice room" is gaining traction. Competition among domestic and overseas startups to dominate the market is also intensifying.
On the 13th, market research firm MarketsandMarkets said the global Digital Twin market is expected to grow at an average annual rate of 47.9%, from $21.1 billion (about 29.5 trillion won) in 2025 to $149.8 billion (about 209.7 trillion won) in 2030. As the use of Digital Twin expands across industries such as manufacturing, automobiles, and logistics, startups developing related technologies are also finding business opportunities.
A Digital Twin is a technology that implements cities and factories in virtual space with high-precision 3D data. It allows repeated testing of various situations, such as severe weather, without exposing vehicles or robots to danger. The completeness of a Digital Twin depends on "how accurately it reproduces reality." The accuracy of "calibration," which aligns information collected by sensors such as cameras, LiDAR, and GPS without mismatch, determines reliability and the performance of physical AI.
As the learning environment for physical AI expands from the real world to virtual space, corporations that handle everything from real-world data collection to virtual implementation are drawing attention. In Korea, the Digital Twin specialist "Mobiltech" is building a data pipeline from building measured datasets to calibration and simulation consolidation.
The flagship solution, "Replica City," uses LiDAR-based high-precision 3D spatial data to implement real cities and industrial sites in virtual space. With its technological prowess recognized, it continues collaborations with Nvidia and Hyundai Motor(005380). It is recently expanding its scope to the robot learning field.
Overseas, competition to dominate the market is also intensifying. U.S.-based Parallel Domain builds Digital Twins with sensor data to provide synthetic data for autonomous vehicles and robots. After collaborating with Google, Continental, and Toyota Research Institute, it is expanding its business by implementing the University of Michigan's Autonomous Driving test site "Mcity" as a Digital Twin.
Israel's Cognata also implements real cities and off-road terrain as Digital Twins to support the training and verification of autonomous vehicles. Israel's Ministry of National Defense and the military adopted Cognata's simulation platform to train and verify algorithms for autonomous military vehicles to be operated in rough terrain.
The growth and competitive edge of startups also depend on their "ability to reproduce reality." If calibration errors or sensor data distortions accumulate, a "Sim-to-Real Gap" can occur, where AI that performs well in virtual environments fails to operate properly in the field. It also costs a significant expense to collect and update high-precision spatial data. On top of that, the fact that global giants such as Nvidia and Siemens are leading the simulation ecosystem is another hurdle to overcome.
An industry official said, "Digital Twin will establish itself as a core foundational technology that determines the performance and safety of physical AI across various industries, including Autonomous Driving, humanoids, manufacturing, logistics, Smart City, and defense," adding, "Another key is how quickly startups can get on board with standards set by global giants and expand partnerships."