延世大学与 LG Uplus 验证 Nvidia AI RAN，但仅为研究/生态进展，无订单或业绩增量

At an engineering lab on the Yonsei University campus in Seoul's Seodaemun district on the 14th, a small experimental robot moved about with a camera and a communications modem mounted on its exterior. The setup looked markedly different from typical robots, which house Wi-Fi modules internally. Video data collected by the robot travels through the modem to a base station built inside the lab and powered by Nvidia graphics processing units. The distance between the robot and the GPU server was only about three meters, with communications equipment placed in between to relay wireless signals without delay. The GPU runs inference the moment it receives the data, then sends the robot's next movement command back over the network. This is a physical AI testbed where AI computing is handled at the edge, on a base station network located at the site itself, rather than routed through a physically distant AI data center.
Building next-generation networks around AI is emerging as a central task for commercializing physical AI. In existing telecom networks, AI has been used in limited fashion to optimize the quality of base station communications. Going forward, industry watchers expect wireless infrastructure itself to evolve into an AI brain. "Physical AI means translating AI algorithms into physical action, and ultimately the key question is where the AI engine sits," said Kim Seong-ryoon, a professor in Yonsei University's School of Electrical and Electronic Engineering, in an interview at the lab. "For robots to evolve into true physical AI agents, implementing network technology may matter more than robotics itself."
That is why Nvidia has singled out telecommunications as a new market for GPUs. The company recently published a detailed account on its website of a GPU-based virtualized radio access network, or RAN, demonstration built by Kim's research team. The industry refers to RAN equipped with computing resources such as GPUs or central processing units as AI RAN. Nvidia said the research demonstrated practical mobility scenarios essential to commercializing physical AI systems including robots and drones. "Robots face sharp limits on power use because of battery constraints, so they need to stay connected to a nearby base station to move without delay," Kim said.
LG Uplus (032640.KS), through joint research with Yonsei University, verified technology that maintains connections even when networks switch in an AI RAN environment. The company linked its own 5G network with the Nvidia GPU-based AI RAN test network developed by Yonsei, then confirmed service continuity as physical AI devices such as robots and drones moved between the two networks. The research also verified whether devices could automatically select the appropriate network as wireless conditions changed and maintain stable connections. "For small and mid-sized robotics firms, building and testing on a dedicated network on their own is extremely difficult in terms of cost and manpower," an LG Uplus official said. "The AI RAN test environment created through cooperation with Yonsei University will serve as a useful testbed for domestic small and mid-sized firms to advance commercial product technology and expand overseas."
AI RAN is drawing growing attention partly because it can also coordinate among robots. Individual robots can carry AI chips and perform their own computing, but in environments where multiple robots operate together, a network that lets them work in concert is essential, according to the assessment. If 100 robots are deployed in a logistics warehouse, for example, they need to be connected collectively to an AI base station to reduce the chance of accidents and work efficiently.
GPU-based next-generation networks can be built using Aerial, the open-source AI RAN platform developed by Nvidia, which some expect will speed adoption. Kim built his lab's AI network by combining Aerial with other open-source software. "Open-source technology allows developers worldwide to participate, which makes it especially attractive in countries that lack domestic telecom equipment makers," he said. "Unlike Europe with Nokia and Ericsson, or China with Huawei, the United States has no global telecom equipment company, which is giving added momentum to AI RAN-based next-generation network technology."
Yonsei University will host the AI RAN Alliance general assembly in mid-November, drawing attention as a new hub for global AI telecom research. The meeting will bring together global companies and domestic industry, academia and research institutes to discuss the direction of AI RAN technology and cooperation on demonstrations. Tech companies including Nvidia and Samsung Electronics (005930.KS), along with telecom carriers and academics, will take part in efforts to develop AI-based 5G and 6G radio access network technology and build an ecosystem around it. "It is significant that we have brought a global event that had mostly been held abroad onto a domestic campus," Kim said. "It will serve as an opportunity for Korea to participate more actively in AI RAN technology development and discussions on international standards."
Still, some point out that robotics companies have been slow to join the next-generation network ecosystem. "The problem is that cooperation among the main participants in physical AI is fragmented by sector — robotics firms, AI foundation model developers, telecom carriers," Kim said. "Experts from various fields need to come together to work through demonstration cases and solve problems jointly so that commercialization of physical AI can be accelerated."