KDDI has launched a joint initiative with partners, including Nvidia, to develop a high-fidelity radio access network (RAN) digital twin aimed at supporting AI-driven optimization and autonomous network operations in the 6G era.
The project is designed to create a virtual replica of real-world RAN conditions, allowing operators and vendors to train, test, and validate AI models without disrupting live commercial networks. KDDI said the platform will support large-scale simulation of changing radio environments, traffic patterns, and network scenarios, giving AI systems a safer and faster environment for development than direct experimentation on production infrastructure.
Under the joint initiative, the companies, which include Keysight Technologies and Samsung, will combine high-fidelity radio propagation and wireless network behavior simulation with scalable computing infrastructure. Nvidia will provide digital twin infrastructure through its Aerial Omniverse Digital Twin and accelerated computing platforms, while Keysight will contribute user equipment emulation technology for validating the RAN digital twin. In addition, Samsung Research America will provide virtualized RAN (vRAN) technology.
KDDI itself will contribute commercial network data and lead commercial trials, while KDDI Research will define technical requirements, develop propagation prediction technologies, and create use cases for the technology. So far, these include a training and evaluation environment for AI-based network optimization, allowing AI algorithms to run through many simulated network conditions in parallel. Another use case is a field-trial platform for new AI-based RAN features.
One target area is what KDDI dubbed the “AI air interface,” where AI is used to assist or predict RAN processes to improve communication quality and reduce power consumption.
KDDI Research added one of its AI propagation models that uses an approach based on a convolutional neural network (CNN) with a residual network (ResNet) structure to capture spatial features from map information. The model estimates propagation paths affected by reflection and diffraction, then uses that path information to estimate path loss. According to researchers, this improved estimation error in unknown regions by about 15 percent compared with a conventional AI model that did not use path estimation.
The companies plan to deliver a prototype by the end of March 2028, while aiming to expand the platform by the end of 2030 to support broader use cases and validate its performance against KDDI’s commercial network.
KDDI plans to use its AI data centers as the computing engine for operating the RAN digital twin at scale. According to Nvidia, the environment will allow multiple autonomous AI agents to safely simulate and validate RAN scenarios, including area optimization strategies, future radio conditions, traffic shifts, and new AI air-interface functions.
Commenting on the venture, Satoshi Konishi, president and CEO at KDDI Research, said: "With the RAN Digital Twin, we can train and validate AI not only against real-world scenarios but even against future ones yet to unfold, accelerating innovation and dramatically shortening time to market."
"6G will be born in simulation and our collaboration ... will pioneer this future. As networks become increasingly autonomous, more complex AI algorithms will be the critical engines driving them. To develop these algorithms safely and at massive scale, the Nvidia Aerial Omniverse Digital Twin provides a high-fidelity 'AI-RAN gym' where operators can rigorously train, simulate, and validate new models before ever deploying them to a live network," added Soma Velayutham, VP of telecoms and AI, Nvidia.
Recent research from Dell'Oro Group projected 6G global wireless capital expenditure to reach $500 billion by 2034, with cumulative 6G RAN revenue expected to top $100bn.
Dell Oro’s report suggested that while 6G wouldn’t expand the overall RAN market, a one percent compound annual growth rate (CAGR) is to be expected between 2030 and 2034, with cumulative 6G RAN investments accounting for almost half of total RAN capex during that period.