SoftBank and Ericsson have demonstrated significant improvements in 5G network performance by deploying artificial intelligence directly inside the radio access network (RAN), achieving up to approximately 50 percent higher downlink user throughput and 25 percent better spectral efficiency in a commercial network trial in Japan.
SoftBank and Ericsson Japan conducted what they describe as Japan’s first validation of Ericsson’s AI-native Scheduler for Link Adaptation on a 5G commercial network. The technology, part of Ericsson AI in RAN software, makes real-time AI decisions within baseband equipment to optimize radio resources according to changing network conditions.
The results provide evidence that telecom operators can use AI-native RAN software to increase the performance and capacity of existing 5G infrastructure without relying solely on additional spectrum or network hardware.
AI-Native RAN Delivers Up to 50% Higher 5G Downlink Throughput
SoftBank and Ericsson evaluated spectral efficiency, user throughput, robustness and stability across SoftBank’s commercial 5G network.
The trial recorded improvements of up to approximately 25 percent in spectral efficiency compared with conventional technology. Downlink user throughput increased by up to approximately 50 percent.
Importantly, the improvements were not limited to the best-performing test locations. Across all evaluated locations, spectral efficiency and downlink user throughput both increased by approximately 10 percent on average.
The results indicate that AI-native RAN could help mobile operators carry more traffic using their existing spectrum while simultaneously improving customer experience.
Ericsson AI Scheduler Optimizes 5G Networks in Real Time
Traditional Link Adaptation optimization generally depends on rule-based algorithms developed through offline analysis and static network parameter configurations.
Ericsson’s AI-native Scheduler for Link Adaptation takes a different approach. The software operates on baseband equipment and uses AI inference in real time to respond to complex and continuously changing radio conditions.
The AI model can optimize link parameters for subscribers at cell edges and locations experiencing high interference. It can also stabilize throughput in congested areas and accommodate additional mobile traffic within existing frequency bands by increasing spectral efficiency.
Ericsson says its telco-grade AI models support ultra-low-latency inference at the microsecond level, enabling AI decisions to be integrated directly into time-sensitive RAN operations.
The approach could become increasingly important as generative AI assistants, autonomous AI agents and immersive applications generate larger and more variable mobile data traffic.
SoftBank Uses Commercial Network Data to Train AI-Native RAN
SoftBank and Ericsson divided responsibilities across network design, AI training and commercial-network implementation.
SoftBank defined evaluation areas according to network traffic characteristics. Ericsson trained its AI-native Scheduler for Link Adaptation model using actual network data and implemented and tuned the software to predict channel capacity and determine the appropriate downlink transmission rate in real time.
The companies also collaborated on parameter design and performance evaluation.
The trial demonstrated that AI-native optimization can operate using commercially deployed network infrastructure, potentially allowing operators to enhance their installed 5G networks through software.
Ericsson AI in RAN Targets 5G Performance and Energy Efficiency
Ericsson AI in RAN is a software subscription designed to introduce telco-grade AI models directly into basebands and radios. Ericsson says the platform can improve network performance, efficiency and energy consumption while helping communications service providers transition toward AI-native RAN without requiring additional hardware.
The software combines real-time AI models with continuous learning based on scalable network data. It also supports agentic AI capabilities aimed at increasing RAN automation and improving network operations.
Ericsson AI in RAN works with Ericsson 5G Advanced across purpose-built RAN and Cloud RAN architectures.
AI inference can run using Ericsson Silicon in radios and the latest generation of RAN Compute, while Cloud RAN software portability allows AI functionality to operate across partner computing platforms.
Ericsson AI in RAN Shows Gains Across More Than 15 Deployments and Trials
Ericsson says its AI in RAN technology has already been proven in more than 15 deployments and trials worldwide.
Across those deployments, the technology has delivered up to 20 percent higher downlink throughput and up to 10 percent better spectral efficiency.
Ericsson also reports that AI in RAN can support up to 2x more high-traffic users, achieve 90–95 percent coverage prediction accuracy and provide up to 5x greater user-positioning precision.
The SoftBank commercial-network validation produced even stronger peak results for the AI-native Scheduler for Link Adaptation, with downlink throughput improving by as much as approximately 50 percent and spectral efficiency increasing by up to approximately 25 percent.
Ericsson Rolls Out AI-Native 5G RAN Features in 2026
The first Ericsson AI in RAN software features became available in Q2 2026, with additional enhancements scheduled for later in the year.
The portfolio includes the AI-native Scheduler for Link Adaptation, AI-powered Macro Positioning, AI-managed Beamforming, AI-powered Multi-layer Coordination, Performance Management Event Schema Files and Augmented Observability for AI in RAN.
The technology is designed to complement Ericsson’s AI-ready radio portfolio as the vendor expands AI capabilities across the mobile network.
For mobile operators, the ability to deploy AI through software across the existing installed base could provide a faster path to AI-native networks while improving the economics of spectrum and infrastructure utilization.
SoftBank and Ericsson Target 5G-Advanced and 6G Networks
SoftBank views AI-native RAN as part of its broader “Activate AI for Society” growth strategy. The operator expects mobile networks to face higher traffic volumes and greater traffic variability as AI becomes embedded in consumer and enterprise applications.
Teruyuki Oya, Senior Vice President and CNO of SoftBank, said the commercial-network results represent an important milestone in advancing the RAN through AI and developing networks capable of supporting the future AI era.
Mårten Lerner, Head of Networks Strategy & Product Management at Ericsson, said the trial demonstrates how telco-grade AI running on Ericsson Silicon in the existing installed base can improve 5G network performance and efficiency in real time.
Jawad Manssour, President and Representative Director of Ericsson Japan, said the trial showed that directly applying AI to the RAN can generate measurable gains using commercially available hardware and software.
The collaboration will now focus on further improving AI-native RAN performance for SoftBank’s operating environment. The companies see the technology as a foundation for increasingly autonomous networks and a pathway from today’s commercial 5G infrastructure toward 5G-Advanced and eventually 6G.
BABURAJAN KIZHAKEDATH
