Ericsson sees AI RAN as a major shift for mobile networks. The company argues that networks must become smarter internally. They must also support demanding AI applications at the same time.
At RCR Wireless News’ Intelligent RAN Forum 2026, Anders Söderlund outlined this dual challenge. He described the model as “AI for networks” and “networks for AI.” That distinction matters for operators planning future network investments.
The first part focuses on network performance. AI can help operators improve coverage, efficiency, and traffic management. This matters because most networks already operate under tight power limits. Many cell sites have little space for extra hardware.
“The way we do it at Ericsson is that we do hardware and software co-design to build really energy efficient systems that also support AI inference then,” he said.
In simple terms, AI inference means running an AI model after training. It allows equipment to make fast decisions in real time. Ericsson says its RAN compute platforms can support these functions efficiently.
The company claims measurable gains from AI-native features in 5G networks. These include higher spectral efficiency and better positioning accuracy. Operators could also improve coverage planning and serve more high-traffic users.
However, these improvements require careful deployment. AI workloads need processing power, energy, and optimized silicon. Operators must balance performance gains against site costs and operational complexity.
The second part addresses future AI applications. Smart glasses, humanoid robots, and physical AI may change traffic patterns. These applications will not only download data. They will also upload video, sensor data, and context information.
“Applications such as physical AI or smart glasses will demand new capabilities from the network, such as a stronger uplink, for example,” Söderlund said.
This point challenges traditional mobile planning. Many networks prioritize downlink capacity for streaming and browsing. AI devices may need more reliable uplink performance instead.
Ericsson has measured uplink performance across more than 100 global networks. The company believes AI applications may need at least 5 Mbps upstream. That requirement could push operators toward stronger FDD spectrum use.
“For that reason, we see that going forward, the uplink needs to be strengthened,” he said.
Low latency will also become critical. Latency means the delay before data receives a response. Smart glasses may feel unusable if responses arrive too late. Network scheduling may need to prioritize time-sensitive AI traffic.
Ericsson also expects AI to spread across the network. Some models will run on devices. Others will run at cell sites, edge locations, or central data centers. Each location serves a different need.
Small models protect smartphone battery life. Cell-site models must respond within very short timeframes. Data centers can handle larger models when seconds are acceptable.
Looking ahead, this distributed approach could shape 6G design. It may also offer new revenue options for service providers. Yet operators must upgrade wisely, especially where uplink demand grows fastest.
