We recently attended the Telco AI Forum hosted by RCR Wireless. It was an excellent event, touching on many AI Telco topics. Some of the key takeaways related to AI RAN include:
- Near-term focus on AI-for-RAN
- Monetization is part of the long-term vision
- Transition will be gradual
- AI from day 1 with 6G
AI in the RAN has been around for decades. But the focus is shifting. Operators now focus on AI-for-RAN, meaning opex, performance, and efficiency improvements. In addition to the role AI can play in the automation journey and with energy savings, operators are increasingly looking at how to squeeze more out of the existing spectrum by using AI for channel estimation, MIMO, CA, and beamforming, among other things. Michael Irizarry, CTO of US Cellular, mentioned they are seeing 10% to 15% of potential efficiency gains in the network.
Monetization is part of the long-term vision. Not surprisingly and consistent with what we discussed in a recent AI RAN blog post, “AI RAN Should We be Excited“, there is strong consensus that AI RAN can improve the user experience, enhance performance, spur efficiency gains, reduce power consumption, and play a critical role in the broader automation journey. However, there is greater skepticism about AI’s ability to turn the RAN sites into profit engines. While many still buy into this concept of improving the overall site utilization by using the workloads for both RAN and AI, the key change over the past year is simply the realization that the GPU business case needs to be justified in the “RAN only” scenario. AI-and-RAN or multi-tenant RAN can improve the ROI, but it should not be a requirement to justify the investment.
The transition from RAN to AI RAN will be an evolution steered by the business case. Rob Hughes, head of wireless marketing at Fujitsu, envisions an approach similar to that of 5G SA. In some cases, the operators start small and go after specific pockets before expanding more broadly. Michael Irizarry with US Cellular believes rip & replace is unlikely. “The competitive nature of this Industry puts a lot of pressure on the margins, and the ROI will guide the investment cycle”. As the natural cycle ends, operators can assess their best options. Guy Turgeon, Senior Principal Industry Specialist at Red Hat, envisions that the operators that have already started the journey by moving to a cloud native architecture leveraging COTS HW might be in a better position from a readiness and skillset perspective. At the same time, there is no question that the asymmetric starting point between appliance-based RAN and COTS HW will impact the transition, and custom HW will likely comprise a significant share of the 5G AI-for-RAN market. The AI-RAN Alliance discussed the hybrid CPU/GPU roadmap.
AI RAN will be a reality from the start with 6G. While 6G does not change the existing site grid dynamics and the splits between C-RAN and D-RAN, the expectation is that AI will be a focus from the start with 6G. Since the base case scenario is that the “anchor” band for 6G will be in the upper 6 GHz+ range and the business case is hinging on carriers’ ability to leverage the existing macro grid, the expectation is that AI is needed for RAN optimization and in the MAC layer for scheduling, beam management, and MIMO optimization. Ofir Zemer, Vice President, Product Management at Qualcomm and responsible for its RAN Automation suite, believes a powerful agentic layer is essential for Level 3+ automation to manage the increased complexity in the networks (According to the TM Forum, the average operator is currently at Autonomous Networks Level 2, Level 3+ is expected by the time 6G comes around).
For more info, please visit the RCR Wireless Telco AI Forum site to watch the webinar on-demand.
