What is it about?

Our work introduces AoRA, a framework that enables artificial intelligence (AI) on RAN infrastructure, rather than relying on external servers. Today, most networks use cloud or edge computing to deliver AI services to users, but these add network delay because the processing happens outside RAN stack. AoRA overcomes this by using the spare computing resources in GPU-based RAN platforms to serve AI models. This reduces network latency by up to 30% compared to edge computing and 70% compared to cloud setups, enabling low-latency AI workloads delivery for time-critical applications like smart cities and connected vehicles.

Featured Image

Why is it important?

AoRA is the first practical realization of the AI-on-RAN vision, making it both timely and unique. While edge computing has become a key enabler of low-latency services, it still introduces extra delays as data must travel outside the RAN environment. By shifting AI execution directly into the RAN, AoRA removes this bottleneck and demonstrates that base stations themselves can act as intelligent service providers. This advance is important because it shows how existing 5G infrastructure can be enhanced without costly new deployments, paving the way for faster, more responsive networks. The ability to cut latency so significantly opens new opportunities for time-critical applications such as autonomous vehicles, real-time safety monitoring, and immersive digital experiences.

Perspectives

From my perspective, this work marks an important step in demonstrating that the RAN can act as AI service provider by hosting and serving AI models directly to external users. AoRA shows the feasibility of this approach and lays the groundwork for rethinking the role of the RAN beyond connectivity. I see this work as the start of a broader effort to explore how such capabilities can be scaled and integrated, and I am excited about the opportunities it opens for future research and development.

Siyavushkhon Kholmatov
Korea Advanced Institute of Science and Technology

Read the Original

This page is a summary of: AoRA: AI-on-RAN for Backhaul-free Edge Inference, August 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3718958.3750517.
You can read the full text:

Read

Contributors

The following have contributed to this page