It’s looking like commercial deployment of sixth-generation (6G) technology will land in the 2029-2030 timeframe. But, meanwhile, the telecom industry is beginning to broaden its focus from early R&D to technology definition and a pre-investment phase.

One influential global technology coordination body, the United Nations’ International Telecommunication Union (ITU), has mapped out its vision for what 6G is expected to look like in 2030. For the ITU, 6G isn't simply a matter of “faster 5G.” Rather, the goal is to transform cellular networks into ubiquitous connectivity platforms that integrate communications, sensing, and artificial intelligence.

For that to happen, some very advanced technologies must coalesce, and none is more critical than AI-RAN. A traditional radio-access network’s architecture looks like:

RF/antenna > radio unit > distributed unit > centralized unit > core network

In this scenario, the RAN’s computing resources are primarily dedicated to executing telecom functions. But an AI-RAN architecture goes more like this:

RF/antenna > accelerated, software-defined RAN + AI compute > network/core/edge

It results in an underlying compute infrastructure that can potentially run RAN baseband processing, AI/ML algorithms for optimizing the RAN, edge-AI applications, and other enterprise workloads.

AI-RAN represents a lot of things, including serving as a testbed for 6G. Researchers and network operators want to experiment now with AI-native networking before 6G exists. Well, they can already deploy AI-RAN capabilities on 5G or 5G Advanced infrastructure. Doing so lets them learn what works, develop models, and build the necessary compute infrastructure. All of that could be carried forward into 6G down the road.

AI-RAN can make existing spectrum more efficient, enabling operators to squeeze more performance from existing infrastructure. According to Nokia, its current AI-RAN platform has demonstrated more than 20% improvement in spectral efficiency. The company expects more software-driven improvements over the next two years.

Meanwhile, AI-RAN has implications for the demands on RF hardware. AI optimization will create the need for more sophisticated beamforming, more antenna elements and RF channels, and more transceivers and RF front ends. There will be greater amounts of data moving between the RF and compute chains and increased demand for high-speed analog-to-digital and digital-to-analog conversions and interconnects.

The heavy AI-compute demands that 6G will impose mean shifts in the competitive landscape. New players from the compute ecosystem, most notably NVIDIA, have become key elements going forward.

One might even ask whether AI-RAN is primarily a telecom-equipment business or does it become part of the accelerated-computing ecosystem? That’s still an open question. Operators won’t want the RAN side of the equation dependent on one or another dominant compute platform.

Most likely, the 6G rollout, much like the 5G rollout, will happen in stages that begin with AI optimization of existing networks. After normative 6G specifications are settled by 3GPP, the industry can move networks forward into AI-native 6G and then blossom into the distributed intelligent infrastructure that the ITU envisions.