The radio access network (RAN) is all the RF part of a mobile network: towers, base stations, the signals between our phones and the towers, phone-to-satellite comms (non-terrestrial network or NTN).

AI-RAN uses AI/ML for adaptive behaviors and optimizations in all these links.

For example, fine-grained RF and modulation details, called the channel state information (CSI), is constantly being exchanged between a phone and a base station. The volume of information creates transmission latencies. Using autoencoder models, this information can be semantically compressed to reduce its volume and decoded with high fidelity on the other side.

That's just one example. In the upcoming 6G, RAN will be "AI-native", using AI/ML everywhere. The standards may require AI accelerator chips in most base stations, NTN satellites, phones, and other elements.