AI and the network traffic challenge

stein@opensky

Happy New Year everyone!

This is the time of year when every self-respecting consultancy company makes predictions for the year to come (i.e. 2026 just now at the start gate). I will not try to do the same – but rather than writing long articles about specific themes (which I have done too often recently), I will issue a set of short commentaries. This one is about 5G, 6G, AI and network traffic.

I have written various articles on 5G and 6G and whether we need 6G or whether 5G is the last G (see, for example, here and here) – and I am not the only one. Many competent people have had similar concerns – but the main telco vendors and their suppliers again (like Ericsson and Qualcomm here), which obviously need something to sell, are working hard betting on 6G. By the way, also Ericsson’s CEO is now also stating that, although Ericssson obviously is betting on 6G, “’evolutionary' 6G puts an end to the G-cycle”.

To cut that discussion short, 6G will come, partly because so much money and effort are brought into it from a standardization and R&D perspective and partly as a technical necessity for cost savings and network optimization. Political support is also strong. The bigger question, however, is: Will there be any significant additional revenues for operators – or will anyone make any money from it? As an example, see my recent commentary on sensing. Some, in addition to the standards community, may think sensing is a potentially great idea (like e.g. Sebastian Barros here, saying “Telco Network-as-a-Sensor Is a Huge Opportunity” – although with some reservations). My main concerns remain, however, on whether any operator will want to build a large sensing network within 6G and if anyone will make money from it. Governments may like it, however.

On the technical side, one of the predictions that seem to be generally viable is that AI will be all around us. This will, however, imply a significant traffic increase which operators are not yet positioned to monetize (from a business model as well as regulatory perspective). AI will generate huge amounts of background traffic, not only across the RAN but, even more importantly, between data centres across the globe. The recent Nokia global network traffic report clearly identifies the challenge. “AI systems are generating traffic autonomously, with machine-to-machine exchanges flowing across metro and core networks and between large compute locations.” It also states that “total global network traffic is projected to grow 3-7x through 2034”.  In addition, traffic patterns are becoming increasingly symmetrical (uplink vs downlink) – which also requires a rethink.

The situation reminds me of a similar discussion we had within a large operator group around 20 years ago – on “the threat of VOIP” and that the operators had unknowingly and consistently made the internet QoS so good that VOIP (basically free of charge) had a comparable and competitive quality to legacy operator voice services.

Without getting into a discussion on two-sided business models or net neutrality, the growth of AI and autonomous agents will create yet another challenge for operators, be it on a 5G or a 6G network. Handling AI network traffic will be a challenge (and of course an opportunity) for operators in the years to come, technically and commercially.