Some quick takes from FutureNet World in London today (April 21), where the telecoms industry said it is still taking incremental steps with AI, rather than giant leaps. Which is not to say the discussion was flat; it felt like an industry getting to grips with the tech in a practical fashion, and buzzing about its potential. Which is just as it should be – an enterprise sector that is keeping its head, while the AI market loses its own. Indeed, a late conversation with Orange suggested that, by accident or design, telcos might just make a virtue of their old caution – as trusted stewards of data, as experienced negotiators of regulation, as experienced integrators of AI. It’s not a given, but it seems like a proper plan. 

At the top of the bill, the question was about the AI demand on mobile networks. There’s not much, as yet – was the response. “We’re not seeing an increase in uplink traffic today,” said Telefonica (Andrea Folgueiras). The traffic ratio remains 90/10; almost everything is still going down the (wireless) internet pipe – for streaming content, rather than up it for AI-content generation. Same, said Orange (Laurent Leboucher); it is watching the patterns “carefully”, and uplink activity is somehow “important” for certain apps, but the “big [uplink] volumes remain small”. Sovereign data and infrastructure, edge compute services, cybersecurity and resiliency – what about these? Will they make telcos any money?

Vodafone (Scott Petty) had the sharpest answer, which reads like a truism, but sounded more profound here: if it doesn’t save money or make money, then what’s the point? “Sovereign cloud or sovereign AI without sovereign networks doesn’t make a lot of sense.” Which is the same as the industry line that AI is useless if it is not connected. Most telling, maybe, in terms of networks-for-AI / AI-for-networks, is the message that telcos should get AI usage right first, before trying to sell it to the wider market (as integrators). It was echoed later by Orange – try the fix first; Vodafone has 60 million conversations between chatbots / agents – in some form, in some time frame. (RCR’s note-taking failed.)

The second session was better. Vodafone (Andrea Donà) was strategic, again: there are three ways to assess progress – what is being used (predictive, generative, agentic), what is being done (efficiency, performance, new revenue), where it is being done (network design, operations, extensions). Vodafone is using it mostly for efficiency and performance – as above. “That’s a tick in the box,” it said; it is moving to “create new revenue streams”. Later, it referenced its ‘5G+’ slicing service in the UK. BT talked about the use of AI in cybersecurity – to move the discipline “from a defense problem to adversarial AI competition”. Eight percent of phishing attacks are AI generated, apparently. 

BT provided an interesting account of sovereignty, as well, which RCR will try to write up fully. But the best commentary in the early sessions was from Swisscom, billed as a “poster child for autonomous networks”, which brought it back to brass tacks: data, and how to sort it, render it, use it. It was the same message from Siemens yesterday – that 80 percent of industrial data is buried in silos. Swisscom has spent six years organising its data into a “data model”, it said. “Every system is producing millions of data points per day. Turning that into knowledge is the goal.” Chatting to people on the floor, the sense is Swisscom, of everyone, has it sussed.

But there were good stories all over the place at FutureNet World, and just maybe, like Orange said / agreed (can’t remember), there is a future where telcos are looked at as a model for AI transformation – and are asked for help.