Carol, let’s play Countdown – and take four from the top and three from the bottom, and do some dodgy maths. (If you’ve lost us – because you’re not British, or you’re under 45 – then hold tight; we’ll get through. Actually, let’s take all five from the top of today’s stack – or from any number of over the past few years. Because the story is always the same, right – about telcos becoming tech-cos. Yawn. If I had a dollar for every… But here we are, early 2026, post-MWC, and every operator and every vendor has the same line about transformation. 

The only difference, versus 2025 or 2024, is that AI is real – or real in the data center, and almost-real outside of it (at the edge, in the hands of enterprises). So now everyone – Huawei, Softbank, TIM, True, as reported below; plus Spirent / Keysight, in a memo to telcos (which makes five) – wants to be some kind of an AI-co, basically. AI-native, ideally; AI-ready, otherwise. Same as all the enterprises they want to sell AI services to on AI networks. All of them have pillars and strategies. Softbank wants to be a “next-gen social infrastructure”.

Thai operator True is following an “AI-first telco-tech model”, apparently. TIM in Italy has a line about how the cloud needs the network and AI needs the cloud, and, so… what? Data has to be delivered? Well, who knew? Huawei reckons there will be a hundred billion agents – at some point, on some (presumably all) networks. Apologies; it’s just we’ve heard this before. Telecoms wants a shot in the arm; it wants new ideas and growth; it wants to be cool again. It needs to change. 

And Softbank has more to say, actually; it will embed AI infra inside its network to build a structural cloud edge for latency, reliability, sovereignty – for 100 billion agents, an 80/20 shift from training to inference, and a line into AI-hungry enterprises. NTT Data, one from the bottom, is doing the same. Huawei says more too – about a fatter uplink (talked about in private 5G for donkey’s years), agent-to-agent telco chatter, and new KPIs for AI QoS. There’s something there, worth chasing: the real story. Telcos must stop talking about change, and make it.

If you do it, it’s done – as I tell my kids about their homework. Easy to say, of course. For telcos, it is a three-way calculation: deploy AI for operations, make networks for AI, try to own the AI customer. Which, to be fair, are stories that also get told. Which brings us to two more links at the bottom of the stack: one, that a bunch of optical vendors (Ciena, Coherent, Marvell etc) are to develop open specs for AI-integrated interconnect solutions; two, that Cisco has new pluggable coherent optics for scale-across networks. 

Both are about this interconnect piece, for distributed training workloads between data centers, which RCR has been harping on about all week. Anyway, point again is that this is where the action is in AI networks, right now. Operators can talk about (infrastructure) change all they want; at some point, soon (likely between their 5G and 6G upgrades), they will have to deliver. Those that have already changed-up their fiber systems will surely be in good stead. And then the biggest challenge of all: how to front-up to enterprises as an AI-co tech-co telco.