From the newsletter (October 21; sign-up or follow if you want it sooner): Nvidia and Nokia are pushing telcos to turn their networks and sites into distributed AI infrastructure, but the business case remains hazy. AT&T’s latest results offer a firmer reminder: connectivity still pays – even while Musk plots noisily in the background.
Plenty of AI-RAN stuff from GTC in Berlin today, courtesy of Nvidia and Nokia – live tests with T-Mobile, BT, Deutsche Telekom, Elisa, Orange, plus others. As a show of strength, a year after the US outfit pumped $1 billion into the Finnish firm, it probably worked. RCR attempted to pull together a round-up, but ran out of time; you’ll have to search through their press notes variously. But most interesting, and the bigger play, is how telcos might make use the assets they own – their radios, for sure; but also their cell sites, fiber networks, switching points, data centers, and other offices and sites – to bring AI closer to the action, in devices, robots, vehicles, machines.
The grand pitch, here, is that telcos can turn their real estate into infrastructure for AI workloads, mostly for enterprises – rather than simply carrying traffic between other people’s training clusters and inference nodes. AI in/on the RAN is part of it, but distributed AI around the RAN seems just as important. Nine operators in the EMEA region are building or expanding their Nvidia-based infrastructure. CRA in Czechia, Etisilat in the UAS, Silknet in Georgia, and TIM in Italy are adding new infrastructure. Deutsche Telekom, Fastweb + Vodafone, Kazakhtelecom, and Serbia-based Orion Telecom are expanding existing deployments.
The GTC news offers few details on capacity, investment, workloads – so the scale of the scheme, and the opportunity, remain blurry. But there are some more tangible examples. Deutsche Telekom is to expand its sovereign ‘industrial AI’ cloud, apparently, hosting open models including Nemotron and Soofi for customers such as Germany-based Agile Robots. A GTC demo with the robotics firm combines device/robot-edge and network-edge processing for “real-time perception and action” and “fine-tuning and coordination” across a robot fleet, respectively. Larger training workloads stay in a central cloud infrastructure. But it all remains very experimental.
And for all the noise about AI infrastructure, the immediate business for telcos is still about selling connectivity, bundling services, and persuading customers to cough up – as AT&T appears to be doing quite well. Service revenue growth hit a four-year high, with more than a million ‘advanced connectivity’ additions for a second consecutive quarter, as fiber and fixed wireless complemented mobile. AT&T is raising its earnings outlook while lowering longer-term capital investment, even as it pushes fiber beyond 60m locations by 2030. For now, AT&T’s strongest proof point is still the old one: customers, convergence, cash flow. Nothing much about hosted AI, as yet.
Telcos doing telco things, then – quite successfully here and there, amid all the AI noise. While Elon Musk burns through capital to clarify his AI vision, and makes out like he will bury the old telco set in the process.
