The biggest telecom operators aren't just using AI to answer support tickets anymore. Over the past year, several have started pointing it directly at the network itself — letting software watch live traffic, spot problems, and act on them without a person in the loop. Two recent moves from Vodafone and Deutsche Telekom show how far that's already gone, and what it's likely to mean for everyone else running a network, at any scale.
A network that turns its own radios down
Vodafone UK and Ericsson ran a trial across live 5G sites in London that cut the daily power consumption of 5G radio units by up to 33%, without affecting the experience for connected customers. The system behind it isn't a single clever trick — it's a suite of AI models working on different timescales: one puts radio units into a low-power "deep sleep" state during quiet traffic periods, another orchestrates sleep behaviour across 4G cells based on a live model of the network, and a third builds a heatmap of underperforming sites so engineers know exactly where to focus next.
None of that is exotic AI in the generative sense — it's closer to continuous, automated decision-making at a scale no operations team could match by hand. That's the pattern worth paying attention to: not a chatbot bolted onto a support queue, but software making thousands of small operational calls a human simply wouldn't have time to make.
From copilot to self-healing network
Deutsche Telekom is pushing the same idea further. Its multi-year partnership with OpenAI, announced in December 2025, is rolling ChatGPT Enterprise out to roughly 200,000 employees — but the stated goal goes beyond productivity tools. DT has been explicit that it wants the collaboration to help move its network toward autonomous, self-healing operations, where systems detect and resolve faults directly rather than routing them to an engineer's queue.
The company isn't starting from zero: its existing AI-driven incident management already resolved around 70% of network incidents automatically in 2024, and it's targeting 90%. At the customer-facing end, a Magenta AI call assistant launched at Mobile World Congress 2026 handles real-time translation, note-taking and call summarisation — though DT's own reporting shows it's still early days, with roughly half of customer issues fully resolved by AI without a human stepping in. That honesty matters: this is incremental progress, not a finished system, and it's a useful reality check against the more breathless AI headlines.
What it means below the scale of a national carrier
Most regional and city-level ISPs will never run a model trained on hundreds of millions of subscriber records, and they don't need to. What matters is the underlying pattern both of these examples point to: closing the gap between "the network reported something" and "the system did something about it," automatically, instead of waiting on a person to notice and act.
That same principle already shows up at ISP scale in a much more grounded form:
- Session state changes acted on instantly. A client's plan expires or a payment lands, and RADIUS enforces the change on the router immediately — no operator logging in to flip a switch.
- Faults surfaced instead of buried. A stuck voucher or a mismatched session status gets flagged for reconciliation rather than sitting unnoticed until a customer calls in.
- Automation that scales with subscriber count. The value of closing that loop only grows as an ISP adds clients, exactly the same reason the largest operators are investing here first.
That's the same real-time enforcement model XpressRADIUS is built on: MikroTik-integrated RADIUS control that acts on client status the moment it changes, tied directly to M-Pesa and Kopo Kopo payments, so growth in subscribers doesn't mean growth in manual admin.
Want your own network to react automatically instead of waiting on manual checks?
Start a free trial