At MWC26, a session moderated by Abe Nejad of The Network Media Group (NMG) brought together leaders from Google, TELUS, Vodafone, and Rakuten Symphony to explore a deceptively simple question: is AI in the telecom network actually delivering? The answer, the panel agreed, is yes – but realizing its full potential requires more than technology. It demands a rethinking of how networks are built, observed, and operated.
Speakers:
Watch the full interview.
The panel was unequivocal: the era of legacy, reactive network management is over. AI is already driving capacity planning, radio optimization, and predictive maintenance across live operator networks. What makes this shift meaningful is the move from responding to problems after the fact to resolving them before they surface. The leaders pointed to deployments where AI agents autonomously detect KPI degradation, trace its root causes across domains, and take corrective action without waiting for a human to intervene. Cross-domain intelligence, the panel noted, is what separates AI that genuinely moves the needle from AI that merely automates the obvious.
With the RAN accounting for more than 80 percent of network energy consumption, the commercial case for AI-driven efficiency is stark. The framework the panelists returned to was straightforward: observe, engage, act. Build real observability into the network. Use that data to predict demand and identify waste. Then act autonomously, powering down underutilized cells, shifting workloads, even adjusting physical cooling systems based on live sensor data. The consensus was that this kind of end-to-end, intelligent energy management is where operators will feel AI's impact most concretely and most quickly.
Closed-loop automation dominated the latter part of the discussion and the framing was direct: it is no longer a strategic ambition, it is a matter of survival. Services today are short-lived and dynamic; operators that cannot respond in near real time will simply be left behind. But the panel was equally candid about why so many AI agent deployments are underperforming: they are built without adequate models of cross-domain dependencies. Without understanding how a change in one part of the network cascades into another, closed-loop systems will always fall short.
The answer, the leaders agreed, is gradual and governed deployment. Digital twin environments allow operators to test AI agents against simulations of their live network before committing to production. Human oversight, quality gateways, and the ability to intervene are not signs of hesitation; they are the architecture of responsible autonomy.
“The RAN network consumes more than 80% of the energy in the network. So the question is – how do you make it autonomous in such a way that, if you can turn off certain things when they are not needed or move workloads to different parts, you can conserve that energy.”
