Artificial Intelligence (AI) in telecom sector is not expected to drive revenues although it can help lower costs and improve efficiency, said OpenSignal after speaking to industry people, who had attended the Mobile World Congress earlier this month in Barcelona.
Noting that AI is becoming a dominant topic even in the telecom discussions, OpenSignal polled the telecom operators, technology providers, experts and industry leaders attending the MWC in a pre-event webinar. OpenSignal provides market insights into the telecom sector.
The polling found that none of the attendees expected AI to drive revenue, while 26 per cent thought it could be cost-effective and improve efficiency. The majority of industry folk or 44 per cent said they found AI useful to optimise network operations like downtime reduction, greater use of preventive maintenance. Further, 23 per cent of attendees said AI will enhance and 8 per cent viewed AI as a technology that could introduce significant risks and challenges.
AI-driven data centres show promise
While telecom operators said they did not find AI to be lucrative in terms of revenue, they found monetisation success in terms of the growing demand for AI-optimised data centres or edge infrastructure.
“Expanding footprint of AI workloads is anticipated by some to be particularly relevant for edge data centres. SoftBank is collaborating with Nvidia, to find use-cases for real-world AI inference applications, to run adjacent to its RAN,” said OpenSignal, noting that Etisalat, a UAE state-owned telco, recently sold a 40 per cent stake in its data centre subsidiary Khazna to an AI company for $2.2 billion, reaping a substantial return on investment.
AI inference on consumer devices
Aside from the use of AI for edge, optimising networks and Intelligent RAN, OpenSignal noted the push to enable AI inference onto consumer devices. This demand is met with challenges like battery constraints and limited capabilities for reasoning by smaller AI models.
“Practically, smaller language models – often around seven billion parameters – achieve results similar to full-sized models when fine tuned for narrowly defined use cases, making them “good enough” for everyday use. The MWC announcements this year suggest that local inference is finding its way into midrange devices, with Samsung’s Galaxy A56 and Realme’s 14 Pro promising affordable AI-enhanced experiences,” said OpenSignal.
