On-device AI: mobile network filler or saviour?

Monetising inference along the network edge has become a priority for telecoms operators to drive returns on 5G infrastructure investments and open up new service capabilities. The narrative around AI is now expanding around how to enable AI workloads on the device itself.

On-device AI is the next frontier for AI deployments in consumer and enterprise segments. Use cases include AI-based search and personal assistants, image analytics on a smartphone, AI-assisted driving and navigation in cars, industrial applications such as automated guided vehicles, and improved IoT analytics. This has implications for the device world and major tech companies whose business models depend on consumer time on the internet and digital platforms. It also matters for operators, given the ripple effect of AI-driven data traffic on mobile and fibre networks, and the direct revenue opportunity.

Working from the hypothesis that most AI use cases will run on a hybrid processing set-up between device and cloud, GSMA Intelligence benchmarked silicon capabilities from a range of leading chipsets to see where the sector is in relation to the technical requirements for on-device AI. In October and November 2025, GSMA Intelligence ran a survey of 100 companies across the world, split between device OEMs and silicon groups. In parallel, analysis was conducted to benchmark a range of leading chipsets from a variety of companies on key capabilities required to deliver different use cases for on-device AI, using publicly available information and sense checking through industry conversations.