AI · SEMICONDUCTOR · FACT CHECK
SKD WIREMostly True
Plans to run very large AI models directly on smartphones are, in principle, feasible, but current mobile hardware imposes a hard constraint: memory capacity around the 50-gigabyte level remains a barrier that today's flagship devices cannot overcome, according to a review of 12 primary sources.
The claim centers on the vision of smartphones carrying AI models on the scale of 100 billion parameters, running locally rather than in the cloud. Such models would need to be held entirely in the device's memory to operate on-device.
The decisive constraint is RAM. A 100-billion-parameter model, even in compressed form, demands tens of gigabytes of memory, and the practical ceiling for on-device AI sits at roughly the 50-gigabyte mark — a level no current smartphone platform reaches. Flagship mobile platforms, including Qualcomm's Snapdragon 8 Elite generation, and memory suppliers such as Samsung Electronics and SK hynix are advancing mobile memory capacity, but the gap between what large AI models require and what phones can carry remains unresolved.
For now, large-scale AI workloads continue to depend on cloud and data-center infrastructure, with on-device AI limited to smaller, distilled models.
The core of the claim — that on-device AI ambitions are real but stalled by the ~50GB RAM barrier — holds up against the available evidence, with the caveat that the timeline for overcoming it remains unconfirmed. Verdict: Mostly True.