Qualcomm's new flagship smartphone chip can reportedly run a 30-billion-parameter AI model entirely on the device, no cloud connection required.
The chipmaker unveiled two new smartphone processors this week, both pitched around AI performance as the headline upgrade. The top-tier chip is built to handle a 30-billion-parameter mixture-of-experts model locally. Mixture-of-experts models split a large network into smaller specialized sub-networks and activate only the ones needed for a given task, which is part of how a phone chip can plausibly handle a model that size without melting the battery. Qualcomm has not yet detailed pricing, release dates, or which phones will carry the new silicon first.
Running a 30B-parameter model locally matters because it shifts work away from the cloud - faster responses, no network dependency, and less user data leaving the device. It also pushes the on-device AI race further: a few years ago, "local AI" on a phone meant simple voice commands, not models in the same size class as some early cloud-hosted assistants.
Bigger local models are still not the same as better ones. Parameter count is a spec sheet number, not a review score, and Qualcomm's own announcement leans hard on that one number rather than showing what the assistant can actually do with it.