Qualcomm Launches Two New Smartphone Chips With AI Focus

Qualcomm has launched two new smartphone chips, placing on-device AI at the centre of the announcement.

The company says its top chip can run a 30B mixture-of-experts model locally. The operative word is locally: inference happens on the handset rather than round-tripping to a data centre.

Mixture-of-experts architectures are well matched to that constraint. The model as a whole is large, but only a fraction of it is activated for any given request, so capacity can grow while per-request compute stays relatively contained.

The 30B figure deserves its own look. Models that ran reliably on phones have generally sat in the single-digit billions, so pushing to 30 billion widens the range of tasks a device can plausibly handle.

For users, the most immediate consequences are privacy and availability. Local inference does not require uploading input to a server and keeps working offline or on poor connections; for device makers, it moves some inference cost off the cloud.

Chip specifications and real experience are still different things. Throughput, power draw, heat and the degree of quantisation applied to the model all determine whether that capability is genuinely usable day to day, which shipping devices will have to demonstrate.

What to watch next is which handset makers adopt the two chips first, and whether developers build applications that exploit a local 30B-class model in ways cloud inference cannot match.