Qualcomm Launches Two AI-Focused Smartphone Chips

What happened

Qualcomm has introduced two new smartphone chips built with a heavier focus on artificial intelligence capability, according to TechCrunch. The company said its flagship chip in the new line-up is powerful enough to run a 30-billion-parameter mixture-of-experts (MoE) model directly on a device, without relying on the cloud.

The announcement marks another step in the shift of generative AI workloads from remote data centres onto local hardware, with Qualcomm positioning on-device processing power as the headline differentiator for the new silicon.

Why it matters

Running a model of this scale locally is a meaningful technical milestone. Mixture-of-experts architectures are designed to deliver large-model performance while only activating the portions of the network needed for a given task, making them more efficient to run than a dense model of similar size. Fitting one at this scale onto a smartphone chip suggests device-level AI is closing the gap with capabilities that, until recently, were only practical via cloud inference.

For technology and experience leaders, the shift matters because it changes the default architecture for AI-powered features. Assistants, translation, image processing and other AI-driven functions can increasingly run without a live connection to a server, which has implications for latency, offline reliability and data handling. Device makers and app developers building on this hardware will need to reassess which features they push to the edge versus the cloud, and what that means for consistency of experience across connectivity conditions.

By the numbers

• 30 billion parameters — the size of the mixture-of-experts model Qualcomm says its new flagship chip can run locally on a smartphone.
• Two — the number of new smartphone chips Qualcomm has launched as part of this update.

The Renascence take

The industry conversation around on-device AI tends to focus on raw benchmarks — parameter counts, tokens per second, chip architecture. The more interesting question for experience leaders is what changes for the person holding the phone once that processing power sits in their pocket rather than in a data centre three continents away.

On-device AI at this scale is less about speed and more about trust and dependability: features that work the same on a train, on a flight or in a basement no longer feel like "AI," they feel like the product simply working. The behavioral shift to watch is that users stop tolerating spinners and connectivity caveats once local processing removes the excuse — so any brand still shipping an AI feature that visibly waits on the network is now competing against a silent, offline-capable baseline. Hardware vendors are handing software and service teams a genuine opportunity to redesign AI features around reliability and privacy rather than novelty; the operators who move fastest will be the ones auditing which of their "smart" features can be pulled entirely onto the device, and treating that as a service-design decision, not just an engineering one.