Qualcomm unveils AI-focused smartphone chips

Qualcomm has introduced two new smartphone processors built around on-device AI. The company says its top-tier chip can run a 30-billion-parameter mixture-of-experts model locally, without sending queries to the cloud. If achieved in real devices, that would be a major advance for privacy, latency, and cost. Users could get faster responses and keep sensitive data on the handset. Cloud dependence would shrink, reducing bandwidth and server expenses. The announcement comes as phone makers look beyond raw CPU speed to differentiate through AI features. Qualcomm named the Snapdragon 8 Elite Gen 6 and an Extreme variant, positioning them for premium devices. On-device AI has become a key battleground because consumers increasingly expect assistants, photo tools, and productivity features to work instantly and privately. Running large models locally is difficult due to memory, power, and thermal limits. A 30-billion-parameter mixture-of-experts model is especially demanding, though MoE designs activate only part of the network at once, which can improve efficiency. Qualcomm's claim suggests it has made gains in NPU performance, memory bandwidth, and software optimization. Still, real-world performance will depend on model compression, quantization, and app integration. Developers must adapt models to the chip's architecture. Handset makers may use the chips to market AI capabilities rather than benchmark scores. The shift could also pressure rivals such as MediaTek and Apple, which are investing heavily in silicon for AI. For Qualcomm, the strategy is to stay central to smartphones even as growth slows. If on-device AI becomes a standard feature, chip designers that offer strong tools and power efficiency will have an advantage. The new processors are expected to appear in flagship phones, with benchmarks and battery tests to follow.