Qualcomm used the Snapdragon Summit keynote to cast the company as a leading architect of on-device artificial intelligence, unveiling a family of chips and software tools designed to run generative and conversational models locally on phones, PCs and cars. The company framed the changes as an effort to combine performance, efficiency and privacy — a thrust that played out across the roughly 21-minute recap published by CNET that condensed the event’s highlights.

“We want secure, efficient AI everywhere, from the pocket to the dashboard,” Qualcomm CEO Cristiano Amon said on stage, emphasizing the integration of neural processing units, advanced modem tech and new power-management features. The headline product is a next-generation mobile System-on-Chip with a beefed-up NPU (neural processing unit) Qualcomm says will accelerate LLM-style workloads while using a fraction of the energy required by cloud inference.

Beyond raw silicon, Qualcomm unveiled expanded developer tools and an SDK aimed at enabling app makers to deploy compressed, optimized models that run offline. The company demonstrated conversational assistants on a phone answering complex queries without pinging servers, and showed a TV experience in partnership with a major app provider that lets users “talk” to their screen — a nod to broader voice and multimodal trends. Qualcomm framed these demos as proof that latency-sensitive features such as real-time translation, multi-camera synthesis and low-latency gaming can be delivered more securely and economically when compute happens on-device.

On the PC side, Qualcomm highlighted new Snapdragon processors for thin-and-light Windows laptops, promising generational gains in sustained performance per watt. Executives argued that combining CPU, GPU and NPU on a single die will let manufacturers ship fanless designs with extended battery life while offering local AI acceleration for productivity and accessibility features. Automotive partners were shown using a car-grade variant for infotainment and driver-assistance workloads, where the company stressed functional safety and automotive lifecycle support.

Qualcomm also addressed connectivity. The keynote previewed modem updates intended to improve 5G throughput and reliability, and suggested future support for hybrid satellite links as an emergency fallback. Security and privacy claims were recurrent: Qualcomm described hardware-backed key stores and encrypted enclaves to keep models and user data on-device, but offered fewer details on certification and auditability.

Analysts and privacy advocates praised the efficiency gains but urged caution. Local AI reduces the amount of data sent to clouds, which can improve privacy, yet it concentrates new kinds of power on device makers and chip suppliers. “On-device models change the balance of trust,” said an industry consultant with knowledge of the space. “We need transparency about model provenance, updates and the potential for bias baked into preloaded assistants.”

Qualcomm’s Sun Valley-style pitch positions it to compete more directly with cloud-heavy incumbents and Apple’s vertically integrated approach. The company’s moves could accelerate a broader shift away from pure cloud inference toward hybrid architectures, reshaping app economics and raising new regulatory and ethical questions about where and how powerful AI runs.