【1991 Archives】

Every Wednesday991 ArchivesFriday, TechNode’s Briefing newsletter delivers a roundup of the most important news in China tech, straight to your inbox.

At the opening day of the Mobile World Congress (MWC) Shanghai 2025, chipmaker Qualcomm laid out its roadmap for shifting AI development from cloud to edge devices, citing evidence that future AI experiences will take place primarily on smartphones, PCs, cars, and other consumer devices.

Speaking at the event’s Agentic AI Summit on Wednesday, Vinesh Sukumar, Qualcomm’s vice-president of AI product management, said the company sees a paradigm shift as AI evolves from generative models to intelligent agents that act autonomously based on user intent. “AI is becoming the new UI (user interface),” he said.

“We were one of the first to demonstrate large models running entirely on edge devices,” Sukumar said, referencing a 2023 showcase in which Qualcomm chips powered image and text generation without a cloud connection.

Edge-based AI, according to Sukumar, brings advantages in speed, energy efficiency, and privacy. To support this, Qualcomm has developed a software stack optimized for small and medium-sized models, with between 7 and 10 billion parameters. He noted that compact models have rapidly caught up with their larger counterparts in performance, narrowing a quality gap once dominated by models like Meta’s Llama 3.

Still, building agentic AI – systems capable of autonomous action – poses challenges, especially around consistency, accuracy, and responsiveness. Qualcomm’s response is a layered architecture: if a user query cannot be fully addressed by a local device, tasks can be offloaded to nearby edge servers and, when needed, escalated to centralized cloud platforms, maintaining user experience without latency spikes.

Key to this is Qualcomm’s Hexagon NPU, a dedicated AI accelerator that handles high-efficiency matrix and tensor operations critical to AI workloads while minimizing battery drainage.

Commercial deployment of edge AI has already begun, Sukumar said, with cases spanning photo editing, content generation, and real-time code creation. Qualcomm claims to have enabled 200 to 300 AI applications annually and expects to scale that further as developer tools and consumer-ready features roll out in the next six months.

As edge AI gains momentum amid hardware limitations and privacy concerns around cloud computing, Qualcomm is positioning itself as a leading infrastructure provider – not just in mobile chips, but in the AI systems that power next-generation user experiences.

The annual conference, held at the Shanghai New International Expo Centre (SNIEC), is hosting over 400 speakers this year, and revolves around four core themes: 5G integration, artificial intelligence, connected industries, and enabling connectivity.

Support TechNode

With a small team, TechNode provides timely news and thoughtfully researched articles for worldwide readers interested in learning more about the Chinese tech industry.

Your contribution is appreciated.