The Zhitong Finance App learned that on September 23, during the 2026 Snapdragon Summit, Qualcomm Technology announced a quadripartite cooperation with Step Yue, Wulianghuo and Jiang Bolong (09976) to perform end-side adaptation and reasoning optimization of the Step StepDge-Omni 30B-MoE (Hybrid Expert) model based on the sixth-generation Snapdragon 8 Super Extreme Edition mobile platform, and to demonstrate an end-side smart assistant with autonomous service and personalization capabilities.
According to reports, the 30b-MoE model can be run on a smartphone and can complete tasks such as email understanding, itinerary planning, calendar synchronization, flight and hotel recommendations, itinerary sharing, and email drafting without calling a cloud service. Through the collaborative optimization of the inference engine, CPU/GPU/NPU heterogeneous scheduling and storage scheme, the memory requirement for model operation was reduced by more than 50% compared to conventional solutions, and the field test pre-filled throughput exceeded 330 token/s, and decoding throughput exceeded 28 token/s.
Among them, Jiang Bolong focuses on leveraging endside storage capabilities. Through storage and computation collaboration, model weights can be loaded from flash memory, reducing the continuous use of operating memory by large models, and providing storage support for the efficient operation of the 30B class model on the mobile phone. Jiang Bolong said that large-scale implementation of end-side AI is inseparable from deep collaboration between computing power and storage, and the company will continue to promote innovation in end-side AI storage solutions.
The analysis points out that this cooperation also further highlights the importance of storage in the end-side AI industry chain. As AI phones evolve from simple question-and-answer to smart devices that plan independently and perform tasks continuously, the demand for high-speed, high-capacity, and low-power storage is expected to further increase, and ecological collaboration between storage vendors and chip, model, and terminal manufacturers will continue to deepen.