As most of DRAM production capacity is allocated to high-margin products such as HBM to support the development of the artificial intelligence industry, the global mobile industry is facing a long-term memory shortage. As a giant in the field of mobile chips, Qualcomm is actively looking for new ways to deal with this challenge.
Recently, news about Qualcomm’s cooperation with Changxin Memory and GigaDevice to develop customized memory and chips has attracted widespread attention. Analyst Ming-Chi Kuo disclosed in-depth details of this cooperation, revealing Qualcomm’s new trends in device-side AI layout.
It is pointed out that Qualcomm is working with GigaDevice to develop a discrete NPU specifically for smartphones. The target customer group of this product is very clear, that is, Chinese smartphone brands, aiming to improve the AI processing capabilities of flagship phones.
This new NPU is expected to be officially shipped at the end of 2026 or early 2027. In terms of product positioning, it will be mainly used in high-end flagship models priced at more than 4,000 yuan, becoming the key hardware to enhance the core competitiveness of mobile phones.
In terms of technical parameters, this NPU can provide a powerful computing power of about 40TOPS and is equipped with 4GB customized 3D memory. It is worth noting that this customized memory is manufactured by Changxin Memory and uses advanced packaging stacking technology.
By applying TSV and hybrid bonding technology, this solution can provide higher memory bandwidth than the current mainstream LPDDR5X. This architectural advantage can effectively solve the data transmission bottleneck in AI computing, making the operation of large end-side models more efficient and power-saving.
However, despite the technical breakthrough, market expectations for the project have been reduced to a certain extent. Ming-Chi Kuo said that due to the sharp increase in memory prices that has pushed up the overall cost of NPU, the shipment expectations of the cooperative products are lower than originally envisaged.
In addition, the specific application scenarios and business models of on-device AI are still in the exploratory stage, and have not yet shown an extremely clear profit path. These uncertainties make this in-depth collaboration between Qualcomm and the supply chain face considerable challenges in the advancement process.