In July 2026, overseas securities firms released a research report on the AI computing power industry chain, indicating a tight supply of advanced packaging capacity worldwide

In July 2026, overseas securities firms released a research report on the AI computing power industry chain, pointing out that the global demand for large-scale model training and inference continues to explode, and advanced packaging has become the core bottleneck restricting the delivery of AI chips. The high-density 2.5D packaging capacity inventory of TSMC and Sunlight, two leading overseas advanced packaging companies, has fallen to the safety supply bottom line of less than two weeks. The massive computing power demand brought by the large-scale construction of AI servers continues to squeeze advanced packaging capacity, further exacerbating the supply-demand imbalance.

In order to seize the dividends of the AI computing power market, TSMC has significantly adjusted its production line structure and allocated 50% -60% of CoWoS advanced packaging capacity to AI chip packaging, corresponding to a monthly packaging capacity of approximately 12000 wafers. With the rapid popularization of Chiplet chip architecture solutions, the new generation of AI chips achieves significant improvement in chip interconnection bandwidth through heterogeneous integration of multiple chips; In the context of large-scale model training, a single high-density AI server requires 9 times more advanced encapsulated chips than traditional servers, resulting in a significant increase in demand. According to calculations, the overall production capacity gap of high-end CoWoS advanced packaging worldwide is as high as 52%.

On the supply side, the existing high-end advanced packaging capacity of overseas original factories has been largely locked in by overseas top cloud manufacturers and AI giants. Downstream AI chip design companies generally face problems such as extended advanced packaging schedules and difficulties in obtaining production capacity, highlighting supply chain risks. This situation also forces the domestic industrial chain to accelerate the pace of advanced packaging technology research and industrialization landing.

Domestic packaging and testing manufacturers are synchronously promoting technological breakthroughs. Changdian Technology's 2.5D silicon bridge advanced packaging products have entered the stage of small-scale trial production. Currently, they are conducting multiple rounds of adaptation and joint verification with domestic AI chip companies such as Cambricon and Haiguang Information. The goal is to achieve small-scale mass production and supply by 2027, gradually filling the gap in local high-end advanced packaging supply and providing chip packaging support for the domestic AI computing power industry chain.