Nvidia faces AI chip supply shortage

Since last year, TSMC, responsible for the manufacturing and packaging of NVIDIA’s AI chips, has experienced a tightening of capacity in advanced packaging. Consequently, TSMC has been relentlessly expanding its 2.5D packaging capacity to meet the soaring demand. Reports suggest that TSMC is fully committed to addressing the high demand for CoWoS packaging, planning to double its capacity this year.

According to UDN, due to the long-term shortage of advanced packaging capacity, NVIDIA’s AI chip supply has been strained. Previously seeking alternative ways to increase advanced packaging capacity, NVIDIA has now turned to Intel as a provider of high-level packaging services to alleviate the tight supply situation. Intel has packaging facilities in the United States and Penang, Malaysia, and has adopted an open model that allows customers to utilize its packaging solutions independently.

Intel is expected to begin providing advanced packaging services to NVIDIA as early as the second quarter of this year, with a monthly capacity of 5,000 wafers. While TSMC will remain NVIDIA’s primary packaging partner, Intel’s entry into the scene is set to significantly boost NVIDIA’s total packaging capacity by nearly 10%. TSMC is not slowing its pace of expanding packaging capacity either, with an expected increase to nearly 50,000 wafers per month in the first quarter, a 25% growth since last December.

The shortage in AI chip supply primarily stems from the inadequacy in advanced packaging capacity, coupled with the tight supply of HBM3. Additionally, excessive orders from some cloud service providers have added pressure to the supply chain. Nonetheless, some server suppliers are benefiting from these orders, accelerating capacity expansion to enable cloud service providers to rapidly deploy equipment.