The high demand for advanced technologies such as artificial intelligence (AI), high-performance computing (HPC), and PCs has driven leading chipmakers to expand their production capacities. TSMC, the largest semiconductor manufacturing company in Taiwan, plans to expand its production capacity to meet the high-end data center GPU demand from NVIDIA, such as the A100 and H100.
According to DigiTimes, the demand for packaging technologies like chip-on-wafer-on-substrate (CoWoS) has reportedly exceeded available capacity. Therefore, TSMC plans to process an additional 10,000 CoWoS wafers specifically for NVIDIA throughout 2023, resulting in approximately 600,000 high-end data center GPUs.
This increase will be achieved by adding 1,000 to 2,000 wafers per month for the remainder of the year. Currently, TSMC produces between 8,000 to 9,000 CoWoS wafers per month, so this addition will significantly boost their production facility utilization.
However, this increase also carries potential negative consequences. It may lead to a scarcity of CoWoS services for other industry players as the demand from other companies also rises, although not to the same extent as NVIDIA.
As widely known, NVIDIA’s AI chips are already extensively used across various industries. For instance, Google recently launched the A3 supercomputer based on NVIDIA H100, which boasts 26 ExaFLOPS of AI performance. Other companies like Microsoft, Oracle, and Elon Musk’s AI project have also ordered tens of thousands of NVIDIA AI GPUs.
