TSMC's chairman, Mark Liu, recently confirmed that the limited availability of compute GPUs for AI and HPC applications stems from constraints in their chip-on-wafer-on-substrate (CoWoS) packaging capacity.
It is forecasted that this deficit will continue for about 18 months due to the rising demand for generative AI applications and the measured expansion of CoWoS capacity at TSMC.
Speaking at Semicon Taiwan, Mark Liu clarified, "The shortage pertains to our CoWoS capacity, not AI chips. We currently can't fulfill all of our customer demands, but we're working towards addressing roughly 80% of them. This is viewed as a transient phase. We anticipate alleviation after the growth of our advanced chip packaging capacity, roughly in one and a half years."
TSMC manufactures key AI processors, among which are Nvidia's A100 and H100 compute GPUs. These GPUs, vital in tools like ChatGPT, find primary application in AI data centers. Comparable to products from AMD, AWS, and Google, these processors use HBM memory, essential for high bandwidth and effective functioning of expansive AI language models, and CoWoS packaging. This elevates the demand on TSMC's advanced packaging infrastructure.
Liu noted the demand for CoWoS had a surprising threefold increase year-over-year earlier this year, resulting in the present supply bottlenecks. TSMC is aware of the surging demand for generative AI tools and the associated hardware. Hence, it's hastening the augmentation of CoWoS capacity to cater to compute GPUs, dedicated AI accelerators, and processors' needs.
The company is now integrating more CoWoS equipment in its current advanced packaging plants. This effort will take time, with expectations to double its CoWoS capacity only by the end of 2024.
TSMC also recently announced a $2.9 billion investment in a new facility focused on advanced chip packaging near Miaoli, Taiwan, which highlights the company's strategy to meet the advanced packaging demand industry-wide and acknowledges the role of advanced chip packaging in the future of semiconductors.
TSMC Chairman Discusses Prolonged Shortage of Nvidia's AI GPUs Due to CoWoS Constraints
