According to a report from Taiwanese media, Taiwan Semiconductor Manufacturing Senior Deputy General Manager Hou Young-ching stated that both the magnitude and the frequency of demand changes triggered by AI are situations not seen in the past 30 years. To break through the AI capacity bottleneck, TSMC has dramatically accelerated its factory construction speed to roughly five times the historical rate. Currently, there are 17 plants being built simultaneously in Taiwan, with an additional 5 to 6 sites progressing in other regions. Even so, "supply still cannot meet demand."
Hou emphasized that to support the next stage of AI growth, the first challenge to address is the capacity gap. The greatest difficulty in this wave of AI demand lies not just in its sheer scale, but in the fact that the speed at which customers continue to raise their demands far exceeds the historical experience of the semiconductor industry. He used Taiwan Semiconductor Manufacturing as an example, noting that at the end of last year, the company estimated the equipment procurement needed to satisfy this year's customer requests based on a baseline of "one time." However, just one quarter later, in response to raised customer demand, that equipment requirement increased to 1.5 times. By July this year, it had climbed further to 1.9 times. In other words, within a short span of about six months, equipment demand nearly doubled compared to the initial estimate.
Hou pointed out that expanding production capacity cannot be indefinitely accelerated simply by having more capital. Taking factory construction as an example, whether in Taiwan or overseas sites like Arizona in the United States, there are constraints related to the availability of construction labor. After the plants are completed, they require a massive influx of equipment. When equipment demand nearly doubles within half a year, it is impossible for equipment suppliers to instantly double their production capacity either.
Comments