The shortage of AI chips will continue until 2026, TSMC predicts.

TSMC management’s caution in assessing the impact of the AI ​​boom on the company’s business, observed in April, has given way to more pronounced confidence in maintaining strong demand for related chips. Chairman of the Board of Directors C.C. Wei said that TSMC’s ability to supply components for artificial intelligence systems will be limited throughout 2025.

Image Source: TSMC

Accordingly, if the deficit recedes a little, this will happen no earlier than 2026. The contract manufacturer’s confidence in maintaining high demand is also indicated by the increase in the lower limit of the range of capital expenditures for this year, as well as the improvement in the forecast for revenue growth. In addition to the actual lines for processing silicon wafers, TSMC is forced to spend more money on packaging chips for AI systems that have a complex spatial layout. The company is ready to look for a more progressive alternative to the CoWoS method, which is currently used for packaging Nvidia chips used as part of computing accelerators. “Demand is very high, supplies will be very limited up to and including 2025, and we hope that relief will come in 2026. We continue to expand production capacity in any place and by any means,” explained CC Wei.

According to TSMC CEO, also CC Wei, the company is currently experimenting with FOPLP (panel fan-out technology) packaging method, but it is not too mature for mass production. This will happen in about three years, as the head of the company suggests. By then, TSMC itself will be ready to master this method of packaging chips in mass production. The head of the company added that in terms of CoWoS, it has more than doubled its core production capacity by this year, and next year it can double it again. Previously, it was believed that the deficit in this area would be eliminated by the end of 2024.

CC Wei explained that the first generation of 2nm chips will enter the TSMC production line in the second half of 2025, and the second will follow in 2026. In the second half of 2026, the company plans to master production of products using the more advanced A16 technical process.

In the field of artificial intelligence, according to the head of TSMC, demand will spread to peripheral devices of computing systems such as smartphones and PCs, but so far this does not in any way affect the quantitative indicators of product deliveries in the relevant market segments. Over the next two years, the development of the market for devices with AI acceleration functions will reduce the life cycle of such devices. Currently, demand from TSMC customers is especially high for products using 5nm and 3nm process technology. Work is already underway to provide customers with adequate quotas for the production of such products with an eye toward 2026.

Despite the controversial statements made in the American information field yesterday by one of the candidates for the post of US President, TSMC management did not revise its plans regarding the construction of its foreign enterprises. In this area, everything is going according to schedule and no changes are currently expected, as CC Wei made clear.

It was previously noted that TSMC management does not exclude the possibility of increasing prices for its services for Nvidia. CC Wei added today that his company is facing increasing pricing pressure. Costs are rising due to increasing complexity of technical processes, rising electricity prices in Taiwan and high capital costs for the construction of overseas plants. However, TSMC insists that it does not take an opportunistic approach to pricing, and builds its relationships with customers by trying to convince them of the adequate value of its services. Analysts at Nomura Global Market Research believe that TSMC will raise prices for its services by 5-10% from January 2025.