Foundry Capacities TSMC: AI-Driven Chip Shortage Will Continue for Years to Come

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

TSMC CEO C.C. Wei sees the current AI boom as a multi-year stress test for manufacturing, lithography and the supply chain. The global foundry market leader is increasing wafer output and wants to keep prices stable. However, the AI-driven chip shortage is likely to persist for years to come.

TSMC does not expect to be able to fully meet the demand for advanced semiconductors for artificial intelligence for years to come. CEO C.C. Wei made it clear at the Annual General Meeting in Hsinchu that the bottlenecks not only affect individual production lines, but also large parts of the upstream supply chain.

Demand is primarily driven by AI accelerators, server processors and customer-specific ASICs for hyperscalers. Nvidia, AMD, Apple and other major customers increasingly require chips from the latest process generations, in which production capacity, yield, packaging and substrates are closely interlinked.

AI Chips Drive Capacity Utilization

TSMC is attempting to increase the utilization of existing plants and build up new capacities at the same time. The production volume is already visibly increasing: after 3.3 million wafers exposed in the first quarter of 2025, it was just under four million in the fourth quarter and around 4.3 million 12 inches wafers or equivalents from older 8 inches lines in the first quarter of 2026.

Despite this ramp-up, demand remains greater than supply. "It will be a long time before we can meet customer demand," said C.C. Wei. Nevertheless, TSMC still expects sales to increase by 30% this year. Furthermore, the company does not want to exploit the supply bottleneck. Wei assured investors that the company wanted to avoid sudden price increases, such as those seen in the memory and memory chip industry, in order to ensure business stability.

Wei emphasized that TSMC is working on not becoming the bottleneck of the global AI value chain itself. At the same time, he pointed out that suppliers of components, materials and equipment are also struggling with the pace of the market. For AI hardware in particular, the bottleneck is increasingly shifting towards the system level. High computing power requires not only logic chips in leading nodes, but also high-bandwidth memory, high-performance interposers, substrates and thermally robust housing technologies.

Arizona Expansion Helps, But Does Not Replace Taiwan

TSMC is investing heavily in new plants outside Taiwan, particularly in the US state of Arizona. Additional capacity for modern chips is to be created there, including further fab phases, packaging units and an R&D site. Nevertheless, Wei made it clear that it will be a long time before US production can meet the needs of American customers to any significant extent. The capacities in the new Arizona plants are already fully utilized and fully booked until the end of 2027.

An earlier target of locating 30 percent of capacity for 2 nm and smaller structures in the USA is now considered difficult to achieve. Among other things, Wei cited lengthy environmental approvals and the lack of qualified construction and skilled workers as reasons for this.

Taiwan will therefore remain the company's most important production location for the foreseeable future. This is where TSMC concentrates its most experienced production teams, central research and development and the majority of volume production. For customers, proximity to this ecosystem remains a key factor in ramp-up, yield learning and process stability.

Although global expansion reduces geopolitical and logistical risks, it does not provide any short-term relief. Establishing a leading semiconductor production facility takes years because clean rooms, process tools, material supply, engineering teams and qualification steps have to be scaled up in parallel.

High-NA EUV Remains An R&D Topic for the Time Being

Wei also commented on TSMC's stance on High-NA EUV. The CEO confirmed that the company has purchased corresponding systems from ASML and is actively using them in research and development. However, TSMC does not yet see any economic compulsion for mass production.

According to Wei, the reason for this is not a lack of interest in the technology, but the cost-benefit calculation. High-NA-EUV systems can cost around 400 million US dollars per system. TSMC therefore only wants to take the technology into production when the costs, throughput, process window and benefits for specific nodes match.

This distinguishes TSMC's strategy from Intel, which wants to integrate High-NA EUV earlier in its roadmap for future process generations. TSMC, on the other hand, is focusing on further exploiting existing EUV processes and combining multiple structuring, process integration and design technology co-optimization for as long as economically viable.

In addition to AI servers, TSMC also sees autonomous vehicles and robotics as growth drivers in the long term. Both applications require powerful, energy-efficient and robust semiconductor platforms. The current capacity pressure is therefore unlikely to be solved by additional fabs alone, but also through better process economy, more efficient designs and closer coordination between foundries, fabless providers, equipment suppliers and system manufacturers.(sg)

Lithography

Imec Receives the World's Most Advanced High-NA EUV System