TSMC Expands Advanced Chip Packaging Capacity, Narrowing Supply Gap

Recent reports indicate that TSMC is accelerating the expansion of its advanced packaging capabilities using CoWoS technology, a key method for assembling high-performance artificial intelligence chips. This technology integrates modern graphics processing units and HBM memory into a single package. With this expansion, the supply-demand gap is projected to shrink from approximately 20% currently to nearly 10% by the end of 2026, with expectations for further improvement during 2027.

A report published on trendforce forecasts that TSMC's monthly CoWoS production capacity will reach between 120,000 and 140,000 chips by 2026. Additionally, specialized OSAT packaging companies are expected to contribute between 50,000 and 60,000 chips monthly, bringing the total global production capacity to nearly 200,000 chips per month. This level of output is unprecedented in this critical sector.

Relieving Supply Chain Pressure in AI Chip Market

The significant increase in capacity is expected to ease the severe strain on the advanced chip market, especially given the rapid demand growth for AI accelerators. Estimates suggest that the shortage crisis in 2.5D three-dimensional packaging capabilities will gradually moderate as new production capacities come online and begin to absorb some of the deferred demand.

Historic Growth in Investment and Industrial Expansion

During a technology event in Taiwan, TSMC highlighted that CoWoS capacity is projected to achieve a compound annual growth rate exceeding 80% between 2022 and 2027. This reflects a structural transformation in the semiconductor industry, with production capacity continuing to expand by more than 60% through 2027.

Development of Next-Generation CoPoS Packaging

TSMC is also developing a new platform called CoPoS (Chip-on-Panel-on-Substrate), designed to overcome current chip size limitations. An experimental research line was established at VisEra in 2025, with material and equipment testing expected to conclude by mid-2026. Pilot production is scheduled for 2027, paving the way for commercial manufacturing between 2028 and 2029.

Anticipated Adoption by Leading AI Companies

Projections indicate that NVIDIA's upcoming platform, known as Feynman, will be the first to adopt this new CoPoS technology. Plans include extensive deployment across the company's manufacturing facilities in Taiwan and the United States.