CONNECT WITH US
Aug 11
TSMC finally catching up to AI packaging demand after years of doubling capacity: 'CoWoS guy' Jun He

TSMC has been roughly doubling its CoWoS and SoIC capacity annually for the past several years, and is running CoWoS at better than 98% yield — sometimes 99% — on multiple AI customer products already in high-volume manufacturing. By Jun He's account, that is finally almost enough.

The global memory industry has embraced a new business model over the past two quarters. Memory manufacturers are not only signing multi-year supply agreements with customers, but also requiring substantial cash deposits or other financial commitments as guarantees against future demand.
Nvidia and Broadcom outlined different paths for scaling AI networks at the OCP APAC Summit in Taipei on Aug. 11, as clusters expand from rack-scale systems to entire data centers and across multiple sites.
DB HiTek posted a 43% jump in second-quarter operating profit as tighter 8-inch foundry supply and stronger power semiconductor demand kept its fabs running near full capacity, with industry reports saying the company plans another round of price increases in the second half of 2026.

South Korean sensor IC designer Haechitech is moving beyond smartphone compass chips into enterprise solid-state drive temperature sensors, with supply to a major South Korean memory maker set to begin in 2026. The shift would end a single-supplier setup previously dominated by Texas Instruments and give Haechitech a foothold in the data center and memory markets.

Passive component distributor Nichidenbo said supply of MLCCs and other products remains tight as AI servers and ASIC applications continue to grow, with lead times for some products stretching to 36 weeks. The company said manufacturers are shifting capacity toward high-end applications, crowding out standard-spec products.

AI computing demand is rapidly driving up chip integration complexity, pushing advanced packaging competition beyond single-device performance and into overall reliability and validation at the packaging, board, system, and even rack levels. Speaking at the 2026 OCP APAC Summit keynote on August 11, 2026, TSMC Vice President of Advanced Packaging Technology and Service Jun He said AI systems now demand far higher quality, reliability and development speed, making traditional component-only qualification insufficient and forcing 3DIC verification methods and development flows to be redefined from the system level.

Siliconware Precision Industries (SPIL), a subsidiary of ASE Technology Holding (ASEH), broke ground on a new plant in Douliu, Yunlin County, Taiwan, amid surging global demand for artificial intelligence (AI) and high-performance computing (HPC). The plant will cover an area of approximately six hectares and is expected to introduce chip-on-wafer-on-substrate (CoWoS) advanced packaging technology, with a total investment of about NT$100 billion (approx. US$3.1 billion). The first phase is scheduled to begin operations in 2028.

South Korea is moving to reinforce its semiconductor materials and components ecosystem as chipmakers accelerate capacity expansion and the government increases financial support for suppliers.

China's effort to advance memory technology has gained a new research avenue, with a Fudan University team developing a room-temperature single-electron non-volatile memory device based on two-dimensional materials.

Lumentum Holdings said it is pulling forward capacity investments across lasers, transceivers, and optical circuit switches (OCS) after fiscal fourth-quarter revenue surged 109% year-over-year to US$1.01 billion, with the company now guiding to hit its long-stated US$1.25 billion quarterly revenue target more than a quarter ahead of schedule.

OCP: ASE unveils AI packaging trilogy
Aug 12, 10:05

Demand for AI infrastructure is forcing chipmakers and data-center operators to rethink how systems are built, with implications for computing costs, energy use, and bandwidth worldwide. At the 2026 OCP APAC Summit, ASE said next-generation packaging, optics, and power delivery will be critical to scaling AI systems for global markets.