TSMC Reportedly Considers 20% Price Increase for CoWoS Advanced Packaging

Potential CoWoS Price Hike

TSMC is reportedly evaluating a price increase of up to 20% for its Chip on Wafer on Substrate (CoWoS) advanced packaging technology. This consideration stems from significant capacity strain, driven by high demand for AI accelerators and extended delivery times for essential chipmaking equipment. CoWoS is a crucial technology for integrating logic dies and high-bandwidth memory (HBM) stacks, which are essential for the data bandwidth required by demanding AI workloads.

AI Chip Demand and Packaging Pressure

The demand for advanced packaging is intensifying with the development of new AI chips. Nvidia's Blackwell GPUs, for instance, combine two large GPU dies with eight HBM3E stacks. Upcoming chips like Rubin and Rubin Ultra are expected to further increase integration, placing additional pressure on packaging capacity. Beyond GPUs, server CPUs and custom ASICs also compete for advanced packaging resources, contributing to an already crowded supply chain.

Testing and Equipment Bottlenecks

Testing processes are emerging as another bottleneck. The testing duration for Nvidia's Blackwell reportedly takes three to four times longer than its predecessor, Hopper, with Rubin potentially requiring up to seven times longer. These extended test cycles tie up equipment, hindering efforts to increase output. Additionally, lead times for chipmaking equipment have reportedly stretched from three to six months to as long as eight months, making timely access to tools a priority for TSMC.

Shift in Equipment Procurement Strategy

Due to the urgency of securing equipment, TSMC has reportedly placed its current year's equipment orders at existing prices, foregoing its usual practice of negotiating discounts or additional features. This shift indicates that obtaining tools on time is currently more critical than cost savings. This situation echoes earlier reports from September, which indicated that TSMC's projected semiconductor equipment requirements had nearly doubled in six months to meet escalating AI demand.