CoWoS Pricing Tops $10,000, Becoming Key Driver of TSMC's AI Profits
TSMC's CoWoS technology integrates computing chips with high-bandwidth memory and is crucial to boosting the performance of AI accelerators from companies such as NVIDIA. Surging demand for AI chips has transformed advanced packaging from a traditionally low-margin assembly and testing segment into a high-value competitive market central to delivery capacity and profitability.
As of July 20, 2026, the average price of a CoWoS wafer had exceeded $10,000, making the technology a new profit engine for TSMC's AI business. TSMC is expanding CoWoS capacity and investing in SoIC and silicon photonics, while Intel is seeking market share with its EMIB packaging technology.
All Coverage
1 original reportsThe Backstory
The history behind this eventTSMC Mass-Produces 5.5-Reticle CoWoS, Targets 14x Scale by 2029
Demand for generative AI and high-performance computing is pushing chipmakers beyond the size limits of individual dies, increasing reliance on advanced packaging. TSMC’s CoWoS technology combines processors and high-bandwidth memory in a single package, making it central to AI accelerator production. Scaling those packages, however, raises system-level challenges including heterogeneous integration, material compatibility, warpage control and sufficient capacity across the supply chain.
TSMC said at the OCP APAC Summit that its 5.5-reticle-size CoWoS package has entered mass production, with yields holding above 98% and reaching as high as 99%. The company plans to scale the technology to 14 times reticle size by 2029 while addressing integration and materials constraints. TSMC also identified ABF substrates as a potential next bottleneck after memory, highlighting shifting supply pressures as AI packages grow larger and more complex.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →