TSMC Expands CoW Packaging Outsourcing as Hanmi Semiconductor and Other Equipment Makers Set to Benefit First

TradingKey - On August 4, Asia-Pacific time, according to a report by South Korea's Electronic Times, TSMC ( TSM) is opening up CoW (Chip on Wafer), the most core packaging process for AI chips, on a large scale to outsourcing partners such as ASE.

CoWoS packaging has long been a critical bottleneck constraining AI chip capacity, and the CoW process, which carries the highest technical difficulty, was previously handled almost exclusively by TSMC. Now that TSMC has decided to open up this process to outsourcing partners, it suggests that the tightest bottleneck on the AI chip supply side may begin to ease.

What Is CoWoS Packaging?

CoWoS is TSMC's 2.5D packaging technology, and almost all AI chips rely on this process. The core is to connect the GPU or AI processor with HBM memory through a silicon interposer. The entire process is divided into two steps: the first step, CoW, attaches the chips to the interposer; the second step, WoS, bonds the interposer to the substrate.

Why TSMC Is Expanding CoW Packaging Outsourcing

In the past, TSMC's strategy was relatively conservative. The WoS process, which has a lower technical barrier, has long been outsourced to packaging and testing plants like ASE and Amkor ( AMKR ), but the CoW process, which requires high precision and presents significant yield control challenges, has been primarily managed by TSMC itself, with a highly limited outsourcing ratio. TSMC's inclusion of the CoW process in its large-scale outsourcing scope this time is intended to relieve its own pressure by leveraging external capacity.

The fundamental driving force behind this shift is the continuous expansion of demand for AI chips. AI chip design companies represented by Nvidia ( NVDA) have full order books, while a large number of AI cloud vendors have begun self-developing custom chips, driving overall demand continuously upward.

TSMC currently accounts for about 90% of the global AI chip manufacturing market. Although its packaging capacity has been continuously ramping up in recent years, the expansion speed of the CoW process has struggled to keep pace with demand growth. Over the past two years, AI chip delivery cycles have been bottlenecked by the packaging stage on more than one occasion. TSMC's decision to ease restrictions on CoW outsourcing this time is essentially dispersing packaging pressure across the entire semiconductor packaging and testing industry chain.

Who Stands to Benefit as TSMC Expands CoW Packaging Outsourcing?

The expansion of outsourcing for the CoW process will most directly benefit equipment suppliers for back-end processes. Outsourcing providers undertaking the CoW process need to build new or expand existing production lines, leading to a subsequent surge in equipment procurement, with demand being most concentrated in equipment for wafer and interposer dicing and bonding processes.

According to South Korean media reports, outsourcing providers such as ASE are already in procurement talks with local South Korean equipment companies regarding laser processing and bonding equipment. The unit cost of such equipment is typically over $1 million, and once these orders are finalized, equipment makers' financial performance will directly benefit.

South Korea has accumulated significant expertise in the semiconductor back-end equipment sector, particularly maintaining competitiveness in steps such as dicing and die bonding, making it poised to be among the first beneficiaries of this expansion in outsourcing. For instance, companies like Hanmi Semiconductor, Avaco, and Wonik are all expected to secure orders in this wave of equipment procurement.

For AI chip buyers, TSMC's outsourcing of the CoW process is a positive signal. As packaging capacity shifts from being solely supported by a single player to being shared among multiple partners, overall supply elasticity will significantly improve. Although shortages will not disappear immediately, supply chain pressures are easing, and the packaging bottlenecks that have long hindered AI chip deliveries are showing signs of loosening.