Damnang
· Damnang’s Substack
· May 11, 2026 at 09:56
· ⏱ 8 min read
| Read on Substack ↗
Summary
The OSAT (Outsourced Semiconductor Assembly and Test) segment is the next undervalued bottleneck in the AI chip value chain because the physical complexity of AI chips (e.g., Nvidia Blackwell with CoWoS-L) forces advanced packaging that exceeds TSMC's capacity, leading to structural overflow and higher-margin work flowing to specialist OSATs. This re-rating opportunity persists through at least 2027-2028 as new CoWoS lines take 18-24 months to build, and hyperscalers actively diversify supply chains away from sole reliance on TSMC.
•TSMC's CoWoS capacity is estimated at 75,000-80,000 wafers/month currently, expanding to 120,000-130,000 by end of 2026, but demand from Nvidia, AMD, Google, and Amazon fully books that capacity, creating a persistent gap.
•Building a new CoWoS line takes 18-24 months, so the capacity shortage will extend through at least 2027-2028, driving overflow to OSATs.
•OSATs are now receiving not just low-margin on-Substrate assembly but also higher-value processes like CoW-related bonding and RDL structures, with sell-side estimates of ~40,000 wafers/month outsourced CoWoS volume.
•The article splits OSAT into four types: Advanced Packaging (CoWoS, 2.5D interposer), Memory Packaging (DRAM/NAND/HBM overflow), Test-Specialized (SLT for chiplet architectures), and Optical Packaging (CPO/silicon photonics, volume post-2028).
•Test-specialized OSATs can achieve EBITDA margins above 40% versus single-digit margins for commodity assembly, highlighting a wide profit disparity within the same segment.
•Hyperscalers (Nvidia, Google, Amazon) are voluntarily diversifying OSAT allocations away from 100% TSMC dependence, creating structural demand separate from overflow.
Article quantifies TSMC's CoWoS capacity tightness (75-80k wafers/month, fully booked through 2027-2028) and notes that building new lines takes 18-24 months, making TSMC's advanced packaging a struct
Article quantifies TSMC's CoWoS capacity tightness (75-80k wafers/month, fully booked through 2027-2028) and notes that building new lines takes 18-24 months, making TSMC's advanced packaging a structural bottleneck that underpins pricing power and revenue growth.
Risk: Capacity expansion may come faster than expected if competitors like Samsung or Intel gain share in advanced packaging.
Amkor is explicitly named as one of the global OSATs directly benefiting from TSMC's CoWoS overflow and is investing in optical packaging for CPO/silicon photonics, positioning it as a Type 1 Advanced
Amkor is explicitly named as one of the global OSATs directly benefiting from TSMC's CoWoS overflow and is investing in optical packaging for CPO/silicon photonics, positioning it as a Type 1 Advanced Packaging OSAT with structural demand through at least 2027-2028.
Risk: Execution risk in qualifying for foundry-grade advanced packaging; customer concentration if a single hyperscaler dominates allocation.
ASE (ASE Technology) is named alongside Amkor and JCET as a global OSAT investing in advanced and optical packaging capabilities, making it a direct beneficiary of the structural shift from commodity
ASE (ASE Technology) is named alongside Amkor and JCET as a global OSAT investing in advanced and optical packaging capabilities, making it a direct beneficiary of the structural shift from commodity assembly to higher-margin CoWoS and SLT work.
Risk: ASE has a large commodity assembly base that could drag overall margins; optical packaging is pre-revenue until post-2028.
This newsletter, published May 11, 2026,
features Damnang
discussing TSM, AMKR, ASE.
3 trade ideas extracted by AI with direction and confidence scoring.