Packaging/CoWoS/HBM shortages
Key Questions
What is the current status of TSMC's CoWoS capacity?
CoWoS capacity of 130k wafers per month is fully sold out with a 30% supply gap, driving structural demand. TSMC is expanding via a new CoPoS pilot line in June and high-volume manufacturing targeted for 2029-2030.
How is TSMC advancing glass substrate technology?
TSMC is scaling glass substrates from 90x90mm to 120x150mm by 2028 and exploring glass core with partners, though first-gen CoPoS is expected to remain glass-free until after 2030. Nvidia's Feynman program may adopt TSMC glass packaging.
What partnerships support TSMC's US packaging expansion?
TSMC signed a 10-year pact with Amkor and secured approval for a $20B Arizona advanced packaging investment as part of a $21B+ US AI ecosystem push with Quanta and Lite-On. This addresses CoWoS and HBM bottlenecks.
How do HBM and memory constraints affect packaging?
Micron's HBM4 revenue exceeding $1B signals gradual easing of memory supply, while TSMC's Winbond DRAM deal provides insurance against AI memory bottlenecks. Goldman Sachs projects TSMC 2027 capex at $78B with CoWoS capacity doubling.
What is the market outlook for glass substrates and panel packaging?
Glass substrates and FO-PLP markets are projected to grow 10x by 2030, with TSMC's CoPoS and TGV yield challenges as key factors. Samsung Electro-Mechanics is forming a competing glass core JV with Sumitomo targeting 2H27.
How does advanced packaging serve as a moat for TSMC?
Citi highlighted advanced packaging as a key differentiator beyond GPUs, expecting raised 2026 growth and $75-80B capex. The AI supply chain bottleneck analysis confirms TSMC's near-monopoly position in CoWoS and related technologies.
What is TSMC's strategy for wafer-on-wafer stacking?
TSMC is collaborating with Winbond on wafer-on-wafer stacking to address AI memory constraints, positioned as risk mitigation rather than market domination. This supports the broader 'second fleet' supply chain ecosystem approach.
How do hyperscaler AI spends reinforce packaging demand?
Projected $650B hyperscaler AI capex in 2026 combined with a 30% CoWoS gap underscores structural constraints in foundry, packaging, and memory. A Digitimes report notes TSMC's capacity bottleneck is spilling demand across the entire supply chain.
CoWoS 130k wpm sold out. CoPoS pilot line June, HVM 2029-2030; TSMC betting on glass. TSMC-Amkor 10-year U.S. pact. Nvidia Feynman may adopt TSMC glass packaging. TSMC touts glass substrate gains: 90x90mm to 120x150mm by 2028. Kuo deep dive on CoPoS: real value in oS for power integrity; timeline divergence TSMC (4Q28-1Q29) vs Ibiden (CY30) key risk. Nvidia CPO roadmap uses TSMC COUPE. Micron HBM4 revenue >$1B signals memory supply easing for packaging. Q1'26 foundry 2.0 market $86B, packaging as key bottleneck. New: TSMC working with Winbond on wafer-on-wafer stacking to fix AI memory bottleneck (reframed as insurance, not domination). TSMC, Quanta, Lite-On inject $21B+ into US AI ecosystem, including $20B Arizona infusion for advanced packaging. Glass substrate and FO-PLP market projected 10x growth by 2030, with TSMC's CoPoS and TGV yield as critical challenges. TSMC's supply chain 'second fleet' strategy reinforces packaging ecosystem moat. New insight from AI Capex Money Map: TSMC's packaging captures real dollars even as Nvidia takes IP margin; bottleneck migration from chip to power/site is key. Updated Jul 5: A supply chain article debunks the 'glass myth' around CoPoS, clarifying that first-gen CoPoS likely won't use glass substrates until after 2030. New data: CoWoS supply gap 30%, $650B hyperscaler spend in 2026 confirms structural demand. Goldman Sachs raised TSMC 2027 capex to $78B, with CoWoS capacity doubling. Earnings preview (Jul 5) reiterates CoWoS supply gap and hyperscaler spend data. Latest (Jul 7): Citi highlights advanced packaging as a key moat, reinforcing the packaging narrative. A detailed article on data center AI bottlenecks (Jul 5) reinforces TSMC's near-monopoly and CoWoS growth. New competitive pressure: Samsung Electro-Mechanics forming glass core JV with Sumitomo, targeting 2H27 production, challenging TSMC's glass substrate roadmap. TSMC's Winbond DRAM deal reframed as insurance, not domination, adding geopolitical and ETF angles. New today (Jul 7): A Digitimes article confirms TSMC's capacity bottleneck spilling demand across the supply chain, reinforcing the structural constraint. Counterpoint Foundry 2.0 report confirms packaging as key bottleneck.