TSMC CoWoS capacity shortage forces orders to competitors

๐กUnderstand how the AI hardware supply chain bottleneck is shifting market share toward Intel and local competitors.
โก 30-Second TL;DR
What Changed
TSMC's CoWoS capacity is currently insufficient to meet AI/HPC market demand.
Why It Matters
The bottleneck in advanced packaging may slow down the deployment of high-end AI hardware. Diversification of supply chains to Intel and other vendors is becoming a strategic necessity for AI chip designers.
What To Do Next
If you are procuring AI hardware, evaluate vendors that utilize alternative packaging solutions to mitigate TSMC supply chain risks.
Key Points
- โขTSMC's CoWoS capacity is currently insufficient to meet AI/HPC market demand.
- โขIntel is capturing overflow orders due to TSMC's supply constraints.
- โขTaiwanese local OSAT (Outsourced Semiconductor Assembly and Test) providers are benefiting from the supply chain shift.
๐ง Deep Insight
AI-generated analysis for this event.
๐ Enhanced Key Takeaways
- โขTSMC has been aggressively expanding its CoWoS (Chip-on-Wafer-on-Substrate) capacity, aiming to more than double output by the end of 2026 compared to 2024 levels to mitigate bottleneck issues.
- โขThe surge in demand is primarily fueled by the transition from H100/B100 series GPUs to next-generation AI accelerators that require larger interposers and higher HBM (High Bandwidth Memory) stack counts.
- โขIntel Foundry's advanced packaging services, specifically Foveros and EMIB (Embedded Multi-die Interconnect Bridge), are being positioned as the primary alternative for high-performance computing clients seeking to diversify supply chains.
- โขMajor OSAT players like ASE Technology (Advanced Semiconductor Engineering) are investing heavily in 'VIPack' and other 2.5D/3D packaging platforms to absorb overflow demand that TSMC cannot accommodate.
- โขThe supply constraint has led to a 'packaging premium,' where clients are increasingly signing long-term capacity reservation agreements (LTAs) to secure CoWoS slots, effectively locking in supply for 18-24 months.
๐ Competitor Analysisโธ Show
| Feature | TSMC (CoWoS) | Intel (EMIB/Foveros) | ASE (VIPack) |
|---|---|---|---|
| Primary Tech | 2.5D/3D Silicon Interposer | EMIB (Bridge) / Foveros (3D) | 2.5D/3D Fan-Out / SiP |
| Maturity | Industry Gold Standard | High (Scaling rapidly) | High (Cost-effective) |
| Best For | Ultra-high performance AI | Heterogeneous integration | High-volume, cost-sensitive |
| Pricing | Premium (Highest) | Competitive (Bundled) | Moderate |
๐ ๏ธ Technical Deep Dive
- CoWoS (Chip-on-Wafer-on-Substrate) utilizes a silicon interposer to connect multiple dies (GPU/CPU + HBM) with high-density wiring, enabling high-speed communication between memory and logic.
- Intel EMIB (Embedded Multi-die Interconnect Bridge) uses a small silicon bridge embedded in the package substrate to connect dies, reducing the need for a large, expensive full-size silicon interposer.
- Foveros 3D packaging allows for face-to-face die stacking, significantly reducing the vertical distance between logic and memory to improve power efficiency and latency.
- ASE VIPack platform supports various configurations including FOCoS (Fan-Out Chip on Substrate) and 2.5D integration, targeting a broader range of thermal and electrical performance requirements.
๐ฎ Future ImplicationsAI analysis grounded in cited sources
โณ Timeline
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: cnBeta (Full RSS) โ


