Greater China Semis: AI Packaging Innovation – CoWoS, CPO and WoW as Structural Growth Multipliers

Core Thesis

The three-pronged AI packaging revolution—CoWoS capacity expansion, CoPo adoption, and WoW 3D stacking—creates a multi-year growth opportunity for Greater China semiconductor supply chain players. By 2030, the combined TAM for these three packaging technologies is projected to exceed $50bn, driven by agentic AI computing demands. The market underappreciates that packaging is becoming a bottleneck as valuable as leading-edge lithography.

AI semi TAM is estimated at $485bn in 2026, and the global semiconductor market could reach $1.5tn by 2030 with AI contributing half. This demand directly feeds packaging innovation: larger chips, higher pin counts, and lower power interconnect requirements.

Market Underappreciation

The prevailing focus is on GPU die shrinks and ASIC design wins. However, packaging innovation is now determining system-level performance and cost. CoWoS capacity, CPO integration timelines, and WoW’s BOM savings are under-modeled in supply chain earnings estimates. The shift from 2.5D to 3D stacking and optical interconnects creates new monopoly-like positions for enablers such as TSMC, ASE, and niche equipment suppliers.

Evidence Chain

CoWoS Expansion – TSMC is expected to double CoWoS+SoIC capacity in 2025, with CoWoS capacity reaching 165k wpm by 2027 (vs. ~80k wpm in 2025). This expansion supports Nvidia’s Blackwell/Ultra, Broadcom’s AI ASICs, and AMD’s MI-series. The larger interposer roadmap (9.5× reticle) enables >4 chips per package, raising test complexity 35% CAGR in test equipment demand (2024–27e).

CPO TAM Growth – Broadcom is first to mass-produce CPO with a switch solution (using TSMC’s COUPE technology). Nvidia’s Rubin server rack system is expected to be the first CPO with an XPU. Base case CPO TAM reaches $9bn by 2030 (172% CAGR from 2023). Key supply chain partners: FOCI (3363.TWO), TSMC, and SPIL.

WoW 3D Stacking – WoW (wafer-on-wafer) replaces 2.5D CoWoS for bandwidth/power improvement. A discrete NPU costing ~$80 can be integrated into SoC via WoW, saving BOM while improving inference performance. WoW TAM base case reaches $6bn by 2030 (257% CAGR 2025–30e). Direct beneficiaries include memory players and foundries; indirect beneficiaries include testing and bonding equipment providers.

Competitive Dynamics – Intel’s EMIB can theoretically support >12 reticles, but TSMC’s CoWoS-S with silicon interposer maintains a cost advantage at scale. Package size trend and rising test socket pin counts favor pure-play testing equipment firms.

Key Risks & Divergence

• Intel EMIB execution: If Intel rapidly scales EMIB capacity, it could fragment the CoWoS supply, depressing pricing and lowering utilization for TSMC’s partners.
• CPO yield and adoption: Optical integration adds cost and complexity; mass adoption hinges on Nvidia’s Rubin ramp. Delay would compress CPO TAM growth.
• WoW memory supply constraints: WoW requires DRAM stacking on logic; memory pricing volatility or technology maturity delays could limit TAM.
• Export controls: U.S. restrictions on advanced packaging equipment to China could disrupt supply chain flows for non-AI applications.

Valuation & Trade Implications

• TSMC (2330.TW) remains the essential platform. Its packaging revenue CAGR of 60% (2024–29e) justifies a premium multiple versus historical ranges.
• ASE Technology Holding (3711.TW) and SPIL are the primary CoWoS/CPO subcontractors. ASE’s AI-related packaging revenue is expected to triple by 2027.
• FOCI (3363.TWO) is the only pure-play CPO fiber optics supplier with Nvidia and Broadcom exposure; base case implies 172% TAM CAGR.
• Test equipment players (e.g., AllRing Tech, MPI, WinWay) benefit from >35% testing equipment CAGR as chip size and complexity grow.
• WoW enablers such as memory foundries and bonding equipment vendors have asymmetric upside if adoption accelerates.

Valuation risk: Premium valuations already reflect some of the growth. Downside sensitivity to CoWoS/CPO volume growth of 10% would reduce aggregate TAM estimates by ~15% in 2027. Investors should focus on companies with proprietary tools or capacity contracts rather than pure exposure.

Appendix Data Summary (Compressed)

Note: TAM figures from industry tracking; capacity data from company disclosures.