PDF 原檔：報告_MS_SEMICON台灣2026觀察_20260906_original.pdf
圖片清單（已驗證 2026-09-09）
ingest 時建立的「眼見為憑」圖片索引，是 lib/ 嵌圖的唯一真相來源。
| 檔名 | size | 分類 | 親眼所見內容 |
|---|---|---|---|
報告_MS_SEMICON台灣2026觀察_20260906_001.png |
401KB | 真資料圖 | 展場實拍：光學平台上的主動式耦光對位設備，含六軸 hexapod 平台、多組微調滑台與鏡頭；背板文字「奈米級主動式耦光對位 / Nano-Precision Active Optical Alignment & Coupling」 |
報告_MS_SEMICON台灣2026觀察_20260906_002.png |
408KB | 真資料圖 | 展場實拍：TOYO（東佑達）與 SURUGA SEIKI（日商駿河）共同展出的「CPO 系列模組」壓克力展示櫃，內為六自由度光耦合對位模組，標示型號 MCRB-XY10Z5-R100；背板列 CPO／矽光子／TAU 應用與 Piezo 等關鍵字 |
報告_MS_SEMICON台灣2026觀察_20260906_003.png |
354KB | 真資料圖 | 展場實拍：Winbond（華邦電）「Customized Memory Solution / CUBE・3DCaaS」看板，並列 3D-CoW（µBump）與 3D-WoW（Hybrid Bonding） 結構圖，下方表格列「4-Layer WoW Hybrid Bonding」與「High-Density Silicon Capacitor」兩項成果 |
報告_MS_SEMICON台灣2026觀察_20260906_004.png |
282KB | 真資料圖 | 展場實拍：Phison（群聯）自研 AI NPU 晶片 Topaz 展示盒與「Topaz U.2 Solution」兩片實體板卡；說明卡列 TSMC 6nm、Arm Cortex-A55 八核、四核 NPU 等規格 |
報告_MS_SEMICON台灣2026觀察_20260906_005.png |
233KB | 真資料圖 | 展場實拍：Toray Engineering 攤位「PLP 製程相關設備」三張海報並列（Panel Level Package 解決方案、TRENG-PLP Coater、Flux-less Bonding UC5000） |
報告_MS_SEMICON台灣2026觀察_20260906_006.png |
362KB | 真資料圖 | 展場實拍：Toray Engineering「Flux-less Bonding Solution Based on UC5000」海報，載明 TCB 接合精度 ±0.8µm、無助焊劑的 safety-first 製程流程圖與 large-format panel 應用 |
原始內容
M September 6, 2026 05:24 PM GMT
Greater China Semiconductors | Asia Pacific
SEMICON Taiwan 2026 key takeaways - we need more
AI semi capex spending in Taiwan remains strong into 2027 to solve the supply chain bottleneck. Keep OW on TSMC, MediaTek, Hon Precision, Allring, MPI, Winway, etc.
Titans of Taiwan AI semi supply chain - key message from the CEO Forum
fireside chat Taiwan is still one of the bottlenecks of AI infrastructure buildout. On the stage, TSMC Deputy Co-COO Mr. Hou said that he had never seen demand growing and changing this frequently as it has over the past six months, while the tool delivery from WFE partners are still not enough. MediaTek CEO Mr. Tsai was asking for more capacity support from ASE, TSMC, and Unimicron.
Why investors came to this trade show in Taiwan: We believe the tech industry and capital markets are finding ways to keep AI hardware spending sustainable in coming years. With memory becoming the major bottleneck and cost item for AI capex, investors are checking whether technologies can improve AI computing efficiency, for example, CPO to replace copper connectors in XPU scale-up. All in all, the goal is to reduce per-token cost. Please see following section for details.
Key focus areas during SEMICON Taiwan:
- Silicon photonics for AI networking and testing process development : In our previous report, AI Supply Chain: GPU Architecture Optimization Given Memory Shortage and Supernode Challenge, we indicated that future GPU/ ASIC scale-up would depend on CPO/NPO ('inter-rack CPO connectivity remains Nvidia's preferred direction'). For example, the NV576 Rubin Ultra in 2027 may use CPO switch trays to connect eight NVL72 server racks. Silicon Photonics and PIC (Photonic integrated circuit), lasers, CPO, and related testing present large investment opportunities.
- Memory optimization from high cost : Instead of adding density, increasing bandwidth and fully utilizing existing memory resources has become the key technology trend. Our recent supply chain checks suggest that Rubin Ultra CPU is highly likely to use 8-Hi HBM4, around one-third fewer than 12-Hi in Rubin. But by increasing HBM4 base-die units and migrating to HBM4e, plus some software optimization, total memory bandwidth should be sufficient for a token output increase.
- "More than Moore"Panel-level packaging and Intel EMIB-T - We visited the Toray (Toray Industries covered by Takato Watabe) booth to check EMIB-T substrate manufacturing yield, because the company supplies the TCB bonder. We expect Intel to provide technical support, and expect MediaTek's TPU v9 (HumuFish) to get sufficient EMIB-T substrate supply in 2028.
- Although it wasn't the main focus of the SEMICON , we think TSMC' s Moore's Law migration (for energy efficiency) and MediaTek' s ASIC enabling (to save on chip budgets) are keys to make AI capex more sustainable.
Idea
| Morgan Stanley Taiwan Limited+ Charlie Chan Equity Analyst Charlie.Chan@morganstanley.com | +886 2 2730-1725 |
|---|---|
| Daniel Yen, CFA Equity Analyst Daniel.Yen@morganstanley.com Morgan Stanley Asia Limited+ | +886 2 2730-2863 |
| Daisy Dai, CFA Equity Analyst Daisy.Dai@morganstanley.com Morgan Stanley Taiwan Limited+ | +852 2848-7310 |
| Tiffany Yeh Equity Analyst Tiffany.Yeh@morganstanley.com | +886 2 7712-3032 |
| Lucas Wang Research Associate Lucas.Wang@morganstanley.com Morgan Stanley Asia Limited+ | +886 2 2730-2875 |
| Henry Zhao Equity Analyst Henry.Zhao@morganstanley.com | +852 2239-7731 |
Greater China Technology Semiconductors
Asia Pacific
Industry View
Attractive
Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision.