PDF 原檔：報告_Aletheia_旺矽6223_20260911_original.pdf
訂閱戶限定素材：不得對外、不上站
原檔頁尾明載：「This report is intended solely for the subscriber. Unauthorized sharing, distribution, or reproduction may violate applicable regulatory requirements and could result in legal action.」 本頁僅供 vault 內部參考。引用至 lib/ 或對外發布前須經使用者確認。 另注意：報告多處描述某未具名台系同業的品質問題與訂單流失，屬敏感內容，對外引用風險高於一般券商報告。
本檔為純圖像 PDF（3 頁、無文字層）
以 150 dpi 轉圖後由人工逐段轉錄，非 OCR。p2 為 Aletheia 研究團隊介紹頁、p3 為免責聲明，均無實質內容。下方為 p1 全文轉錄。
原始內容（p1 人工轉錄）
Aletheia Capital｜11 September 2026｜Technology Hardware｜Taiwan MPI Corp — #Flashnote — Buy — α Portfolio Analyst: Angus Lin (+852 9250 7088, angus.lin@aletheia-capital.com)
Winning Bigger: A Clean Sweep of Vera CPU and Mellanox
MPI's winning streak continues, as it takes further share at both Nvidia and Mellanox, amplified by share losses at a Taiwanese peer. That vendor – formerly the lead probe card supplier – has already lost Nvidia's Vera CPU (TB500) business outright and is now excluded from the next-gen Vera CPU (TB600). At Mellanox, MPI has secured near-exclusive share across all three networking devices (ConnectX, SpectrumX, QuantumX) at the same competitor's expense, and is already engaged on the next-gen projects – from which that competitor is again excluded. MPI is our Alpha Portfolio stock on the rising chip probing trend; our TP remains at NT$12,000.
Vera CPU: In our early-June report we flagged MPI's meaningful share gain in Nvidia's Vera CPU (TB500). MPI was initially allocated only 30%, with a Taiwanese peer holding the remaining 70%. Following quality issues at that peer — stemming from weaker support by its probe needle partner and its lack of turnkey experience — MPI has become the near-monopoly supplier on this project. Our recent checks indicate the peer will ship its final batch of Nvidia orders this month, with no follow-on orders, while MPI continues to ramp deliveries. On the next-generation Vera CPU (TB600), MPI is likely to be the sole probe card vendor, with the peer excluded entirely for now. Nvidia did approach a second Taiwanese vendor for Vera CPU probe cards, but that vendor was disqualified on quality grounds.
Mellanox: The current generation of Mellanox devices (ConnectX, SpectrumX, QuantumX) was originally split roughly evenly between MPI and the same Taiwanese competitor. For the same quality and support reasons, MPI has since absorbed nearly all of that allocation. On the next generation, we expect MPI to share SpectrumX with TSMC's in-house probe cards and to be the sole vendor on ConnectX and QuantumX, with the peer again excluded for now.
MPI is our top pick to capture the massive chip probing proliferation: TSMC is more than doubling its leading-edge (N3 and below) capacity additions next year (+150kwpm in FY27E vs +70kwpm in FY26E). That incremental wafer volume reaches the chip probing stage in FY28E, given the c.9-month lag from front-end capacity installation to wafer-out. We regard probe cards — a semi-consumable — as essential strategic exposure within semiconductor testing, and MPI as the best way to own it.
We also believe the Street materially underestimates MPI's FY27E earnings power: excluding CPO, we see EPS of NT$170-180, assuming capacity covers only 80% of total demand (10-11mn VPC/MEMS monthly pins, versus company guidance of 8.5mn by end-FY27E, which we view as a guide-to-beat). Including CPO, and assuming a smooth ramp that fully meets pent-up demand, EPS would reach NT$206 in FY27E — leaving MPI on just 25-30x our pre- and post-CPO earnings base, versus a peak multiple of 50x.
Furthermore, we believe consensus still fails to capture (1) the meaningful cantilever probe card ASP hike and (2) the significant ASP uplift from a more favourable VPC/MEMS mix (Nvidia/AMD/Mellanox), leaving Street FY27E earnings estimates 30% below ours.
Lastly, despite the negative rumblings around MPI's CPO progress, our latest checks suggest the progress remains on track: MPI is currently validating its full insertion-2 system (probe cards plus probers) at the foundry site, while its closest competitors may only complete their gen-2 prober by year end and would still face mass-production qualification thereafter — a timeline likely no earlier than MPI's.