公司

AI-specialized hyperscaler. NVIDIA holds 47.21M shares ($3.66B at Q1-2026), NVDA's second-largest disclosed stake. Runs NVIDIA GPU fleets for hyperscaler customers; benefits from Rubin generation rollout.

Dutch AI-cloud provider (Yandex spin-off). NVIDIA stake ~$123M at Q1-2026. Operates GPU clusters across EU for AI training; emerging CoreWeave-style customer of Rubin generation.

Leading EDA vendor (chip-design software + IP). NVIDIA Q1-2026 13F disclosed stake. Synopsys tools are used to design Rubin, Vera CPU, and ConnectX-9 NIC silicon — strategic supplier on the design side.

Low-Dk glass-cloth maker. 據傳 passed 台光電 certification as a second Low-Dk2 glass-cloth source (after Japan''s Nittobo) for Rubin-gen CCL; 雲林虎尾 + Thailand capacity expansion. RUMOR-grade attribution per CTEE 2026-01-17.

Copper-clad laminate (CCL) leader. Rubin-gen mix shifts to high-margin M9 for backplane/switch tray + high-volume M8.5 for UBB/compute tray; described as the first/only M9-certified supplier (CTEE / Goldman Sachs, 2026-01).

InP / GaAs epitaxial-wafer maker for optical comms. Led Taiwan semiconductor revenue-growth rankings on CPO momentum: +76.2% YoY cumulative to NT$1.97bn through Nov 2025 (Digitimes/TechNews).

Optical passive components / fiber-connector maker; named in the Wiwynn CPO optical scale-up rack Taiwan ecosystem (UDN 經濟日報, Computex 2026).

Fiber-array / optical-connector maker for CPO light engines; named in the Wiwynn CPO optical scale-up rack Taiwan ecosystem (UDN 經濟日報, Computex 2026).

AI server ODM; named in NVIDIA''s Vera Rubin full-production partner roster (Computex/GTC Taipei 2026). Rubin NVL72 rack + sub-assembly.

Compound-semiconductor MBE epitaxy. +53.8% YoY (NT$0.97bn) through Nov 2025; share-swap (5.1:1) with optical-component maker EZconn/光聖 (6442) to deliver end-to-end silicon-photonics CPO from epi growth to module integration.

Cable + connector for NVLink copper interconnect.

Designer of the TPU programme; co-architects v7p/v8t/v8i and v9 with Broadcom + MediaTek; Cloud Next 2026 split TPU into training (8t) and inference (8i) SKUs.