Second Quarter 2025 Financial Results July 29, 2025 Forward-Looking Statements This presentation contains forward-looking statements that reflect the Company's current views with respect to future events and financial performance. You can identify forward-looking statements by the fact that they do not relate strictly to current or historic facts. Forward-looking statements are identified by the use of the words "aim," "believe," "expect," "anticipate," "intend," "estimate," "project," "outlook," "forecast" ...

Investment Rating - The report maintains an "Overweight" rating for GigaDevice and AP Memory, while upgrading AP Memory's price target to NT$390.00 from NT$235.00 [6][12][13]. Core Insights - The wafer-on-wafer (WoW) stacking technology is expected to significantly enhance edge AI computing capabilities by improving memory bandwidth and reducing power consumption, thus facilitating the adoption of advanced AI tools in various devices [8][10][39]. - The total addressable market (TAM) for WoW technology is projected to grow from US$10 million in 2025 to US$6 billion by 2030, indicating a compound annual growth rate (CAGR) of 257% [10][40][84]. - Specialty memory players are anticipated to benefit the most from WoW technology due to their technological readiness and market positioning [12][13][40]. Summary by Sections Industry View - The report highlights a shift in investment ratings for specific companies, with GigaDevice's price target raised to Rmb169.00 and AP Memory upgraded to "Overweight" [6]. Key Takeaways - WoW technology is set to unlock the potential of edge AI devices by addressing limitations in size, shape, and layout through 3D memory stacking [8]. - The current mainstream solution, high-bandwidth memory (HBM), is not suitable for edge devices due to cost and form factor constraints, making WoW a more viable alternative [9][10]. Market Potential - The TAM for WoW is expected to reach US$622 million by 2027, with significant growth anticipated in the automotive and consumer electronics sectors [40][84]. - The report identifies key beneficiaries of WoW technology, including niche memory players like Winbond and GigaDevice, which are well-positioned to capitalize on the emerging market [12][13][40]. Technology and Adoption - WoW stacking technology is compared favorably against HBM, with potential improvements in memory bandwidth by 10-100 times and a significant reduction in power consumption [10][39]. - The report outlines the main hurdles for edge AI deployment, including power consumption, memory bandwidth, and cost, which WoW technology aims to overcome [38][46].

Core Insights - TechInsights released early-stage findings on NVIDIA's Blackwell HGX B200 platform, highlighting its advanced AI and HPC capabilities in data centers [1] - The GB100 GPU features SK hynix's HBM3E memory and TSMC's advanced packaging architecture, marking significant technological advancements [1][2] HBM3E Supplier - The GB100 GPU incorporates eight HBM3E packages, each with eight memory dies in a 3D configuration, achieving a maximum capacity of 192 GB [2] - The per-die capacity of 3 GB represents a 50% increase over the previous generation of HBM [2] CoWoS-L Packaging Technology - The GB100 GPU utilizes TSMC's 4 nm process node and features the first instance of CoWoS-L packaging technology, which significantly enhances performance compared to the previous Hopper generation [3] - The GB100's design includes two GPU dies, nearly doubling the die area compared to its predecessor [3] HGX B200 Server Board - Launched in March 2024, the HGX B200 server board connects eight GB100 GPUs via NVLink, supporting x86-based generative AI platforms [4] - The board supports networking speeds up to 400 Gb/s through NVIDIA's Quantum-2 InfiniBand and Spectrum-X Ethernet platforms [4] TechInsights Overview - TechInsights provides in-depth intelligence on semiconductor innovations, aiding professionals in understanding design features and component relationships [6][7] - The TechInsights Platform serves over 650 companies and 125,000 users, offering extensive reverse engineering and market analysis in the semiconductor industry [8]