Group 1: AI Industry Development - The AI industry follows a "technology-hardware-terminal-application" development model, with a shift from communication networks to large model theoretical research[1] - Domestic chip manufacturers are accelerating technological breakthroughs, enhancing the application ecosystem, and driving the deep integration of generative AI across multiple industries[2] - The global large model technology is entering a deep competitive phase, with differentiated development paths between China and the US[2] Group 2: AI Chip and Hardware Investment - AI chips are the cornerstone of the large model industry, characterized by long R&D cycles, high technical barriers, and significant investment costs[2] - China has established a basic layout in GPU, ASIC, and FPGA chips, meeting standards for various application scenarios[2] - Investment opportunities exist in the AI industry chain, including optical modules, power distribution technology, and liquid cooling technology[2] Group 3: Market Trends and Opportunities - The domestic AI industry is experiencing a strategic transformation from "software-hardware decoupling" to "full-stack collaboration"[2] - The market for AI software ecosystems is still dominated by foreign open-source frameworks, but domestic companies are accelerating their AI ecosystem layout[2] - The procurement rate of domestic large models in key industries like finance and telecommunications has exceeded 45%[2] Group 4: Risks and Challenges - Risks include slower-than-expected technological iterations, industry growth rates, and potential policy risks[2] - The need for high-quality data and standards in model training remains a challenge for the domestic AI industry[2]

Summary of Mu Xi Co. Conference Call Company Overview - Mu Xi Co. offers a product line that spans from low-end to high-end GPUs, including the C7,000, intelligent computing series, and the G100 series, primarily shipping in module and board card forms, with a small number of servers [2][3] Core Business and Product Features - The main focus is on the GPU industry, with three product series: 1. Training and inference integrated GPUs (C series) 2. Intelligent computing series for inference and video transcoding 3. Graphics rendering GPUs (G100 series) for low-end applications [3] - The C series GPUs are designed for AI inference and small to medium model training, featuring self-developed Link protocol and domestic supply chain [2][5] Financial Performance - As of June 2025, Mu Xi has sold a total of 25,000 GPU chips, with 2024 revenue reaching 700 million RMB, primarily from training and inference integrated cards, and an additional 200 million RMB from servers [2][6] - The gross margin for 2024 was 53.4%, with Q1 2025 showing a slight increase to 55.3%, which is close to competitors but lower than others in the industry [9][10] Research and Development - R&D investment for 2024 was approximately 900 million RMB, with a high R&D expense ratio of 121.2% [4][11] - The company is focused on high-performance general-purpose GPU development, with significant ongoing investments [11] Inventory and Prepayments - As of Q1 2025, prepayments reached 1.17 billion RMB, and inventory was 802 million RMB, reflecting increased material orders [12] Customer Concentration and Supplier Structure - The top five customers accounted for 88.35% of sales, indicating high customer risk [4][13] - Major suppliers include companies related to wafers and HBM, which have been consistent over the past three years [14] Industry Trends - The AI chip demand is shifting from pre-training to post-training and inference, with domestic cards gaining opportunities in the inference phase [15] - Domestic internet giants are expected to increase capital expenditures, although short-term impacts from supply chain issues may affect this [17] Global and Domestic Market Dynamics - The global GPU market is dominated by NVIDIA (80% market share) and AMD (20%), with significant growth in shipments [20] - Domestic policies are encouraging the use of local chips, with plans for 100% self-sufficiency in computing centers by 2027 [20] Future Outlook - The Chinese AI chip market is projected to reach 50 billion USD by 2027, indicating strong growth potential [21] - Domestic AI chip companies are transitioning to selling complete systems and clusters, supported by government policies [22] Emerging Trends - The computing power leasing market is expected to grow significantly, benefiting from improvements in the domestic supply chain [23]

Market Overview - On June 19, the Hang Seng Index fell by 473 points or 2.0%, closing at 23,237, marking a new low since June 2[1] - The Hang Seng Tech Index dropped 2.4%, closing at 5,088, the lowest since April 30[1] - The total market turnover was over HKD 220.1 billion, indicating that selling pressure was not excessively high[1] - Despite the decline, the Stock Connect saw a net inflow of HKD 1.43 billion[1] Economic Insights - The Federal Reserve maintained interest rates during the June FOMC meeting, emphasizing inflation control as a priority[2] - Economic forecasts indicate concerns over stagflation, with GDP growth predictions lowered and inflation and unemployment rates raised[2] - The divergence among Fed officials regarding interest rate cuts has increased, with the number of officials opposing cuts rising from 4 to 7[2] Industry Developments - Black Sesame Technologies (2533 HK) announced plans to acquire an AI chip company, which could enhance its product line for smart vehicles[3] - The Hang Seng Healthcare Index fell by 3.2%, with Innovent Biologics (1801 HK) experiencing a smaller decline due to positive clinical data for its drug candidate[4] Company Highlights - Cao Cao Travel (2643 HK) reported a GTV of RMB 17 billion for 2024, a year-on-year increase of 38.8%, with a market share of 5.4%[5] - The company aims to expand its fleet to over 34,000 customized vehicles by 2024, with customized vehicles accounting for approximately 25.1% of GTV[5] - Cao Cao's AI-driven system, "Cao Cao Brain," is designed to optimize order matching and reduce idle mileage, contributing to GTV growth[6] Financial Projections - Stone Pharmaceutical (1093 HK) signed a strategic agreement with AstraZeneca, receiving an upfront payment of USD 110 million (approximately RMB 790 million) and potential milestone payments totaling up to USD 1.62 billion[8] - Revenue forecasts for Stone Pharmaceutical for 2025-2027 have been adjusted upward by 1.3%, 1.2%, and 0.5% respectively, reflecting the anticipated income from the AstraZeneca agreement[9] - The target price for Stone Pharmaceutical has been raised to HKD 8.15, maintaining a "neutral" rating[11]

Investment Rating - The report maintains an "Increase" rating for the industry, emphasizing the explosive demand for ASICs and the investment opportunities in self-developed chip industries by CSPs like Amazon and Google [5][7][31]. Core Insights - The demand for custom ASICs is experiencing a comprehensive explosion, with significant growth expected in the custom acceleration computing chip market, projected to reach $42.9 billion by 2028, growing at a CAGR of 45% from 2023 to 2028 [14][31]. - Major North American CSPs are accelerating their self-developed ASIC layouts, with Google and Amazon leading the progress in custom chip development [14][19]. - Broadcom's guidance indicates that XPU deployments will exceed expectations in 2026, driven by strong demand for customized AI accelerators [20][22]. - Marvell is set to start 3nm chip production in 2026, with significant progress in custom AI XPU projects for large-scale data center clients [3][26]. - Wistron reported a significant revenue increase in May 2025, indicating a robust growth phase for ASIC demand, with AI inference servers expected to account for nearly 50% of the market [30][31]. Summary by Sections ASIC Demand and CSP Developments - The custom acceleration computing chip market is projected to grow from $6.6 billion in 2023 to $42.9 billion by 2028, with a CAGR of 45% [14]. - Google has launched the TPU v6 Trillium chip, expected to replace the TPU v5 by 2025, while Amazon is focusing on the Trainium v2 chip for generative AI applications [15][19]. - Meta is developing the MTIA v2 chip in collaboration with Broadcom, focusing on energy efficiency and low-latency architecture [18][19]. - Microsoft is accelerating the development of its Maia series chips for Azure cloud applications, with the next generation Maia v2 in progress [18][19]. Broadcom's Performance and Projections - Broadcom's AI semiconductor revenue reached over $4.4 billion in Q2 2025, marking a 46% year-over-year increase [20][22]. - The company expects to deploy at least 1 million AI accelerator clusters for three clients by 2027, with a significant portion being custom XPUs [20][22]. Marvell's Advancements - Marvell's revenue reached a record $1.895 billion in Q1 FY26, driven by strong AI demand, with expectations for continued growth in custom AI XPU revenue [3][24]. - The company is collaborating with NVIDIA to enhance its custom platform capabilities, integrating NVLink Fusion technology [26]. Wistron's Revenue Growth - Wistron reported a revenue of 208.4 billion New Taiwan Dollars in May 2025, reflecting a 162% year-over-year increase, driven by ASIC demand [30][31]. - The shift from AI training to AI inference is expected to boost the demand for AI servers significantly [30][31]. Analog Chip Market Insights - The global analog chip market is projected to reach $84.3 billion by 2025, with a strong focus on domestic production to meet rising demand [40][43]. - The domestic market for analog chips is expected to grow steadily, with a CAGR of 11% from 2025 to 2029 [43][46].