Investment Rating - The report does not explicitly provide an investment rating for the industry or specific companies involved in AI ASIC servers. Core Insights - The AI ASIC market is expected to grow significantly, with Meta's MTIA AI servers potentially marking a milestone in 2026. The total AI ASIC volume could exceed nVidia's AI GPU volume by 2026 [1][3][12] - nVidia currently dominates the AI server market with over 80% market value, while ASIC AI servers hold approximately 8-11% market value share. However, unit comparisons suggest that Google and AWS's AI ASICs could reach 40-60% of nVidia's GPU volume by 2025 [1][3] - Meta's MTIA AI server is projected to ramp up significantly, with expectations of 1 to 1.5 million units of MTIA V1 and V1.5 by late 2025 to 2026 [10][12] Summary by Sections AI ASIC Market Dynamics - The AI ASIC market is experiencing aggressive growth, with more cloud service providers (CSPs) developing their own ASIC solutions, including Meta and Microsoft [1][3] - nVidia is responding to competition by unveiling NVLink Fusion, which allows inter-chip connections between its GPUs and third-party CPUs, indicating a proactive approach to maintain its market position [2] Meta's MTIA AI Server Development - Meta's MTIA AI server is set to launch its first ASIC (MTIA T-V1) by late 2025, with subsequent versions (V1.5 and V2) expected in mid-2026 and 2027, respectively [9][10] - The MTIA T-V1.5 is anticipated to be significantly more powerful than V1, with a larger interposer size and complex design [9][10] Supply Chain and Component Insights - Key suppliers for Meta's MTIA projects include Quanta, Unimicron, EMC, WUS, and Bizlink, which are expected to benefit from the growing demand for AI ASICs [12][13][14][15][19] - The report highlights the importance of baseboard management controllers (BMCs) in Meta's ASIC AI server development, with an estimated 23 BMCs per rack [17][18] Competitive Landscape - Despite nVidia's current leadership in AI computing, the report suggests that the gap is narrowing as ASIC solutions improve in specifications and performance [3][7] - The specifications of AI ASICs from companies like Google and AWS are catching up to nVidia's offerings, although nVidia still holds advantages in connectivity and ecosystem [7][8]
野村:Meta 在 ASIC 服务器方面雄心勃勃,其 MTIA AI 服务器有望在 2026 年成为一个里程碑