Summary of Conference Call on Huawei's Super Node and Domestic AI Chip Market Industry Overview - The conference call discusses the domestic AI chip market, focusing on Huawei's 384 Super Node and its competitive landscape against international players like NVIDIA [1][2][8]. Key Points and Arguments - **Huawei 384 Super Node**: - Utilizes high-speed interconnect and optical module technology, enhancing cost-effectiveness and deployment certainty in inference scenarios, serving as a domestic alternative to NVIDIA's Nvlink bottleneck [1][2]. - Supports SP8, providing superior performance in inference tasks compared to many domestic cards lacking this feature [2]. - **Domestic AI Chip Companies**: - Companies like Moer, Muxi, and Suiyuan are narrowing the gap with international leaders through technological advancements, with expectations to increase market share by 2026 [1][4]. - New products from these companies, such as Moer S5000 and Muxi C600, are expected to meet the demands of major internet companies, enhancing their competitive position [4]. - **Market Demand and Supply**: - The demand for domestic computing cards is not limited to internet companies but extends to telecom operators, financial institutions, and state-owned enterprises, indicating a supply-demand imbalance [3][16]. - The market is projected to focus on 2026 valuations, with optimism surrounding the growth of domestic graphics cards [3][13]. - **High-Density Super Point Solutions**: - Companies like Haiguang are developing high-density, high-performance super point solutions, such as a silent liquid cooling solution supporting up to 660 cards, to improve overall computing efficiency [5][6]. - **GPGPU Architecture Development**: - Domestic GPGPU architecture is evolving, with companies like Huawei Ascend transitioning to GPGPU to reduce interconnect development risks [7]. - The integration of DSA functions into GPUs by manufacturers like Moer indicates a trend towards technological convergence [7]. - **Future Market Dynamics**: - The competitive landscape is expected to become more complex by 2026, with new product launches and testing processes for internet companies [10][12]. - Despite the anticipated changes, the overall market structure is expected to remain stable, with Huawei leading, followed by Cambricon and Haiguang [10]. Additional Important Insights - **Inference Demand Growth**: - The increasing demand for inference capabilities, especially in the context of the multi-modal era, is driving the need for domestic computing cards [9][12]. - **AI Application Revenue**: - Significant growth in AI application revenues is anticipated, with estimates for companies like Keling reaching $200 million in 2025 [10]. - **Comparison with U.S. Market**: - The U.S. market is dominated by a few major clients, while China's demand is diversified across numerous long-tail customers, indicating a robust growth potential for domestic computing cards [16]. This summary encapsulates the critical insights from the conference call, highlighting the competitive landscape, technological advancements, and market dynamics within the domestic AI chip industry.

Core Insights - The article discusses the anticipated performance and market release timelines for the AI semiconductor products 910D and 920, highlighting their architectural improvements and expected advantages over previous models [1]. Group 1: 910D Overview - 910D is designed with four dies, an upgrade from the two dies in 910C, enhancing its performance to exceed that of the H100 [1]. - The expected market release for 910D is optimistic, with potential availability by Q2 2024 and no later than Q2 2026 [1]. - The performance of 910D is expected to support certain training applications, although its cost-effectiveness may decline for models exceeding 400 billion parameters due to weaker ecosystem capabilities [1]. Group 2: 920 Overview - The 920 will also have multiple versions, with the first version adopting a dual-die design and optimized processes, transitioning to a GPGPU architecture for better ecosystem integration with NVIDIA [1]. - The first batch of 920 chips is expected to be available by the end of 2025, with larger scale production anticipated around mid-2027 [1].