Core Viewpoint - The company, Lihexing, is gaining significant attention in the market as a long-term partner within the Huawei ecosystem, leveraging its collaborations in emerging fields to drive new growth momentum [1] Group 1: Partnership with Huawei - Lihexing has been a qualified supplier for Huawei since 2013, with over 10 years of collaboration, initially focusing on testing equipment for Huawei's Mate series smartphones, capturing over 10% market share in the smart terminal testing sector [2] - The partnership has expanded beyond mobile devices to include 5G base stations, new energy vehicles, and digital energy, with Lihexing becoming a core manufacturer for Huawei's liquid cooling ultra-fast charging equipment [2] - The company is also involved in Huawei's solar storage supply chain, providing precision components for photovoltaic inverters and energy storage systems, benefiting from the acceleration of new energy infrastructure [2] Group 2: Semiconductor Component Collaboration - In response to the accelerated domestic semiconductor localization, Lihexing has partnered with semiconductor equipment company Xinkailai, entering the high-margin precision components sector and becoming a core supplier [3] - Xinkailai currently has a robust order book, serving leading domestic wafer fabs such as Shenzhen Pengxinwei and SMIC, indicating a phase of explosive growth [3] Group 3: Liquid Cooling Technology and AI Computing - The demand for AI applications is driving a surge in computing power needs, with liquid cooling technology becoming a standard for AI servers [5] - Lihexing has developed an automatic maintenance testing system for liquid cooling servers and has successfully entered Huawei's Ascend server supply chain, addressing high-power cooling challenges [5] - The global liquid cooling market is projected to exceed 100 billion yuan by 2026, quadrupling from 2025, positioning Lihexing to benefit from technological advancements and infrastructure acceleration [5] Group 4: Humanoid Robot Development - Leveraging its expertise in precision machinery, automation, and machine vision, Lihexing is advancing into the humanoid robot sector [6] - The company has developed high-precision transmission testing equipment for robotic joints and is involved in the joint drive development for Huawei's Jimo robot, while also supporting Tesla's humanoid robot supply chain [6] Group 5: Growth Potential - Lihexing's strategy of being a core supplier for Huawei, entering the semiconductor component market, capitalizing on liquid cooling technology, and expanding into humanoid robotics presents a significant growth potential [7] - The company is expected to continue breaking through in multiple high-growth sectors, ushering in a new phase of accelerated growth [7]

Core Insights - SPDB is accelerating the application of large model technology in its digital transformation strategy, having established an artificial intelligence center to enhance AI technology iteration and application [1][2] - The bank is building a data center in Hohhot, which will support its large-scale, low-cost computing needs and facilitate the deployment of intelligent computing capabilities [1][2] Group 1: AI Application and Infrastructure - SPDB is constructing a "four-in-one" intelligent foundation that integrates computing power, algorithms, platforms, and knowledge to transition AI from pilot projects to large-scale applications [1] - The bank has developed a heterogeneous computing cluster based on Huawei Ascend servers, overcoming the "computing power bottleneck" and enabling extensive deployment of large model scenarios [1][2] Group 2: Data Center and Technical Support - The Hohhot data center will provide a unified management platform for intelligent computing clusters, ensuring comprehensive coverage of various scenarios to support SPDB's intelligent transformation [2] - The design of the data center focuses on safety, energy efficiency, intelligence, and flexible management to meet the needs of intelligent computing deployment [1][2] Group 3: Model Evaluation and Algorithm Development - SPDB has integrated over 30,000 scenario data points to create a multi-modal evaluation system that assesses large model algorithms across four dimensions: general capabilities, financial knowledge, security capabilities, and technical levels [2] - The bank has introduced various large models, such as DeepSeek and Qwen, to build a matrix of multi-modal, multi-capability algorithms for tasks like financial report analysis and document generation, significantly enhancing operational efficiency and service quality [2]

Group 1 - Huawei will showcase the Ascend 384 super node at the 2025 World Artificial Intelligence Conference, marking its first offline exhibition of this technology [1] - The demand for PCBs is expected to grow due to the recovery in downstream electronic demand and the ongoing boom in AI and high-speed communication sectors [1] - Domestic PCB manufacturers are gradually enhancing their position in the high-end market, with an increasing share in AIPCB supply, driven by technological advancements and product competitiveness [1] Group 2 - Companies like Qichuang Data focus on IoT smart terminals and data storage devices, leveraging their storage technology to enter the computing hardware market [2] - Sichuan Changhong, through its joint venture Huakun Zhenyu, has designed Ascend servers and won a 650 million yuan AI server procurement contract from China Mobile [2]

虎嗅注： "大模型江湖，落地为王。"这句话的含金量还在提升。随着DeepSeek V3/R1在春节期间一夜爆火， 基于超大规模MoE（Mixture of Experts）架构的大模型正在从训练开发转向推理应用的落地。 对于MoE推理部署来说，效率一直是一个痛点。谁能将部署计算效率提升至最高，才能真正获得大 模型商业成功。但受限于庞大的模型容量与计算需求，传统部署方案通常依赖于多张数据中心级 GPU（如H20）。你我都知道，英伟达不仅贵，而且不断受到地缘政治摩擦的影响，不断降低自己的 性能来满足监管需求。 而在最近，华为全面揭秘超大规模MoE模型推理部署技术，不仅实现了国产的进一步突破，更全面 超越了基于英伟达Hopper架构的推理部署性能。 他们是怎么做到的？ 数学补物理，极致提升计算效率 "数学补物理"，这种通过数学理论、工具、算法和建模等方式，来弥补硬件和工艺的局限性，实现最 大化发挥芯片和系统能力效果。华为轮值董事长孟晚舟曾在2025年新年致辞中提到： "华为十多个实验室与伙伴们的工程师组成"大杂烩"团队，面对天成AI集群系统和单芯片性能的严峻 工程挑战，他们创造性应用数学补物理、非摩尔补摩尔、系统补 ...