Core Viewpoint - The article discusses the challenges and advancements in the Mixed Expert Model (MoE) technology, particularly focusing on the load balancing issues and the introduction of the OmniPlacement strategy by Huawei to enhance inference performance [2][4][12]. Group 1: Challenges in MoE Models - The MoE models face significant challenges, particularly the "cold and hot expert" phenomenon, where some experts are frequently called (hot experts) while others are rarely used (cold experts), leading to uneven load distribution [2][4]. - This imbalance results in increased inference latency and limited throughput, as underutilized resources restrict overall system performance [3][14]. Group 2: OmniPlacement Strategy - Huawei's OmniPlacement strategy addresses these challenges through expert reallocation, inter-layer redundancy deployment, and near-real-time dynamic scheduling, significantly improving MoE model inference performance [4][12]. - The strategy includes a joint optimization algorithm that reduces load imbalance by analyzing expert activation data and optimizing deployment order based on call frequency and computational needs [5][14]. Group 3: Key Features of OmniPlacement - OmniPlacement employs inter-layer redundancy deployment to alleviate the pressure on hot experts by allocating additional redundant instances, thus enhancing system throughput [5][12]. - The framework supports dynamic resource allocation based on real-time resource usage and expert call frequency, allowing for predictive resource distribution to minimize performance discrepancies between hot and cold experts [6][9]. Group 4: Testing and Results - Comprehensive testing on the DeepSeek-V3 model demonstrated that OmniPlacement reduces average inference latency by approximately 10% compared to baseline methods, primarily due to dynamic expert allocation and communication domain optimization [12][14]. - The system's throughput improved by about 10%, reflecting a significant increase in resource utilization, especially in high-concurrency scenarios [14]. Group 5: Future Directions - Future research will focus on developing smarter scheduling algorithms and adaptive expert selection mechanisms to further enhance the system's adaptability to complex inputs [15][16]. - The OmniPlacement framework aims to expand its functionality to support more types of MoE models, increasing its versatility and applicability in various industrial settings [16].

Key Points - CATL prefers institutional investors over retail investors for its IPO, limiting retail participation to 7.5% despite high demand [4][5][6] - Xiaomi has invested over 13.5 billion RMB in its semiconductor project, aiming for top-tier performance with its new chip, the Xiaomi Xuanjie O1, which utilizes 3nm technology [8][9] - Huawei launched two HarmonyOS computers, including a foldable model starting at 23,999 RMB, marking a significant step in its ecosystem development [11][12][14] - Ant Group's international business, Ant International, generated nearly 3 billion USD in revenue last year and is preparing for a potential IPO [14][15] - Xiaomi reported a criminal gang manipulating nearly 10,000 accounts to defame the company, leading to multiple arrests [19][20] - Tesla's autonomous taxi service in Austin will operate on an invitation-only basis with a limited fleet of 10 to 20 vehicles [27] - Google I/O is set to showcase new AI models and updates to Android 16, emphasizing the integration of AI in search and other applications [25][26]

" 华为不只是「官宣」一下而已，后面更会是全面开源。 " 作者丨李希 推理部署，成为大模型落地重中之重 从 2017 年 Google 提出 Transformer —— 这一人工智能 中最常用的 神经网络架构，到 DeepSeek V3/R1 在 202 5 年 春节 一夜爆火，超大规模 MoE 架构大模型的重点逐渐从训练开发转向推理支撑的 应用落地。 推理场景是大模型认知能力的 " 试金石 " ，是大模型商业化落地的核心能力，从抢先上线 DeepSeek 模 型到 API 服务价格战，在推理为王的时代，谁能最极致的提升推理部署计算效率，谁才能真正获得大模型 商业成功。 数学补物理，极致提升计算效率 " 数学补物理 " ，通常指通过数学理论、算法和建模方法，弥补传统物理设备开发在复杂系统分析、大规 模计算或多场耦合问题中的局限性。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、系统补单点等思想，在散热、供电、高速、高密 及大芯片在板可靠性等工程 ...

" 连续七季度保持3位数增长的阿里AI，依旧没能让二级市场满 意。 " 作者丨赵之齐 编辑丨胡敏 宣称要在云和AI硬件基础设施投入超3800亿元的阿里 ， 上周 交 出了2025财年Q4的财报。与此同时， 股价下跌了8%。 财报中，阿里的整体营收同比增长7%。阿里云收入301.27亿元，加速增长至18%，为三年来最快增 速，"主要由更快的公共云业务收入增长所带动"，其中便包括了AI相关产品采用量的提升。 而其AI相关产品收入，则连续七个季度保持三位数增长，调整后EBITA约24亿元。全财年来说，阿里云收 入达到1180亿元，突破双位数增长。 尽管，在财报电话会上，管理层表示，"客户需求的增长是确定的"，且"阿里巴巴投入云和 AI 基础设施的 信心和决心不会改变"。但 对于阿里云比上个季度少了约22%的Capex，"AI搞不定了"的怀疑声还是开始 此起彼伏 。 在云和AI投入上野心勃勃的阿里云，为什么没能令市场满意？ 01 阿里云在AI上做了什么？ 阿里云近年在AI上抢占的市场，大体可以分为算力供给、开源基础模型服务以及云和AI人才储备。 算力储备上，阿里云已经依托全球29个地域、87个可用区的分布式数据中心， ...

「 RTK+双目 」成割草机方案香饽饽，公司A和B上演 「 始乱终弃 」 九号割草机出了一个 「 RTK+双目」的方案，A公司认为这个方案可行，直接在内部优化了这个方案，并 投入了200人，甚至还找了外部协作团队B一起做这个事情。但等B把方案做出来，A觉得这个方案自己也 能做，加上B的工作效率不如预期，开价100万价格太高，就又自己做了。 据说公司A没有用B代工的还有一个顾虑是，B用了一家大厂的算法，担心后续会在法庭相见。（B公司长 于算法，劣于渠道，在做方案还是做品牌，做何种方案之间均有摇摆，A公司今年承压，尤其在九号侵入 其腹地之后，更多行业动态可添加微信 MOON_ERS 交流） 扫地机公司C一级代理商吃肉，二级代理商被吃干抹净 扫地机公司C采取 「一省一代理 」的渠道策略，C给代理商是30个点，达成后月返季返，最多能拿到32 个点。C的一级代理如果要拓展市场，会找二级代理商，但最多只能给22个点，如果按照国补15个点去 卖，合起来相当于30个点。所以如果没有国补，C的产品基本卖不动了。有人评价，C的一级代理商吃肉 喝血，二级代理商则是直接被吃干抹净。（今年扫地机市场的代理体系暗流涌动，C公司重投线上， ...

" 内容型产品「缺位」，平台型产品「补位」。 " 作者丨胡家铭 编辑丨董子博 随着 Sensor Tower 发布2025年4 月中国手游发行商全球收入榜，中国游戏厂商在全球范围内的吸金实 力再度被放在聚光灯下。 从数据来看，33家手游发行商吸金20亿美元，占据全球发行商收入近四成份额。但到目前为止，占据各大 发行商流水前列的产品，仍然以大厂的长青内容型产品+大DAU产品，和中小厂商的平台型产品（小游 戏）为主。各大厂商新近上线的"重磅手游"，并未在市场端复刻前辈们的成功，在市场表现上有所"缺 位"。 一位游戏分析师向雷峰网表示，从重度手游平均2-3年的研发周期来看，近期上线的手游多为2021-2022 年期间的立项产品。 彼时正值版号寒冬，行业未来的不确定性，致使厂商会在内容型产品的开发上"留力"，由此而来的"后 劲"则是，近一年上线的重度手游产品，在经历开服的宣发期后，也"凑巧"出现了高开低走的态势。从行 业大盘来看，则呈现出"手游小年"的尴尬状况。 以上海某游戏厂商为例，原定在今年上线的模拟经营类项目，其完成度在内部评审上不达预期，内部对于 是否将其上线公开，目前仍然在讨论阶段。其核心冲突点，则是纠结于 ...

Core Viewpoint - Despite a downgrade in net profit expectations by some brokerage firms, Alibaba generally received a "buy" rating from analysts [1] Group 1: Financial Performance - In Q4 of FY2025, Alibaba achieved revenue growth of 7% year-on-year, reaching 236.45 billion RMB, and a non-GAAP net profit increase of 22%, amounting to 29.85 billion RMB [2] - The growth was driven by the core e-commerce and cloud businesses, with all segments showing year-on-year improvement in EBITA [2] - The Chinese retail segment of Taotian Group saw an 8% revenue increase, totaling 95.58 billion RMB, surpassing the previous quarter influenced by Double 11 [4] Group 2: E-commerce Strategy - Alibaba has made significant adjustments to its e-commerce operations, including divesting from offline retail and streamlining its business lines [4] - The introduction of the "All-Station Promotion" product aims to enhance monetization and compete effectively against rivals like Pinduoduo [5] - The company is focusing on stabilizing market share and increasing GMV, which has seen a decline from 64% to 49% of the national online retail market share from FY2024 Q4 [5] Group 3: User Engagement and Membership - Alibaba is enhancing its 88VIP membership program, which has grown to over 50 million users, a 43% increase year-on-year [6] - The company is also targeting high-value users with tailored shopping benefits while introducing low-threshold monthly cards to attract cost-conscious consumers [6] Group 4: Cloud Business Performance - Alibaba Cloud's revenue growth has been underwhelming, with less than 10% growth for nine consecutive quarters, attributed to increased competition and reduced enterprise demand [9] - Recent quarters have shown improvement, with revenue growth of 13% and 18%, driven by AI-related demand [10] - The CEO noted that AI-related product revenue has seen triple-digit growth for seven consecutive quarters, indicating a shift in enterprise cloud adoption [10] Group 5: Market Comparison - Year-to-date, Alibaba's stock has risen approximately 46% in the US and 52% in Hong Kong, outperforming Microsoft and Amazon [11] - Despite this growth, Alibaba's current market valuation is only half that of Tencent, indicating potential for further recovery [11]

Group 1 - Deep Blue CEO's controversial statement about employees taking a two-month leave after purchasing a car sparked discussions, with the CEO clarifying it was not intended as a layoff strategy but to allow employees to enjoy their time off [4][5] - Lei Jun's internal speech at Xiaomi highlighted the significant impact of a recent traffic accident on the company's reputation, emphasizing the need for Xiaomi to become a leader in automotive safety [7][8] - Liu Qiangdong's intensive training sessions for JD's management, aimed at improving operational efficiency and addressing food safety in the delivery sector, have reportedly led to health issues due to his dedication [9][10] Group 2 - ByteDance has implemented new policies restricting employees from taking food home and prohibiting lights off during nap time, aiming to address resource management issues [12][14] - The launch of "Wuxiang AI," a high-level security AI system, marks a significant advancement in cybersecurity, capable of autonomous threat detection and response [15] - Nvidia's new legislation requires AI GPUs to have built-in location tracking to prevent unauthorized exports, reflecting increasing regulatory scrutiny in the tech industry [29][30] Group 3 - Tesla appointed a senior executive from Chipotle to its board, indicating a strategic move to enhance financial management amid declining electric vehicle sales [30][31] - OpenAI plans to assist the UAE in developing one of the world's largest data centers, showcasing its commitment to expanding AI infrastructure in the Middle East [34]

Core Viewpoint - The article discusses the recent leadership changes at Geely, highlighting the rise of An Conghui as the new CEO of Geely Holding Group, marking a significant shift in the company's internal structure and strategy [1][5]. Group 1: Leadership Changes - An Conghui has been appointed as the CEO of Geely Holding Group, taking over from Li Donghui, who will now serve as the vice chairman [5]. - The restructuring aims to consolidate Geely's operations under a unified leadership to enhance efficiency and reduce costs [9]. - An Conghui's previous role included leading Zeekr, which was established under his guidance, showcasing his capability in managing new ventures [2][4]. Group 2: An Conghui's Background and Influence - An Conghui has a strong reputation within Geely, having been groomed by founder Li Shufu since his early career [8]. - He is known for his practical approach and dedication, often working late hours and demonstrating a strong sales acumen [8][9]. - His leadership style is characterized by an engaging communication ability that inspires confidence among stakeholders [9]. Group 3: Strategic Goals Post-Reorganization - The integration of Zeekr, Lynk & Co, and Geely aims to achieve three main objectives: a cost reduction target exceeding 3% in production, a 10-20% optimization in R&D, and a 10-20% increase in management efficiency [9].

Core Viewpoint - The recent departures of key executives, including Zhou Jiang, have raised uncertainties for Neta Auto's overseas business and overall operations [2][4][5]. Group 1: Executive Departures - Zhou Jiang, the president of Neta Auto's overseas division, has recently left the company, with his future plans currently unknown [2]. - Other recent departures include Zhang Panpeng, the general manager of Neta Auto's Indonesia company, who has joined Jietu Auto [2]. - Zhou Jiang had over 25 years of experience in the automotive industry, previously holding significant positions at Changan Automobile before joining Neta Auto in 2019 [2][4]. Group 2: Impact on Business Operations - Zhou Jiang's departure has created uncertainty in Neta Auto's overseas operations, with reports of multiple high-level exits from the overseas business unit [4]. - Some overseas dealers are reportedly struggling to receive parts and after-sales support from the manufacturer, leading to inventory issues with models like Neta V and Neta X [4]. - Neta Auto is facing challenges in its domestic business, including a bankruptcy application from an advertising company due to unpaid debts, although Neta Auto claims it is not seeking bankruptcy itself [4]. Group 3: Company Status and Future Options - The company currently has over 1,000 employees, but many are on hold, with only a few remaining active, particularly in the marketing division [4]. - Internal discussions suggest two potential paths for the company: bankruptcy restructuring or seeking investment, both of which are expected to take considerable time [4].