Core Viewpoint - Kuaishou's capital expenditure (CapEx) is projected to reach 26 billion yuan in 2026, representing an increase of approximately 11 billion yuan compared to 2025 [1] Group 1: Capital Expenditure Details - The anticipated CapEx for 2026 includes investments in large models and other foundational models, as well as expenditures for offline data storage, processing, and conventional server procurement [1] - The CapEx also encompasses investments in the construction of data computing centers [1]

Core Insights - Kuaishou Technology's CEO Cheng Yixiao announced that the annualized revenue run rate for Keling AI has exceeded $300 million as of January this year, which is approximately 2.07 billion RMB at current exchange rates [1] - The company maintains strong confidence in Keling AI achieving over 100% year-on-year revenue growth by 2026 based on current growth momentum and commercialization progress [1] Financial Performance - Kuaishou's CFO Jin Bing revealed that the group's capital expenditure (Capex) is expected to reach approximately 26 billion RMB in 2026, with an additional 11 billion RMB investment compared to 2025 [1] - The increased Capex will primarily support Keling's large model and other foundational model computing power, as well as conventional server procurement and data/computing center construction [1] - The recently released financial report indicates that Kuaishou's average daily active users reached 410 million in 2025, with total revenue growing by 12.5% year-on-year to 142.8 billion RMB [1] - The adjusted net profit for the full year 2025 was 20.6 billion RMB, reflecting a year-on-year increase of 16.5%, with an adjusted net profit margin of 14.5% [1]

Group 1: Company Performance - Xirui (02507) reported a revenue of $1.354 billion for 2025, a year-on-year increase of 13.13%, with a net profit of $139 million, up 15.02% [1] - Jiantao Laminates (01888) achieved a revenue of HKD 20.4 billion, a 10% increase year-on-year, and a profit attributable to shareholders of HKD 2.442 billion, up 84.16% [1] - H&H International Holdings (01112) reported a revenue of RMB 14.354 billion, a 10% year-on-year increase, with an adjusted net profit growth of 22.7% [2] - China Jinmao (00817) posted a gross profit of RMB 9.221 billion, a 7% increase year-on-year, with a gross margin improvement of 1 percentage point [3] - Kunlun Energy (00135) reported a revenue of RMB 193.979 billion, a 3.71% increase year-on-year, but a net profit decrease of 10.3% [3] - China Nonferrous Mining (01258) achieved a revenue of $3.42 billion, a decrease of 10.4% compared to 2024 [4] Group 2: Market Trends - Lithium stocks saw renewed activity, with Tianqi Lithium (09696) up 4.26% and Ganfeng Lithium (01772) up 1.61%, influenced by a rise in lithium carbonate prices [2] - The storage sector stocks showed gains, with companies like Lanke Technology (06809) and Zhaoyi Innovation (03986) experiencing increases [5] - The drone sector saw significant gains, with Swarmer (SWMR.US) up 34.22% following its IPO, highlighting the growing interest in drone technology [6] - Oil and gas stocks collectively rose, driven by a rebound in international crude oil prices, with Brent crude up over 3% [7] - The AI infrastructure expansion faces potential bottlenecks due to storage chip shortages and energy supply constraints [5]

Group 1 - Alibaba's new flagship processor, Xuantie C950, is claimed to be the highest-performing RISC-V CPU globally, achieving a single-core performance score exceeding 70 in the SPEC Cint2006 test [2] - Amazon is accelerating its Leo satellite launch schedule, planning to exceed 20 launches per year, having already deployed over 200 satellites with hundreds more ready for launch [2] - Toyota announced a $1 billion investment in its factories located in Kentucky and Indiana, part of a larger plan to invest up to $10 billion in the U.S. over five years [2] - TSMC's Fab 21 construction in Arizona is ahead of schedule, with the P2 wafer fab expected to start 3nm process mass production as early as the second half of 2027 [2]

Investment Rating - The report maintains a positive outlook on the investment opportunities in the domestic large model sector, particularly focusing on MiniMax as a key player [5][6]. Core Insights - The large model technology path has converged, with domestic models emphasizing cost-effectiveness while overseas commercialization accelerates. The emergence of applications like OpenClaw is driving significant increases in token consumption, indicating a strong demand for domestic large models in high-frequency scenarios such as programming and office tasks [4][5][6]. - MiniMax is positioned as a pioneer in self-developed multimodal large models, with a global strategy driving high revenue growth. The company achieved total revenue of $79.04 million in 2025, a year-on-year increase of 159%, and its annual recurring revenue (ARR) surpassed $150 million by February 2026 [4][5][6]. - MiniMax's M2.5 model offers extreme cost-effectiveness, making it suitable for agent scenarios like programming and office automation. The model's API output price is 1/10 to 1/20 of that of leading overseas models, enhancing its competitive edge [4][5][6]. Summary by Sections 1. Industry: Overseas Performance Leading, Domestic Models Provide Cost Advantages - The competition in large models has shifted from finding the right path to optimizing efficiency on mainstream paths. The pre-training paradigm has converged to a Decoder-Only + MoE architecture, focusing on mid-training and post-training optimizations to enhance model capabilities [12][14]. 2. MiniMax: Dual Drive of Self-Developed Models and Commercial Applications - MiniMax is a leading global company in multimodal models, established in 2021, focusing on general large models and multimodal capabilities (text, speech, video). The company has a high proportion of R&D personnel, with over 70% of its workforce dedicated to research [34][40]. 3. MiniMax: Rapid Model Iteration Drives Revenue Growth - MiniMax's revenue structure is diverse, with AI-native applications accounting for 67.2% of total revenue. The company has seen a significant increase in token consumption, with daily token usage from its M2 series models growing over six times from December 2025 [40][41].

将投中网设为"星标⭐"，第一时间收获最新推送 海淀的创新生态藏不住了。 本文字数3904 手搓量丨 100% AI含量丨0% 3月还未结束，AI赛道的资本热度已经被北京海淀的三家企业推向顶峰。 先是3月2日，具身智能公司银河通用宣布，完成了25亿元A+轮融资，这不仅是目前中国具身智能领域单轮融资的最高纪录，更释放了一个关键信号 ——国家大基金罕见出手，首次布局具身智能赛道。 紧接着，物理AI仿真基础设施企业光轮智能宣布，已于近日完成A++与A+++两轮融资，累计金额达10亿元。至此，这家"AI卖铲人"成功跻身独角兽行 列，成为全球首家具身数据领域的独角兽。 而最新的消息来自月之暗面。近日有媒体报道称，该公司正以180亿美元估值进行新一轮最高10亿美元（约合人民币69亿元）的融资。若消息属实，月 之暗面将打破自己在今年初创下的近一年国内大模型融资纪录，成为国内估值最快突破百亿美元的"独角兽"公司。 三个现象级项目，超104亿元的真金白银，将聚光灯引向了同一个坐标——北京海淀。 这不禁让人发问：为什么是海淀？在全国各大城市争抢AI高地、各地政府开出天价"绣球"的当下，这片土地究竟藏着怎样的魔力，能让资本如此密集地 ...

Investment Rating - The report maintains a positive outlook on the investment opportunities in the domestic large model sector, particularly focusing on MiniMax as a key player [5][6]. Core Insights - The large model technology path is converging, with domestic models emphasizing cost-effectiveness while accelerating commercialization overseas. The leading global models have established successful business models through enterprise-level APIs and subscription tools [4][5]. - MiniMax is positioned as a pioneer in self-developed multimodal large models, with a global expansion strategy driving significant revenue growth. The company achieved total revenue of $79.04 million in 2025, a year-on-year increase of 159%, and its ARR surpassed $150 million by February 2026 [4][6]. - The demand for agent applications is surging, leading to a substantial increase in token consumption, particularly in high-frequency scenarios such as programming and office automation, where MiniMax's cost advantages make it a preferred choice [4][5]. Industry Overview - The competition in large models has shifted from finding the right path to optimizing efficiency on mainstream paths. The pre-training paradigm has converged to a Decoder-Only + MoE architecture, with core efforts focusing on mid-training and post-training optimizations [15][17]. - Domestic models have certain performance gaps compared to overseas counterparts in text and programming capabilities, but they are closing the gap rapidly, especially post-2026, with a competitive edge in cost-effectiveness [17][24]. MiniMax Overview - MiniMax is a leading global company in multimodal models, established in 2021, focusing on text, speech, and video capabilities. The company has a high proportion of R&D personnel, with over 70% of its workforce dedicated to research [39][44]. - The company has successfully built a diverse revenue structure, with AI-native applications accounting for 67.2% of its revenue and a rapid growth in enterprise services [46]. - MiniMax's M2.5 model, which utilizes a sparse MoE architecture, offers a significant cost advantage, with API output prices being 1/10 to 1/20 of those of leading overseas models, making it suitable for high-frequency token-consuming scenarios [4][6]. Investment Analysis - The report suggests focusing on domestic large models, particularly MiniMax, due to their multimodal capabilities and extreme cost-effectiveness, which are expected to drive widespread application in high-frequency demand scenarios [5][6].

Core Viewpoint - Southbound funds have significantly increased their holdings in China Software International (00354.HK), indicating strong investor interest in the company [1] Group 1: Southbound Fund Activity - On March 23, southbound funds increased their holdings by 5.512 million shares of China Software International [1] - Over the past 5 trading days, there have been 4 days of net increases, totaling 8.722 million shares [1] - In the last 20 trading days, there were 14 days of net increases, amounting to 48.742 million shares [1] - Currently, southbound funds hold 949 million shares of China Software International, representing 34.73% of the company's total issued ordinary shares [1] Group 2: Company Overview - China Software International is an investment holding company that provides global technology software and IT services [1] - The company operates through two main segments: Technology Professional Services and Internet Information Technology Services [1] - Its primary business focuses on the development of generative artificial intelligence (AIGC), large model software and hardware sales, and digital transformation consulting services for enterprise resource planning (ERP) [1] - Key products include the "Question Series" solutions, large model application integrated machines, and Lingxi AI application platforms [1] - The company serves various sectors, including water conservancy, transportation, government platforms, military, energy, education, and finance, operating in both domestic and international markets [1]

Investment Rating - The report rates the computer industry as "Outperform" [1][34]. Core Insights - The report highlights that the industry is approaching a critical point in embodied intelligence, with three key challenges to overcome before reaching the "ChatGPT moment" [1][10]. - Xiaomi has launched its self-developed MiMo-V2 series large models, with the flagship model MiMo-V2-Pro achieving top-tier performance globally in various assessments [1][13][15]. - The demand for AI is driving a price increase among major domestic cloud service providers, while JD Cloud is adopting a counter-cyclical pricing strategy [1][19][21]. Summary by Sections Industry Investment Rating - The computer industry is expected to outperform the market index over the next 6-12 months [34]. Key Developments in Embodied Intelligence - Yu Shu Technology's CEO discussed the need to enhance model generalization, data utilization efficiency, and scalable reinforcement learning to achieve the "ChatGPT moment" [1][10]. - The industry is currently facing significant challenges in developing models that can perform reliably in unfamiliar scenarios [10][12]. Xiaomi's MiMo-V2 Series Launch - Xiaomi introduced three models: MiMo-V2-Pro, MiMo-V2-Omni, and MiMo-V2-TTS, with MiMo-V2-Pro being optimized for high-intensity agent tasks and achieving top-tier global performance [1][13][15]. - The pricing strategy for MiMo-V2-Pro is aggressive, with a tiered billing model aimed at attracting developers [16][17]. Cloud Service Pricing Trends - Major cloud providers like Alibaba Cloud and Baidu Cloud have raised prices by up to 34% due to surging AI demand and supply chain pressures [19][21]. - JD Cloud is maintaining stable pricing and even reducing prices on several products to capture market share during this inflationary period [19][21].

本文来自微信公众号： 字母AI ，作者：毕安娣，题图来自：视觉中国 实在搞不定，就借借谷歌的兵。 最近Meta不顺，最典型的就是新模型"牛油果"项目推迟。 新模型推迟的原因很简单，效果不及预期，比得上去年3月的Gemini 2.5，但是比不上去年11月的 Gemini 3.0。 以下文章来源于字母AI ，作者毕安娣 字母AI . 聚焦前沿科技，抢先看到未来。 在相关爆料中，有一个容易被忽略的细节是，相关人士称，Meta高层曾讨论在"牛油果"能支棱之 前，暂时使用Gemini顶上。 就在今年1月，苹果宣布和谷歌达成合作，下一代Siri和Apple智能核心将基于Gemini模型和谷歌云技 术。 巨头之间的AI竞争有多激烈，已不必多费口舌去描述。在如此激烈的竞争中，谷歌Gemini却默默混 成了硅谷"托底王"。 谷歌帮帮忙 如果只看表面，Gemini成为业内的"托底王"多少有些反常。 毕竟无论是Meta还是苹果，过去几年都在把人工智能能力视为自己的核心战场，按常理说，真到关 键时候，最不愿意依赖的恰恰应该是竞争对手的模型。 但今年以来接连出现的两个案例，却把同一个事实摆到了台面上： 当自家路线暂时撑不住，或者核 心 ...