在胡润中国500强TOP10阵容中，被称为深圳创新"四姐妹"的华为、比亚迪、腾讯、中国平安均上榜， 深圳也成为榜单TOP10总部企业数量最多的城市。 伴随《2025胡润中国500强》出炉， 中国500家非国有企业最新座次明朗，一批深圳企业也引来关注。 水涨船高成为看点。南都湾财社记者看到，今年胡润中国500强的上榜门槛比上一年上升75亿元，达到 340亿元，涨幅接近三成。 沉浮起落成新常态。按照榜单，有95家新上榜企业，这些新面孔主要来自消费电子、AI算力、新能源 等领域；16家企业的价值与上一年持平；102家企业的价值较上一年有所下降；与此同时，上一年榜 单"座上宾"中，有99家企业在今年落榜。其中，房地产行业上榜企业数量从去年的19家减少至12家。 | 排名 | 公司 | 价值(亿元人民币) | 涨幅 | | --- | --- | --- | --- | | 1-5 | 台积电 | A - 1 | 50% | | 2- | 腾讯 橙股 | 53,300 | - 56%5 | | 3- | 字节跳动 | 34,000 | 109% | | 4- | 阿里巴巴 | 27,000 | 75% | | 025- ...

格隆汇2月5日丨西子洁能(002534.SZ)在投资者关系中表示，有被问到：公司国内AIDC领域有什么新进 展？ 第二，合作开展面向云厂商的绿色算力中心服务。清微智能负责提供算力底座与建设支撑，西子洁能贡 献先进的储能技术与零碳建设方案，共同帮助云厂商优化PUE(能源使用效率)指标与可持续发展目标， 拓展高增长市场。公司熔盐储能技术在AIDC场景中展现出独特的适配性：一方面能实现电力的"移峰填 谷"，有效平抑电网负荷波动，保障算力中心供电的稳定性与经济性；另一方面，其储热特性可与液冷 等先进散热技术深度结合，为高密度算力集群提供高效、精准的温控解决方案，真正达成"电—热— 算"一体化协同。 公司此次与清微智能的战略携手，是公司主动布局"AI+能源"交叉领域的关键一步。展望未来，双方将 通过组建专项工作小组、建立常态化沟通机制，将协议蓝图转化为具体项目，共同探索算力中心储能的 创新模式，为我国人工智能产业的绿色可持续发展，注入一股强劲而智慧的"零碳"动力。 答：1月16日，公司与北京清微智能科技有限公司正式签署战略合作协议。双方将围绕人工智能数据中 心(AIDC)建设，推动算力和能源建设等全方位合作，共同开启" ...

Core Insights - A collaboration worth 1.68 billion yuan has been signed between Shenzhen Tiandun Data Technology Co., Ltd. and Jiangsu Wukong Digital Industry Group Co., Ltd. to enhance high-performance computing capabilities in Lianyungang City, Haizhou District, marking a significant step in developing the regional AI industry cluster [1][3] Group 1: Collaboration Details - The partnership aligns with Haizhou District's development strategy of "computing power hub + scenario-driven + ecological aggregation," aiming to accelerate the layout of the AI industry [3] - The collaboration will establish an inclusive computing service platform to enhance R&D innovation and intelligent transformation for local enterprises, thereby attracting high-end digital industry elements and promoting industrial ecosystem aggregation [3] Group 2: Technical and Market Integration - Tiandun Data will leverage its core technical advantages in data center construction, system integration, and computing platform operation to support the project [3] - Wukong Digital Group will utilize its extensive government and enterprise client channels, quality market resources, and mature industry application scenario excavation capabilities to create a complete business loop from infrastructure to service application [3] Group 3: AI and Biopharmaceutical Focus - The collaboration includes the establishment of an "AI + Biopharmaceutical Empowerment Center," focusing on deep integration of AI and biomedicine, particularly in drug development [4] - The center aims to create a professional large model system for drug research and development, ultimately serving as a regional benchmark platform for AI and biopharmaceutical innovation [4] Group 4: Company Background and Achievements - Wukong Digital Group is a leading enterprise in Lianyungang's intelligent computing industry, focusing on AI computing infrastructure and ecosystem construction [4] - The company has received authoritative certifications and has successfully established significant projects, including the first public safety AI training center in the country [4] - In January 2026 alone, Wukong Digital Group's signed computing cooperation orders exceeded 600 million yuan, indicating robust growth in the industry [4]

(3)华为在2025年9月的全联接大会上宣布3年算力计划，正在开发Ascend 950、未来布局960和970系列， 其中950 8192卡超节点有望在26Q4正式推出。 智通财经APP获悉，广发证券发布研报称，1月22日，ODCC举办2026超节点大会，本次大会聚焦于AI 算力的关键演进方向：Scale-Up(即超节点)，旨在应对万亿参数大模型训练与推理带来的单机性能受限 等核心挑战。聚焦Scale-Up变化，阿里、腾讯、华为等云厂商/算力厂商超节点渗透率有望持续超预期。 现阶段的Scale-Up网络大多采用铜缆互联的方案，但在系统设计、传输距离上有明显局限性，建议关注 交换设备、NPO、DAC/AEC等赛道。 广发证券主要观点如下： 当前模型推理的核心特征(长上下文、高并发、实时交互)推动Scale-Up升级 模型推理逐渐具备更加复杂的多步骤决策，还会与用户进行多轮对话交互，产生连续的、超长上下文、 实时的推理需求。长上下文推理依赖KV Cache 缓存机制，但缓存数据在分布式环境下需频繁跨节点传 输，若通信带宽或延迟不足，GPU会因等待数据而空闲，导致效率下降。同时，这些KV通常存在多张 GPU的显存里 ...

来源：科技日报 科技日报记者 崔爽 日前，北京智谱华章科技股份有限公司通过其官方公众号发布GLM Coding Plan限售公告。公告指出， 随着GLM-4.7系列模型上线，用户数量迅速增长，导致算力资源出现阶段性紧张。 不过，国产AI算力近两年也在持续突破。据咨询公司摩根士丹利2025年发布的研究报告估算，中国人 工智能GPU自给率已从2020年不足10%提升至2024年约34%，并有望在2027年升至约82%。在这一趋势 下，算力产业发展正从过去单点硬件的性能追赶，转向更加务实、高效的系统级创新。 第二问：算力资源供需矛盾如何造成？ 这是AI产业算力吃紧的一个缩影。随着GPT、DeepSeek等大模型的算法突破和应用普及，算力需求水涨 船高。数据显示，我国AI芯片市场规模预计2028年将超一万亿元，约占全球市场的30%。面对庞大的市 场需求，自主可控的高质量AI算力供给已成为我国抢占人工智能产业应用制高点、全方位赋能千行百 业的前提条件。 目前，我国算力资源供给面临多重限制。高端芯片进口受限，国产GPU芯片在绝对计算性能、能效比、 工艺方面与国际旗舰产品仍有差距；技术创新能力不足，如在芯片设计工具、底层算 ...

Core Insights - The transition of generative AI from experimental to essential for enterprise survival highlights the challenges faced in deploying AI applications, including high computational costs and response delays [2][3][4] Group 1: AI Deployment Challenges - 37% of enterprises deploying generative AI report that over 60% experience unexpected response delays in real-time applications, with significant computational costs leading to losses upon deployment [2][4] - The demand for computational power is growing exponentially, with enterprise AI systems requiring an annual growth rate of 200%, far exceeding hardware technology iteration speeds [3] - The complexity of AI applications has evolved from simple Q&A to intricate tasks, resulting in a paradox where non-scalability leads to no value, while scalability incurs losses [2][3] Group 2: Market Growth and Projections - The global AI server market is projected to reach $125.1 billion in 2024, increasing to $158.7 billion in 2025, and potentially exceeding $222.7 billion by 2028, with generative AI servers' market share rising from 29.6% in 2025 to 37.7% in 2028 [3] - The financial sector's AI applications require millisecond-level data analysis, while manufacturing and retail sectors demand real-time processing capabilities, further driving the need for advanced computational resources [3] Group 3: Cost and Efficiency Issues - The cost of token consumption is rising sharply, with ByteDance's model usage increasing over tenfold in a year, and Google's platforms processing 43.3 trillion tokens daily by 2025 [6] - High operational costs are evident, with AI programming token consumption increasing by approximately 50 times compared to the previous year, while the cost of computational power is decreasing at a rate of tenfold annually [6][7] - The average utilization of computational resources is low, with some enterprises reporting GPU utilization rates as low as 7%, leading to high operational costs [9] Group 4: Structural and Architectural Challenges - The mismatch between computational architecture and the demands of AI applications leads to inefficiencies, with over 80% of token costs stemming from computational expenses [8][9] - Traditional architectures are not optimized for real-time inference tasks, resulting in significant resource wastage and high costs [9][10] - Network communication delays and costs are significant barriers to scaling AI capabilities, with communication overhead potentially accounting for over 30% of total inference time [11] Group 5: Future Directions and Innovations - The future of AI computational cost optimization is expected to focus on specialization, extreme efficiency, and collaboration, with tailored solutions for different industries and applications [16] - Innovations in system architecture and software optimization are crucial for enhancing computational efficiency and reducing costs, with a shift towards distributed collaborative models [13][14] - The industry is moving towards a model where AI becomes a fundamental resource, akin to utilities, necessitating a significant reduction in token costs to ensure sustainability and competitiveness [14][16]

中新网北京1月13日电 (记者 孙自法)记者1月13日从中科曙光获悉，在最近举办的"越来越好"国际设计 大赛中，其推进科技与文化融合的scaleX万卡超集群，从全球69个国家和地区的15691份参赛作品中脱 颖而出，成功摘得大赛最受瞩目的"产品至尊奖"。 目前，人工智能已从辅助工具跃迁为核心生产要素，科学大模型、世界模型等前沿方向的迅猛发展，对 底层算力提出了前所未有的规模与性能要求，而高端算力供给不足、成本高昂已成为人工智能产业发展 的主要壁垒，scaleX万卡超集群的诞生，正是对这一产业痛点的精准回应。 据介绍，作为中科曙光面向大模型前沿方向、科学智能等复杂任务场景打造的大规模智能算力基础设施 方案，scaleX万卡超集群在超节点架构、高速互连网络、存储性能优化、系统管理调度等方面实现多项 创新突破，引领国产大规模算力集群技术进入新阶段。 更值得关注的是，scaleX万卡超集群将工业设计与前沿技术深度融合，运用模块化设计思路使功能区域 划分清晰明了，杜绝繁琐的形式化堆砌，用纯粹的人体工学造型排列成独特的韵律感，在破解产业难题 的同时，也呈现出大国重器简约、厚重、强劲的整体形象视觉美感。 中科曙光国家级工业 ...

Core Viewpoint - Pingao Co., Ltd. is set to launch the Pingyuan AI integrated liquid-cooled workstation series, AI Station, on December 25, 2025, as part of its strategy to accelerate the domestic AI infrastructure, focusing on a comprehensive autonomous computing solution from desktop AI terminals to future cluster training [1] Product Launch - The AI Station series features the Jiangyuan D20 AI accelerator card, which is the first domestically produced cloud AI inference card, with a peak INT8 computing power of 320 TOPS and 256GB of memory, supporting multiple precision calculations [2] - The workstation series includes single-card (D20-S), dual-card (D20-D), and quad-card (D20-Q) models, designed to meet varying computing power needs, with the quad-card model capable of running large models like Qwen 235B and DeepSeek 685B [2] Performance and Design - The AI Station series utilizes a liquid cooling architecture that improves thermal conductivity efficiency by 40% compared to traditional air cooling, allowing for continuous operation without performance throttling [3] - The series operates at a noise level of 28dB, making it suitable for quiet environments such as offices and laboratories, while maintaining a compact design even for the quad-card model [3] Strategic Layout - The launch is part of Pingao's broader strategy to create a comprehensive domestic AI computing ecosystem, featuring a full range of hardware and software solutions to support various AI application needs [4] - The hardware lineup covers edge, desktop, and server cluster levels, with deep compatibility with mainstream domestic CPUs, ensuring supply chain security [4] Software Ecosystem - The company has developed three core software platforms: BingoAIInfra for GPU management, BingoAIStack for model training and deployment, and BingoAIDriver for industry application integration, facilitating AI development and operations [5] Collaborative Ecosystem - Pingao has established strategic collaborations with domestic chip manufacturers and AI algorithm vendors, aiming to promote the large-scale implementation of domestic AI computing ecosystems across various sectors [7] Industry Recognition - The Pingyuan AI integrated workstation series has received multiple industry awards, including the "2025 Annual AI Innovation Product" at the 27th China International Software Expo, highlighting its performance and market impact [7] Future Outlook - In 2026, Jiangyuan Technology plans to launch the T800 AI chip, targeting large-scale AI model training and inference, with significant performance advantages over international competitors [8] - The T800 chip will support advanced computing techniques and is designed to meet the future demands for large-scale computing, aligning with the AI Station's capabilities [9]

Market Overview - The market experienced a strong upward trend with the Shanghai Composite Index rising for six consecutive days, and the Shenzhen Component Index increasing by nearly 1%. Over 4,100 stocks in the market saw gains [1] Spring Market Outlook - The spring market may unfold in two ways: first, capital may rush in to buy on dips, leading to a generally strong market; second, if incremental capital is exhausted and negative news arises, a "deep squat jump" may occur. Currently, there is a strong willingness among A-share investors to capitalize on the spring market, with limited visibility of negative factors [2] - Historical data suggests that sectors with high returns in the first half of the year may see a pullback at the end of the year, while underperforming sectors may experience a rebound. The internal demand sector is highlighted as having sufficient attractiveness and increasing win rates, supported by year-end industry allocation patterns and policies aimed at boosting domestic demand [2] Key Sectors to Watch - Focus on sectors such as insurance, brokerage, non-ferrous metals, AI computing/power semiconductors, retail/personal care/social services/dairy products, aviation, and new energy [3] North American AI Power Supply Issues - North America faces a significant power supply gap, exacerbated by the growing demand for AI Data Centers (AIDC). Traditional rapid energy replenishment methods are limited, making AIDC energy storage solutions more economically viable and quicker to deliver. The demand for AIDC energy storage is projected to increase from 9.6 GWh in 2025 to 21 GWh by 2028, with storage duration extending from 4 hours to 6-8 hours [4] - The global AIDC transformer market is expected to grow significantly, with estimates of 60 billion yuan in 2024 and 264 billion yuan in 2027, reflecting a compound annual growth rate (CAGR) of approximately 64% [4] Transformer Export Data - According to customs data, China's transformer exports totaled 579 million yuan from January to November, marking a year-on-year increase of 36%, indicating sustained high demand in the transformer export market [5] AIDC Concept Stocks - AIDC concept stocks focus on core areas such as computing infrastructure, liquid cooling, power distribution, and network equipment. Key players include: - **Core Computing and IDC Operations**: Companies like Zhongke Shuguang and Inspur Information are leading in liquid cooling and AI server markets [6] - **Liquid Cooling Technology**: Companies such as Yingweike and Qiu Tianwei are key suppliers in the liquid cooling sector, catering to AI server needs [7] - **Power Distribution and Storage**: Companies like Zhongheng Electric and Kehua Data are positioned to meet AIDC power supply demands [8] - **Network and Server Support**: Companies such as Xinyi Sheng and Zhongji Xuchuang are critical suppliers for AI computing network transmission [8]

Core Viewpoint - The Chinese computing power industry is at a strategic crossroads, deciding between continuing a closed technology stack approach or pioneering a new competitive model based on open collaboration [1][3]. Industry Challenges - The current domestic AI computing power industry faces a dilemma of "full-chain internal competition" and "dual barriers," leading to significant industry anxiety. Companies have invested heavily in creating isolated "technology islands," resulting in fragmentation and high adaptation costs for users [4][5]. - The performance gap and ecological barriers present deeper challenges, with domestic chips lagging behind international standards and NVIDIA's CUDA ecosystem creating a strong lock-in effect [4]. Strategic Shift - The solution lies in transitioning from a "closed full-stack" to an "open layered" competitive logic, emphasizing collaboration among various manufacturers to create an industry platform that can systematically challenge dominant players [6][8]. - The establishment of the Photonic Organization serves as a platform to balance competition and cooperation, allowing companies to focus on their strengths while sharing results for mutual benefit [6][8]. Implementation of Open Architecture - Leading IT companies in China are moving away from a "large and comprehensive" model to a "focused and strong + ecological empowerment" approach, concentrating resources on their core competencies while opening other areas to ecosystem partners [8]. - The scaleX super cluster exemplifies this open architecture, achieving significant breakthroughs in system architecture and energy efficiency, with a 20-fold increase in computing density and a PUE of 1.04 [9]. Market Engagement - The open architecture aims to lower the barriers for users transitioning from closed ecosystems, enhancing cost efficiency and optimization for clients, particularly benefiting small AI chip design and software companies [9][10]. - The shift from standardized supply to joint customization is crucial for domestic computing power systems to penetrate mainstream commercial markets [10]. Future Outlook - The competition in the AI computing power industry is evolving into a battle between centralized control models and distributed innovation models based on open standards [14]. - The open path chosen by the Chinese industry reflects a deep understanding of its structural and innovative characteristics, aiming to harness the full potential of its comprehensive electronic information manufacturing chain and vibrant small to medium enterprises [14].