"我们将直接在芯片、软件、系统乃至AI（人工智能）工厂层面与OpenAI合作，帮助他们成为一个完全 自运营的超大规模公司。" 近日，英伟达（Nasdaq：NVDA）CEO黄仁勋在一档播客节目中透露，英伟达与OpenAI达成一项新合 作，后者希望建立自己的完整技术栈。 AI浪潮扑面而来，全球科技巨头正在加大AI基础设施建设力度，全栈AI布局正成为科技巨头角逐AI的 核心壁垒。 《每日经济新闻》记者（以下简称"每经记者"）注意到，在9月底落幕的云栖大会上，中国科技巨头阿 里云在核心定位上正悄然发生转变。阿里巴巴集团CEO吴泳铭表示，"在这个崭新的时代，阿里云的定 位是全栈人工智能服务商"。 基于这一论断，黄仁勋解释，在传统的AI规模定律（预训练、后训练）之上，引入了全新的"思考"推理 定律——即在回答前深度思考、研究和学习。这将使推理能力呈指数级增长（百万倍乃至十亿倍），最 终将智能推向新高度。 事实上，当前全球能实现全栈自研的科技公司仍屈指可数。全栈AI布局，意味着从底层AI基础设施， 譬如芯片、数据中心，云计算层以及模型层等，都需要全栈AI布局厂商有足够的研发和资本支出力量 支撑。眼下，全球AI领域的军备竞赛 ...

以下文章来源于投中网 ，作者刘燕秋 投中网 . 投中网是领先的创新经济信息服务平台，拥有立体化传播矩阵，为创新经济人群提供深入、独到的智识和洞见，在私募股权投资行业和创新商业领域拥有 权威影响力。官网：www.chinaventure.com.cn 融了超过30亿美元。 文 ｜ 刘燕秋 来源｜ 投中网（ID：China-Venture） 封面来源 ｜ 视觉中国 当英伟达宣布达成跟 OpenAI 最高 1000 亿美元的合同，它的竞争对手， AI 芯片初创公司 Groq 也刚刚宣布完了一笔 7.5 亿美元（约合人民币 50 亿元） 的最新融资，融资后估值为 69 亿美元（约合人民币 490 亿）。这一数字超过了 7 月间的传闻。当时有报道称， Groq 的融资额将达到约 6 亿美元，估值 接近 60 亿美元。 资本正高度关注 AI 推理芯片赛道—— Groq 曾于 2024 年 8 月以 28 亿美元的估值融资 6.4 亿美元，这意味着，在短短一年多的时间里，估值翻了一倍 多。本轮融资由 Disruptive 领投，此外也获得了来自贝莱德、 Neuberger Berman 集团有限责任公司和德国电信资本的"重 ...

本文来自微信公众号：投中网 （ID：China-Venture），作者：刘燕秋，题图来自：AI生成 当英伟达宣布达成跟OpenAI最高1000亿美元的合同，它的竞争对手，AI芯片初创公司Groq也刚刚宣布 完了一笔7.5亿美元（约合人民币50亿元）的最新融资，融资后估值为69亿美元（约合人民币490亿）。 这一数字超过了7月间的传闻。当时有报道称，Groq的融资额将达到约6亿美元，估值接近60亿美元。 资本正高度关注AI推理芯片赛道——Groq曾于2024年8月以28亿美元的估值融资6.4亿美元，这意味着， 在短短一年多的时间里，估值翻了一倍多。本轮融资由Disruptive领投，此外也获得了来自贝莱德、 Neuberger Berman集团有限责任公司和德国电信资本的"重大投资"，以及包括三星电子、思科、D1 Capital和Altimeter在内的现有投资者的出资。 根据半导体产业研究，全球AI芯片市场正处于高速增长期，2023年市场规模只有231.9亿美元，预计至 2029年将以31.05%的复合年增长率攀升至1175亿美元。随着大语言模型从研发走向应用，AI产业重心 正从训练阶段转向推理环节。英伟达财 ...

将投中网设为"星标⭐"，第一时间收获最新推送 融了超过30亿美元。 作者丨 刘燕秋 来源丨 投中网 当英伟达宣布达成跟 OpenAI 最高 1000 亿美元的合同，它的竞争对手， AI 芯片初创公司 Groq 也刚刚宣布完了一笔 7.5 亿美元（约合人民币 50 亿元）的最新融资，融资后估值为 69 亿美元（约合人民币 490 亿）。这一数字超过了 7 月间的传 闻。当时有报道称， Groq 的融资额将达到约 6 亿美元，估值接近 60 亿美元。 资本正高度关注 AI 推理芯片赛道—— Groq 曾于 2024 年 8 月以 28 亿美元的估值融资 6.4 亿美元，这意味着，在短短一 年多的时间里，估值翻了一倍多。本轮融资由 Disruptive 领投，此外也获得了来自贝莱德、 Neuberger Berman 集团有限 责任公司和德国电信资本的"重大投资"，以及包括三星电子、思科、 D1 Capital 和 Altimeter 在内的现有投资者的出资。 根据半导体产业研究，全球 AI 芯片市场正处于高速增长期， 2023 年市场规模只有 231.9 亿美元，预计至 2029 年将以 31.05% 的复合年增 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 过去几年，云厂商为了训练大模型投入巨资购买芯片，如今也到了利用推理实现变现的时候了。根据麦肯锡报告，全球AI推理市场规模预计2028年将达1500亿美 元，年复合增长率超40%，远高于训练市场的20%。推理支撑着各类应用的实时推理需求，包括智能推荐、内容生成、虚拟助手等。可以说，推理阶段才是实现实 际应用和商业化的关键。 这场推理之战，随着华为、英伟达和谷歌三大巨头相继发布了各自的推理芯片之后，已经将正式打响！ 华为Ascend 950PR： 成本优化下的推理利器 9月18日，在2025年华为全联接大会上，华为宣布了昇腾芯片的规划和进展。未来3年，也就是到2028年，华为在开发和规划了三个系列，分别是Ascend 950系列、 Ascend 960、Ascend 970系列。华为表示，将以几乎一年一代算力翻倍的速度，同时围绕更易用，更多数据格式、更高带宽等方向持续演进，持续满足AI算力不断 增长的需求 以往每年9月，都是手机发烧友的狂欢月，因为这时期苹果、小米、华为等都会发新机。然而，今年的9月，一个更深层次的产业变革正在暗流涌动。当所有人都在 对iphone ...

资料来源：聚源，中邮证券研究所 公司基本情况 | 最新收盘价（元） | 77.85 | | --- | --- | | 总股本/流通股本（亿股）4.64 | / 4.44 | | 总市值/流通市值（亿元）361 | / 346 | | 52 周内最高/最低价 | 87.66 / 26.25 | | 资产负债率(%) | 60.4% | | | 深圳市领信基石股权投 | | | 资基金管理合伙企业 | | 第一大股东 | (有限合伙)-深圳市领泰 | | | 基石投资合伙企业(有 | | 限合伙) | | 研究所 证券研究报告：电子 | 公司点评报告 股票投资评级 分析师:吴文吉 SAC 登记编号:S1340523050004 Email:wuwenji@cnpsec.com 香农芯创(300475) 企业级 SSD 需求提升，分销受益 DDR4 存储涨价 l 投资要点 企业级 SSD 需求提升。AI 推理需求催生庞大数据存储需求， 企业级 SSD 需求将快速增长，根据华经产业研究院，2023 年全球 数据量约为 120ZB，预计到 2028 年，全球数据量将上涨至 393.8ZB，带动中国企业级 SSD 行业 ...

Group 1 - The core viewpoint of the articles indicates a significant shift in demand towards Nearline SSDs due to the rising needs for AI inference services, as traditional HDDs face severe supply shortages [2] - The DRAM market is experiencing price increases, with conventional DRAM prices expected to rise by 8-13% in Q4 2025, driven by prioritization of advanced process capacities for high-end server DRAM and HBM [3][4] - The production capacity for LCD TV panels is projected to decrease, with an estimated drop in utilization rates from manufacturers' planned levels, indicating a slowdown in demand [7] Group 2 - NAND Flash prices are anticipated to rise by 5-10% in Q4 2025, influenced by a shift in storage demand towards QLC enterprise SSDs due to HDD supply shortages and extended lead times [8]

Mergers and Acquisitions - Hualing Cable plans to acquire control of San Bamboo Intelligent for no more than 270 million yuan, with San Bamboo being a leading provider of industrial connection product solutions in the industry [1] Share Buybacks - Shennma Co. intends to repurchase shares worth between 100 million and 200 million yuan [2] - Fenghuo Communication plans to repurchase shares worth between 75 million and 150 million yuan and will cancel the repurchased shares [2] Investment Cooperation and Operational Status - Yuntian Lifei's self-developed AI-driven product, Luka Doctor AI plush toy, is expected to launch by the end of 2025. The company is also developing the next-generation "brain" chip, DeepXBot series, to enhance humanoid robots' perception, cognition, decision-making, and control tasks [3] - Zhejiang Haideman is in the prototype trial phase for its humanoid robot and robotic dog projects [3] - Haimuxing is among the first in the industry to achieve the "oxide + lithium metal anode" technology route and has completed the commercial closed loop for lithium metal solid-state batteries, which are currently being delivered in batches [3] - Guomai Culture is investing in the film "The Starry Sky of the Three Kingdoms Part One," set to premiere on October 1 [3] - Visionox plans to use Suzhou Guoxian as the project company to carry out preliminary work for the Kunshan Global New Display Industry Innovation Center project [4] - Fola New Materials has collaborated with several domestic and international dexterous hand and humanoid robot companies [5] - Jiuwu High-Tech signed a procurement contract for a membrane treatment system worth 81.5 million yuan with Guotou Xinjiang Lithium Industry Co., Ltd. [6] - Wolker Materials' controlling subsidiary plans to invest 1 billion yuan in the construction of the Kote (Suzhou) New Materials Project [7] - Heng Rui Pharmaceutical's SHR7280 tablet application for marketing approval has been accepted by the National Medical Products Administration, with no oral GnRH antagonists approved in the domestic and international assisted reproductive fields [7] - Ganli Pharmaceutical signed a "Technology Transfer and Supply Agreement" with FZ and BIOMM for a production partnership plan in Brazil, with a total contract value expected to be no less than 3 billion yuan (including tax) [7] - Sinovac Biotech's human interferon α1b inhalation solution has been included in the list of breakthrough therapeutic varieties [8] - Baijin Medical's radiofrequency ablation system has entered the special review process for innovative medical devices [8] - Jiangbolong anticipates an increase in demand for server NAND market inventory in the second half of the year, with storage market prices expected to rise in the fourth quarter. The company's enterprise-level PCIe SSD and RDIMM products have begun bulk imports to leading domestic enterprises [8] - Jingzhida has completed its annual target ahead of schedule for the high-speed FT testing machine, with steady progress on the KGSD CP testing machine, continuously iterating and upgrading in conjunction with storage industry technology solutions [8] - Fuhuang Steel Structure signed a strategic cooperation agreement with the Twenty-Second Metallurgical Group [9] - Poly Development plans to issue company bonds not exceeding 15 billion yuan [10] - Shangwei New Materials has completed the transfer of shares and the controlling shareholder has changed to a holding platform established by an affiliate of Zhiyuan Robotics [11]

Core Viewpoint - AMD has transformed from a laggard to a strong competitor in the data center CPU market, driven by advantages in CPU and a shift towards AI accelerators, despite short-term investor focus on volatility [1][2]. Group 1: Data Center Growth - AMD has established a multi-engine growth infrastructure encompassing CPU, GPU, AI PC, and future rack systems, which will drive long-term growth in the CPU market [2]. - The data center is the key battleground for AMD, with inference task computing power now surpassing training as the primary driver of demand [2]. - AMD's data center revenue reached $3.24 billion in Q2 2025, a 14% year-over-year increase, attributed to record EPYC CPU sales [2]. Group 2: AI Accelerator Performance - AI GPU revenue has declined due to the transition from MI300 to MI350 and export controls, leading to an $800 million inventory write-down [4]. - The upcoming MI355 is expected to outperform NVIDIA's B200 by processing 40% more tokens per dollar, a critical metric for large-scale companies [4][6]. Group 3: Long-term Differentiation - AMD's acquisition of ZT Systems enhances its capabilities in rack-level design, allowing it to compete directly with NVIDIA's offerings [9]. - The focus on autonomous AI projects across various markets positions AMD as a viable alternative to NVIDIA, especially in government projects requiring domestic infrastructure [12]. Group 4: Financial Outlook - AMD's expected P/E ratios for FY2025 and FY2026 are 40.4 and 26.2, respectively, aligning with industry medians, indicating potential for growth despite initial high valuations [13]. - Revenue is projected to grow from $33 billion in 2025 to $40.1 billion in 2026, with expected EPS increasing from $3.90 to $6.01, a 55% rise [13]. Group 5: Investment Perspective - AMD is viewed as a stable investment, with EPYC CPUs providing consistent revenue and GPUs and rack systems poised for significant growth in inference and autonomous AI demand [16]. - Recent stock price adjustments are seen as healthy corrections rather than trend reversals, with potential for long-term shareholder returns if AMD successfully executes its GPU roadmap [16].

Core Insights - The focus of AI infrastructure development in the next two years will shift towards supporting efficient inference services, leading to a demand for Nearline SSDs designed specifically for inference AI due to a severe shortage of traditional large-capacity HDDs [2][3] - Major NAND Flash suppliers are accelerating the validation and introduction of Nearline QLC NAND Flash products, which can store more data at a lower cost, addressing the growing demand for large capacity [2][3] - The competition among NAND Flash suppliers is driving the diversification of NAND Flash applications, moving from simple storage to deeper integration with AI computing [4] Group 1 - The demand for Nearline SSDs is expected to continue until 2027, with a tight supply of Enterprise SSDs anticipated in 2026 due to the expansion of inference AI applications [2] - Suppliers are actively developing new Nearline SSD products that not only exceed mainstream HDD capacities but also optimize costs and significantly reduce power consumption [3] - NAND Flash suppliers are also launching High Bandwidth Flash (HBF) products to capture market share in AI training applications, indicating a dual technology route in the industry [4] Group 2 - The competition in NAND Flash technology is characterized by two main camps: one led by SanDisk, focusing on a hybrid design of HBM and HBF, and the other represented by Samsung and Kioxia, which aims to provide more cost-effective solutions with SCM and XL-Flash technologies [4] - This technological competition is expected to enhance the integration of NAND Flash in AI operations, indicating a trend towards more diverse applications in the future [4]