21世纪经济报道记者彭新 不谈游戏显卡，英伟达在本届GDC（游戏开发者大会）上的重点是推动更多用户使用AI制作内容，进 一步降低开发门槛。 性能优化方面，英伟达称，自去年9月以来，ComfyUI在RTX GPU上的体验性能已提升40%；在GeForce RTX 50系列GPU上，使用NVFP4数据格式时性能最高可提升2.5倍，并减少60%的显存占用，使用FP8数 据格式的性能最高可提升1.7倍，并减少40%显存占用。 英伟达还尝试提高制作4K视频的效率，其介绍，此前制作4K视频时，创作者通常会先生成低分辨率预 览以保证效率，随后再进行画质提升，目前处理10秒4K视频通常耗时数分钟。对此，英伟达称，借助 RTX Video超分辨率技术，用户现在能极速将视频提升至 4K，该技术显存占用极低，速度比其他主流 本地分辨率提升工具快30倍。 英伟达大力推动本地AI视频制作流程的背景是，AI生成内容大幅增长。2024年初，OpenAI的Sora以惊 人的画面逼真度让"AI生成视频"从概念走入现实，被市场接受，随后字节跳动和快手迅速跟进，推出即 梦和可灵，快速笼络一批专业创作者，制作成本低廉的AI短片。2025年，"AI生 ...

Group 1 - Zhaoyi Innovation (03986) saw an intraday increase of 1.72% amid rising shipping risks in the Hormuz Strait affecting the supply chain of key raw materials like helium [1] - Rongchang Bio (09995) rose over 3.6% after its RC288 injection application was accepted by NMPA, showing excellent anti-tumor activity and safety in preclinical studies [1] - Swire Properties (01972) increased by over 2.2% following the release of its 2025 full-year results, reporting revenue of HKD 16.041 billion, a year-on-year increase of 11%, and a basic earnings per share of HKD 1.49 [1] Group 2 - Health 160 (02656) surged over 10%, reaching a new high of HKD 140.5, with a nearly doubled stock price since March 9, despite a previous drop of about 30% on February 11 [2] - Gaming stocks collectively rose, with notable increases in companies like Boyaa Interactive (00434) up 4.44% and Tencent (00700) up 1.1%, following Apple's announcement of a commission rate adjustment for the App Store in mainland China [2] - Qidian Guofeng (01280) experienced a significant rise of over 26% after announcing a sales contract for AI servers with an independent third party [2] Group 3 - Hong Kong bank stocks fell again, with Standard Chartered (02888) down 4.67% and HSBC Holdings (00005) down 3.8%, amid ongoing tensions in the Middle East affecting transactions involving Asian balance sheets [3] - Cement stocks saw a general increase, with China National Building Material (03323) up 2.62%, as construction activity picked up post-Lantern Festival, leading to a steady recovery in cement market demand [3] - Domestic property stocks rebounded, with CIFI Holdings (00884) up 2.9% and Sunac China (01918) up 5.45%, as recent data indicated a 3.3% year-on-year decline in second-hand housing listings in Shenzhen [4] Group 4 - Yao Cai Securities (01428) saw a significant rise of over 39% after extending the acquisition offer deadline with Ant Group to March 25, 2026 [4] - PayPay (PAYP.US), a digital wallet operator backed by SoftBank, debuted on the US stock market with a 13.5% increase, achieving a market cap of nearly USD 12 billion [5] - Chinese electric vehicle companies NIO (NIO.US) and Xpeng Motors (XPEV.US) saw stock increases of 1.46% and 3.58%, respectively, amid discussions of potential collaborations with European automotive giant Stellantis [5] Group 5 - Storage stocks collectively declined, with SanDisk (SNDK.US) down 5.59% and Micron Technology (MU.US) down 3.19%, following negative sentiment from short-seller Citron Capital [6] - Optical communication stocks fell, with Applied Optoelectronics (AAOI.US) down 16.39%, despite securing a bulk order for a new data center transceiver [6] - Agricultural input stocks continued to rise, with CF Industries Holdings (CF.US) up 13.21%, driven by supply chain disruptions in the Middle East affecting fertilizer transportation [7] Group 6 - Oil stocks rose sharply, with Battalion Oil (BATL.US) up 15.48% as international oil prices surged, with WTI crude rising over 8% to USD 94.66 [8] - EHang Intelligent (EH.US) reported total revenue of RMB 509.5 million (approximately USD 72.9 million) for the fiscal year 2025, marking an 11.7% year-on-year increase, despite a net loss of RMB 231 million [8]

这是东方港湾创始人但斌为中信出版集团今年 3 月出版的新书《 " 十五五 " 开局》所写的一篇文章。 隐藏福利： 在评论区分享"未来十年最看好什么"，点赞前 5 名的读者朋友，将获赠 《 " 十五五 " 开局》 。 中奖公示： 2026 年 3 月 28 日前 《 " 十五五 " 开局》，中信出版集团 3 月出版 放在当下来看，这篇文章有它特别的现实意义。市场对于人工智能的分歧依然很大，但斌态度一贯的鲜明：如果把时间拉长，人工智能并不是一轮普通的主 题投资，而很可能是未来十年最重要的产业变量之一。 这些年但斌确实也是在用自己的持仓来表达态度。 经出版社授权，我们将这篇文章分享给大家。 根据最新披露的 13F ，截至去年底，东方港湾美股持仓总市值约 13.16 亿美元，持仓标的缩减至 10 只，组合进一步集中。 汇集了过去一年多他的一些言论，也有一些新的表达，可以完整地看到一个投资从业 30 多年的"老价投"如何迭代认知、穿越周期并在 AI 时代中激流勇进 的过程。 谷歌取代英伟达成为第一大重仓股，英伟达位列第二，同时微软、苹果、 Meta 、亚马逊等美国科技巨头也都是核心配置。 延伸阅读： 如何看待高成长与经 ...

Group 1 - ByteDance is collaborating with Southeast Asian firm Aolani Cloud to deploy approximately 500 Nvidia Blackwell computing systems in Malaysia, which will utilize around 36,000 B200 chips [1] - This move signifies ByteDance's efforts to enhance its computing power outside of China, particularly in the context of its operations related to TikTok [1] - The partnership with Aolani Cloud indicates a strategic expansion for ByteDance in the AI and cloud computing sectors [1]

Core Insights - TikTok's parent company aims to expand its global presence and compete with major players like Google and OpenAI by providing a variety of AI applications tailored for everyday users [1] Company Strategy - The company is focusing on developing AI applications that cater to the needs of general consumers, indicating a shift towards more accessible technology solutions [1]

此后板块就没什么催化了。 所以近期可能确实会是AI板块最好的一次做多机会。 转，AI接下来两个催化 OFC 2026（光纤通信大会） 技术会议：3 月 15 日–3 月 19 日OFC 展览：3 月 17 日–3 月 19 日OFC 地点：美国洛杉矶会议中心OFC NVIDIAGTC 2026（GPU 技术大会） 大会 / 展览：3 月 16 日–3 月 19 日NVIDIA 黄仁勋主题演讲：3 月 16 日（周一）11:00 预判：从这 转，AI接下来两个催化 OFC 2026（光纤通信大会） 技术会议：3 月 15 日–3 月 19 日OFC 展览：3 月 17 日–3 月 19 日OFC 地点：美国洛杉矶会议中心OFC NVIDIAGTC 2026（GPU 技术大会） 大会 / 展览：3 月 16 日–3 月 19 日NVIDIA 黄仁勋主题演讲：3 月 16 日（周一）11:00 预判：从这两天板块走势来看，只要后面没有大的地缘政治利空，那么资金的想法应该是顶着做到下周的OFC和 GTC大会，这两场大会均有非常高的关注度。 2.NV，微软，meta都计划采用OCS，下周OFC大会OCS会是热点议题之一 ...

公众号记得加星标⭐️，第一时间看推送不会错过。 为了降低每比特能耗并提高带宽，人工智能数据中心开始用共封装光器件取代铜线。 使用CPO的优势在于，它可以将光连接更靠近ASIC、GPU或CPU，从而减少对长而低效的电线的需 求。虽然CPO本身并非低功耗，但它能提供极低的每比特传输能耗——这是衡量数据中心效率的关键 指标。 CPO 没有采用传统的远距离可插拔收发器，而是将光子引擎直接集成到同一封装基板上或同一模块 内，使得电信号在多芯片组件中只需传输几毫米，而不是像通常那样从芯片到前面板需要通过 PCB 传输 15 到 30 厘米。 这是人工智能发展中至关重要的一环。据麦肯锡公司预测，到2030年，满足全球人工智能需求将需要 5.2万亿美元的数据中心投资。因此，解决电力和带宽挑战对于确保超大规模企业获得最佳投资回报 至关重要。在采用CPO（计算机化产品）方面处于领先地位的公司包括博通、英伟达、英特尔、 Marvell和Ayar Labs，它们拥有各种技术和产品，并由GlobalFoundries、IBM、英特尔晶圆代工、 Tower Semi和台积电等代工厂提供支持。EDA工具供应商Cadence、Keysigh ...

Core Viewpoint - The release of the GeForce 3 GPU marked the beginning of the artificial intelligence revolution, transitioning NVIDIA from fixed-function accelerators to programmable shaders, allowing developers greater control over game aesthetics [2][3]. Group 1: Historical Context and Technological Evolution - Jensen Huang highlighted that the limitations of fixed-function accelerators in the late 1990s led to a significant transformation in NVIDIA's approach, enabling a shift to programmable vertex and pixel shader architecture [2]. - The introduction of CUDA was facilitated by this new programming paradigm, which introduced parallel computing capabilities for GPU computing [2][3]. Group 2: Artistic Expression in Gaming - Huang emphasized that gaming serves as a medium for artistic expression, necessitating the ability to programmatically convey artistic elements, which was previously constrained by the lack of robust compiler technology [3]. - The transition to programmable hardware pipelines was crucial for NVIDIA's evolution into a computing company [3]. Group 3: Advancements in Graphics Technology - NVIDIA's pioneering efforts in ray tracing technology, which is computationally intensive, have paved the way for advancements like DLSS (Deep Learning Super Sampling), introducing generative capabilities to computer graphics [3]. - Continuous improvements in computing and rendering capabilities are seen as foundational for the development of generative AI technologies [3]. Group 4: Future Outlook - Despite the advancements in computer graphics technology, the outlook for the gaming industry and NVIDIA's prospects appears less optimistic at present [4]. - NVIDIA is actively working to break traditional computing barriers by developing AI-driven techniques for frame generation, reducing the need for extensive hardware [4].

公众号记得加星标⭐️，第一时间看推送不会错过。 随着人工智能集群规模的不断扩大，它们开始使用光互连来实现横向扩展连接。然而，它们 可能很快就会需要使用光互连来实现纵向扩展连接。为了应对这一挑战，超大规模数据中心 运营商微软、Meta 和 OpenAI与硬件设计公司 AMD、博通和英伟达合作，共同开发了一种 与协议无关的、适用于人工智能集群的纵向扩展互连技术。 为此，本周这些公司成立了光计算互连（OCI:optical compute interconnect）多源协议（MSA： Multi-Source Agreement ）小组，旨在为大型人工智能系统和机架内部的扩展互连定义一个开放的 光连接规范，使超大规模数据中心能够使用光缆而非铜缆，以高速和可预测的功耗连接更多加速器。 实际上，这意味着该联盟将开发一个通用的光物理层（PHY）和统一的组件，以支持各种协议，例如 AMD和博通的UALink协议以及英伟达的NVLink协议。 用于人工智能机架和扩展集群的短距离光链路的OCI连接技术将定义一个基于NRZ信令和波分复用 (WDM)的通用物理层(PHY)，初始配置为4个波长×50 Gb/s（单向200 Gb/s） ...

Core Insights - The demand for tokens and AI computing is experiencing explosive growth, driven by advancements in model capabilities and rapid development of intelligent workflows, leading to a surge in user adoption and total token demand [3] - Anthropic has added up to $6 billion in annual recurring revenue (ARR) in February, primarily due to the widespread application of its AI coding platform, Claude Code [3] - Despite significant investments in AI infrastructure over the past few years, available computing resources remain scarce, with rising prices for on-demand GPUs [3][5] Group 1: AI and Semiconductor Demand - The demand for TSMC's N3 logic wafers is primarily driven by consumer electronics, but by 2026, AI will become the main source of demand for N3 wafers as the industry transitions to this technology [10][18] - By 2026, AI-related applications are expected to account for nearly 60% of total N3 chip production, with the remaining 40% for smartphones and CPUs [18] - The transition to N3 technology is being accelerated by major companies like NVIDIA, AMD, Google, and AWS, all of which are moving their AI accelerators to N3 nodes [11][17] Group 2: Supply Chain Constraints - TSMC is facing a silicon chip shortage that is limiting its ability to meet the growing demand for N3 wafers, despite plans to expand capacity [5][23] - The effective utilization rate of N3 processes is expected to exceed 100% by the second half of 2026, as TSMC maximizes its existing production lines [23] - The shortage of memory, particularly DRAM and HBM, is becoming a critical constraint, with HBM capacity experiencing rapid growth due to increased memory requirements for AI accelerators [30][36] Group 3: Market Dynamics - The smartphone market may become a release valve for N3 wafer demand, as expected low growth in smartphone shipments could free up capacity for AI accelerators [26] - If smartphone N3 wafer production is reduced, it could potentially allow for the production of additional AI chips, such as NVIDIA's Rubin GPUs and Google's TPU v7 [26][27] - The competition for HBM and DRAM is intensifying, with memory suppliers needing to adjust their production strategies in response to changing market demands [38][40]