Core Viewpoint - Nvidia is planning to release a modified version of the H20 GPU, referred to as a "special version" for China, based on the Blackwell architecture instead of the original Hopper architecture [1]. Group 1: GPU Specifications and Market Impact - Jefferies analysis suggests that the US may impose a memory bandwidth limit of 1.7-1.8TB/s on GPUs, which could lead Nvidia to replace HBM memory with GDDR6 in the downgraded H20, potentially outperforming gaming GPUs like the RTX5090D [2]. - Another report from GF Securities indicates that the new GPU, possibly named 6000D or B40, is expected to be released in early July, featuring GDDR7 memory with a bandwidth of approximately 1.7TB/s, compared to H20's 4TB/s [3]. - The expected shipment volume for the B40 GPU is projected to reach around 1 million units by the end of 2025, supported by NVLink and CUDA technologies [3]. Group 2: Market Dynamics and Historical Context - Prior to the ban on H20, there were indications that it would be discontinued in favor of a Blackwell-based special version, which was not fully prepared at the time of the ban [3]. - The rapid development and release of the new GPU can be attributed to Nvidia's established design framework and TSMC's advanced manufacturing processes, allowing for a quick turnaround following the H20 ban [3].

星球中每日还会更新Seeking Alpha、Substack、 stratechery的精选付费文章， 现在星球中领券后只需要 390元，即可每天都能看到上百篇外资顶尖投行科技行业的分析报告和每天的精选报告，无论是我们自 己做投资，还是对行业有更深入的研究，都是非常值得的。 还有专注于半导体行业分析的SemiAnalysis的全部分析报告： 想要看外资研报的同学，给大家推荐一个星球，在星球中每天都会上传几百篇外资顶尖投行的原文研 报：大摩、小摩、UBS、高盛、Jefferies、HSBC、花旗、BARCLAYS 等。 ...

Group 1 - The article recommends a platform where users can access hundreds of top-tier foreign investment bank research reports daily, including those from firms like Morgan Stanley, UBS, Goldman Sachs, Jefferies, HSBC, Citigroup, and Barclays [1] - The platform also provides comprehensive analysis reports focused on the semiconductor industry from SemiAnalysis, along with daily updates of selected paid articles from Seeking Alpha, Substack, and Stratechery [3] - The subscription to the platform is currently available for 390 yuan, offering access to numerous technology industry analysis reports and selected articles daily, which is deemed valuable for both personal investment and deeper industry research [3]

Core Viewpoint - The report from Mizuho Securities provides an analysis of companies including Broadcom, NVIDIA, AMD, Supermicro, and Huawei, highlighting the expected growth and challenges in the AI ASIC and GPU markets. Group 1: Broadcom and NVIDIA - Mizuho expects Broadcom's custom ASIC chips (TPUv7p/MTIA2) to accelerate in deployment by 2026, potentially being used in OpenAI's Strawberry and Apple's Baltra projects in the second half of 2026 [1] - In 2024, Broadcom's custom ASIC chips are projected to account for 70%-80% of usage, establishing it as a leader in AI ASICs, excluding self-manufactured AI ASICs like Google's TPU [1] - The UMAIN project in Saudi Arabia plans to deploy 4,000 GB200 NVL72 servers, corresponding to 280,000 NVIDIA GPUs and 350,000 AMD GPUs over the next five years [1] - The G42 project in the UAE has committed to importing 500,000 NVIDIA GB200 GPUs annually, valued at $15 billion, although the sustainability of this figure is questioned [1] Group 2: Huawei - The report anticipates that Huawei's Ascend 910 orders will exceed 700,000 units by 2025, with the next-generation Ascend 920 expected to launch in 2026 [2] - However, the current yield rate for the Ascend 910 is low at only 30%, a figure corroborated by previous reports [2][3] - Other estimates suggest that the shipment volume for the Ascend 910 series could be over 700,000 units this year [5]

还有专注于半导体行业分析的SemiAnalysis的全部分析报告： 想要看外资研报的同学，给大家推荐一个星球，在星球中每天都会上传几百篇外资顶尖投行的原文研 报：大摩、小摩、UBS、高盛、Jefferies、HSBC、花旗、BARCLAYS 等。 星球中每日还会更新Seeking Alpha、Substack、 stratechery的精选付费文章， 现在星球中领券后只需要 390元，即可每天都能看到上百篇外资顶尖投行科技行业的分析报告和每天的精选报告，无论是我们自 己做投资，还是对行业有更深入的研究，都是非常值得的。 ...

Core Viewpoint - Nvidia's introduction of NVLink Fusion aims to enhance flexibility and customization in AI infrastructure while maintaining its technological advantage in the market [8][17]. Group 1: Nvidia's Development and Products - Nvidia has evolved from focusing on GPUs to becoming a giant in AI infrastructure, with significant milestones such as the launch of CUDA in 2006 [1]. - The GB300 chip, set to launch in Q3, boasts a 1.5x improvement in inference performance, HBM memory, and a 2x increase in network bandwidth, while maintaining physical compatibility with previous generations [6]. - The Project DIGITS personal AI computer, DGX Spark, is now in full production, with availability expected by Christmas [6]. Group 2: NVLink Fusion Technology - NVLink Fusion extends Nvidia's NVLink technology to third-party CPUs and accelerators, allowing for a more open ecosystem while still requiring Nvidia chips in the system [8][10]. - The technology includes two components: a semi-custom CPU connection via NVLink C2C and the integration of NVLink 5 Chiplet into third-party accelerators [9][10]. - NVLink Fusion is designed as a "either/or" technology, allowing for either a semi-custom CPU or GPU but not both simultaneously, ensuring Nvidia's presence in the system [10]. Group 3: Market Implications and Partnerships - Current partners for NVLink Fusion include Alchip and AsteraLabs, with Fujitsu and Qualcomm developing new CPUs compatible with Nvidia GPUs [11]. - The limited openness of NVLink Fusion may accelerate diversification in AI computing infrastructure and provide pathways for third-party chips to enter the high-performance computing market [11][17]. - Nvidia's strategy reflects an understanding that a fully closed NVLink could limit market expansion, particularly among cloud service providers and sovereign AI projects [17]. Group 4: NVLink Advantages - NVLink 5 offers a dual bandwidth of 1.8 TB/s, significantly outperforming PCIe 5.0, which is crucial for scaling AI model training and inference [20]. - The NVLink Switch chip enables rack-level scalability, supporting up to 72 GPUs with a total bandwidth of 130 TB/s, a capability that competitors struggle to match [20]. - The integration of NVLink with Nvidia's SHARP protocol and Mission Control software optimizes AI workload throughput and latency, enhancing overall performance [20].

Core Viewpoint - Citi believes that 2026 will be the year when 800G optical modules dominate the market, with a projected sales volume of 37 million units, representing an 85% year-on-year growth driven primarily by demand from overseas cloud service providers [1][3]. Group 1: Market Demand and Supply Dynamics - The demand for 800G optical modules is expected to be robust, with over 90% of this demand coming from overseas cloud service providers (CSPs) [3]. - The deployment speed of 1.6T Ethernet switches may slow down, leading to a downward revision of demand forecasts for 1.6T modules [3]. - The market share distribution will be influenced by production capacity and delivery capabilities, with a preference for second-tier suppliers among overseas customers [3]. Group 2: Company Performance and Predictions - Citi's rating hierarchy for companies in the sector is as follows: Eoptolink > Innolight > T&S > TFC, with a particular focus on Eoptolink due to its strong performance in 800G/1.6T products and capacity [4]. - Eoptolink is expected to benefit significantly from the strong demand for 800G, with projections indicating that 75% of its sales will come from this segment by 2026 [15]. - Innolight is also anticipated to capture a substantial share of the 800G demand due to its superior supply capabilities and ongoing silicon photonics migration [17][18]. Group 3: Financial Forecasts and Market Trends - The 2025-2027 shipment forecasts for 400G, 800G, and 1.6T modules have been adjusted, reflecting an increase in 800G shipments and a decrease in 1.6T due to industry demand delays [17]. - The overall market valuation is expected to recover, with industry price-to-earnings ratios projected to rise from 8-10 times to 15-20 times by 2025, driven by cloud infrastructure upgrades and higher optical module integration rates [9]. - The anticipated strong demand for 800G and the potential delay in 1.6T migration may pose risks for CPO (Co-Packaged Optics) deployment, which could be pushed to 2027 [9]. Group 4: Competitive Landscape - The competitive landscape is shifting, with more second-tier suppliers entering the supply chain due to insufficient supply from first-tier suppliers [13]. - Companies like Suzhou Taicheng Light are facing challenges due to lower-than-expected InfiniBand penetration and weak GB200 rack numbers, which may impact their 1.6T demand [20][21]. - The ongoing silicon photonics migration is expected to provide cost advantages for companies like Innolight and Eoptolink, allowing them to maintain higher profit margins compared to competitors who rely on external design sources [18].

还有专注于半导体行业分析的SemiAnalysis的全部分析报告： 星球中每日还会更新Seeking Alpha、Substack、 stratechery的精选付费文章， 现在星球中领券后只需要 390元，即可每天都能看到上百篇外资顶尖投行科技行业的分析报告和每天的精选报告，无论是我们自 己做投资，还是对行业有更深入的研究，都是非常值得的。 想要看外资研报的同学，给大家推荐一个星球，在星球中每天都会上传几百篇外资顶尖投行的原文研 报：大摩、小摩、UBS、高盛、Jefferies、HSBC、花旗、BARCLAYS 等。 ...

这篇文章，我们结合国盛证券的一份研报，来看下华为昇腾产业链上的公司们。主要讲下面4个产业链 的方向：整机、电源、散热和连接。本文中所有的图片都是来自国盛证券。 昇腾整机硬件伙伴要求拥有自有品牌 产品，能在昇腾产品基础上二次开发或加工生产，并销售与服务 至最终用户，伙伴类型 分领先级、优选级、认证级。 昇腾整机硬件伙伴： 一、整机 根据国盛证券，2024年新增算力规模约为2万Pflops，2028年中国智算中心市场投资规模有望达到 2886 亿元。2023年中国智算中心市场投资规模达879亿，同比增长 90%以上。未来，AI大模型应用场景不断 丰富，商用进程加快，智算中心市场增长动力 逐渐由训练切换至推理，市场进入平稳增长期，预计 2028 年中国智算中心市场投资规 模有望达到2886亿元。截至2024年8月，中国智算中心项目超过300 个，已公布算力 规模超50万PFlops。从已投用、在建、规划的智算中心项目来看，全国各省智算中心 总 计300 余个，约三分之一智算中心项目规划算力大于500PFlops，主要为政府或基础电 信运营商投建 项目，2024年当年投运项目数量超过50个，60%以上为地方政府、国资 ...

前几天第一时间转载了路透社的报道， 外媒：H20阉割版预计在7月推出，性能或大幅缩水 。 根据这个 报道，英伟达计划将在未来2个月推出H20的阉割版，网友调侃这是"丐中丐"版。Jefferies也第一时间对 这款"丐中丐"做解读分析。 报告的核心观点为： H20 因内置 HBM3 内存易受限，总带宽 4.0TB/s 。美国或设 GPU 内存带宽上限 1.7-1.8TB/s ，若如 此英伟达 H20 可能弃用 HBM 内存改用 GDDR6 ，降级版 H20 性能仍可能强于使用 GDDR6 的游戏 GPU ，如 RTX5090D 报告内容 Jefferies在之前的研究中，就提到过H20 容易受到限制，因为它内置了 HBM3 内存，其总内存带宽为 4.0TB/s（高于 H800）。去年 12 月底，拜登政府对向中国出售独立的 HBM3 及以上产品实施了限制。 科技媒体Tom's hardware上周也披露，英伟达在中国已停止接受游戏GPU RTX5090D 的订单， RTX5090D 内置 32GB 的 GDDR7 内存，总带宽为 1.79TB/s。因此，当英伟达在 4 月 16 日宣布对 H20 的限制措施（需 ...