公众号记得加星标⭐️，第一时间看推送不会错过。 微软不仅是OpenAI模型的全球最大用户，也是OpenAI构建最新GPT模型时为其提供计算、网络和存储支持的最大合作伙伴。这给了微软两个理由 去打造更强大的Maia AI加速器，而微软也刚刚宣布他们已经完成了这项工作。 所有的大型云服务和超大规模云服务商，以及四大GenAI模型开发商中的三家——OpenAI、Anthropic和Meta Platforms——都在竭力打造各自的 定制AI XPU，以降低GenAI推理工作负载的单代币成本。第四家独立模型开发商xAI似乎准备采用特斯拉与Dojo合作开发的任何产品（如果其可扩 展性足够强，并能适应GenAI的训练和推理任务），但目前看来，xAI对Nvidia的GPU感到满意。 虽然仍有一些玩家对人工智能训练感兴趣，但实际上，英伟达几乎垄断了这个市场。随着全球企业和政府直接或间接通过云技术将人工智能推理投 入生产，预计人工智能推理所需的计算能力将比人工智能训练高出一个数量级，因此，一百多家人工智能计算引擎初创公司有机会开辟一片新天 地，并从中获利。 与所有超大规模数据中心运营商一样，微软希望在部署人工智能驱动的自动驾驶 ...

Core Insights - The report emphasizes the importance of "speed + power" as the core contradiction in the future development of the AI industry, highlighting significant market movements in both speed and power sectors over the past year [1][9] - For 2026, the focus should be on observing the commercial closure rhythms of CSPs and large model vendors to grasp the overall industry beta, while actively seeking value expansion and capital expenditure shifts in specific segments [1][10] - The report suggests that capital expenditure (Capex) and return on investment (ROI) are critical variables in understanding computing power demand, which is primarily driven by token counts and Capex [1][10] Investment Strategy - The computing power industry is viewed as the foundation of technology, with a long-term positive outlook. The report recommends actively seeking value expansion and capital expenditure shifts in specific segments, maintaining the focus on "speed + power" [3][12] - Key areas of investment include domestic computing power, semiconductor equipment, storage, and AI terminals [3][12] Capital Expenditure Analysis - Major cloud service providers (CSPs) have significantly increased their capital expenditures, with the top five CSPs' combined Capex reaching $308.1 billion in Q3 2025, a 75% year-on-year increase [24][27] - Google, Microsoft, Amazon, Meta, and Oracle are leading this trend, with Google and Microsoft showing particularly aggressive Capex growth to support AI infrastructure [27][28] - The report highlights that Google’s Capex for 2024 is projected to be $52.5 billion, a 63% increase year-on-year, while Microsoft’s Capex is expected to reach $75.6 billion, an 84% increase [27][28] AI Model and Chip Development - The report discusses the rapid iteration of Google's Gemini model family, which has introduced significant advancements in AI capabilities, including multi-modal understanding and enhanced reasoning abilities [36][41] - NVIDIA is identified as a key player in the computing power landscape, with its customer base including CSPs, large model vendors, and government clients, driving substantial revenue growth [24][30] - The report notes that the demand for AI chips is expected to grow, with companies like OpenAI forming strategic partnerships with major chip manufacturers to enhance their infrastructure [62][63] Domestic Computing Power Growth - The report anticipates a breakthrough year for domestic computing power in 2026, driven by the acceleration of domestic large models and positive capital expenditure outlook from cloud vendors [2][6] - The supply side is expected to transition from single-point breakthroughs to multi-point developments, indicating a robust growth trajectory for domestic computing power vendors [2][6] Semiconductor and Storage Opportunities - The semiconductor sector is highlighted as benefiting from an AI-driven storage supercycle, with equipment manufacturers poised to gain from original factory expansions [2][8] - The report emphasizes the importance of AI in driving growth in the storage industry, predicting rapid expansion in this sector [2][8]

当地时间1月26日，微软宣布推出第二代人工智能芯片Maia 200。 微软云与人工智能执行副总裁斯科特·格思里表示，Maia 200采用台积电3纳米工艺制造，每颗芯片包含超过 1400 亿个晶体管，为大规模 AI 工作负载量身打造，同时兼顾高性价比。他称Maia 200是微软迄今为止部 署的最高效推理系统，其每美元性能比微软目前部署的最新一代硬件提升了30%。他同时表示，Maia 200的 FP4性能是第三代 Amazon Trainium的3倍。 目前，Maia 200已部署在微软位于 爱荷华州 得梅因附近的美国中部数据中心区域，接下来将部署位于亚利 桑那州凤凰城附近的美国西部3数据中心区域，未来还将部署更多区域。 2026.01. 27 本文字数：1180，阅读时长大约2分钟 作者 | 第一财经 陆涵之 萨蒂亚·纳德拉则在财报中表示，将继续加大对人工智能领域的投资，包括资金和人才，以把握未来巨大的 机遇。他同时表示，微软旗下所有产品中，人工智能功能的月活跃用户已达9亿。 在应用方面，微软超级智能团队将利用 Maia 200 进行合成数据生成和强化学习，以改进下一代内部模型。 在去年11月，微软宣布成立超级 ...

Scott Guthrie, executive vice president of cloud and enterprise at Microsoft, speaks at the Microsoft Build developer conference in Seattle on May 7, 2018. The Build conference, marking its second consecutive year in Seattle, is expected to put emphasis on the company's cloud technologies and the artificial intelligence features within those services.Microsoft announced the next generation of its artificial intelligence chip, a potential alternative to leading processors from Nvidia and to offerings from cl ...

黄仁勋终于得到了他最想要的东西。 本周美国政府正式批准英伟达向中国以及其他"经批准的客户"出售高端的H200 GPU芯片，但需要向美国政府缴纳25%的销售提成。这一提成比 例同样适用于AMD、英特尔等其他美国芯片巨头。不过，英伟达最新的Blackwell和未来的Rubin系列GPU仍然禁止出口。 这标志着黄仁勋长达数月的游说取得成功。过去半年时间，他不断造访佛罗里达与华盛顿，随着特朗普总统一道出访和出席国宴，向白宫宴会厅 建设工程捐款，就是为了这一刻。就在上周，他再一次来到白宫会见总统，终于如愿以偿得到了解锁禁运令。 受这一利好消息推动，英伟达股价盘后应声上涨。受美国政府连续多道芯片加码禁运令限制，过去两年时间，英伟达一步步失去迅猛增长的中国 市场，丢掉了在AI GPU市场原先高达95%的份额。在英伟达最核心的数据中心业务，中国市场的营收占比也从原先的四分之一急剧下滑。 心急如焚的黄仁勋在两个月前公开抱怨， "我们已经失去了全球最大的市场之一，在中国市场完全出局，市场份额已经归零。 "即便是向美国政府 缴纳四分之一的提成，对英伟达的业绩营收也意义重大，因为中国AI GPU今年规模估计高达200亿-300亿美元 ...

Core Insights - The U.S. government has officially approved NVIDIA to sell high-end H200 GPU chips to China and other "approved customers," requiring a 25% sales commission to the U.S. government, which also applies to other U.S. chip giants like AMD and Intel [2][24] - This approval marks a significant victory for NVIDIA CEO Jensen Huang, who has lobbied for months to lift the export ban, which had severely impacted NVIDIA's market share in China [2][24] - NVIDIA's stock price rose following this news, as the company had lost a substantial portion of its market share in the AI GPU market, dropping from 95% to nearly zero in the past two years due to U.S. export restrictions [2][24] Group 1: NVIDIA's Market Position - NVIDIA is a leading company in the generative AI era, dominating the AI chip market with over 80% market share due to its performance advantages and the CUDA platform [3][25] - The company's data center business generated $130 billion in revenue in the most recent fiscal year, but it faces risks due to high customer concentration, with the top two customers accounting for 39% of revenue [3][25] - Huang has expressed concerns about losing the Chinese market, which is estimated to be worth $20 billion to $30 billion in AI GPUs this year [3][24] Group 2: Competition from Major Tech Giants - Major cloud service providers like Google, Amazon, and Microsoft are accelerating the development of their own chips, posing a significant threat to NVIDIA's market position [3][24] - Amazon's new AI chip, Trainium 3, is designed to be a low-cost alternative to NVIDIA's GPUs, claiming to reduce training costs by 50% compared to similar GPU systems [6][27] - Google has released its seventh-generation TPU, Ironwood, which boasts a performance increase of 10 times over its predecessor and is optimized for high-throughput, low-latency inference tasks [10][31] Group 3: Future Market Dynamics - The competition is expected to intensify in 2026, with a focus on a "performance vs. cost" showdown as Google, Amazon, and Microsoft release their latest self-developed chips [38] - Amazon aims to increase its self-developed chip share to 50% and grow its AI cloud market share from 31% to 35% [40] - Google's TPU market share has reportedly climbed to 8%, with plans to sell its previously internal-use TPUs to external customers, further diversifying the chip supply landscape [41][40]

Core Viewpoint - Microsoft is adjusting its internal AI chip development strategy to focus on less aggressive designs by 2028, aiming to overcome delays in development while maintaining competitiveness against Nvidia [2][4]. Group 1: Development Delays and Strategic Adjustments - Microsoft has faced challenges in developing its second and third-generation AI chips, leading to a strategic shift towards more pragmatic and iterative designs [2][4]. - The Maia 200 chip's release has been postponed from 2025 to 2026, while the new Maia 280 chip is expected to provide a 20% to 30% performance advantage per watt over Nvidia's 2027 chip [2][4][5]. - The company acknowledges that designing a new high-performance chip from scratch each year is not feasible, prompting a reduction in design complexity and an extension of development timelines [2][5]. Group 2: Chip Development Timeline - The Braga chip's design was completed six months late, raising concerns about the competitiveness of future chips against Nvidia [5]. - A new intermediate chip, Maia 280, is being considered for release in 2027, which will be based on the Braga design and consist of multiple Braga chips working together [5][6]. - The Maia 400 chip, initially known as Braga-R, is now expected to enter mass production in 2028, featuring advanced integration technologies for improved performance [6][7]. Group 3: Impact on Partners - The revised roadmap has negatively impacted Marvell, a chip design company involved in the Braga-R project, leading to a decline in its stock price due to project delays and economic factors [9]. - Not all of Microsoft's chip projects are facing issues; CPU projects, which are less complex than AI chips, are progressing well [9][10]. - Microsoft's Cobalt CPU chip, released in 2024, is already generating revenue and is being used internally and by Azure cloud customers [10].

Core Insights - Microsoft is adjusting its ambitious AI chip development strategy due to delays, shifting towards a more pragmatic and iterative design approach to remain competitive with Nvidia in the coming years [1][4] - The release of the Maia 200 chip has been postponed from 2025 to 2026, with plans to launch less aggressive designs by 2028 [1][4] - Microsoft aims to reduce its dependency on Nvidia's chip procurement, which costs the company billions annually [1] Group 1: Strategic Adjustments - The delays in the development of Microsoft's second and third-generation AI chips have prompted a strategic overhaul [4] - The Braga chip's design was completed six months later than planned, raising concerns about the competitiveness of future chips against Nvidia [4] - Microsoft is considering an intermediate chip, Maia 280, to be released in 2027, which will be based on the Braga design [4][5] Group 2: Future Chip Plans - The chip initially known as Braga-R will now be called Maia 400, expected to enter mass production in 2028 with advanced integration technology [5] - The release of the third-generation AI chip, Clea, has been delayed until after 2028, with uncertain prospects [5] Group 3: Impact on Partners - The revised roadmap negatively affects Marvell, which was involved in the Braga-R project, leading to a decline in its stock price [6] - Marvell had anticipated earlier revenue from Microsoft, but delays and economic factors have impacted its performance [6] Group 4: Other Projects - Not all of Microsoft's chip projects are facing issues; the CPU project, Cobalt, is progressing well and has already generated revenue [8] - The next generation of Cobalt, Kingsgate, has completed its design and will utilize chiplet architecture and faster memory [8]

Group 1 - Nvidia currently holds over 80% of the market value share in the AI server sector, while ASIC AI servers account for approximately 8%-11% [1][3][4] - Major cloud service providers like Meta and Microsoft are planning to deploy their own AI ASIC solutions, with Meta starting in 2026 and Microsoft in 2027, indicating potential growth for cloud ASICs [1][4][10] - The total shipment of AI ASICs is expected to surpass Nvidia's AI GPUs by mid-2026, as more cloud service providers adopt these solutions [4][10] Group 2 - Meta's MTIA AI server project is anticipated to be a significant milestone in 2026, with plans for large-scale deployment [2][13] - Meta aims to produce 1.5 million units of MTIA V1 and V1.5 by the end of 2026, with a production ratio of 1:2 between the two versions [21][22] - The MTIA V1.5 ASIC is expected to have a larger package size and more advanced specifications compared to V1, which may pose challenges during mass production [23][19] Group 3 - Companies like Quanta, Unimicron, and Bizlink are identified as potential beneficiaries of Meta's MTIA project due to their roles in manufacturing and supplying critical components [24][25][26] - Quanta is responsible for the design and assembly of MTIA V1 and V1.5, while Unimicron is expected to supply key substrates for Meta and AWS ASICs [24][25] - Bizlink, as a leading active cable supplier, is poised to benefit from the scaling and upgrading connections in Meta's server designs [26]

Investment Rating - The report maintains a "Buy" rating for several key companies in the semiconductor and AI sectors, including 中芯国际 (SMIC), 海光信息 (Haiguang), and others, indicating strong growth potential in the domestic AI and computing landscape [5][6]. Core Insights - The domestic AI landscape is witnessing significant advancements with the emergence of models like 豆包 (Doubao) and DeepSeek, which are leading the charge in multi-modal and lightweight AI model development, respectively [1][2]. - The report highlights a shift towards domestic computing power solutions, with chip manufacturers rapidly adapting to the evolving AI ecosystem, particularly through advancements in semiconductor processes and AI training capabilities [2][3]. - There is a notable increase in capital expenditure among cloud computing firms, driven by the rising demand for AI computing infrastructure, which is expected to lead to a "volume and price rise" scenario in the cloud computing market [3][4]. Summary by Sections Section 1: Breakthroughs in Domestic AI Models - 豆包 has emerged as a leading multi-modal model, enhancing capabilities in speech, image, and code processing, with a significant release of its visual understanding model in December 2024 [1][11]. - DeepSeek focuses on lightweight model upgrades, achieving a remarkable cost-performance ratio with its DeepSeek-V3 model, which has 671 billion total parameters and costs only 557.6 million USD, positioning it among the world's top models [1][12]. - The rapid iteration of domestic models, including updates from 通义千问 and others, reflects a competitive landscape that is accelerating the development of AI applications [1][34]. Section 2: Advancements in Domestic Computing Power - 中芯国际 is advancing its semiconductor processes, with N+1 and N+2 technologies being developed to support the growing demand for AI chips, achieving significant performance improvements [2][56]. - The report notes that the domestic chip industry is evolving, with companies like 昇腾 (Ascend) and others making strides in AI training and inference capabilities, thereby reducing reliance on international competitors [2][59]. - The cloud computing sector is experiencing a capital expenditure boom, with companies like 华勤 and 浪潮 rapidly deploying servers that are compatible with domestic computing power solutions [3][4]. Section 3: Infrastructure and Supply Chain Developments - The report emphasizes the need for enhanced computing infrastructure to meet the surging demand for AI applications, with significant investments being made in server and power supply innovations [3][4]. - Innovations in power supply and cooling systems, particularly the shift from traditional air cooling to liquid cooling, are becoming essential to support the increasing power density in data centers [4]. - The report identifies key players in the supply chain, including companies in power supply, cooling, and server manufacturing, that are poised to benefit from the growth of the AI and computing sectors [5].