最近我们花了很多精力在H200/B200这些数据中心的服务器上，只能说坑很多，套路很深，但好事多 磨，最近的收货让我们觉得做件事是值得的。 这篇文章我们就来简单聊一下数据中心的投资现状，综合TD Cowen报告、The Information/BBG文章 及多位行业专家访谈，看下国外的大厂对IDC的态度，后面我们还有专门写一期 国内IDC 投资现 状。 微软数据中心投资放缓 相信大家也都看到这个新闻，微软正经历数据中心投资需求的显著放缓或调整。自去年起退出超 1GW的数据中心交易，并终止部分土地合同。放缓国际扩张步伐，并暂停/推迟了多个国内外项目， 包括美国（亚特兰大、威斯康星二期、圣安东尼奥、堪萨斯城、锡达拉皮兹）及欧洲、印度、英 国、澳大利亚等地，涉及规划租赁需求减少近1.98GW（原计划4年完成，年均约500MW）。 导致调整的原因是多方面的： 1. 资源消化：消化2024年已大量租赁的资源，避免过度建设。 2. 建设复杂性：超大规模数据中心设计和建设本身复杂，导致客观延迟。 3. OpenAI战略转移：OpenAI不再完全依赖微软，转向甲骨文、CoreWeave等第三方并大力推进自 建，导致微软为其规 ...

近日，Meta首席执行官马克·扎克伯格接受了媒体采访，全程信息量满满。访谈中， 扎克伯格谈到了 Meta如何看待下一步AI发展格局，并回应了外界认 为"DeepSeek吊打Meta"的质疑。 他表示，通过比较Llama 4 模型与 DeepSeek 的能力可知， 尽管 DeepSeek 可能在特定领域取得了显著进展，但Llama 4模型能够提供更高的效率和更广泛 的功能。 以下为采访内容（有删节）： 马克·扎克伯格：在我看来，世界会变得更加有趣、甚至有些奇特。根据我的经验，如果你觉得别人做的事情不好，但他们自己却认为很有价值，那么通 常是他们对，你错了。 主持人Patel： 我们似乎正在消除技术利用奖励机制来完全操纵我们的所有障碍。 马克·扎克伯格：我们正在努力构建能推进 Llama 研究的编码代理。我估计 在未来 12 到 18 个月内，我们将达到一个阶段，届时这些研发工作所需的大部 分代码都将由 AI 编写。我倾向于认为，至少在可预见的未来，这反而会增加对人类工作的需求，而非减少。如果你将提供服务的成本降至原来的十分之 一，那么现在去做这件事实际上可能是有意义的。 主持人Patel：你上次来的时候，发布了 ...

4月29日，阿里新一代的大模型Qwen3发布。以DeepSeek+Qwen的中国开源组合，正在以中国开源生态"双 子星"之势，取代过去Llama为主，Mistral为辅的开源生态。Qwen3发布后，Founder Park采访了不同行业 的AI创业公司代表，就开源闭源如何选择、模型的微调、模型能力瓶颈，以及大模型创业的坑等方面进行 了访问。心言集团开源布道师、高级算法工程师"左右"在接受采访的过程中，从心言集团的实践视角剖析 了中国开源模型的崛起路径，并分享了自己在开源生态赋能下的业务探索与技术思考。 以下为采访实录： 开源、闭源模型如何选择？ 左右：从我们目前的业务角度看，绝大部分(可能90%以上)需要使用我们自己微调(fine-tuned)的模型，因 此主要是本地化部署后供线上使用。同时，对于某些特定任务，比如数据构造、数据蒸馏，或者需要利用 特定大模型(如GPT、豆包、Qwen)的能力时，我们会直接调用它们的API。 本地部署的主要是基于Qwen的微调模型，根据不同的业务入口，我们会使用不同量级的模型。常用的是 7B、32B和72B这几个版本。此外，我们还有一些具身智能的业务，考虑到用户隐私和多模态处理 ...

证券代码：300229 证券简称：拓尔思 二、参会机构提问及回答 （一）贵司的核心竞争力主要体现在哪些方面？ 拓尔思信息技术股份有限公司 投资者关系活动记录表 编号：2025-001 投资者关系活 动类别 特定对象调研 □分析师会议 □媒体采访 □业绩说明会 □新闻发布会 □路演活动 □现场参观 □其他（电话会议） 参与单位名称 及人员姓名 国新证券：孙一夫、刘洪伟、辛奉来、黄宇飞； 国信证券：李楠； 东吴证券：臧佳楠； 中信银行：孙赛楠； 北京盛运德诚投资：张雪峰； 中信医疗基金：徐立晨。 时间 2025 年 4 月 29 日 14:00-16:30 地点 拓尔思信息技术股份有限公司一层会议室 上市公司接待 人员姓名 副总经理、董事会秘书 李党生； 证券事务代表 毛文仪。 投资者关系活 动主要内容介 绍 一、公司副总经理、董事会秘书李党生先生就公司业务发展情况、近期经 营情况及未来战略规划进行介绍。 2024 年公司营业总收入 7.77 亿元，较上年同期同比下降 0.59%；实现归 属于上市公司股东的净利润-9,414.56 万元，其主要原因是本年度公司对子公司 天行网安计提商誉减值 9,067 万元，对子 ...

中新网北京4月30日电 (记者 夏宾)清华大学人工智能研究院常务副院长、欧洲科学院外籍院士孙茂松近 日在北京接受中新网记者专访时称，中国科技公司在大模型领域掀起的开源浪潮向全球发出了中国"强 音"，其技术在获得国际认可的同时，悄然扭转了全球大模型文化。 来源：中国新闻网 最新消息显示，4月29日凌晨，新一代通义千问模型Qwen3(千问3)宣布开源，总共涉及8款不同尺寸的 千问3模型。据悉，阿里通义已开源200余个模型，全球下载量超3亿次，其衍生模型数超10万个，超越 美国Llama，成为全球第一开源模型。 以DeepSeek、Qwen为代表的中国开源模型实现先进模型的参数权重、推理逻辑和工具链条的全开源， 正在打开人工智能商用的新局面。 "尽管DeepSeek总体上是一个'从1到2'的创新，但在人工智能反馈强化学习方面是开源大模型中走得最 远的，将人类反馈变成了人工智能反馈。"谈到DeepSeek时，孙茂松说。 孙茂松特别强调了小模型的重要价值。从应用的角度，小模型可降低成本，拓展应用的普及度；从研究 的角度，小模型可有助于高校科研机构应对资源约束带来的研究挑战，这些都有很强的必要性。 在他看来，大模型做得越 ...

Core Viewpoint - Li Auto's decision to open-source its self-developed automotive operating system, "Xinghuan OS," aims to address industry-wide challenges and promote collaboration within the automotive sector [2][3]. Group 1: Initial Motivation and Industry Impact - The initial motivation for developing the operating system was to solve internal issues related to chip shortages and operational reliability, which were also common industry problems [3]. - By open-sourcing the operating system, the company aims to prevent resource wastage from redundant development efforts across the industry, fostering a collaborative environment for automotive innovation [3]. Group 2: Interest from Other Manufacturers - There is significant interest from two main types of manufacturers: suppliers (such as chip and solution providers) who see value in leveraging validated technologies to reduce R&D costs, and automotive manufacturers who are currently observing the open-source initiative [4]. Group 3: Responsibility and Risk Management - In the event of issues arising from the open-source system, responsibility is expected to lie primarily with the users and service providers who derive commercial benefits, while contributors focus on industry advancement [5]. - The company plans to establish a systematic process to review external contributions, ensuring project health and quality through automated and manual testing [7]. Group 4: Open-Source Development Timeline - The Xinghuan OS project was initiated in 2021, with the first deployment expected in 2024. The open-source announcement was made in March 2023, with the first batch of code released in April 2023 and further updates planned for June 2023 [6]. Group 5: Benefits and Strategic Goals - The open-source initiative is not aimed at direct financial returns but rather at fostering a collaborative ecosystem where all participants can contribute and innovate together, ultimately striving for the vision of an Intelligent OS [7]. - The operating system is positioned as a critical factor for future competitiveness in intelligent driving, addressing complexities in system performance, hardware iteration cycles, and AI computing needs [8].

Core Insights - Meta is focusing on integrating AI capabilities into its social media platforms like Facebook and Instagram, distinguishing itself from competitors like OpenAI and Google, which have standalone AI applications [1][3] - The launch of the Meta AI App at the LlamaCon developer conference marks a significant step in enhancing user experience through personalized AI interactions [1][6] Group 1: Meta AI Features - Meta AI emphasizes two main features: integration with social media accounts for personalized experiences and compatibility with Meta's Ray-Ban smart glasses for direct interaction [3][4] - The AI assistant can remember user preferences and context, utilizing information from Facebook and Instagram to provide relevant answers [4][6] - Meta AI supports both voice and text interactions, with improvements in the underlying model for more personalized and friendly responses [5][6] Group 2: Market Position and Strategy - Meta's CEO, Mark Zuckerberg, highlighted the importance of seamless and low-latency interactions in the company's AI development strategy [3] - The company aims to explore the commercialization of open-source models, with the Llama API service being a key initiative to enhance user engagement and potentially create new revenue streams [6] - Meta's open-source models have seen significant adoption, with downloads increasing from 1 billion to 1.2 billion in a short period, indicating strong market interest [6]

Core Insights - Li Auto showcased new models at the Shanghai International Auto Show, including the Li MEGA Home Family Special Edition and upgraded versions of Li MEGA Ultra and Li L6 [1][8] Group 1: Product Launches - The Li MEGA Home Family Special Edition is priced at 559,800 yuan, while the Li MEGA Ultra is priced at 529,800 yuan, both available for order immediately [1] - The Li L6 Smart Upgrade version is priced between 249,800 and 279,800 yuan and is also available for pre-order [1] Group 2: Features and Innovations - The Li MEGA Home features zero-gravity rotating seats in the second row, enhancing family usability with various configurations for easy access and comfort [3] - The vehicle includes a new front-row silent electric door with a damping design and 11 pieces of double-layer laminated glass, improving noise reduction by 3.8 decibels at 40 km/h on concrete roads [3] - Both the Li MEGA Home and Li MEGA Ultra are equipped with the upgraded NVIDIA Thor-U chip, providing 700 TOPS of computing power for enhanced safety and intelligent driving experiences [3] Group 3: Li L6 Smart Upgrade - The Li L6 Smart Upgrade features ATL all-weather laser radar and upgraded AD Pro driving assistance, now compatible with Horizon Journey 6M chip for advanced driving capabilities [5] - The Li L6 Pro Smart Upgrade can stop from 120 km/h in front of an accident vehicle and includes an automatic emergency steering (AES) function for night driving scenarios [5] - New aesthetic options for the Li L6 include a special sky blue pearl paint and dual-tone interior choices, along with redesigned 20-inch and 21-inch wheels [5] Group 4: Software Development - Li Auto's self-developed operating system, "Li Star Ring OS," has begun to open source some modules, marking a step towards ecosystem collaboration [7]

鱼羊 一水 发自 凹非寺 量子位 | 公众号 QbitAI 开源大模型新王者，正在受到空前关注。 Qwen3预告一出，直接开启不眠夜模式。 △ 来自编辑部本部 等到深夜正式上线并宣布登顶全球最强开源模型，更是瞬间引爆全网热议。 | | | Hope you enjoy our new models! | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 22B | Qwen3-32B Dense | OpenAl-o1 2024-12-17 | Deepseek-R1 | Grok 3 Beta BB Think | QwQ-32B | Qwen3-4B Dense | Qwen2.5-72B-Instruct | Gemma3-27BIT | | | 93.8 | 92.1 | 93.2 | | 89.5 | 76.6 | 81.2 | 86.8 | | | 81.4 | 74.3 | 79.8 | 83.9 | 79.5 | 73.8 | 18.9 | 32.6 | | | 72.9 | 79.2 | 70.0 | ...

Group 1 - Shanghai has entered a new phase with an economic scale exceeding 5 trillion yuan, becoming a leading hub for artificial intelligence innovation, application, industry aggregation, and talent [1] - The city is focusing on enhancing its technological innovation capabilities and high-end industry leadership to build a globally influential technology innovation center [1] - The AI talent pool in Shanghai accounts for one-third of the national total, with 250,000 AI professionals concentrated in the city [2] Group 2 - Shanghai has established a development pattern with significant AI innovation communities, such as "Mosu Space" in Xuhui and "Moli Community" in Pudong, attracting numerous innovative enterprises and young talents [2] - The "Mosu Space" has gathered nearly 400 AI companies and facilitated the implementation of 43 registered large models, accounting for approximately 61% of the city's total [8] - The city aims to build a world-class AI industry ecosystem by 2025, covering computing power, data, models, and applications [10] Group 3 - The local government is committed to creating a supportive environment for AI development through policy guidance, comprehensive industry chain layout, and strong data and computing power support [5][6] - Shanghai's AI industry is projected to exceed 450 billion yuan in 2024, with a year-on-year growth of over 7.8%, having already met its "14th Five-Year Plan" goals ahead of schedule [7] - The city is fostering an open-source ecosystem to enhance community cohesion and trust in AI technology, which is crucial for its development [7][11]