Workflow
DeepSeek
icon
Search documents
观察 | 智谱AI的钱到底花哪儿了?
▲ 戳蓝 色字关注我们! "投资的本质,是对未来价值的提前下注。"——沃伦·巴菲特 上次那篇智谱招股书的分析发出去之后,评论区有个留言让我印象特别深。 他说:"你只看到亏损23亿,但你有没有想过,这些钱可能根本没烧掉,而是变成了算力?AI公司的亏损和传统公司完全不一样。" 这个问题问得好。 说实话,我做了这么多年投资,见过太多"讲故事"的公司,所以看到亏损数字第一反应就是警惕。 但这位朋友说的也有道理——我是不是只看到了问题,没看到硬币的另一面? 所以今天这期,我们换个角度,用投资人的眼光重新审视:智谱这23亿,到底是烧掉了,还是变成了别的东西? 这个问题的答案,直接关系到你怎么判断这家公司的未来。 所以有了这篇智谱招股书解读第二弹。 一、钱变成了什么?数据里的秘密 然后我昨晚又仔细研究了一遍招股书的研发支出明细,发现了一些第一次看时没注意到的细节。 2025年上半年,智谱研发费用15.9亿,其中11.45亿用于购买云服务和硬件设备——占比71.8%。 更值得关注的是趋势:2022年这个比例只有17.3%,三年时间飙升到七成以上。 | | | | 截至12月31日止年度 | | | | | 截至6月30日止六 ...
AI竞赛进入“能效“新阶段:前Facebook隐私主管警示千亿美元基建热潮暗藏电网危机
智通财经网· 2025-12-24 00:54
Core Insights - The next phase of the AI industry's development will focus on enhancing the efficiency of technology applications, as stated by Chris Kelly, former privacy chief at Facebook [1] - Major AI companies are competing to build infrastructure to support AI computing power demands, with a need to optimize high-energy-consuming infrastructure [1] - Companies that achieve breakthroughs in reducing data center costs are expected to emerge as winners in the AI sector [1] Industry Trends - According to S&P Global, a surge in global data center construction is anticipated in 2025, with infrastructure-related transaction volumes exceeding $61 billion for the year [1] - OpenAI has committed over $1.4 trillion to AI investments in the coming years, including significant partnerships with Nvidia, Oracle, and CoreWeave [1] Energy Concerns - The rapid data center construction has raised concerns about the ability of the already strained power grid to support these computing infrastructures [2] - A collaboration between Nvidia and OpenAI announced in September requires at least 10 gigawatts of power, equivalent to the annual electricity consumption of 8 million American households [2] - This power requirement is comparable to the total demand during New York City's peak summer electricity usage in 2024 [2] Cost and Competition - The AI industry is increasingly focused on cost concerns, especially following the announcement of a free, open-source large language model by DeepSeek, which has a development cost of less than $6 million [2] - Chris Kelly anticipates the emergence of several Chinese companies in the AI space, particularly after the recent approval for the sale of Nvidia's H200 chips to China [2] - Open-source models, especially those from China, are expected to provide foundational computing power and capabilities for generative and agentic AI [2]
AI诈骗入侵电商领域,假图骗取退款,“洗脑”驯化大模型
Nan Fang Du Shi Bao· 2025-12-23 23:15
| 厨师视章 | 文件一圈 | 구 | MAKA | 금융 | I 天工4 | 11.200 | AMERIA | 随机的量 | 120 | 8 12 | kiral | 展场地名 | 14 图小图 | Despoak | HAS | 120 | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 非法贯议安徽 | 000 | CIL | 28 | 015 | 2000 | 10% | 29 | COL | 1 | 03 | 1 | 695 | ਲ | 40% | 1004 | 496 | | 周户数据从国家副书馆。 | 2056 | NOW | 1000 | 5296 | 2004 | 2096 | 3 DO4 | 2016 | 2006 | 40% | 20% | 5005 | 1 6004 | 12774 | 20% | 406 | | 生成内容的成形已量 | ecry | 0000 | 000 | 133 | 005 | 100% | 1023 ...
X @Bloomberg
Bloomberg· 2025-12-23 19:42
RT Bloomberg Opinion (@opinion)Remember China’s DeepSeek moment? China had moments like that in biotech and defense in 2025, too, @shuli_ren explains.That’s why underestimating the country is a huge mistake 🎥 https://t.co/5lFbe4LWR9 ...
招股书里的MiniMax:当聪明人决定不再为巨头打工
华尔街见闻· 2025-12-23 14:03
在过去三年的"百模大战"中,一级市场流行着一种近乎迷信的观点:大模型是吞金兽,没有百亿美金的入场券,就没有资格谈AGI。OpenAI累计数百亿美 元的投入,似乎就是逃不开的案列。 如果说,去年是国内大模型的应用元年,那么2025年无疑是行业的分化之年。此刻你选择的叙事,决定了你看到的世界。 当幻方背后的DeepSeek年初异军突起,当字节跳动的豆包以巨量流量覆盖C端,当阿里与蚂蚁系快速占领广告牌,市场的共识似乎正在凝固:AI最终将是 一场属于巨头的游戏。在这个叙事里,AI创业公司要么成为巨头的附庸,要么在垂直赛道里做一个小而美的"工具"。 但年末,闪击港股的MiniMax,将"头部通吃"的叙事撕开了一道裂缝。这家成立不足4年、员工规模仅385人、核心团队平均年龄29岁的独角兽公司,以 一种近乎反直觉的姿态站在了IPO的门口。 在没有背靠任何单一互联网大厂流量池的情况下,它跑通了从模型到商业化的闭环,靠Talkie、海螺等产品在 C 端跑出真实规模,既有用户,也有收入; 在全行业深陷亏损时,它手握11亿美元现金储备,足以在零融资状态下支撑50个月 ;它用不到OpenAI 1%的资金投入,将全模态模型做到了全球第 ...
深度|狂飙、徘徊与转身:2025 年中国AI从业者浮世绘
Z Potentials· 2025-12-23 06:19
Core Viewpoint - The emergence of DeepSeek-R1 signifies a pivotal moment for Chinese innovation on the global stage, indicating a shift towards accelerated technological advancement and a redefined direction for the industry [1]. Group 1: Collective Narrative Amidst Chaos - The year witnessed parallel experiences of success and confusion within the same industry, with some companies achieving explosive growth while others reassess their strategies [5][6]. - DeepSeek maintained a strong position in application downloads, while competitors like Doubao also saw significant user engagement [6]. - Companies like Baichuan Intelligence and Zero One Matter are pivoting towards specific sectors, showcasing strategic differentiation among the so-called "AI Six Tigers" [7]. - The rapid iteration of AI models has left many entrepreneurs in a state of indecision, grappling with whether to embrace new trends or stick to established paths [7]. Group 2: Value Inquiry and Choices in 2025 - Practitioners are increasingly questioning the essence of technology and its practical applications, moving beyond theoretical discussions to real-world implications [13]. - The discourse on platforms like Zhihu reflects a shift in focus from immediate outcomes to long-term technological potential and sustainable practices [14][20]. - The concept of "density" in AI development is emerging, emphasizing quality over sheer size, as highlighted by industry experts [17]. Group 3: Observations from Practice - The interactions on Zhihu illustrate a dynamic exchange of ideas, where individuals document their experiences and uncertainties, contributing to a collective understanding of the industry's evolution [21][30]. - The narrative of personal journeys in the tech space reveals a transition from theoretical exploration to practical application, with many professionals sharing their challenges and decisions [20][31]. - The ongoing discussions emphasize the importance of individual contributions to the broader AI landscape, highlighting the need for continuous questioning and adaptation in an uncertain environment [33].
亚洲资产“大反攻”:亚太股市5年来首次年度同时跑赢欧美,机构预测股债汇2026年延续辉煌
智通财经网· 2025-12-23 00:10
智通财经APP获悉,今年以来,亚洲股市表现优于美国和欧洲,信贷市场强劲,货币走强,投资者预计 这股势头将延续至2026年。以美元计价,MSCI亚太指数(涵盖亚太地区股票,含股息)今年已上涨27%。 这也是自2020年以来,亚洲股市首次在一年内同时跑赢美国和欧洲的基准指数。 随着美国和欧洲经济增长放缓,亚洲对寻求更快增长的投资者的吸引力日益增强,这反映出亚洲的复苏 势头强劲。美元走软提升了亚洲资产的吸引力,而该地区与塑造全球经济的技术之间的紧密联系也增强 了其投资价值。 澳大利亚 Colonial First State 首席投资官Jonathan Armitage表示,对中国科技的重新关注增强了该基金管 理公司对新兴市场股票到 2026 年的展望。 当然,这波上涨行情也伴随着风险。亚洲经济复苏并不均衡,美元走强可能会损害外国投资者的回报。 此外,人们还担心人工智能相关科技股的上涨行情过于拥挤,一旦经济增长放缓或市场情绪逆转,股价 可能会受到冲击。 即便如此,一些投资者表示,这些风险并不会改变整体趋势。该地区的跨资产上涨被视为长期估值重估 的早期阶段——随着增长前景的改善,市场估值也会随之提高。 Vantage ...
X @Bloomberg
Bloomberg· 2025-12-22 12:04
Chase for DeepSeek Moments Fires Up Chinese IPO Market https://t.co/uhnIq62AcK ...
“烧钱” “抢时间” 详解智谱IPO招股书
Sou Hu Cai Jing· 2025-12-22 11:22
12月19日,港交所官网显示,北京智谱华章科技股份有限公司(以下简称"智谱")已通过聆讯并公布了 IPO(首次公开募股)招股书等资料,不仅成为业界所称的"大模型六小龙"中率先步入IPO流程的企 业,而且也抢在OpenAI、Anthropic等海外大模型明星企业前面,有望成为"全球大模型第一股"。随着 488页的招股书披露,公众终于得以一窥智谱大模型生意的"成色"。 中国城市专家智库委员会常务副秘书长、浙大城市学院副教授林先平认为,从财务数据来看,智谱当前 收入规模难以支撑长期高研发投入,需尽快提升收入或控制成本。依赖企业本地化部署,市场空间或受 限,且面临定制化成本高、毛利率压力等问题。智谱的商业化路径可持续性面临挑战,不过仍有调整空 间,高研发投入若转化为产品差异化,可能提升竞争力,这取决于未来一两年内能否实现收入规模化, 并优化研发效率。 需要指出的是,就在12月21日一早,"大模型六小龙"中另一家MiniMax也在港交所披露了聆讯资料集。 《中国经营报》记者注意到,不同于此前多数企业在递表时会先公示一个"申请版本"的招股书,智谱与 近日通过聆讯的壁仞科技、天数智芯等,均是直接披露聆讯后资料集,这是港交所 ...
详解智谱IPO招股书:抢跑上市换取规模增长时间
招股书显示,智谱在2022年至2024年的亏损额为1.44亿元、7.88亿元、29.58亿元,2025年上半年录得亏 损为23.58亿元。 智谱方面称,亏损主要是由于对研发作出重大投资。在报告期内,智谱研发投入逐年增加,2022年至 2024年分别是8440万元、5.29亿元、21.95亿元,占收入的百分比分别是147%、425%、703%。2025年上 半年研发支出为15.95亿元,占总收入的百分比高达835%,而2024年上半年这一百分比数字为1913%。 三年半的时间里累计研发投入超过44亿元。 从毛利率来看,智谱在2022年至2024年、2025年上半年分别是54.6%、64.6%、56.3%和50%。智谱的商 业模式是通过MaaS(模型即服务)平台提供大模型服务,收入主要来源于本地私有化部署,贡献占比 从2022年的95.5%降至2024年的84.5%,到2025年上半年实现收入1.62亿元,在总收入中的占比为 84.8%。来自云端部署的收入为2910万元,贡献占比为15.2%。值得注意的是,智谱MaaS的本地化部署 业务毛利率在2024年全年、2025年上半年分别为66.0%、59.1%;而其云端 ...