每经记者｜李宇彤 每经编辑｜段炼 陈俊杰 大模型行业的生存竞赛，已悄然进入新阶段。 12月31日，《每日经济新闻》记者获悉，月之暗面（Moonshot AI）"90后"创始人杨植麟发布内部信，其透露："公司近期完成了5亿美元C轮融资且大幅超 募，当前现金持有量超过100亿元。" 图片来源：视觉中国（资料图） 2025年底，行业正驶入快车道，上市潮起之际，月之暗面却明确表示不急于叩响资本市场大门。杨植麟在信中称："相比于二级市场，我们判断还可以从 一级市场募集更大量资金。事实上，我们B/C轮融资金额就超过绝大部分IPO募资及上市公司的定向增发。所以我们短期不着急上市，也不以上市为目 的。" 就在2025年12月中旬，大模型独角兽智谱AI与MiniMax已相继通过港交所聆讯。两家企业近两日先后开始招股，预计分别于2026年1月8日、1月9日登陆香 港交易所。 对月之暗面旗下的Kimi来说，2025年是技术持续演进的一年。杨植麟在内部信开头即祝贺团队取得"SOTA成绩"（业界顶尖水准）。他写道，K2和K2 Thinking的发布标志着公司在AGI道路上走出重要一步，并列举了"中国首个万亿参数基座模型""第一个开源的 ...

财经摆渡人 精研出品 破浪前行，共探财富新局 大家好欢迎收看【古今财鉴】 AI科技圈最热闹的事，莫过于MiniMax和智谱AI同时冲刺港股IPO。 这两家公司，一个拼命往海外跑赚美元，一个扎根国内接政企订单，活生生把港股变成了中国AI大模 型的"路径对决场"。 今天咱们就来聊聊，这两家公司到底走了啥路子，谁的日子更好过，又给整个行业提了哪些醒。 MiniMax，出海狂奔的"烧钱机器" MiniMax这几年在AI圈算是个"话题人物"。 文本、语音、音乐、视频，啥火就做啥，尤其视频生成工具"海螺AI"，全球评测排第二，生成的视频超 过5.9亿个。 靠着这些花里胡哨的应用，它在海外圈了2.12亿用户，欧美市场占了大头，海外用户比例超80%。 用户多了，钱也确实进来了。 2023年营收346万美元，2024年就涨到3050万美元，前九个月直接干到5340万美元，同比增长170%还 多。 这个增长速度，放在整个AI圈都算能打的。 但问题是，它赚的钱根本不够花。 2023年亏2.69亿，2024年亏4.65亿，前九个月又亏了5.12亿，累计下来，这三年烧了快13.3亿美金，换成 人民币就是92亿左右。 它走的路特别简单， ...

Core Insights - Meta has acquired Manus, a Singapore-based AI startup with Chinese roots, for over $2 billion to enhance its AI capabilities [2][3] - The acquisition is part of Meta's strategy to catch up in the competitive AI landscape, following a significant investment of $14.3 billion in AI training data startup Scale AI earlier this year [3][5] - Manus will continue to operate independently under Meta, with a focus on integrating its AI technology into Meta's suite of products [4][6] Company Strategy - Meta's shift towards AI comes after a pivot from its initial focus on the metaverse, with plans to invest $600 billion in U.S. AI tech and infrastructure by 2028 [5] - The acquisition of Manus is expected to bolster Meta's AI product offerings across platforms like Facebook, Instagram, and WhatsApp [3][8] Operational Changes - Following the acquisition, Meta plans to wind down Manus's operations in China and relocate remaining employees, ensuring no access to first-party user data from Meta's existing products [7][9] - Manus's CEO expressed that joining Meta will allow the company to build on a stronger foundation without altering its operational structure [4]

OpenAI is paying its employees at levels that have little precedent in Silicon Valley history — $1.5 million in stock-based compensation per employee in 2025. That’s according to financial projections shared with investors and first reported by the Wall Street Journal. Such compensation dwarfs what earlier generations of tech giants paid their workforces before going public. Adjusted for inflation, OpenAI's equity payouts are multiples above the norm for large tech firms over the past two decades, and eve ...

This is CNBC's Morning Squawk newsletter. Subscribe here to receive future editions in your inbox.Happy Wednesday and New Year's Eve. I've decided that there are three main groups of holiday observers this year: Those going to parties or watching the Times Square ball drop at home; those doing late-night workout classes or races; and those going to AMC viewings of the "Stranger Things" finale.This newsletter normally walks you through five things to know before the market opens. Today, that list would inclu ...

Core Insights - The year 2025 marks a significant turning point for AI, transitioning from curiosity in 2024 to profound societal impact [1] - Predictions from industry leaders suggest that advancements in AI will continue to accelerate, with Sam Altman forecasting the emergence of systems capable of original insights by 2026 [1][3] - The debate around the Scaling Law continues, with some experts asserting its ongoing relevance and potential for further evolution [12][13] Group 1: Scaling Law and Computational Power - The Scaling Law has shown resilience, with computational power for training AI models growing at an exponential rate of four to five times annually over the past fifteen years [12][13] - Research indicates a clear power-law relationship between performance and computational power, suggesting that a tenfold increase in computational resources can yield approximately three times the performance gain [13][15] - The concept of "AI factories" is emerging, emphasizing the need for substantial computational resources and infrastructure to support AI advancements [27][31] Group 2: Breakthroughs in AI Capabilities - The SIMA 2 project at DeepMind demonstrates a leap from understanding to action, showcasing a general embodied intelligence capable of operating in complex 3D environments [35][39] - The ability of AI models to exhibit emergent capabilities, such as logical reasoning and complex instruction following, is linked to increased computational power [16][24] - By the end of 2025, AI's ability to complete tasks has significantly improved, with projections indicating that by 2028, AI may independently handle tasks that currently require weeks of human expertise [41] Group 3: Future Challenges and Considerations - The establishment of the Post-AGI team at DeepMind reflects the anticipation of challenges that will arise once AGI is achieved, particularly regarding the management of autonomous, self-evolving intelligent agents [43][46] - The ongoing discussion about the implications of AI's rapid advancement highlights the need for society to rethink human value in a world where intelligent systems may operate at near-zero costs [43][46] - The physical limitations of power consumption and cooling solutions are becoming critical considerations for the future of AI infrastructure [31][32]

Core Insights - The founder and CEO of Kimi, Yang Zhilin, announced that the company currently holds over 10 billion yuan in cash and is not in a hurry to go public [1][2] - Kimi recently completed a $500 million Series C funding round, led by IDG with a $150 million investment, and the post-money valuation reached $4.3 billion [1][2] - Kimi's paid user base saw a month-over-month growth rate of 170% from September to November 2025, potentially reaching around 1.7 million users by the end of the year [2][5] Financial Performance - Assuming an initial paid user count of 100,000 at the beginning of 2025, the estimated monthly revenue could reach approximately 85 million yuan by year-end, with API revenue potentially bringing total monthly revenue close to 100 million yuan [2][5] - The company has a significant cash reserve, which allows it to avoid rushing into an IPO, indicating a strong financial position to face competition in 2026 [2][5] Product Development - Kimi plans to launch the K2 and K2 Thinking models in September and November 2025, focusing on explainability in reasoning processes and complex logical reasoning [1][2] - The company has been actively releasing new agent functionalities since May 2025, contributing to a substantial increase in commercial performance [5][6] Strategic Goals - Kimi aims to surpass leading companies like Anthropic to become a world leader in AGI, with plans to enhance the K3 model's capabilities significantly [6][7] - The company is focusing on vertical integration of model training and agent products, aiming for a unique user experience and substantial revenue growth [7][8] Future Plans - A reward scheme for the K2 Thinking model and subsequent products is expected to be established before the 2026 Spring Festival, with average incentives projected to be 200% of 2025 levels [2][6] - The company intends to utilize the Series C funding to aggressively expand GPU resources and accelerate the training and development of the K3 model [6][7]

编译：Haina、Arlene 在 AI 创业与投资迅速升温的这两年里，一个微妙但普遍的困惑正在出现。 许多 AI 公司在真实使用中展现出惊人的价值密度：用户愿意为它们付出远高于传统软件的预算， 它们深入到工作流、决策乃至执行层；但当这些公司被放进熟悉的 SaaS 模型里审视时，却往往显 得"不够好看"：毛利率偏低、成本结构复杂、增长路径难以线性外推。 与此同时，另一类公司在融资与叙事层面几乎完美，却迟迟无法在真实环境中建立持续、高频的使 用。 问题也许不在公司本身，而在我们仍然使用的那套判断坐标系。 过去十年，VC 行业高度依赖一组被反复验证的标准：毛利率、单位经济模型、可预测的扩张曲 线。这套体系在 SaaS 时代行之有效，但当 AI 开始直接介入人类的认知、判断与执行环节时，这些 指标是否仍然足以解释价值创造，正在变得不那么确定。 Benchmark 新合伙人 Everett Randle 的这场访谈，正是在这样的背景下展开的。他从自己在 Founders Fund、Kleiner Perkins 与 Benchmark 的经历出发，反思了为何一些看似理性的判断在 AI 时 代失灵，也尝试提出一种尚未完 ...

AI淘金热进入深水区，卖铲子的人，开始下场挖矿了。 当英伟达被曝出以20亿-30亿美元洽谈收购AI21 Labs，这是提前锁定「下一代AI主导权」，而不是一笔普通的技术并购。 这笔交易尚未官宣，但行业认为「大概率会发生」。 如果最终成交价落在30亿美元，这将成为英伟达历史上金额最高的一笔AI并购。 这也解释了为什么很多业内人第一反应不是「AI21要起飞了」，而是：这是一场典型的高价「人才收割」。 在算力称王的时代，这家全球最强芯片公司，第一次把巨额支票直接投向模型与人才本身，信号异常明确：AI战争，已经从训练阶段，转向推理与系统 整合的终局之战。 AI21 Labs 在2022年生成式AI爆发前，AI21就是以色列AI赛道的「门面担当」，技术路线颇具学术气质： 强调可控文本生成 更让人吃惊的是，AI21 Labs全职员工规模约200人，折算下来，人均「身价」高达1000万至1500万美元，远高于大多数独角兽并购案例。 曾与OpenAI唱对台戏的独角兽 2021年，AI21 Labs的位置并不尴尬。 AI21由Amnon Shashua教授、Yoav Shoham教授与Ori Goshen于2017年联合创立， ...

当Meta收购Manus消息扩散开时，整个科技圈的反应既在意料之中，又充满了微妙的张力。 整场谈判由扎克伯格亲自操盘，且仅持续了10多天就谈拢了价格。因此它看起来像是小扎在面对2026年更加激烈的AI战争时，急切打出的一张牌。 Manus虽然被AI圈戏称是"套壳"公司，但是毋庸置疑，它仍能补足Meta的战略短板。 而对于小扎来说，这也不是他第一次用钱解决问题。 不过，小扎真的能通过Manus来改变现在Meta的颓势吗？我看未必。 01 简而言之，小扎很可能相中了Manus的赚钱能力。因为这家公司仅在8个月内，就达到了1.25亿美元的年化收入。 Manus的增长主要来自于两方面，一方面体现在企业本身对于工具的整合能力，另一方面体现在企业获客上。 其实Manus的技术并不难理解，它的核心是一个多Agent编排系统。名为多智能体协同架构 (Multi-Agent System, MAS)，由规划 (Planner)、执行 (Execution)、验证 (Verification) 和知识 (Knowledge) 四大核心Agent组成。 就像人类一样的大脑一样，不同区域掌管不同的任务。这4个Agent分别完成用户下 ...