广告
加载中

唐杰杨植麟师徒对决:Kimi K3炸榜 智谱蒸发3000亿

任雪芸 2026-07-23 13:40
任雪芸 2026/07/23 13:40

邦小白快读

EN
全文速览

本文核心讲清华系大模型领域师徒唐杰与杨植麟的行业竞争事件,梳理了大模型行业当前的资本与竞争现状,核心干货信息如下:

1. 事件核心:杨植麟创办的月之暗面推出2.8万亿参数的Kimi K3,上线后登顶多个全球AI权威榜单,48小时就因用户请求过多挤爆服务器,带动合作公司股价暴涨,同时其老师唐杰创办的智谱,两个交易日股价跌超42%,市值蒸发超3000亿港元。

2. 资本与商业化动态:月之暗面半年完成三次融资超32亿美元,已启动港股IPO,目标估值300亿美元,Kimi K3发布后日销售额翻6倍,年度经常性收入达3亿美元。

3. 行业现状:大模型行业没有稳定技术护城河,迭代速度快,竞争是资金与算力的军备竞赛,行业座次可能一夜洗牌。

本文梳理了大模型行业的竞争、用户与资本现状,给大模型及相关领域品牌商提供了不少参考干货:

1. 用户端:开发者用户更看重模型的综合性能、真实落地能力与口碑,性能领先的模型更容易引爆用户热度,不过价格是影响企业客户选择的核心因素之一,价格过高会抵消部分性能优势。

2. 品牌与资本化路径:技术迭代打造领先性,再借助资本放大估值,是当前大模型品牌通用的成长路径,清晰的技术领先地位能够快速获得市场与资本的认可。

3. 竞争趋势:同定位产品的技术突破会直接冲击对手的市场信心与市值,大模型没有永久技术壁垒,品牌需要持续投入研发维持迭代,才能保住自身市场位置。

本文梳理了大模型行业的最新变化,给AI相关卖家明确了当前的机会、风险与可参考方向,干货如下:

1. 机会提示:当前企业开发者对高性能大模型存在旺盛需求,只要实现技术突破,就能快速获得用户与资本认可,当前国内头部大模型座次尚未稳定,仍有抢占市场的空间,技术领先的创业项目也能获得资本支持,登陆资本市场路径已经走通。

2. 风险提示:当前API调用赛道价格战极为惨烈,收入命脉容易被对手的技术迭代冲击,背靠顶级资源的大厂正在加速布局,创业公司的技术代差很容易被快速追平。

3. 经验参考:大模型行业比拼的是资金储备与迭代速度,需要提前储备充足弹药,才能留在竞争牌桌,上市后也要持续兑现商业化能力,否则容易遭遇估值大幅缩水。

本文披露的大模型行业发展动态,给相关工厂带来了商业机会与数字化转型的启示,干货如下:

1. 生产需求变化:大模型行业当前进入算力军备竞赛,头部模型用户增长快,急需扩容算力与基础设施,对GPU、服务器等相关硬件的需求快速增长,给上游生产工厂带来了大量订单机会。

2. 商业机会:大模型行业聚焦长程编程、复杂Agent任务、百万级上下文的企业需求,对应配套的软硬件产品需求也在增长,相关工厂可以调整生产方向对接新需求。

3. 数字化转型启示:大模型行业迭代速度快,不持续跟进技术就会被快速淘汰,这一规律也提示传统工厂加快推进数字化转型,持续优化技术与生产能力,才能跟上市场变化,保住自身竞争力。

本文梳理了大模型行业的最新趋势,给AI相关服务商明确了客户痛点与行业机会,干货如下:

1. 行业发展趋势:大模型行业技术迭代速度极快,不存在稳定的技术护城河,行业座次随时可能重新洗牌,竞争已经升级为算力与资金的军备竞赛,行业准入门槛在指数级抬高,最终能留在牌桌的玩家会越来越少。

2. 核心客户痛点:大模型创业公司的核心痛点是算力供给不足,用户爆发式增长后很容易出现服务器卡崩的问题,同时企业需要持续投入巨额资金支付算力成本,维持技术迭代,API赛道价格战激烈,企业有较强的降本需求。

3. 业务机会:算力基础设施服务、融资配套服务、大模型成本优化服务都有旺盛的市场需求,围绕高性能大模型的开发者配套服务也存在较大的增长空间。

本文披露的大模型行业动态,给各类平台梳理了行业需求、机会与风险,干货如下:

1. 行业需求:大模型创业公司对算力基础设施平台的需求急剧提升,用户爆发式增长后,企业普遍面临算力不足的问题,平台需要提前扩容提升服务承载能力,才能满足客户需求。

2. 招商机会:当前大模型行业优质项目的上市需求旺盛,月之暗面已经启动港股IPO筹备,智谱已经成功上市,后续会有更多头部大模型创业项目排队对接资本市场,给港股等融资平台带来了优质招商机会。

3. 风险规避提示:当前大模型行业普遍存在估值泡沫,很多头部公司估值严重脱离商业化兑现能力,常年高额亏损,平台在招商审核环节需要重点核查企业实际营收与盈利能力,规避泡沫风险。

4. 运营提示:大模型技术迭代快,竞争格局变化快,平台需要及时调整对大模型项目的评估与服务标准,跟进行业变化。

本文披露了国内大模型产业的最新竞争动态,暴露了行业存在的新问题,为产业研究提供了丰富的新鲜素材,干货如下:

1. 产业新动向:国内大模型行业出现清华系师徒对决的新竞争格局,头部座次快速更迭,Kimi K3上线后登顶多个权威榜单,月之暗面延续智谱的技术+资本路径,启动港股IPO,证明该模式已经可复制。

2. 行业新问题:当前行业估值泡沫问题突出,头部公司估值严重脱离自身商业化兑现能力,普遍保持高额亏损,行业进入算力军备竞赛,门槛指数级抬高,且不存在稳定的技术护城河,竞争格局随时可能重构。

3. 商业模式研究启示:当前主流大模型商业化以API调用为主,该赛道价格战激烈,企业极度依赖融资维持运营,技术迭代速度直接决定企业生死,这种重资产高消耗的商业模式的可持续性,仍然值得进一步研究。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

This article centers on the industry rivalry between Tang Jie and Yang Zhilin, two Tsinghua-affiliated mentors and mentees in China's large language model (LLM) sector, while outlining the current capital and competitive landscape of the industry. Key takeaways are as follows:

1. The core event: Moonshot AI, founded by Yang Zhilin, launched its 2.8-trillion-parameter model Kimi K3. The model topped multiple global AI benchmark rankings immediately after launch, and its servers crashed within 48 hours due to overwhelming user traffic. This event drove up share prices of Moonshot's partner companies, while Zhipu AI—founded by Tang, Yang's former mentor—saw its share price drop more than 42% over two trading days, erasing over HK$300 billion in market capitalization.

2. Capital and commercialization updates: Moonshot AI closed three funding rounds totaling over $3.2 billion in six months, and has initiated a Hong Kong IPO targeting a $30 billion valuation. Following the Kimi K3 launch, its daily sales surged sixfold, pushing its annual recurring revenue to $300 million.

3. Industry status quo: The LLM sector lacks stable technological moats, with extremely fast product iteration. Competition has become an arms race of capital and computing power, and industry rankings can be reshuffled overnight.

This article outlines the competition, user dynamics and capital landscape of the LLM industry, providing actionable insights for brands in the LLM and related sectors:

1. User side: Developer users prioritize a model's overall performance, real-world deployment capability and market reputation. Performance-leading models can easily generate viral user traction, but pricing remains a core factor influencing enterprise clients' choices—excessive pricing can offset part of a model's performance advantage.

2. Brand growth and capitalization path: Building technological leadership via rapid iteration, then scaling valuation through capital support, is the dominant growth path for LLM brands today. A clear leading position in technology can win rapid recognition from both the market and investors.

3. Competitive trends: Technological breakthroughs by similarly positioned competitors can directly erode a rival's market confidence and market valuation. There are no permanent technological barriers in LLMs; brands must sustain continuous R&D investment to maintain iteration speed and retain their market position.

This article summarizes the latest developments in the LLM industry, clarifying current opportunities, risks and actionable directions for AI-related sellers:

1. Opportunity outlook: Enterprise developers have strong, unmet demand for high-performance LLMs. A meaningful technological breakthrough can quickly win user and capital recognition. The ranking of China's top LLM players remains unsettled, so there is still room for new entrants to capture market share. Technologically leading startups can secure capital support, and a clear path to public markets has already been established.

2. Risk warnings: The API services segment is already experiencing an extremely fierce price war. A player's revenue stream can be severely disrupted by competitors' technological iterations. Large tech incumbents with top-tier resources are accelerating their布局, and the technological gap held by startups can be closed very quickly.

3. Key takeaways: Competition in the LLM sector ultimately comes down to capital reserves and iteration speed. Companies need to stockpile sufficient resources in advance to stay in the game. Even after going public, they must continuously deliver on commercialization goals, otherwise they will face sharp valuation markdowns.

The LLM industry developments outlined in this article bring new business opportunities and digital transformation insights for relevant manufacturing players:

1. Shifting production demand: The LLM industry has entered an arms race for computing power. Top models are seeing explosive user growth and urgently need to scale up computing power and infrastructure, driving rapidly rising demand for related hardware such as GPUs and servers. This brings a surge of new order opportunities for upstream manufacturing factories.

2. New business opportunities: The LLM industry is increasingly focused on enterprise needs such as long-context coding, complex agent tasks and million-token context windows, which is boosting demand for matching software and hardware products. Relevant factories can adjust their production focus to tap into these new demand streams.

3. Digital transformation insights: Just as LLMs players that fail to keep up with rapid technological iteration get eliminated quickly, this same lesson applies to traditional manufacturing: accelerating digital transformation and continuously optimizing technology and production capabilities is required to keep up with market shifts and retain competitiveness.

This article summarizes the latest trends in the LLM industry, clarifying core client pain points and industry opportunities for AI-related service providers:

1. Industry development trends: The LLM industry sees extremely fast technological iteration, with no stable technological moats. Industry rankings can be reshuffled at any time. Competition has escalated into an arms race of computing power and capital, pushing industry entry barriers exponentially higher, and the number of viable players that can remain in the market will keep shrinking.

2. Core client pain points: The biggest pain point for LLM startups is insufficient computing power supply. When user growth surges, server outages and performance degradation are common. Companies also need to continuously inject massive capital to cover computing costs and sustain technological iteration. Fierce price wars in the API segment have left enterprises with strong demand for cost reduction.

3. Business opportunities: There is robust market demand for computing infrastructure services, financing supporting services and LLM cost optimization services. Developer supporting services centered on high-performance LLMs also have large room for growth.

The LLM industry developments outlined in this article clarify industry demand, opportunities and risks for all types of platform players:

1. Industry demand: LLM startups' demand for computing infrastructure platforms has spiked sharply. After explosive user growth, most companies face significant computing shortages, so platforms need to scale capacity in advance to meet client demand.

2. Investment promotion opportunities: High-quality LLM startups have strong demand for public listings. Moonshot AI has already initiated preparations for a Hong Kong IPO, and Zhipu AI is already publicly traded. More top LLM startups will queue up to access capital markets in the future, bringing high-quality investment promotion opportunities for markets such as the Hong Kong Stock Exchange.

3. Risk mitigation guidance: There is widespread valuation froth in the current LLM industry, with many leading players' valuations far outpacing their actual commercialization capacity, and most operate with sustained heavy losses. Platforms need to prioritize verification of enterprises' actual revenue and profitability during the review process to avoid froth-related risks.

4. Operational guidance: Given the fast pace of technological iteration and competitive shifts in the LLM sector, platforms need to adjust their evaluation and service standards for LLM projects in a timely manner to keep up with industry changes.

This article discloses the latest competitive dynamics of China's domestic LLM industry, exposes emerging industry problems, and provides abundant fresh materials for industrial research:

1. New industrial dynamics: A new competitive landscape of "Tsinghua lineage mentor-mentee rivalry" has emerged in China's LLM sector, with rapid turnover among top-ranked players. After launch, Kimi K3 topped multiple authoritative AI benchmarks. Moonshot AI has followed the same "technology + capital" growth path that Zhipu AI pioneered and initiated a Hong Kong IPO, proving this model is replicable.

2. New emerging industry problems: Valuation froth is a prominent problem in the current sector. Top players' valuations are severely disconnected from their actual commercialization capacity, and most operate with persistent heavy losses. The industry has entered a computing power arms race that pushes entry barriers exponentially higher, with no stable technological moats, so the competitive landscape can be restructured at any time.

3. Implications for business model research: The mainstream commercialization model for LLMs today centers on API access, a segment marked by fierce price wars. Companies are extremely dependent on financing to sustain operations, and technological iteration speed directly determines survival. The sustainability of this asset-heavy, high-consumption business model warrants further in-depth research.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

文|任雪芸

五天之前,月之暗面创始人杨植麟带着2.8万亿参数的Kimi K3低调上线。没有发布会,甚至在公告第一句就主动承认“仍落后于全球最顶尖模型”。

但随后一周市场的反应远超预期。Kimi K3上线登顶ArenaAI榜单,所有人都迫切地想体验一把这个综合性能仅落后于Fable 5和GPT-5.6 Sol的开源模型,上线48小时用户请求直接把服务器挤爆,月之暗面紧急叫停新用户订阅。

资本也迅速用脚投票,与其有Token合作的中国软件国际7月20日单日暴涨35%,紧接着,科技热搜几乎被Kimi K3霸榜,甚至传出了“导致美股半导体崩盘”的说法,华尔街将其称为 “DeepSeek 2.0时刻”。

资本市场的另一端,先一步登陆港股、被称为“全球大模型第一股”的智谱,却走出了截然相反的走势。Kimi K3发布后的两个交易日,智谱股价连续暴跌,累计跌幅超42%,市值蒸发超过3000亿港元,相当于一个二线AI公司的估值。

很少有人注意到,站在这冰火两重天背后的两个男人,其实是一对师徒。智谱董事长唐杰,是杨植麟在清华计算机系读本科时的导师,从如何找问题、做实验到写论文,手把手带他入了NLP(自然语言处理)的门。甚至杨植麟本人,也是智谱GLM系列模型早期论文的共同作者之一。

老师的公司半年股价涨了24倍,一度站上万亿市值神坛;而学生的新模型一发布,老师公司的股价就遭遇了接近腰斩的暴跌。这样的局面,或许只是巧合。

不过,清华系的这对师徒,就这样以相当残酷的方式,站在了竞争舞台中央的聚光灯下。

一场挤爆服务器的技术秀,和一盘估值300亿的资本局

五天内,Kimi K3横扫三大权威榜单。

权威评测机构Artificial Analysis的智能指数榜单上,Kimi以57分登顶前三;而在Arena AI的Code Arena榜单上Kimi K3得分1679,位列第一。DeepSWE里67.5%的任务完成率同样跻身全球前三。

这三个榜单分别从综合能力、开发者真实口碑、复杂项目落地能力三个维度定义了K3。一位用户在深度体验后评价称:“除了慢和贵,的确好用。”

Kimi K3作为技术面的火热,只是月之暗面商业化叙事的起点。对资本市场来说,真正支撑高估值的核心,是一张“全球前沿模型厂商”的身份标签,它需要用K3告诉投资者,月之暗依旧在牌桌上。

月之暗面的资本引擎,这半年一直在加速运转。去年底刚完成5亿美元C轮,今年2月又拿下超7亿美元,5月再融约20亿美元。密集的弹药,瞄准的都是同一个靶心。

于是,在K3热度冲顶的第三天,消息如期而至:月之暗面已启动港股IPO筹备,目标估值300亿美元,IPO最快可能在未来六个月内进行。

与此同时,K3带来的商业转化立竿见影:Kimi K3发布后,月之暗面的日销售额至少翻了6倍;公司年度经常性收入在今年6月达到3亿美元,而就在两个月前,这个数字还停留在2亿美元。一条陡峭的收入曲线,远比任何技术论文都更能撬动估值。

同样的一幕,也曾在智谱身上完整地预演过一遍。

2026年1月,智谱抢滩港股,摘下“全球大模型第一股”的头衔。而在上市前的一年里,它依托GLM系列大模型的持续技术迭代,讲出了一个从底层算法研发到商业化应用全栈能力的故事,把市场期待一步步拉满。

而在资本层面,股东名单俨然一张中国科技投资的顶级玩家,阿里、腾讯、美团、红杉、高瓴,以及多地国资,悉数落座。

技术负责抬高天花板,资本把天花板变成更高的市值。这条路径,智谱已经走过一次。现在,似乎要轮到杨植麟带着月之暗面沿着同样的地图,再走一遍。

智谱坐上万亿过山车,月之暗面正排队入场

智谱今年1月登陆港股后,只用了五个月,股价涨了约20倍,一度跻身港股科技股第三大市值公司,排在它前面的只有腾讯和阿里。6月22日,市值摸到万亿港元关口,那是智谱最高光的时刻。

资本愿意买单的理由很清晰:GLM系列模型持续迭代,产品涨价后调用量不降反增,技术壁垒和商业化预期在线。

但高光消失得比想象中快。7月8日,智谱迎来首批限售股解禁,抛压随之而来。到7月17日,股价已从高峰时的2980港元跌至1107港元,跌幅超60%,总市值从1.3万亿缩水至5154亿港元,整整蒸发了约8100亿。

股价坠落时的斜率,远比当初上涨时更陡。

戳破智谱万亿市值泡沫的最终是智谱自身的业务,财报显示,2025年智谱营收7.24亿元,经调整净亏损31.82亿元。月亏近3亿的速度,让年初那笔43亿港元的IPO募资显得杯水车薪。

7月9日,智谱再次以每股1588港元配售新股,紧急补血约314亿港元,而配售公告中同时明确披露:这次配售的313亿预计2027年底前全部用完。

技术护城河遭遇挑战,市销率却严重脱离商业化兑现能力,这就是智谱当下的真实处境。

同样的局面,月之暗面一旦上市成功,或许大概率也要经历一遍。

而且,月之暗面的处境比智谱更复杂。K3上线后用户暴增,服务器被挤爆的同时,算力账单也在飞涨。它需要通过融资和资本市场拿到更多钱,去填算力采购和基础设施建设的无底洞。

而基础设施建设的规划和建设周期动辄数月,在这段漫长的窗口期里,月之暗面还能否保持领先,谁都不敢打包票。

月之暗面目前B端收入中,API调用占比高达70%,而API是当前大模型行业价格战最惨烈的阵地。无论是竞争对手的降价抢客,还是技术代差被迅速追平,都可能直接动摇这条收入命脉。Kimi需要靠持续迭代保持吸引力,才能从不断收紧的夹缝里,为自己留出更多用户。

同时,外部竞争压力也在加剧。Kimi K3的发布,让“国内大模型第一”的座次一夜易主,直接动摇了市场对智谱的信心,甚至冲击它的市值。

月之暗面也难言高枕无忧。在它身后,字节、阿里、腾讯各自在Agent方向加速推进。大厂的资源厚度,远不是一家创业公司能比的。一旦技术代差被追平,市场信心的转移同样可以在一夜之间发生。

大模型市场,没有护城河?

尽管唐杰和杨植麟坐到了牌桌的对面。但这对师徒都清楚:不同于互联网时代的产品,大模型这场牌局上,从来就没有护城河。

6月13日,Anthropic旗下王牌模型Claude Fable 5刚被全面限制访问,国内智谱AI的全新旗舰模型GLM-5.2便火速官宣全量开放。

在大模型盲测平台Code Arena上,GLM-5.2拿到“排名总榜第二、全球可用模型第一”。但这份“第一”的保质期极为短暂。仅仅一个月后,唐杰的学生出手。杨植麟的Kimi K3发布,同样的Code Arena榜单,1679分,把GLM-5.2从榜首位置上拽了下来。

而且Kimi K3与GLM-5.2在产品定位上重合,两者均聚焦于长程编程、复杂Agent任务及百万级上下文(1M Token)场景,争夺的是企业开发者。

不过,Kimi K3还远没到可以松一口气的时候。

单从价格角度来看,GLM-5.2每百万Token输入1.4美元、输出4.4美元;K3输入3美元、输出15美元,输出成本是前者的三倍还多。当能力差距没有拉出代差,价格就是企业客户最先看到的东西。

一来一回的攻守,向市场重复印证了一件事:AGI的护城河无法在某一个公司扎深,更便宜的GPU、更高质量的数据、更顶尖的研究员、更高效的烧钱速度,当这四个变量里的任何一个重新排列,都可能让牌桌上的座次一夜洗牌。

往更远处看,下一代模型的竞争门槛还在抬高。

更大的参数规模、更长的训练周期、更高的算力预算,每一项都在指数级推高门槛。训练成本在涨,推理需求在涨,算力军备竞赛没有终点。能撑到下一轮的人只会越来越少。

对于智谱、月之暗面、MiniMax们来说,现在的问题或许根本不是谁能领先一个身位,而是谁能先攒够弹药,确保自己还有资格坐在牌桌上。

注:文/任雪芸,文章来源:Tech星球(公众号ID:tech618),本文为作者独立观点,不代表亿邦动力立场。

文章来源:Tech星球

广告
微信
朋友圈

FAQ回顾

Kimi K3大模型的性能表现怎么样?

Kimi K3参数达2.8万亿,上线后登顶ArenaAI Code Arena榜单得分1679位列第一,Artificial Analysis智能指数榜、DeepSWE任务完成率均跻身全球前三,综合性能仅落后于Fable 5和GPT-5.6 Sol。

智谱市值大幅下跌的原因是什么?

Kimi K3与智谱GLM-5.2产品定位重合,均聚焦长程编程、复杂Agent任务等企业开发者场景,动摇了市场对智谱的信心,叠加智谱自身业务亏损、限售股解禁抛压,Kimi K3发布后两个交易日智谱累计跌幅超42%,市值蒸发超3000亿港元。

大模型行业是否存在稳定的技术护城河?

当前大模型行业不存在稳定的技术护城河,更便宜的GPU、更高质量的数据、更顶尖的研究员、更高效的烧钱速度,任一变量变化都可能引发行业座次洗牌,头部玩家需持续投入才能保留竞争资格。

月之暗面Kimi K3的商业化进展如何?

Kimi K3发布后月之暗面日销售额至少翻6倍,公司2025年6月年度经常性收入达3亿美元,目前已启动港股IPO筹备,目标估值300亿美元,IPO最快将在未来六个月内进行。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0