一年亏了47亿股价一天涨了1000亿智谱这笔账怎么算的？｜财报解析

胡镤心 2026-04-02 11:48

胡镤心 2026/04/02 11:48

邦小白快读

全文速览

智谱财报关键点：收入大增但亏损扩大，股价却暴涨，核心在于业务转型。

1. 收入与亏损：全年收入7.24亿元，同比增长132%，净亏损47.18亿元扩大近六成，股价单日涨31.94%，市值飙升近1000亿港元。

2. 业务亮点：开放平台及API业务收入1.90亿元增幅292.6%，年度经常性收入达17亿元增长60倍；智能体业务收入1.66亿元增幅248.8%，毛利率52.3%。

3. 问题挑战：企业级通用大模型业务增速放缓（70.5%），毛利率从69.6%下滑至47.0%；综合毛利率降至41.0%，亏损因研发开支高（31.8亿元增44.9%）和算力瓶颈。

4. 转型策略：从项目制向平台型转变，API业务量价齐升（定价提83%需求增），驱动因素是AI向智能体工程转型导致Token消耗激增。

智谱案例揭示品牌在AI领域的营销、定价和趋势洞察。

1. 品牌定价：API业务提价83%后需求不降反增，显示高端定价策略在供不应求市场中的溢价空间，毛利率从3.4%跃升至18.9%。

2. 产品研发：密集迭代模型（GLM-4.5到GLM-5 Turbo），提升全球竞争力（OpenRouter付费模型排名第一），支持品牌渠道建设。

3. 消费趋势：AI行业从Vibe Coding转向Agentic Engineering，用户行为导致Token消耗指数增长，引导品牌聚焦云端服务和智能体业务。

4. 品牌竞争：对标Anthropic路径，强调模型能力上限突破，风险在于云巨头竞争（阿里、腾讯、百度）可能削弱品牌优势。

智谱财报提供市场机会、风险提示和可学习商业模式。

1. 增长市场：API业务年度经常性收入17亿元增长60倍，显示云端调用服务的高增长机会；智能体业务增幅248.8%提供合作渠道。

2. 需求变化：消费需求从私有化部署转向API调用，Token消耗激增驱动量价齐升，启示事件应对如把握算力紧张窗口期。

3. 风险提示：亏损绝对值扩大（47.18亿元），现金流压力大；算力瓶颈可能限制扩展；通用大模型业务毛利率下滑显示买方议价增强。

4. 可学习点：转型平台型服务商，API为核心商业模式，扶持政策如调整算力采购缓解现金流；合作方式包括云服务整合。

智谱策略启示产品生产、数字化机会和商业潜力。

1. 生产需求：推动国产芯片适配软硬协同设计，解决算力瓶颈，显示工厂需优化硬件生产支持AI效率提升。

2. 商业机会：API业务爆发增长提供参与AI产业链机会；智能体业务高毛利（52.3%）启示开发自主智能系统产品。

3. 数字化启示：从私有化部署转向云端服务，引导工厂拥抱电商化API调用；资本开支大降83.8%展示服务采购模式降低投入风险。

行业趋势新技术、客户痛点和解决方案。

1. 行业趋势：AI向Agentic Engineering转型，Token消耗激增驱动API需求；模型迭代高频（GLM系列）提升服务能力上限。

2. 客户痛点：算力供给紧张（实际需求是支持算力的1-2倍），导致交付成本增加和毛利率下滑；市场供需错配带来短期溢价窗口。

3. 解决方案：智谱调整算力采购模式（服务采购替代资本开支），国产芯片优化适配；API业务对标Anthropic路径提供高端服务模式。

商业需求平台做法、招商运营和风险规避。

1. 平台需求：API业务增长显示平台需支持高扩展服务；企业转向云端调用需求平台招商吸引开发者（OpenRouter排名第一）。

2. 最新做法：转型平台型服务商，API为核心（收入占比26.3%）；运营管理调整算力采购（资本开支降83.8%）缓解现金流压力。

3. 风险规避：算力瓶颈和竞争加剧（如谷歌、亚马逊降价）需监控；市场风向规避如通用大模型业务毛利率下滑（69.6%到47.0%）表明平台优化交付多元化。

产业动向新问题、政策启示和商业模式。

1. 新动向：API业务年度经常性收入17亿元增长60倍，标志AI向平台型转型；智能体工程需求激增形成新增长点。

2. 新问题：亏损扩大（47.18亿元）与高增长矛盾，算力约束突出；私有化部署转向买方市场，厂商议价能力下降。

3. 政策启示：推动国产芯片适配软硬协同设计，建议支持算力国产化；研发投入占收入439%引发政策资助讨论。

4. 商业模式研究：从项目制到API平台型路径对标Anthropic，探讨高定价策略可持续性及与云巨头竞合关系。

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

Zhipu's financial report highlights a paradox: surging revenue amid widening losses, yet a stock price surge driven by business transformation.

1. Revenue and Losses: Full-year revenue reached ¥7.24 billion, up 132% year-over-year, while net loss expanded by nearly 60% to ¥4.718 billion. The stock price surged 31.94% in a single day, adding nearly HK$100 billion in market cap.

2. Business Highlights: Open platform and API revenue soared 292.6% to ¥190 million, with annual recurring revenue hitting ¥1.7 billion (a 60-fold increase). Agent business revenue grew 248.8% to ¥166 million, boasting a 52.3% gross margin.

3. Challenges: Enterprise-grade general model growth slowed (70.5%), with gross margin dropping from 69.6% to 47.0%. Overall gross margin fell to 41.0%, and losses were driven by high R&D spend (¥3.18 billion, up 44.9%) and compute bottlenecks.

4. Transformation Strategy: Shifting from project-based to platform model, API business saw volume and price rise (83% price hike with increased demand), fueled by the industry's shift from Vibe Coding to Agentic Engineering causing token consumption to explode.

The Zhipu case offers insights into branding, pricing, and trend spotting in the AI sector.

1. Brand Pricing: An 83% API price increase did not curb demand, demonstrating premium pricing power in a supply-constrained market; gross margin leapt from 3.4% to 18.9%.

2. Product R&D: Intensive model iteration (GLM-4.5 to GLM-5 Turbo) enhances global competitiveness (ranked #1 on OpenRouter's paid model list), supporting brand and channel development.

3. Consumption Trends: The AI industry's pivot from Vibe Coding to Agentic Engineering drives exponential token consumption growth, guiding brands to focus on cloud services and agent businesses.

4. Brand Competition: Emulating Anthropic's path by pushing model capability limits; risk lies in competition from cloud giants (Alibaba, Tencent, Baidu) potentially eroding brand advantage.

Zhipu's report reveals market opportunities, risks, and a replicable business model.

1. Growth Markets: API annual recurring revenue of ¥1.7 billion (60x growth) signals high-growth potential for cloud call services; the 248.8% agent business growth offers partnership channels.

2. Demand Shift: Consumption is moving from private deployment to API calls, with surging token usage driving volume and price increases, suggesting strategies like capitalizing on compute shortage windows.

3. Risk Warnings: Absolute losses expanded (¥4.718 billion), creating cash flow pressure; compute bottlenecks may limit scaling; declining gross margin in general models indicates stronger buyer bargaining power.

4. Learnings: Transitioning to a platform service provider with API as the core model; supportive policies include adjusting compute procurement to ease cash flow; partnership models involve cloud service integration.

Zhipu's strategy offers insights for product manufacturing, digital opportunities, and commercial potential.

1. Production Demand: Pushing for domestic chip adaptation and hardware-software co-design to solve compute bottlenecks indicates factories need to optimize hardware production for AI efficiency gains.

2. Commercial Opportunities: The explosive growth of the API business provides entry points into the AI supply chain; the agent business's high margin (52.3%) suggests potential for developing proprietary intelligent system products.

3. Digital Insights: The shift from private deployment to cloud services guides factories towards adopting e-commerce-like API calls; an 83.8% reduction in capital expenditure demonstrates how service procurement lowers investment risk.

Industry trends, client pain points, and potential solutions.

1. Industry Trends: AI's shift towards Agentic Engineering drives surging token consumption and API demand; frequent model iteration (GLM series) raises the ceiling for service capabilities.

2. Client Pain Points: Tight compute supply (actual demand is 1-2x supported capacity) increases delivery costs and erodes margins; market supply-demand mismatch creates short-term premium windows.

3. Solutions: Zhipu adjusted its compute procurement model (service procurement replacing capital expenditure) and optimized for domestic chips; its API business, following the Anthropic playbook, offers a high-end service model.

Platform requirements, merchant operations, and risk mitigation strategies.

1. Platform Needs: API growth indicates platforms must support highly scalable services; enterprises shifting to cloud calls require platforms to attract developers (e.g., #1 ranking on OpenRouter).

2. Latest Practices: Transitioning to a platform service model with API as the core (26.3% of revenue); operational management adjusted compute procurement (capex down 83.8%) to ease cash flow pressure.

3. Risk Mitigation: Monitor compute bottlenecks and intensifying competition (e.g., price cuts by Google, Amazon); avoid pitfalls like the gross margin decline in general models (69.6% to 47.0%), indicating a need for diversified delivery optimization.

Industry movements, emerging questions, policy implications, and business models.

1. New Movements: API annual recurring revenue of ¥1.7 billion (60x growth) signals AI's shift to a platform model; surging demand for Agentic Engineering creates a new growth frontier.

2. Emerging Questions: The contradiction between expanding losses (¥4.718 billion) and high growth, with compute constraints prominent; the shift from private deployment to a buyer's market weakens vendor pricing power.

3. Policy Implications: Promoting domestic chip adaptation and hardware-software co-design suggests support for compute localization; R&D spend at 439% of revenue sparks debate on policy funding.

4. Business Model Research: The path from project-based to API platform,对标Anthropic, raises questions about the sustainability of high-price strategies and competition/cooperation with cloud giants.

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.

【亿邦原创】3月31日晚，智谱交出了上市后的第一份全年成绩单。全年收入7.24亿元，同比增长132%，在国内大模型公司中稳居营收榜首；年内净亏损47.18亿元，同比扩大近六成。

这份“增收不增利”的财报发布次日，智谱股价大涨31.94%，单日市值飙升近1000亿港元，总站上4079亿港元。

一年亏出一座城，一天涨出一片海。投资者究竟在这份亏损扩大的财报中看到了什么？答案藏在智谱业务结构的微妙变化里。这家公司正在从一个卖私有化大模型解决方案的“项目制”厂商，加速转型为一家以API为核心的“平台型”智能服务商。

1、API业务，全村的希望

智谱将收入重新划分为三大业务线，这一分类本身就透露了战略意图。

表现最亮眼的是开放平台及API业务，全年收入1.90亿元，增幅同比292.6%，占总收入比重从2024年的15.4%提升至26.3%。更有看点的是，截至2026年3月，API业务的年度经常性收入（ARR）已达到约17亿元（2.5亿美元），过去12个月增长了60倍。毛利率则从2024年的3.4%跃升至18.9%，虽然绝对值不算高，但增幅够大。

企业级智能体业务同样可圈可点，收入从0.47亿元增至1.66亿元，增幅248.8%，占比22.9%。智能体被定义为“以通用大模型为核心控制单元，结合企业知识库与工具调用能力的自主智能系统”，从“卖模型”升级为“卖能完成任务的数字员工”，毛利率52.3%，产品化程度较高，定制化交付成本相对可控。

作为昔日主力的企业级通用大模型业务，营收3.66亿元，同比增长70.5%，仍是第一大收入来源（占比50.4%），增速显著低于整体（132%），更远低于API和智能体业务。毛利率从69.6%下滑至47.0%，财报解释为“交付成本随部署需求多元化而阶段性增加”。私有化大模型部署正在从“卖方市场”转向“买方市场”，厂商越来越多，客户议价能力增强，毛利率的进一步下滑几乎是必然的。

智谱财报的看点集中API业务的ARR爆发式增长，智能体业务的高增长与高毛利。问题同样明确，综合毛利率从56.3%降至41.0%，净亏损绝对值仍在扩大，现金流压力不容忽视。

2、API业务能“量价齐升”，是结构性变化还是短期窗口？

API业务ARR增长60倍，调用定价提升83%后需求不降反增，这种“量价齐升”在AI行业中极为罕见。我们的问题是：这是结构性红利，还是市场供需错配下的短期窗口？

从模型侧看，智谱产品迭代节奏高频，2025年至2026年初，密集发布了GLM-4.5、4.6、4.7、GLM-5到GLM-5 Turbo，在代码竞技场LMArena上获第一，SWE-bench Verified开源最高分，OpenRouter上GLM成为付费模型排名第一。

从需求端看，AI行业从“Vibe Coding”（氛围编程）转向“Agentic Engineering”（智能体工程），一个Agent任务可能需要数百万Token来执行多步操作，Token消耗量的指数级增长，带来API业务ARR暴增。

可以说，Agent转型的驱动是长期的，但定价权的“量价齐升”存在阶段性特征。当前API市场供需偏紧，部分原因是全球算力资源紧张，高性能模型供给有限。智谱电话会中承认，实际需求是当前算力支持的1-2倍。在这种“供不应求”的格局下提价83%，需求依然旺盛，更多反映的是短期稀缺性溢价。随着算力供给逐步释放，以及谷歌、亚马逊等巨头的同类模型降价竞争，API价格大概率会回落。

3、亏损扩大与算力瓶颈如何解？

如果说API高增长是智谱的A面，那么持续扩大的亏损和算力约束就是财报的B面。

2025年，智谱年内亏损47.18亿元，经调整净亏损31.82亿元。核心是研发开支31.8亿元，同比增长44.9%，占收入比重高达439%，还有向投资者发行的金融工具账面价值变动带来9.37亿元的非现金亏损，源于带有回购权利的股权融资。

高研发投入是所有大模型公司的共同特征，智谱的研发投入增速（44.9%）低于收入增速（131.9%），研发强度已经相对下降。

更值得关注的是算力供给，智谱通过调整算力采购模式（从资本开支转为服务采购）来缓解现金流压力，2025年资本开支7470万元，同比大降83.8%。同时智谱推动国产芯片的“Day 0”适配与软硬协同设计，希望在国产算力上跑出接近国际顶尖芯片的效率。

有意思的是，智谱将API业务对标Anthropic，即专注于模型能力上限的突破，以API为主要交付形态，走高端定价路线。

Anthropic的ARR从2024年底的10亿美元升至2025年底的90亿美元，主打“最强的模型通过API交付”，同时Anthropic的成功离不开亚马逊和谷歌的云生态输血与渠道支持。智谱在国内直面阿里、腾讯、百度等云计算巨头的竞争，API业务能否在国内复制Anthropic的奇迹，还需要观察其与云厂商的竞合关系演变。

这份财报没有给出盈利承诺，但是讲了了一个足够有吸引力的故事：模型越强，Token越好卖。智谱希望成为那个定义Token价值标准的人。

文章来源：亿邦动力