广告
加载中

大语言模型“收入榜”出炉 阿里云以46.6%份额居首

龚作仁 2026-10-29 15:11
龚作仁 2026/10/29 15:11

邦小白快读

EN
全文速览

市场格局与规模

1.2026年上半年中国大语言模型MaaS市场按营收口径统计,阿里云份额46.6%居首,智谱10.6%、火山引擎10.2%位列第二第三。

2.中国MaaS市场上半年规模106.3亿元,同比增长889.4%,占中国公有云PaaS市场比重由2025年下半年5.2%升至21.0%。

需求变化与实操启示

1.大语言模型占中国MaaS市场营收47.0%,是最大模态;企业知识问答、代码生成和智能体编排需求大多以文本为中心,大模型从辅助办公走向直接承担生产任务。

2.阿里云领先靠模型、芯片、云三层投入:真武AI芯片大规模部署,自研AI超节点支撑超2万亿参数高效推理,真武V900性能为M890的3倍、单集群可扩至50万卡;Qwen迭代至2.4万亿参数,并计划训练5万亿至10万亿参数新模型。

3.IDC判断MaaS是云服务商未来几年最重要增长来源,竞争最难复制的是模型能力、推理效率,以及算力规模、单位成本和供应链确定性。

用户与需求趋势

1.上半年中国MaaS市场106.3亿元、同比增889.4%,占公有云PaaS比重升至21.0%;大语言模型占MaaS营收47.0%,是最大模态,说明企业文本类AI需求正在放量。

2.企业需求集中在知识问答、代码生成和智能体编排,且大多以文本为中心;大模型从辅助办公走向直接承担生产任务,品牌应关注生产场景中的用户行为和任务型需求。

产品研发与竞争

1.阿里云以46.6%份额居首,背后是模型、芯片、云三层投入:真武AI芯片大规模部署,自研AI超节点支撑超2万亿参数高效推理,真武V900性能为M890的3倍、单集群可扩至50万卡。

2.Qwen迭代至2.4万亿参数Qwen3.8-Max,计划训练5万亿至10万亿参数新模型;算力上计划到2032年全球数据中心规模扩至20GW以上。

3.IDC认为竞争最难复制的是模型能力、推理效率以及算力规模、单位成本和供应链确定性,品牌定价和竞争需围绕这些能力与成本展开。

增长市场与机会

1.2026年上半年中国MaaS市场规模106.3亿元,同比增长889.4%,占公有云PaaS市场比重从2025年下半年5.2%升至21.0%,是高速增长市场。

2.大语言模型占MaaS营收47.0%,为最大模态;企业知识问答、代码生成、智能体编排需求以文本为中心,大模型从辅助办公走向直接承担生产任务,卖家可关注文本类AI应用和企业生产任务场景。

模式与风险提示

1.阿里云以46.6%份额居首,智谱10.6%、火山引擎10.2%位列其后;阿里云靠模型、芯片、云三层投入,真武AI芯片大规模部署,自研AI超节点支撑超2万亿参数推理,真武V900性能为M890的3倍、单集群可扩至50万卡。

2.Qwen迭代至2.4万亿参数,计划训练5万亿至10万亿参数新模型,并计划到2032年全球数据中心扩至20GW以上,显示云厂商正以算力和模型迭代争夺MaaS增长。

3.IDC认为MaaS是云服务商未来几年最重要增长来源,竞争关键在模型能力、推理效率、算力规模、单位成本和供应链确定性,进入者需评估这些难复制能力带来的机会与压力。

生产与基础设施需求

1.阿里云在大语言模型MaaS市场份额46.6%居首,背后是模型、芯片、云三层同时投入;平头哥AI芯片真武已在阿里云大规模部署,自研AI超节点支撑超过2万亿参数大模型高效推理,反映AI生产对专用芯片和集群基础设施的需求。

2.新一代AI芯片真武V900性能达到真武M890的3倍,单一集群可扩展至50万卡,可支撑前沿模型训练和推理;Qwen模型迭代至2.4万亿参数,还在计划训练5万亿至10万亿参数新模型。

商业机会与数字化启示

1.中国MaaS市场上半年规模106.3亿元,同比增长889.4%,占公有云PaaS市场21.0%;大语言模型占MaaS营收47.0%,企业知识问答、代码生成、智能体编排需求以文本为中心,大模型从辅助办公走向直接承担生产任务。

2.阿里巴巴计划到2032年将阿里云运营的全球数据中心规模扩大至20GW以上,AI云与算力服务收入同比增长45%至484.4亿元,显示算力基础设施扩张带来商业机会。

3.工厂可关注AI芯片、超节点、数据中心等生产与设计需求,并思考模型承担生产任务后的数字化和电商应用启示。

行业趋势与客户需求

1.2026年上半年中国MaaS市场规模106.3亿元,同比增长889.4%,占公有云PaaS市场比重升至21.0%;IDC认为MaaS正成为云服务商未来几年最重要增长来源。

2.大语言模型占MaaS营收47.0%,是最大模态;企业知识问答、代码生成和智能体编排需求大多以文本为中心,大模型从辅助办公走向直接承担生产任务,服务商需围绕这些任务型场景理解客户痛点。

新技术与解决方案

1.阿里云以46.6%份额居首,智谱10.6%、火山引擎10.2%位列第二第三;阿里云领先靠模型、芯片、云三层投入。

2.平头哥AI芯片真武已在阿里云大规模部署,自研AI超节点支撑超过2万亿参数大模型高效推理;真武V900性能为M890的3倍,单一集群可扩展至50万卡。

3.Qwen迭代至2.4万亿参数Qwen3.8-Max,并计划训练5万亿至10万亿参数新模型;阿里巴巴计划到2032年全球数据中心扩至20GW以上。

4.IDC指出竞争最难复制的是模型能力、推理效率以及算力规模、单位成本和供应链确定性,服务商可据此设计效率、算力成本与供应链保障方案。

平台增长与运营做法

1.2026年上半年中国MaaS市场规模106.3亿元,同比增长889.4%,占公有云PaaS市场比重从5.2%升至21.0%;IDC认为MaaS正成为云服务商未来几年最重要增长来源,平台需把握这一增长来源。

2.阿里云以46.6%份额居首,智谱10.6%、火山引擎10.2%位列其后;阿里云通过模型、芯片、云三层投入形成平台能力:真武AI芯片大规模部署,自研AI超节点支撑超2万亿参数高效推理,真武V900性能为M890的3倍、单集群可扩至50万卡。

需求与风向规避

1.大语言模型占MaaS营收47.0%,是最大模态;企业知识问答、代码生成、智能体编排需求以文本为中心,大模型从辅助办公走向直接承担生产任务,平台招商和运营应关注企业任务型AI需求。

2.Qwen迭代至2.4万亿参数,计划训练5万亿至10万亿参数新模型;阿里巴巴计划到2032年全球数据中心扩至20GW以上,AI云与算力服务收入同比增长45%至484.4亿元。

3.竞争关键在模型能力、推理效率、算力规模、单位成本和供应链确定性,平台需围绕这些难复制能力做运营管理和风险规避。

产业新动向与市场结构

1.2026年上半年中国MaaS市场规模106.3亿元,同比增长889.4%,占中国公有云PaaS市场比重由2025年下半年5.2%升至21.0%,显示MaaS成为云服务重要增量。

2.按营收口径统计,大语言模型占中国MaaS市场营收47.0%,是规模最大模态;阿里云份额46.6%居首,智谱10.6%、火山引擎10.2%位列第二第三,市场集中度与竞争格局值得研究。

商业模式与新问题

1.IDC认为MaaS正成为云服务商未来几年最重要增长来源;阿里云领先背后是模型、芯片、云三层同时投入,真武AI芯片大规模部署,自研AI超节点支撑超2万亿参数高效推理,真武V900性能为M890的3倍、单集群可扩至50万卡。

2.Qwen迭代至2.4万亿参数Qwen3.8-Max,并计划训练5万亿至10万亿参数新模型;阿里巴巴计划到2032年全球数据中心扩至20GW以上,AI云与算力服务收入同比增长45%至484.4亿元,体现垂直整合与算力基础设施投入。

3.企业知识问答、代码生成、智能体编排需求以文本为中心,大模型从辅助办公走向直接承担生产任务;竞争最难复制的是模型能力、推理效率、算力规模、单位成本和供应链确定性,构成产业研究的新问题。

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

Market landscape and scale

1. In 1H 2026, China's LLM MaaS market by revenue was led by Alibaba Cloud with 46.6%, followed by Zhipu at 10.6% and Volcano Engine at 10.2%.

2. China's MaaS market reached RMB 10.63 billion in 1H, up 889.4% year on year, and its share of China's public cloud PaaS market rose from 5.2% in 2H 2025 to 21.0%.

Demand shifts and practical implications

1. LLMs accounted for 47.0% of China's MaaS revenue, making them the largest modality. Enterprise demand for knowledge Q&A, code generation and agent orchestration is mostly text-centric, and large models are moving from assisting office work to directly taking on production tasks.

2. Alibaba Cloud's lead rests on three layers of investment: models, chips and cloud. Zhenwu AI chips have been deployed at scale, and its proprietary AI supernode supports efficient inference for models with more than 2 trillion parameters. Zhenwu V900 delivers 3x the performance of M890 and a single cluster can scale to 500,000 accelerators. Qwen has iterated to 2.4 trillion parameters, and Alibaba Cloud plans to train new models with 5 to 10 trillion parameters.

3. IDC sees MaaS as cloud providers' most important growth source over the next few years. The hardest capabilities to replicate are model capability, inference efficiency, plus compute scale, unit cost and supply-chain certainty.

User and demand trends

1. In 1H 2026, China's MaaS market reached RMB 10.63 billion, up 889.4% year on year, and its share of the public cloud PaaS market rose to 21.0%. LLMs accounted for 47.0% of MaaS revenue, making them the largest modality and indicating that enterprise demand for text-based AI is scaling.

2. Enterprise demand is concentrated in knowledge Q&A, code generation and agent orchestration, and is mostly text-centric. Large models are moving from assisting office work to directly taking on production tasks. Brands should watch user behavior and task-based needs in production scenarios.

Product R&D and competition

1. Alibaba Cloud leads with a 46.6% share, underpinned by three-layer investment in models, chips and cloud. Zhenwu AI chips have been deployed at scale, and its proprietary AI supernode supports efficient inference for models with more than 2 trillion parameters. Zhenwu V900 delivers 3x the performance of M890 and a single cluster can scale to 500,000 accelerators.

2. Qwen has iterated to the 2.4 trillion-parameter Qwen3.8-Max, and Alibaba Cloud plans to train new models with 5 to 10 trillion parameters. On compute, it plans to expand global data center capacity to more than 20 GW by 2032.

3. IDC believes the hardest capabilities to replicate are model capability, inference efficiency, compute scale, unit cost and supply-chain certainty. Brand pricing and competition should center on these capabilities and costs.

Growth market and opportunities

1. In 1H 2026, China's MaaS market reached RMB 10.63 billion, up 889.4% year on year, and its share of the public cloud PaaS market rose from 5.2% in 2H 2025 to 21.0%, making it a high-growth market.

2. LLMs accounted for 47.0% of MaaS revenue, making them the largest modality. Enterprise demand for knowledge Q&A, code generation and agent orchestration is text-centric, and large models are moving from assisting office work to directly taking on production tasks. Sellers can focus on text-based AI applications and enterprise production-task scenarios.

Models and risk notes

1. Alibaba Cloud leads with a 46.6% share, followed by Zhipu at 10.6% and Volcano Engine at 10.2%. Alibaba Cloud relies on three-layer investment in models, chips and cloud. Zhenwu AI chips have been deployed at scale, and its proprietary AI supernode supports inference for models with more than 2 trillion parameters. Zhenwu V900 delivers 3x the performance of M890 and a single cluster can scale to 500,000 accelerators.

2. Qwen has iterated to 2.4 trillion parameters, and Alibaba Cloud plans to train new models with 5 to 10 trillion parameters. It also plans to expand global data centers to more than 20 GW by 2032, showing that cloud providers are competing for MaaS growth through compute and model iteration.

3. IDC sees MaaS as cloud providers' most important growth source over the next few years. Competition hinges on model capability, inference efficiency, compute scale, unit cost and supply-chain certainty. Entrants must assess the opportunities and pressure created by these hard-to-replicate capabilities.

Production and infrastructure demand

1. Alibaba Cloud leads the LLM MaaS market with a 46.6% share, backed by simultaneous investment across models, chips and cloud. T-Head AI chip Zhenwu has been deployed at scale on Alibaba Cloud, and its proprietary AI supernode supports efficient inference for large models with more than 2 trillion parameters, reflecting AI production demand for specialized chips and cluster infrastructure.

2. The next-generation AI chip Zhenwu V900 delivers 3x the performance of Zhenwu M890, and a single cluster can scale to 500,000 accelerators, supporting frontier model training and inference. Qwen has iterated to 2.4 trillion parameters and is planning to train new models with 5 to 10 trillion parameters.

Business opportunities and digitalization implications

1. China's MaaS market reached RMB 10.63 billion in 1H, up 889.4% year on year, accounting for 21.0% of the public cloud PaaS market. LLMs accounted for 47.0% of MaaS revenue. Enterprise demand for knowledge Q&A, code generation and agent orchestration is text-centric, and large models are moving from assisting office work to directly taking on production tasks.

2. Alibaba plans to expand global data center capacity operated by Alibaba Cloud to more than 20 GW by 2032. AI cloud and computing power services revenue grew 45% year on year to RMB 48.44 billion, indicating business opportunities from compute infrastructure expansion.

3. Factories can watch production and design demand in AI chips, supernodes and data centers, and consider the digitalization and e-commerce implications as models take on production tasks.

Industry trends and customer needs

1. In 1H 2026, China's MaaS market reached RMB 10.63 billion, up 889.4% year on year, and its share of the public cloud PaaS market rose to 21.0%. IDC sees MaaS becoming cloud providers' most important growth source over the next few years.

2. LLMs accounted for 47.0% of MaaS revenue, making them the largest modality. Enterprise demand for knowledge Q&A, code generation and agent orchestration is mostly text-centric. Large models are moving from assisting office work to directly taking on production tasks, and service providers need to understand customer pain points around these task-based scenarios.

New technologies and solutions

1. Alibaba Cloud leads with a 46.6% share, followed by Zhipu at 10.6% and Volcano Engine at 10.2%. Alibaba Cloud's lead rests on three-layer investment in models, chips and cloud.

2. T-Head AI chip Zhenwu has been deployed at scale on Alibaba Cloud, and its proprietary AI supernode supports efficient inference for large models with more than 2 trillion parameters. Zhenwu V900 delivers 3x the performance of M890, and a single cluster can scale to 500,000 accelerators.

3. Qwen has iterated to the 2.4 trillion-parameter Qwen3.8-Max and plans to train new models with 5 to 10 trillion parameters. Alibaba plans to expand global data centers to more than 20 GW by 2032.

4. IDC notes that the hardest capabilities to replicate are model capability, inference efficiency, compute scale, unit cost and supply-chain certainty. Service providers can design efficiency, compute-cost and supply-chain assurance solutions accordingly.

Platform growth and operating practices

1. In 1H 2026, China's MaaS market reached RMB 10.63 billion, up 889.4% year on year, and its share of the public cloud PaaS market rose from 5.2% to 21.0%. IDC sees MaaS becoming cloud providers' most important growth source over the next few years, and platforms need to capture this growth source.

2. Alibaba Cloud leads with a 46.6% share, followed by Zhipu at 10.6% and Volcano Engine at 10.2%. Alibaba Cloud builds platform capability through three-layer investment in models, chips and cloud: Zhenwu AI chips have been deployed at scale, its proprietary AI supernode supports efficient inference for models with more than 2 trillion parameters, and Zhenwu V900 delivers 3x the performance of M890 with a single cluster scalable to 500,000 accelerators.

Demand and risk avoidance

1. LLMs accounted for 47.0% of MaaS revenue, making them the largest modality. Enterprise demand for knowledge Q&A, code generation and agent orchestration is text-centric, and large models are moving from assisting office work to directly taking on production tasks. Platform merchant recruitment and operations should focus on enterprise task-based AI demand.

2. Qwen has iterated to 2.4 trillion parameters and plans to train new models with 5 to 10 trillion parameters. Alibaba plans to expand global data centers to more than 20 GW by 2032, and AI cloud and computing power services revenue grew 45% year on year to RMB 48.44 billion.

3. Competition hinges on model capability, inference efficiency, compute scale, unit cost and supply-chain certainty. Platforms need to manage operations and mitigate risks around these hard-to-replicate capabilities.

Emerging industry trends and market structure

1. In 1H 2026, China's MaaS market reached RMB 10.63 billion, up 889.4% year on year, and its share of China's public cloud PaaS market rose from 5.2% in 2H 2025 to 21.0%, showing that MaaS has become an important growth driver for cloud services.

2. By revenue, LLMs accounted for 47.0% of China's MaaS market revenue, making them the largest modality. Alibaba Cloud leads with a 46.6% share, followed by Zhipu at 10.6% and Volcano Engine at 10.2%. Market concentration and the competitive landscape warrant research.

Business models and new questions

1. IDC sees MaaS becoming cloud providers' most important growth source over the next few years. Alibaba Cloud's lead reflects simultaneous investment across models, chips and cloud. Zhenwu AI chips have been deployed at scale, its proprietary AI supernode supports efficient inference for models with more than 2 trillion parameters, and Zhenwu V900 delivers 3x the performance of M890 with a single cluster scalable to 500,000 accelerators.

2. Qwen has iterated to the 2.4 trillion-parameter Qwen3.8-Max and plans to train new models with 5 to 10 trillion parameters. Alibaba plans to expand global data centers to more than 20 GW by 2032, and AI cloud and computing power services revenue grew 45% year on year to RMB 48.44 billion, reflecting vertical integration and compute infrastructure investment.

3. Enterprise demand for knowledge Q&A, code generation and agent orchestration is text-centric, and large models are moving from assisting office work to directly taking on production tasks. The hardest capabilities to replicate are model capability, inference efficiency, compute scale, unit cost and supply-chain certainty, forming new questions for industry 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.

10月29日,国际数据公司(IDC)最新发布的《中国AI软件市场半年度追踪,26H1》显示,2026年上半年,阿里云在中国大语言模型MaaS市场的份额为46.6%,智谱、火山引擎分别以10.6%、10.2%位列第二、第三。

这份数据按营收口径统计,并将大语言模型与图像、视频等模态分开计算。上半年,大语言模型占中国MaaS市场营收的47.0%,是规模最大的模态。IDC指出,企业对知识问答、代码生成和智能体编排的需求大多以文本为中心,大模型正从辅助办公走向直接承担生产任务。

中国MaaS市场整体仍在快速增长。上半年市场规模达106.3亿元人民币,同比增长889.4%,在中国公有云PaaS市场中的占比从2025年下半年的5.2%升至21.0%。

阿里云在大语言模型市场的领先,背后是阿里巴巴在模型、芯片和云三个层面的同时投入。平头哥AI芯片真武已经在阿里云上大规模部署,自研AI超节点支撑超过2万亿参数大模型的高效推理。9月发布的新一代AI芯片真武V900,性能达到真武M890的3倍,单一集群可扩展至50万卡,可支撑前沿模型的训练和推理。Qwen模型快速迭代至2.4万亿参数的Qwen3.8-Max,还在计划训练参数规模达5万亿至10万亿的新模型。

阿里巴巴也在持续加大算力投入,计划到2032年将阿里云运营的全球数据中心规模扩大至20GW以上。最新财报显示,截至6月30日的季度,其AI云与算力服务分部收入同比增长45%至484.4亿元。

IDC认为,MaaS正成为云服务商未来几年最重要的增长来源。在这一市场的竞争中,模型能力、推理效率,以及算力的规模、单位成本和供应链确定性是厂商最难被复制的能力。

注:文/龚作仁,文章来源:Laborer,本文为作者独立观点,不代表亿邦动力立场。

文章来源:Laborer

广告
微信
朋友圈

FAQ回顾

阿里云在中国大语言模型MaaS市场占据多少份额?

2026年上半年,阿里云在中国大语言模型MaaS市场的营收份额为46.6%,排名第一;智谱和火山引擎分别以10.6%、10.2%位列第二和第三。该数据由IDC按营收口径统计,并将大语言模型与图像、视频等模态分开计算。

中国MaaS市场规模有多大?

2026年上半年,中国MaaS市场规模达106.3亿元人民币,同比增长889.4%;其在中国公有云PaaS市场中的占比从2025年下半年的5.2%升至21.0%,显示MaaS市场仍在快速增长。

大语言模型在中国MaaS市场中处于什么地位?

2026年上半年,大语言模型占中国MaaS市场营收的47.0%,是规模最大的模态。企业对知识问答、代码生成和智能体编排的需求大多以文本为中心,大模型正从辅助办公走向直接承担生产任务。

阿里云为什么能在大语言模型MaaS市场保持领先?

阿里巴巴在模型、芯片和云三个层面同时投入:平头哥AI芯片真武已在阿里云大规模部署,自研AI超节点支撑超过2万亿参数大模型推理;Qwen迭代至2.4万亿参数的Qwen3.8-Max,并计划训练5万亿至10万亿参数新模型,同时持续扩大算力与数据中心规模。

MaaS市场竞争的关键能力有哪些?

MaaS正成为云服务商未来几年最重要的增长来源。竞争中最难被复制的能力包括模型能力、推理效率,以及算力的规模、单位成本和供应链确定性。

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