吴泳铭：机器正成为思考主力，智能将成规模化商品

作者：亿邦动力胡镤心文章来源：亿邦动力

2026-09-22 10:58

作者：亿邦动力胡镤心文章来源：亿邦动力

2026-09-22 10:58

邦小白快读

全文速览

这是阿里巴巴CEO吴泳铭在2026云栖大会上分享的关于机器智能时代的核心判断与产业布局，是普通人了解AI发展趋势、把握时代变化的重要参考信息。

1. 可快速理解的核心认知：大家不必把AI理解为模仿人类的“人造人”，它是和蒸汽机类似的全新生产力物种，目前帮人写PPT、做编程只是初级应用阶段，未来机器承担的思考总量会达到人类的1000倍以上，当前机器思考量还不到人类的3%，还有上万倍增长空间。

2. 和个人发展相关的趋势：未来每个人都能借助AI获得万倍级的智力杠杆，不用被重复繁杂的脑力劳动束缚，能把更多时间放在自己真正感兴趣、有创造力的事情上，AI越强大，人类的创造力空间反而越大。

3. 可感知的落地进展：目前阿里已经发布性能领先的国产AI芯片、可本地部署的开源大模型、移动端千问AI能力，普通人很快就能在各类设备上用到更流畅、低成本的AI服务。

这篇演讲透露的机器智能发展趋势，能为品牌商把握未来消费方向、调整研发与营销策略提供清晰参考。

1. 消费趋势新判断：智能会成为可规模化供给的常规商品，传统咖啡、汽车等物理商品的消费受人口规模限制存在天花板，但智能类服务的消费增长只和智能供给规模挂钩，没有明确的人口绑定上限，是品牌可以提前布局的长坡赛道。

2. 用户行为新变化：未来每个普通用户都能借助AI获得万倍智力杠杆，成为具备强创意落地能力的超级个体，用户的决策逻辑、需求颗粒度会发生根本变化，品牌要提前适配更个性化、定制化的用户需求，不能再依赖标准化大众爆品的运营思路。

3. 产品研发新机会：未来低成本的充沛AI算力，能支撑各个细分领域的长期研发，此前因为研发成本过高被放弃的小众圈层、特殊群体需求，也能获得足够的智力资源支持，品牌可以挖掘这类细分赛道的新机会。

吴泳铭披露的阿里AI长期战略与智能时代趋势，是电商卖家寻找新增长机会、对接平台资源、调整经营策略的核心参考。

1. 明确的增长机会提示：当前AI产业还处在类似电力时代刚发明电灯的早期阶段，真正的现象级AI应用还未出现，机器思考总量仅为人类的3%，未来有上万倍的增长空间，智能相关商品、适配AI能力的周边服务都是确定性的高增长赛道。

2. 可对接的平台资源：阿里正在全力搭建AI模型、AI芯片、AI云三大基础设施，已经推出可本地化部署的开源千问模型、千问移动端AI解决方案，卖家可以对接这些能力改造客服、运营、内容生产环节，大幅降低经营成本。

3. 经营风向提示：未来消费者借助AI会做出更理性、更精准的消费决策，套路化营销、同质化爆品的生存空间会被挤压，卖家可以借助AI能力承接小批量定制化订单，匹配用户的个性化需求。

本次演讲披露的智能产业布局与供应链需求，为工厂寻找新订单机会、调整产品设计方向、推进数字化转型提供了明确指引。

1. 确定性的供应链商业机会：当前全球AI数据中心相关供应链处于全球性紧缺状态，阿里计划到2032年将运营的全球数据中心规模扩充到20T瓦，自研的AI芯片、M890 AI超节点已经开始规模化上架，未来几年芯片出货量会大幅增长，相关服务器、网络设备、硬件配套的生产工厂有明确的大额订单机会。

2. 产品设计的新方向：未来智能会成为无处不在的规模化服务，目前已有可本地部署的开源桌面大模型、面向移动设备的千问移动端AI方案，工厂可以对接这些成熟AI能力，设计生产搭载本地化AI能力的桌面硬件、智能终端等新品类。

3. 数字化转型启示：工厂可以借助未来低成本的充沛AI算力，用大量AI智能体承接生产调度、质检、工艺优化等工作，把少数专家的生产经验转化为可规模化复制的能力，提升生产效率。

本次演讲清晰披露了AI产业的发展阶段、技术演进方向与现存行业痛点，是服务商找准服务定位、打磨解决方案的重要决策依据。

1. 行业发展趋势判断：当前AI产业尚处早期，类似电力时代刚点亮电灯的阶段，真正改变行业的代表性应用还未出现，机器思考总量仅为人类的3%，未来有上万倍增长空间，面向模型训练、推理、场景落地的服务会有长期确定性需求。

2. 技术落地的明确方向：阿里当前重点攻坚可递归自我改进的长程任务大模型、多模态理解生成一体化模型、全栈自研AI芯片、适配AI负载的专用云服务，服务商可以围绕这些技术方向，开发行业应用插件、场景化落地解决方案、多模态交互服务。

3. 现存的客户痛点提示：当前AI算力供给受全球供应链限制存在较大缺口，大量企业、开发者有大模型本地化部署、移动端AI适配的需求，服务商可以围绕算力调度优化、本地化部署服务、移动端适配等方向打磨产品。

吴泳铭作为头部数字平台阿里的CEO，提出的机器智能时代平台战略布局，为各类平台商的长期发展提供了可参考的实践路径。

1. 平台长期核心建设方向：机器智能时代的核心底层基础设施是AI模型、AI芯片、AI云，平台不能只停留在应用层布局，要提前攻坚底层能力：模型层要研发可自我迭代、支持长程复杂任务、多模态自然交互的能力，芯片层要做全栈算力布局与软硬件协同优化，云层要建设适配大模型持续训练、推理的专用AI云架构。

2. 生态建设与需求对接：当前企业、开发者对AI能力的需求非常强劲，但算力供给受全球供应链限制存在较大缺口，平台要联动上下游合作伙伴共同扩充电力、算力等基础设施，同时通过开源模型、开放移动端AI解决方案等方式，吸引开发者共建生态。

3. 风险规避提示：平台在推进AI能力落地时，要做好AI和人类文化、审美、价值观的对齐，规避技术伦理风险，承担相应的平台责任。

吴泳铭在云栖大会上提出的机器智能时代系列判断、阿里的全栈AI布局，为研究者观察智能产业演进、总结产业发展规律提供了最新的一手产业实践素材。

1. 产业认知的新动向：当前产业界对AI的认知已经从模仿人类的人工智能，升级为将机器智能视为全新生产力物种，AGI只是发展起点，下一阶段目标是可自我迭代的超级智能ASI，产业界已形成共识：未来机器承担的思考总量将达到人类的1000倍以上，智能会成为可规模化供给的商品。

2. 产业实践的新进展：头部企业阿里已形成AI模型、AI芯片、AI云全栈布局，模型侧由天问团队研发面向长程复杂任务的新架构模型，芯片侧发布性能达3890型号3倍的昇5B 900芯片，M890 AI超节点已规模化部署，云侧规划2032年全球数据中心规模超20T瓦，但当前面临全球AI供应链紧缺的现实约束。

3. 值得关注的新研究命题：机器智能规模化供给后的社会结构变化、超级个体的行为模式、AI基础设施的公共属性、智能商品的流通机制等，都是值得深入研究的新方向。

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 speech, delivered by Alibaba CEO Eddie Wu at the 2026 Apsara Conference, outlines core judgments on the machine intelligence era and Alibaba’s industrial layout, serving as a key reference for general audiences to understand AI development trends and navigate ongoing technological shifts.

1. Accessible core takeaways: AI should not be misunderstood as a human-like "artificial person" that mimics human behavior. Rather, it is an entirely new productive force analogous to the steam engine. Today’s use cases—such as generating presentations or assisting with coding—represent only its earliest stage of application. In the future, the total volume of thinking carried out by machines will exceed that of humans by more than 1,000 times; currently, machine thinking accounts for less than 3% of human thinking volume, leaving tens of thousands of times room for growth.

2. Trends relevant to individual development: In the future, every person will be able to access a 10,000-fold intellectual leverage through AI, freeing them from repetitive, tedious mental labor and allowing them to devote more time to creative, personally meaningful pursuits. The more powerful AI becomes, the greater the space for human creativity.

3. Tangible implementation progress: Alibaba has already released high-performance domestic AI chips, locally deployable open-source large models, and mobile-based Qwen AI capabilities. General users will soon have access to smoother, lower-cost AI services across a wide range of devices.

The machine intelligence development trends outlined in this speech provide clear guidance for brands to identify future consumption directions and adjust R&D and marketing strategies.

1. New assessment of consumer trends: Intelligence will become a routinely available commodity supplied at scale. While consumption of physical goods such as traditional coffee and automobiles faces a ceiling tied to population size, growth in intelligent services is linked only to the scale of intelligence supply, with no explicit population-linked upper limit—making it a long, high-potential track for brands to position early.

2. Shifts in user behavior: In the future, every ordinary user will gain 10,000-fold intellectual leverage via AI, becoming a "super individual" with strong capabilities to turn creative ideas into reality. User decision-making logic and the granularity of demand will fundamentally change. Brands must adapt in advance to more personalized, customized user needs, moving away from reliance on standardized mass-hit product strategies.

3. New opportunities in product R&D: Low-cost, abundant AI computing power in the future will support long-term R&D across niche segments. Previously abandoned demand from small, specialized circles and underserved groups—set aside due to prohibitive R&D costs—will now have access to sufficient intellectual resources, allowing brands to tap new opportunities in these niche tracks.

Alibaba’s long-term AI strategy and intelligence-era trends disclosed by Eddie Wu serve as a core reference for e-commerce sellers to identify new growth opportunities, leverage platform resources, and adjust business strategies.

1. Clear growth opportunity signals: The AI industry is currently in an early stage analogous to the invention of the electric lightbulb at the dawn of the electrical age. Truly breakout, transformative AI applications have yet to emerge. With machine thinking volume at only 3% of human levels, there is tens of thousands of times growth potential ahead, making intelligence-related products and AI-enabled supporting services definite high-growth tracks.

2. Accessible platform resources: Alibaba is fully building three core AI infrastructure layers: AI models, AI chips, and AI cloud. It has already launched the locally deployable open-source Qwen model and mobile Qwen AI solutions. Sellers can integrate these capabilities to upgrade customer service, operations, and content production, significantly reducing operating costs.

3. Shifts in business direction: Future consumers will make more rational, precise purchasing decisions with AI assistance, squeezing the viability of formulaic marketing and homogeneous hit products. Sellers can use AI capabilities to fulfill small-batch customized orders and match users’ personalized demand.

The intelligent industry layout and supply chain demand disclosed in this speech provide clear guidance for factories to identify new order opportunities, adjust product design directions, and advance digital transformation.

1. Definite supply chain business opportunities: The global supply chain for AI data centers is currently facing worldwide shortages. Alibaba plans to expand its global operational data center scale to 20 terawatts by 2032, and its self-developed AI chips and M890 AI supernodes have already begun large-scale deployment. Chip shipment volumes will rise sharply in the coming years, creating clear, large order opportunities for factories producing related servers, network equipment, and supporting hardware.

2. New directions for product design: Intelligence will become a ubiquitous, large-scale service in the future. Already, locally deployable open-source desktop large models and Qwen mobile AI solutions for end devices are available. Factories can integrate these mature AI capabilities to design and produce new product categories such as desktop hardware and smart terminals with on-device local AI functionality.

3. Digital transformation insights: Factories can leverage future low-cost, abundant AI computing power to deploy large numbers of AI agents for production scheduling, quality inspection, and process optimization, translating the production expertise of a small number of specialists into scalable, replicable capabilities to improve production efficiency.

This speech clearly outlines the development stage of the AI industry, technology evolution directions, and existing industry pain points, serving as a critical decision-making reference for service providers to define their market positioning and refine solutions.

1. Industry trend assessment: The AI industry is still in its early stage, comparable to the moment the first electric light was switched on in the electrical era. Truly transformative, industry-defining applications have yet to emerge. With machine thinking volume at just 3% of human levels, there is tens of thousands of times growth potential ahead, creating long-term, certain demand for services covering model training, inference, and scenario-based implementation.

2. Clear directions for technology implementation: Alibaba is currently focusing on key technology priorities: large models for recursive self-improvement and long-horizon tasks, integrated multimodal understanding and generation models, full-stack self-developed AI chips, and dedicated cloud services optimized for AI workloads. Service providers can build industry application plugins, scenario-based implementation solutions, and multimodal interaction services around these technology directions.

3. Highlighted existing customer pain points: Current AI computing power supply faces a large gap due to global supply chain constraints, and a large number of enterprises and developers have demand for local large model deployment and mobile AI adaptation. Service providers can refine products focused on computing power scheduling optimization, local deployment services, and mobile adaptation.

As CEO of Alibaba, a leading global digital platform, Eddie Wu’s outlined strategic layout for the machine intelligence era provides actionable reference pathways for the long-term development of all types of platform operators.

1. Core long-term platform construction priorities: The core underlying infrastructure of the machine intelligence era consists of AI models, AI chips, and AI cloud. Platforms cannot limit their layout to the application layer; they must build underlying capabilities in advance: at the model layer, they need to develop capabilities for self-iteration, support for long-horizon complex tasks, and natural multimodal interaction; at the chip layer, they need full-stack computing power layout and hardware-software co-optimization; at the cloud layer, they need to build dedicated AI cloud architectures adapted to continuous large model training and inference.

2. Ecosystem building and demand alignment: Currently, enterprises and developers have very strong demand for AI capabilities, but computing power supply faces a large gap due to global supply chain constraints. Platforms should collaborate with upstream and downstream partners to expand infrastructure such as power supply and computing capacity, while attracting developers to co-build the ecosystem through open-source models and open mobile AI solutions.

3. Risk mitigation guidance: When rolling out AI capabilities, platforms must ensure alignment between AI and human culture, aesthetics, and values to mitigate technology ethics risks and fulfill corresponding platform responsibilities.

The series of judgments on the machine intelligence era and Alibaba’s full-stack AI layout put forward by Eddie Wu at the Apsara Conference provide the latest first-hand industry practice materials for researchers to observe the evolution of the intelligent industry and summarize industrial development patterns.

1. New trends in industry perceptions: The industry’s understanding of AI has evolved from the framing of "artificial intelligence" that imitates humans, to framing machine intelligence as an entirely new species of productive force. AGI is viewed only as a starting point for development; the next stage target is self-iterating artificial superintelligence (ASI). An industry consensus has formed that the total volume of thinking undertaken by machines will exceed that of humans by more than 1,000 times in the future, and that intelligence will become a commodity supplied at scale.

2. New progress in industry practice: Leading enterprise Alibaba has formed a full-stack layout across AI models, AI chips, and AI cloud. On the model side, its Tianwen team is developing a new architecture model for long-horizon complex tasks; on the chip side, it has released the Sheng 5B 900 chip, which delivers three times the performance of the 3890 model, and the M890 AI supernode has been deployed at scale; on the cloud side, it plans to expand global data center scale to over 20 terawatts by 2032, though this rollout faces real constraints from the global AI supply shortage.

3. New research propositions worth exploring: Changes to social structures following the large-scale supply of machine intelligence, behavioral patterns of super-empowered individuals, the public attribute of AI infrastructure, and the circulation mechanism of intelligent commodities are all new directions worthy of 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.

【亿邦原创】9月22日上午，2026云栖大会开幕，阿里巴巴集团CEO吴泳铭发表主题演讲，系统阐述了他对机器智能时代的判断，以及阿里巴巴在AI模型、AI芯片和AI云上的长期战略。

他表示，过去一年AI技术迭代远远超过原有设想，AGI只是起点，AI将继续走向能够自我迭代的超级智能ASI。一个深刻变化正在发生：机器正在成为思考的主力，智能正在成为一种可以规模化供给的商品。

1、机器智能不是“人造人”，而是新物种

吴泳铭首先回应了人们对AI的惯性理解。他说，大家习惯把AI叫作人工智能，但人工在应用中更像人造的意思，就像人造钻石、人造皮革。过去人们总在问：AI像不像人？能不能像人一样思考？图灵测试也一度成为衡量标准。

但他认为，这种思路可能低估了机器智能。蒸汽机早期替代人和马做矿井抽水、搬运等工作，马力一词也由此而来。但很快，机器动力能做的事远远超过人和动物。再多的人和马也无法让火车飞奔，更无法让飞机和火箭上天。机器动力不是简单的人造体力，而是创造了一个完全不同的物种。

“今天的机器智能，也不是人类智能的替代，而是完全不同的工作。”吴泳铭说，今天AI帮人编程、写报告、写PPT，只是机器智能的初级阶段。当机器智能供给变得无限充沛，它对人类社会的意义将远远超出对现有脑力劳动的替代。这种变化相比工业革命来得更加深刻。

2、两大判断：思考总量与代表性产品

吴泳铭对机器智能时代提出两个判断。

第一，未来机器供给的思考总量将是人类思考总量的1000倍以上。

他回顾物理世界的历史：从蒸汽机发明开始，机械动力承担了世界几乎99.9%的物理工作。未来，人类思考总量也许还会继续增长，但机器思考总规模将以更快速度扩张，并承担整个世界思考总量的99.9%。

他从供给和需求两方面进行了一段推演。在供给端，人类顶级思考者供给十分有限，培养一名顶级科学家或行业专家需要数十年学习、时间、天赋和运气，许多领域长期缺少足够智力配合。例如新生儿早衰症这样的罕见病，全球有记录患者可能只有几百个，很难获得充足研究资源，很少有公司愿意耗费几十年、几百人研发一款新药。但当AI具备专业研究能力后，只要算力足够多、足够便宜，每个细分领域都可能同时有上百万个AI持续研究、模拟和验证。顶级思考将从奢侈品变成可规模化供给的商品。

在需求端，工业时代人类消费的很多商品取决于人口数量，咖啡、汽车、手机等物理商品消费总有上限。但当AI能力越来越强，关系会发生根本变化。比如让AI去造一艘飞船去火星，AI可以将复杂任务分解成成千上万个子任务，由上百万个agent不眠不休工作，直到任务完成。人只需要提供创意和目标，就能调动庞大智力资源。每一个超级个体，未来都可以拥有上万倍的智力杠杆。思考消耗量将不再与人口数量成正比，而是与机器智能的思考深度和供给规模成正比。

吴泳铭说，今天机器思考总量大概不到人类的3%。如果未来机器思考总量是人类的1000倍，那么机器思考的规模和需求还有几万倍以上增长空间。

第二，机器智能时代的代表性产品还没有出现。他用电的历史作类比：电最早用于替代煤油灯和蜡烛照明。1882年，爱迪生建成珍珠街电站，点亮附近约400盏灯。但后来改变世界的电器产品，如洗衣机、电冰箱等，是很多年后才进入家庭。第一台计算机要到1946年，距离珍珠街电站已过去60多年。

“今天的AI coding，非常像机器智能时代早期的电灯。”吴泳铭说，它替代的是人类现在刻意做的工作，比如编程、写报告。但替代现有工作，不可能创造一个真正的新时代。今天很难预测机器智能爆发后会出现什么新产品、新发明，但无论后来出现多少新发明，第一步都是要建足够多的发电站、铺足够广的电网。机器智能时代同样需要更强大的基础设施。

3、三大基础设施：AI模型、AI芯片、AI云

吴泳铭明确，机器智能时代的三大基础是AI模型、AI芯片和AI云，它们是机器思考能够规模化供给的前提。阿里巴巴将坚定投入这三大基础，这是长期战略。

在AI模型方面，吴泳铭表示，通往ASI的技术路径正变得更加清晰。从编码、办公到科学研究，模型正在承担越来越多真实任务，这些真实任务和反馈会让模型越来越强。AI研究员逐渐探索出超级人工智能的具体技术路径——递归式自我改进，即行业所称的AISI。模型从真实反馈中发现短板，自己设计实验、构建数据、评价结果，循环推动自身进化。

他透露，千问团队正在推进模型架构和数据优化研究，计划训练全新模型，目标是完成更复杂、更长程的任务，向ASI迈进。同时，多模态能力是另一个重要方向。一个足够聪明的大脑还不够，AI还需要掌握与人类交互、感知的能力，理解声音、图像、表情和动作，对齐人类文化、审美和价值观。未来人们可以像跟人一样与AI交流，不必面对复杂交互界面。因此，理解与生成一体化的模型也是重要方向。

在AI芯片方面，吴泳铭说，如果说Token是AI时代的电，芯片就是关键装备。未来机器智能时代对Token需求几乎没有上限，芯片需要不断提高性能、扩大供给。平头哥已在数据中心芯片上全面布局，包括GPU、CPU、智能网卡和ASIC互联芯片等，并正在对芯片、服务器、超节点、网络、模型和推理进行联合优化，提升AI集群功能和效率。

阿里自研的M890 AI超节点已成功支撑超大参数规模的大模型推理，是全球极少数具备这样能力的公司之一。本季度开始，基于M890的超节点已在阿里数据中心规模化上架。当天，阿里还发布新一代AI芯片昇5B 900，称其是目前中国算力性能最强的AI芯片，性能达到3890的3倍。基于它构建的单一集群可扩展至50万，支撑前沿模型训练和推理。随着平头哥芯片产品线成熟和客户广泛应用，预计未来几年平头哥年出货量将大幅增长。

在终端侧，吴泳铭表示，智能应无处不在。桌面端，阿里开源的千问27B模型是全球最受开发者欢迎的模型之一，未来将持续优化，支持开发者和企业本地化部署。移动端，当天正式发布千问移动端，为合作伙伴提供移动端AI解决方案，让移动设备能够完成各类复杂任务。

在AI云方面，吴泳铭说，芯片负责生产Token，云负责把Token送到每一个需要它的地方。模型需要持续训练和推理，推理需要长时间执行任务，AI开发的软件也会越来越多，它们都会成为AI云上的居民，持续调用算力和资源。阿里需要围绕这些新工作负载，建设为模型和推理持续运行而设计的AI云。

吴泳铭透露，目前客户需求非常强劲，阿里云收入也在持续加速增长，但行业中长期需求远超阿里云供给能力。当前AI数据中心相关供应链全球性紧缺，限制了阿里云算力增长速度。为此，阿里巴巴将与所有合作伙伴一起，全力投入AI基础设施建设，打造面向机器智能时代的AI云。目标是到2032年，阿里云运营的全球数据中心规模超过20T瓦，以满足快速增长的AI需求。

演讲最后，吴泳铭回到人与机器的关系。他说，工业革命让机器承担越来越多体力劳动，人类从繁重体力劳动中解放后，发展出足球、篮球等更多运动，奥运会、世界杯、NBA都是工业革命后的产物。印刷术兴起让文化创作爆发式增长；照相机出现后，绘画并没有消失，画家不再追求画得多像，反而促进印象派和抽象艺术诞生；留声机出现也没有让歌手失业，反而出现更多大众流行歌手，让音乐成为最广泛的大众艺术。

“当机器承担更多不能做的事情，人便有时间去做真正想做的。”吴泳铭说，未来机器智能将为人类的好奇心和创造力打开更大空间。今天我们很难具体描绘未来，就像300年前田间农夫很难想象，未来会有人走进健身房付费进行体力劳动。

“去年我们说AI越强大，人类更强大，今天我们依旧这么认为。”吴泳铭表示。

本文首发于亿邦动力官方网站

【版权提示】亿邦动力倡导尊重与保护知识产权。未经许可，任何人不得复制、转载、或以其他方式使用本网站的内容。如发现本站文章存在版权问题，烦请提供版权疑问、身份证明、版权证明、联系方式等发邮件至run@ebrun.com，我们将及时沟通与处理。

已点赞