广告
加载中

不比拼Token总量 华为云选择了一条更难的路

牛慧 2026-06-08 13:02
牛慧 2026/06/08 13:02

邦小白快读

EN
全文速览

本文总结了华为云首届INSPIRE创想者大会的核心信息,干货如下:

1. 当前云计算已迈入Agent原生时代,技术重心从模型训练转向大模型推理,行业内多数厂商比拼Token总量、比拼规模,内卷严重,华为云明确走差异化路线,定位为全栈国产化、生态开放、AI普惠,不追求Token绝对数值,更关注每个Token在ToB场景带来的真实生产效率提升。

2. 本次大会发布了全栈重构的Agentic云新品,基于华为自研芯片打造的灵衢智算集群,单千卡吞吐是全球最大超节点的两倍,代码智能体码道已正式商用,用户数突破十万,华为云还推出了安全运行底座、混合云AI解决方案等多款产品。

3. 华为云构建了多层级开放生态,上线四大行业AI梦工厂专区,推动AI落地到县域医疗等下沉场景,实实在在实现了AI普惠。

本文能为ToB科技品牌的定位、营销和发展提供参考,干货如下:

1. 当前Agent原生时代云服务行业竞争激烈,多数品牌陷入比拼Token总量的规模内卷,华为云选择差异化定位,聚焦全栈国产化和真实场景生产效率提升,成功避开红海竞争,为品牌差异化打造提供了可参考的成功案例。

2. 行业趋势层面,当前AI在国内行业的渗透率极低,传统企业Token支出占IT总支出不足5%,央国企仅1%左右,同时政企、国计民生行业对数据安全、本地化部署有强需求,市场增长空间巨大。

3. 品牌建设层面,华为云通过开放生态聚合伙伴,落地真实行业场景打造普惠价值,比如和瑞金医院合作的病理大模型落地县域医院,解决真实痛点,有效强化了负责任的科技品牌认知,值得品牌参考。

本文梳理了智能体时代云服务领域的最新变化,对卖家把握机会、规避风险有指导意义,干货如下:

1. 当前行业已进入Agent规模化落地阶段,云技术栈全面转向大模型推理,行业整体增长空间充足,目前AI在国内各行业渗透率仍然很低,传统企业Token支出占IT总支出不足5%,央国企仅1%左右,存在大量未被满足的市场需求。

2. 机会层面,华为云构建了全链路开放生态,从底层芯片架构到操作系统、开发工具全部开源开放,还推出了行业AI梦工厂开放专区,中小卖家、模型企业可以入驻共享算力、数据和场景资源,大幅降低开发和运营成本。

3. 风险提示层面,当前行业普遍陷入比拼Token规模的无效内卷,卖家应避开单纯的规模竞争,聚焦真实场景的生产效率提升,深耕垂直行业,才能获得长期增长。

本文对制造工厂把握AI转型机会、推进数字化升级有诸多启示,干货如下:

1. 商业机会层面,智能制造已经成为华为云行业AI梦工厂的核心新增专区,该专区以开放社区模式聚合开发者、模型、数据和解决方案,工厂可以低成本接入生态资源,不需要单独投入高额的算力和开发成本,就能推进自身的AI数字化转型。

2. 转型方案层面,华为云主推公有云加混合云、私有云并重的模式,推荐采用本地数据加远端公有云模型调用的方案,既满足工厂对生产数据安全、本地化部署的核心需求,又能享受最新的AI技术迭代成果,平衡了安全、成本和性能,非常符合国内制造工厂的实际需求。

3. 安全层面,华为云提供全栈国产化的AI基础设施,基于自研芯片的智算集群能满足大规模AI推理需求,工厂选择国产方案可以保障技术供应链安全,还能获得完整的生态支持,降低转型风险。

本文梳理了智能体时代云服务行业的发展趋势和客户痛点,给出了可行的解决方案方向,干货如下:

1. 行业发展趋势:云计算已经完成从云原生、AI原生到Agent原生的进化,技术重心全面转向大模型推理,全栈国产化是国内市场明确的发展方向,客户不再只关注规模,更关注AI带来的真实生产价值,单纯比拼Token总量的规模竞争已经不可持续。

2. 行业核心痛点:国计民生行业客户对数据安全、本地化部署有强需求,传统方案无法满足;大量中小AI服务商缺算力、缺行业资源,难以快速实现商业化;传统头部标杆复制的AI落地模式,跟不上AI高速迭代的节奏,落地效率太低。

3. 可行解决方案:打造全栈国产软硬芯协同的基础设施,推出混合云加AI的解决方案满足安全合规需求;构建多层级开放生态,开放核心技术代码,打造开放行业专区社区聚合资源,降低中小服务商的参与门槛,有效提升AI落地效率。

本文分享了华为云在智能体时代的平台运营经验,对云平台商的发展有参考价值,干货如下:

1. 当前客户核心需求:ToB类客户尤其是国计民生行业客户,对全栈国产化、安全可控的云服务有强需求,需要满足数据安全合规、本地化部署的要求;不同场景需要适配不同模型,需要低成本高效率的模型调用能力;中小开发者和企业需要低成本的开发资源,加快AI落地速度。

2. 平台创新运营做法:华为云构建从底层芯片到应用层的全链路开放体系,开放各类底层技术开源,对接全市场所有合作模型,推出模型路由,精准率超95%,调用成本平均下降20%;创新推出行业AI梦工厂专区的开放社区模式,替代传统的头部标杆复制模式,大幅加快AI向行业渗透的速度。

3. 风险规避方向:当前行业普遍陷入比拼Token总量、收入规模的内卷,平台不要盲目追求规模增长,应聚焦真实场景的生产效率提升,深耕行业落地,走差异化路线,才能避免无效竞争,获得长期发展。

本文披露了智能体时代云产业的最新发展动向,为产业研究提供了最新的一手素材,干货如下:

1. 产业最新动向:全球云计算已经正式进入Agent原生时代,技术重心全面从模型训练转向大模型推理,全球头部云厂商开启了Agent基础设施的军备竞赛,国内云市场出现明确的差异化路线,华为云主打全栈国产化,不拼Token规模和收入总量,更关注真实生产效率提升,和其他厂商的路线形成鲜明差异。

2. 产业新问题:当前AI在国内行业的渗透率极低,传统的先做头部标杆再向中腰部渗透的落地模式,已经跟不上AI高速迭代的需求;国计民生行业对数据安全和本地化部署的强需求长期没有得到很好满足;行业内对AI泡沫的争论不断,需要探索可持续的落地路径。

3. 商业模式创新:华为云推出全链路开放生态加行业AI梦工厂专区的新模式,通过开源开放聚合各类生态伙伴,依托开放社区加快行业AI落地,依托多年积累的政企资源推动AI普惠,为云产业探索ToB落地提供了新的商业模式参考。

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 summarizes the core takeaways from Huawei Cloud's first INSPIRE Innovator Conference:

1. Cloud computing has now entered the Agent-native era, with the technical focus shifting from model training to large model inference. Most industry players are locked in cutthroat internal competition, vying to outdo each other in total token volume and scale. Huawei Cloud has clearly chosen a differentiated positioning: it focuses on full-stack localization, open ecosystem, and accessible AI for all. Rather than chasing absolute token numbers, it prioritizes the actual productivity gains that each token delivers for B2B scenarios.

2. At the conference, Huawei Cloud launched its fully stack-rebuilt Agentic Cloud new products. Its Lingqu intelligent computing cluster, built on Huawei's self-developed chips, delivers 2x the throughput per 1,000-accelerator-card unit compared to the world's largest supernode. Its code intelligence product "Maodao" is now officially commercially available and has surpassed 100,000 users. Huawei Cloud also released multiple other products, including a secure operating foundation and hybrid cloud AI solutions.

3. Huawei Cloud has built a multi-level open ecosystem and launched four industry AI Dream Factory zones, which are driving AI adoption in underserved scenarios such as county-level healthcare, delivering on the promise of accessible AI for all.

This article offers actionable insights for positioning, marketing and growth of B2B technology brands:

1. In the highly competitive cloud service market of the Agent-native era, most brands are trapped in fruitless internal competition over total token volume. Huawei Cloud has chosen a differentiated positioning focused on full-stack localization and productivity gains in real-world scenarios, successfully avoiding red-ocean competition. It serves as a replicable success case for brands looking to build their own differentiation.

2. In terms of industry trends, AI penetration across Chinese domestic industries remains extremely low today. Token spending accounts for less than 5% of total IT spending at traditional enterprises, and only around 1% at central state-owned enterprises. Meanwhile, government and public interest sectors have strong demand for data security and on-premises deployment, leaving enormous room for market growth.

3. In terms of brand building, Huawei Cloud aggregates partners through an open ecosystem, delivers inclusive value by rolling out solutions in real industry scenarios – for example, its pathology large model developed in partnership with Ruijin Hospital has been deployed at county-level hospitals to solve real pain points – which effectively reinforces its brand perception as a responsible technology player. This approach is well worth referencing for other brands.

This article outlines the latest changes in the cloud service sector in the agent era, offering guidance for sellers to capture opportunities and mitigate risks:

1. The industry has now entered the stage of large-scale agent deployment, with the entire cloud technology stack shifting toward large model inference. The overall industry boasts ample room for growth, but AI penetration across Chinese domestic industries still remains very low. Token spending accounts for less than 5% of total IT spending at traditional enterprises, and only around 1% at central state-owned enterprises, leaving a large volume of unmet market demand.

2. In terms of opportunities, Huawei Cloud has built a full-link open ecosystem, with open-source access to everything from underlying chip architecture to operating systems and development tools. It has also launched open industry AI Dream Factory zones. Small and medium-sized sellers and AI model companies can settle in to share computing power, data and scenario resources, significantly cutting their development and operating costs.

3. In terms of risk warning, the industry is currently widely trapped in fruitless internal competition over token scale. Sellers should avoid pure scale competition, and instead focus on improving productivity in real scenarios and深耕垂直行业, to achieve long-term growth.

This article offers multiple insights for manufacturing factories to capture AI transformation opportunities and advance digital upgrade:

1. In terms of business opportunities, intelligent manufacturing has become a core new zone of Huawei Cloud's Industry AI Dream Factory. The zone aggregates developers, models, data and solutions through an open community model. Factories can access ecosystem resources at low cost, and advance their own AI and digital transformation without having to invest heavily in standalone computing power and development.

2. In terms of transformation roadmaps, Huawei Cloud promotes a balanced model combining public cloud with hybrid cloud and private cloud, and recommends a solution that combines local data storage with remote public cloud model invocation. This approach meets factories' core requirements for production data security and on-premises deployment, while also allowing them to benefit from the latest AI technology iterations, striking a good balance between security, cost and performance that aligns well with the actual needs of Chinese domestic manufacturing factories.

3. In terms of security, Huawei Cloud provides full-stack localized AI infrastructure. Its intelligent computing cluster built on self-developed chips can meet large-scale AI inference requirements. Choosing a domestic solution allows factories to secure their technology supply chain, while also gaining access to complete ecosystem support and reducing transformation risks.

This article outlines industry development trends and customer pain points for the cloud service sector in the agent era, and outlines actionable solution directions:

1. Industry development trend: Cloud computing has completed its evolution from cloud-native to AI-native and now to Agent-native, with its technical focus fully shifting to large model inference. Full-stack localization is a clear development direction for the Chinese domestic market. Customers no longer focus only on scale, but prioritize the actual production value delivered by AI. Pure competition over total token volume is no longer sustainable.

2. Core industry pain points: Customers in public interest sectors have strong demand for data security and on-premises deployment, which traditional solutions cannot meet. A large number of small and medium-sized AI service providers lack computing power and industry resources, making it difficult to achieve commercialization quickly. The traditional AI go-to-market model of building a flagship case first and replicating it cannot keep up with the rapid iteration pace of AI, resulting in very low deployment efficiency.

3. Actionable solutions: Build full-stack localized infrastructure integrating software, hardware and chips, and launch hybrid cloud + AI solutions to meet security and compliance requirements; build a multi-level open ecosystem, open source core technology code, and create open industry zone communities to aggregate resources, lower the entry barrier for small and medium-sized service providers, and effectively improve AI deployment efficiency.

This article shares Huawei Cloud's platform operation experience in the agent era, offering reference value for the development of cloud platform players:

1. Current core customer needs: B2B customers, especially those in public interest sectors, have strong demand for full-stack localized, secure and controllable cloud services that meet requirements for data security compliance and on-premises deployment; different scenarios require adaptation to different models, so there is demand for low-cost, high-efficiency model invocation capabilities; small and medium-sized developers and enterprises need low-cost development resources to speed up AI deployment.

2. Innovative platform operation practices: Huawei Cloud has built a full-link open system from underlying chips to the application layer, open-sourced various underlying technologies, connected to all cooperative models across the market, and launched model routing with an accuracy rate exceeding 95% and an average 20% reduction in invocation cost; it has pioneered the open community model of industry AI Dream Factory zones, replacing the traditional flagship case replication model and significantly accelerating the pace of AI penetration into industries.

3. Risk mitigation direction: The industry is currently widely trapped in internal competition over token volume and revenue scale. Platforms should not blindly pursue scale growth, but instead focus on productivity improvements in real scenarios, deepen industry deployment, and pursue a differentiated route to avoid ineffective competition and achieve long-term development.

This article discloses the latest development trends of the cloud industry in the agent era, providing up-to-date first-hand material for industry research:

1. Latest industry developments: Global cloud computing has officially entered the Agent-native era, with the technical focus fully shifting from model training to large model inference. Leading global cloud vendors have launched an arms race for agent infrastructure. A clear differentiated route has emerged in China's cloud market: Huawei Cloud focuses on full-stack localization, does not compete on token scale or total revenue, and prioritizes real productivity improvements, which stands in stark contrast to the route adopted by other vendors.

2. New industry problems: AI penetration across Chinese domestic industries is currently extremely low. The traditional go-to-market model of building a flagship case first before penetrating to mid-market and smaller players can no longer keep up with the rapid iteration demand of AI; the strong demand for data security and on-premises deployment from public interest sectors has not been well met for a long time; there is ongoing debate over an AI bubble in the industry, and there is a need to explore a sustainable deployment path.

3. Business model innovation: Huawei Cloud has launched a new model combining a full-link open ecosystem with industry AI Dream Factory zones. It aggregates various ecosystem partners through open source and open access, accelerates industry AI deployment through open communities, and leverages its years of accumulated government and enterprise resources to promote AI accessibility, offering a new business model reference for the cloud industry to explore B2B deployment.

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.

华为云对外传递差异化定位:全栈国产化、生态开放、实现AI普惠。

文|牛慧

编|赵艳秋

6月5日,上海西岸国际会议中心,“INSPIRE”渐变色字体闪烁,背景视频中播放着华为云已深入30多个行业、500多个场景解决行业难题的画面。会场内挤满了企业代表与生态伙伴,华为云首届INSPIRE创想者大会就此拉开帷幕。

今年以来,Agent规模化落地,让高价值Token“一夜之间”变得抢手。云计算正从云原生、AI原生迈入Agent原生时代——过去几年云的技术栈更看重模型训练,现在开始全面转向大模型推理,各大云厂商官宣庞大复杂的技术栈和产品服务体系重构,设下高Token销售KPI,市场空前活跃和内卷。

在此背景下,此次华为云INSPIRE创想者大会,也官宣了庞大的技术产品体系更新,“是从去年开始,面向AI发布新品最多的一次。”华为公司董事、华为云CEO周跃峰介绍。在产品之外,周跃峰也对外传递了华为云与其他大厂有所不同的战略和打法:

——不追求Token的绝对数值,更关注每一个token背后面向ToB场景带来的生产效率提升;

——由于全栈技术“用国产硅基泥土一点点堆积起来的”,要让土壤足够肥沃,开放是必经之路,大会也官宣了不同层面的开源开放共建举措;

——深耕行业、用AI解决行业难题,并通过开放生态,加快实现AI普惠。

01

国产化技术,打造一朵智能体原生云

大会的重心是Agentic云全栈新品官宣,华为云围绕“软硬芯”协同,进行了全栈重构。不过,全球云大厂都在对云的技术栈和产品体系全面迭代,华为云的差异化在哪里?

周跃峰认为,“硅基黑土地”首先以硅为基础,华为云采用自主芯片构建算力集群,“目前是全球独一无二的”。周跃峰坦承,不能采用“万国牌”技术,给华为公司和华为云发展带来诸多困扰,“我们只能努力让自己的硅基持续进步,这条道路显然会更加曲折,但对国家工业安全和全球AI技术生态繁荣,具有意义”。比如,此次在Agentic Infra层推出AICS灵衢智算集群,基于华为自研灵衢网络和芯片,连接10万卡集群、单千卡吞吐超是全球最大超节点的两倍,实现了极致效率Token工厂。

其次,华为几十年在ICT的积累,尤其在数学、算法和软件工程上的积累,能打造出与众不同的算法和软件体系。比如推出的通智一体化调度CCE Volcano Next,调度颗粒度达到了1%;华为云训练和推理平台、企业智能体开发平台,能够为企业开发专属安全的智能体,“如果没有ICT的积累,是很难快速具备的”。

在安全层面,此次推出的AgentSphere,为Agent创造了安全运行底座。截止6月5日,华为云已连续稳定运行1037天,创造云产业纪录。

同时,华为云另一个差异化是坚持公有云与混合云、私有云并重。考虑到中国市场的现实——政府机构、央国企等国计民生企业,对数据安全和算力本地化部署高度关注,同时模型和算力技术迭代又极快,“我们更推荐本地数据+远端公有云模型服务调用模式。”周跃峰说,此次大会同步发布了混合云+数据+AI的解决方案,以及《企业如何构建面向智能体的混合云》白皮书,并推出了机密推理、机密训练等技术,目标是确保国计民生行业尽快用上AI,达到“善政、惠民、兴业”。

Coding是当下最具商业价值的赛道,华为云也宣布,其代码智能体——码道(CodeArts)用户数突破十万、正式商用。周跃峰强调,代码智能体是智能体平台核心基础能力,是“碳基世界与硅基世界对话的翻译器”,关乎tokens能否实现高价值,“我们愿意投入重兵和重要资源”。大会期间,华为云与中国高校共建AI编程教学活动,投入资源、专家和免费算力,让高校开发者和科研者来合力提升码道原子级能力。

02

智能体时代,成为最开放的一朵云

在Agentic时代,生态变得更为重要。全球云大厂在生态策略上都投入重兵——微软正将智能体部署到Windows、GitHub、Azure和新硬件中,构建全方位AI生态;AWS走“全模型货架”路线、谷歌云通过全栈技术+流量,争夺AI Native市场。同时,国内外云大厂都在追随Anthropic的逻辑,推出通用、专用智能体来触达用户。

从此次大会来看,华为云正在构建AI多层级开放生态体系。“我希望华为云能够在智能体时代,成为一朵最开放的云。”周跃峰说。

在大会上,他系统阐述了华为包括华为云,从底到上的开放策略:底层芯片上,鲲鹏通用算力与昇腾智算的CANN(异构计算架构)保持开放;操作系统欧拉已开源开放;云基础设施容器平台Volcano同样开源;模型即服务层ModelArts所采纳的工具链亦保持开源开放。

“我们也花了更多力气去对接所有愿意与我们合作的模型,共同打造模型服务。”周跃峰说。此次大会上华为云举办了“百模千态,云聚共赢”合作发布,连几乎不太参与外部活动的DeepSeek也亮相合作仪式;展区中,除了智谱、Minimax等国内大语言模型,以及大量文生视频大模型企业之外,连京东、美团、B站等也携带自家模型亮相。智能体需要不同模型适配不同场景,为提升智能体模型调用效率,华为云推出了全新模型路由,精准率超95%、调用成本平均下降20%以上。

在模型服务之上,华为云推出的企业级智能体开发平台AgentArts,也同时上线了开源版本openJiuwen。周跃峰透露,其内核代码与商业版几乎一致,还开放了更多MCP工具,以及面向各行业场景的数十万项skills。

在智能体产品层面,除了重点发布代码智能体码道之外,华为云并未推出眼花缭乱的智能体和Claw,而是将很大一部分时间放在“行业AI梦工厂”四大专区的发布和入驻上。行业AI梦工厂专区由周跃峰于2025年底提出,今年2月首发智慧医疗专区。每个专区汇聚开发者、垂类行业模型、智能体、数据集与解决方案,类似一个行业开放社区。这次大会上还新上线具身智能、智能制造、科学计算专区。

值得关注的是,此前各大企业向行业渗透的主流方式,是先与行业头部企业打造标杆案例,形成一定成果后,再向行业的中腰部企业渗透。但这种由企业按计划推进的落地节奏,已满足不了今天AI高速迭代、向行业扩散的需求。“开放社区形态,可以加速行业落地节奏。”展区中,一位华为云人士对数智前线分析。“只有这样,这片硅基黑土地才能长出更多参天大树。”周跃峰说。

03

AI普惠:让县城医院用上三甲诊断能力

2026年,海外三巨头AWS、微软Azure、谷歌云今年合计资本开支高达5650亿美元左右,一场围绕Agent基础设施的“军备竞赛”全面展开,同时各家也在加速向智能体应用场景推进。AI竞赛的喧嚣之下,一个根本的问题是,AI技术究竟为谁而存在?

“经常有人问我一个问题,今天AI是不是一个泡沫?”周跃峰说,“我觉得AI这个方向一定是正确的。”

他进一步举例分析,在智慧医疗专区中,一个重要版块是瑞金医院联合华为打造的RuiPath病理大模型。“由于历史原因,中国病理医生只有2万多人,很多中小医院无法判断某一个患者是不是得了癌症,而这是最基础的诊断。”周跃峰说,很多偏远医院由于经验不足,病理诊断存在效率和准确率不足的问题。如今,在智慧医疗专区上线后4个月中,已有20余家医院接入开展病理诊断,实际上共享了瑞金医院高水准医疗服务。这些医院中除了三甲医院,还包含区域医联体、尤其是县域医院,实现了AI普惠。在大会的一个视频中,远在贵州安龙县的县域医院,也用上了该病理诊断服务,患者不再需要千里迢迢赶到上海就医。

“我觉得这样的AI技术落地,才是我们应该去追求的。只有这样踏踏实实去做,才不至于让我们对AI的未来发展没有信心。”周跃峰透露,6月30号华为云还将上线医疗AI平台,将有更多医疗服务通过AI来普惠。

有行业人士预估,目前中国传统企业Token支出占IT总支出仍在5%以下、央国企在1%左右,AI在行业中的渗透依然很低。周跃峰认为,“AI面向行业去生根,需要更多耐心、更多投入,这个远比ToC问答和给个人带来情绪价值要难得多。”华为云构建的行业AI梦工厂,其实也是为了聚合行业从业者联合创新,加快AI落地速度。

比如,具身智能专区部署了全球首个全链路具身智能开发平台CloudRobo,覆盖数据准备、模型训练到强化学习全链条。“中国目前300多家具身智能创业企业规模普遍不大,这个平台让中小企业以极低算力费用接入,快速实现企业商业化目标,并在专区共享数据、模型与案例。”目前该专区已汇集30多家企业。

多年来,华为云在政企市场积累深厚,尤其在政府、金融、央国企等国计民生行业,连续多年保持份额第一,与其他大厂的“流量入口”路径截然不同。华为云这次也发挥了其在政企市场的影响力,大量行业企业和研究机构参与了四大专区的发布入驻。而今年下半年还将有更多行业专区上线,在更多行业加速AI落地,实现AI普惠。

“我们一定努力做好在中国TOP1的全栈国产化云。”周跃峰再次强调,华为云不关注token绝对数值,因为不能简单用多少Token来衡量产生的价值。而在当前国产化算力正在发展的背景下,华为云也不比拼收入总量,而是更关注每一个由国产化算力生产的token背后,生产效率是否得到实在的提升。“我们希望AI技术能为各行业,尤其是国计民生行业带来生产力跃升。”

注:文/牛慧,文章来源:数智前线(公众号ID:MzkwNDMyOTA1NA==),本文为作者独立观点,不代表亿邦动力立场。

文章来源:数智前线

广告
微信
朋友圈

这么好看,分享一下?

朋友圈 分享

APP内打开

+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