广告
加载中

“死磕”鲲鹏昇腾生态的极客们 要搞点大事情

胡珈萌 2026-05-27 08:45
胡珈萌 2026/05/27 08:45

邦小白快读

EN
全文速览

本文核心介绍了国产AI算力生态鲲鹏昇腾的发展变化,以及当前AI落地的最新进展,普通读者可从中获得这些核心干货:

1. 此前AI落地一直存在痛点,国内算力紧张优先流向大厂,国产算力生态不完善,中小开发者需要花大量时间填补生态缺口,现在这一情况已经彻底改变,中小开发者也能用上和大厂同等级的算力,鲲鹏昇腾生态已经变得好用易用,开放开源程度大幅提升。

2. 已经有大量成功落地案例,创业团队AIGCode基于昇腾把混合专家模型预训练的算力利用率做到65%,达到行业平均水平的两倍,DeepSeek依托昇腾生态把大模型Pro版降到原价的四分之一,清华大学、中科大的科研项目也基于鲲鹏昇腾落地了顶尖AI应用。

3. 普通开发者现在使用该生态迁移成本很低,开源社区已经积累了大量现成解决方案,不用从零开始造轮子,开发者可以把精力放在产品研发和创意落地,不用反复在适配优化上踩坑。

本文围绕国产AI算力生态鲲鹏昇腾的发展,给布局AI的品牌商带来了很多值得参考的干货,相关信息如下:

1. 当前消费和产业端的AI需求已经从概念走向实用,用户希望AI替代劳动提升效率,企业希望借助AI盈利,甚至非技术用户也希望通过AI快速落地个人创意,这给品牌结合AI开发新产品、开辟新赛道提供了明确的方向。

2. 国产算力生态已经跨过发展拐点,成本优势十分明显,品牌做AI研发不再需要依赖高价进口算力,依托成熟的国产生态可以大幅降低研发和算力成本,DeepSeek大模型降价就是典型案例,成本优势可以转化为产品的价格优势,提升市场竞争力。

3. 品牌做AI研发可以采用和生态深度共创的模式,遇到技术问题可以直接对接华为的技术专家团队快速解决,沟通效率远高于海外封闭生态,同时依托开源社区可以大幅降低适配研发的时间成本,加快产品落地速度。

本文披露了国产AI算力生态的最新发展情况,给AI领域相关卖家带来了明确的机会提示和可参考经验,具体干货如下:

1. 当前AI已经进入大规模落地阶段,市场机会非常多,AI能力已经在金融、能源、教育、交通、科研等多个领域落地,非技术人群的AI创意落地需求也在快速增长,中小卖家切入AI相关赛道的时机已经成熟。

2. 中小卖家的算力痛点已经得到解决,原来中小团队资金有限买不起高价进口算力卡,现在可以用上和大厂同等级的国产算力,还可以通过生态优化提升算力利用率,做到一张卡抵两张用,大幅降低了初创项目的算力投入门槛。

3. 中小卖家做AI项目可以学习深度共创模式,主动向生态方反馈需求,可以快速获得技术支持解决问题,同时目前开源社区已经积累了大量现成代码和解决方案,开发适配的门槛大幅降低,中小团队不用花费大量时间填补生态缺口,可以专注在产品和市场层面。

本文介绍的国产AI算力生态发展,给工厂推进AI数字化转型带来了新的商业机会和启示,具体干货如下:

1. 当前全栈自主AI的发展已经给工厂带来了新的商业机会,千行百业都在落地AI能力,工厂推进生产智能化、产品智能化已经有了成熟稳定的国产算力基础,不用再受限于海外算力的价格限制和供应限制,发展自主性更强。

2. 工厂推进数字化和AI转型获得了新的路径,原来工厂做AI转型需要花费大量人力物力做底层适配,现在鲲鹏昇腾生态开放开源,有大量现成的解决方案,代码迁移成本很低,工厂可以把主要精力放在自身生产场景的AI优化上,不用在底层适配层面消耗过多资源。

3. 工厂如果有个性化的AI生产定制需求,可以依托国产生态的共创模式,直接对接技术专家团队解决问题,相比海外封闭生态更灵活,沟通响应速度更快,更适合工厂结合自身生产场景做AI创新,打造差异化的智能化生产能力。

本文梳理了国产AI算力生态的最新发展动向,给AI相关服务商提供了行业趋势、客户痛点和解决方案方向的干货,内容如下:

1. 当前AI行业的明确发展趋势是国产算力生态加速替代海外生态,鲲鹏昇腾已经跨过生态拐点,进入了快速发展的阶段,越来越多的开发者、企业开始转向国产生态布局,服务商提前布局国产生态相关的适配、开发、咨询服务,将会获得大量的市场增长空间。

2. 做AI落地的客户核心痛点已经十分清晰,过去客户普遍面临算力不足成本高、生态不完善需要自己填坑适配、遇到底层问题求助无门的痛点,这些痛点在鲲鹏昇腾生态中已经得到了有效的解决,服务商可以围绕国产生态为客户解决这些痛点。

3. 服务商可以依托鲲鹏昇腾的开放生态降低自身服务成本,生态已经有大量开源代码和现成解决方案,服务商不需要从零开发适配方案,同时遇到客户的底层定制化需求,可以对接华为的技术专家团队协作解决,能有效提升自身的服务能力,承接更多类型的客户需求。

本文介绍了华为鲲鹏昇腾生态的建设经验,给做AI生态的平台商提供了运营管理、生态建设的参考干货,具体内容如下:

1. 开发者对AI算力平台的核心需求可以总结为两点:好用、易用,既要求硬件性能足够,也要求降低开发者的学习、迁移适配成本,平台做生态建设需要把复杂的底层优化留给自己,把简单便捷的使用体验留给开发者,这样才能吸引更多开发者入驻。

2. 生态建设的有效经验是开放开源+快速响应用户反馈,鲲鹏昇腾通过和开发者深度共创,快速响应开发者的问题补齐生态缺口,只用一年多时间就把CANN生态的覆盖率提升到百分之八九十,全面开源后5个月内开源项目从零增长到65个,代码量突破1200万行,下载量破千万,这种快速迭代模式值得平台借鉴。

3. 平台生态发展可以覆盖多类主体,吸引创业团队、科研机构、行业企业等不同类型的参与者入驻,形成繁荣共生的生态,同时需要紧跟技术动向,针对Agentic AI等新方向提前做好架构升级,提前布局构筑生态优势。

本文披露了国产AI算力生态发展的最新产业动向,给产业研究者提供了很多新的研究方向和素材,干货内容如下:

1. 当前国产AI算力产业出现了明确的新动向,鲲鹏昇腾生态已经跨过生态拐点,进入了高速发展阶段,不仅在硬件层面实现了技术创新,软件生态也通过开放共创快速完善,已经能够支撑大模型训练和全行业AI落地,大量开发者开始从进口生态向国产生态迁移,这是国内AI产业非常重要的新变化。

2. 国内走出了一种新的生态建设商业模式,也就是开发者和生态方深度共创的模式,开发者反馈实际问题,生态方快速迭代解决,这种模式比海外封闭生态更灵活,迭代速度更快,目前业内已经认为CANN生态的发展速率和可掌控性已经优于传统的CUDA生态,这种新模式非常值得深入研究。

3. 当前产业也出现了一些值得研究的新问题,早期国产生态的算力不足、生态不完善等痛点已经通过开放共创得到解决,但目前全栈自主创新的完整链路还需要进一步打通,如何持续保持生态的快速迭代、实现智能普惠,这些都是AI产业研究值得关注的新方向。

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 development and latest progress of China's domestic Kunpeng and Ascend AI computing ecosystem, sharing the following key insights for general readers:

1. The long-standing pain points in AI adoption have been fully resolved. Previously, limited computing resources were prioritized for large tech companies, and the immature domestic ecosystem forced small and medium-sized developers to spend massive effort filling gaps. Today, the Kunpeng-Ascend ecosystem has become accessible and user-friendly with greatly improved open-source openness, enabling small developers to access the same tier of computing power as large enterprises.

2. A growing number of successful real-world deployments prove the ecosystem's maturity. Startup AIGCode achieved 65% computing utilization for MoE model pre-training on Ascend, double the industry average. DeepSeek cut the price of its Pro-version large model to one quarter of its original cost by leveraging the Ascend ecosystem. Leading AI research projects from Tsinghua University and the University of Science and Technology of China have also been deployed on the Kunpeng-Ascend platform.

3. Migration to the ecosystem now comes at extremely low cost for ordinary developers. The open-source community has accumulated a large library of ready-to-use solutions, eliminating the need to build infrastructure from scratch. Developers can now focus on product development and creative execution, rather than wasting time debugging compatibility and optimization work.

This article outlines the development of China's domestic Kunpeng and Ascend AI computing ecosystem, with key takeaways for brands looking to integrate AI into their business:

1. AI demand among consumers and industrial clients has shifted from hype to practical application: end users want AI to replace labor and boost efficiency, businesses aim to generate profit via AI, and even non-technical users want to bring their creative AI ideas to life quickly. This trend creates clear opportunities for brands to develop new AI-powered products and open up new market segments.

2. The domestic AI computing ecosystem has passed a key inflection point, with significant cost advantages. Brands no longer need to rely on expensive imported computing power; the mature domestic ecosystem can drastically cut R&D and computing costs. DeepSeek's large model price cut is a prime example of this: lower infrastructure costs can be translated into more competitive product pricing to gain market share.

3. Brands can adopt a deep co-creation model with the ecosystem. Any technical issue can be quickly resolved by direct access to Huawei's expert team, delivering far higher communication efficiency than closed overseas ecosystems. Plus, the open-source community drastically cuts the time required for compatibility and R&D, enabling faster product launch.

This article shares the latest updates on China's domestic Kunpeng and Ascend AI computing ecosystem, with clear opportunity alerts and actionable insights for AI-focused sellers:

1. AI has entered the phase of large-scale commercial deployment, creating abundant market opportunities. AI capabilities have already been adopted across finance, energy, education, transportation, scientific research and other sectors, while demand from non-technical users to bring AI ideas to life is growing rapidly. This is the right time for small and medium-sized sellers to enter AI-related segments.

2. The core pain point of limited access to affordable computing for small sellers has been resolved. Previously, small teams with limited capital could not afford expensive imported computing cards. Today, they can access the same tier of domestic computing power as large companies, and ecosystem optimizations have improved utilization enough that one card can deliver the work of two, drastically lowering the barrier to entry for early-stage AI projects.

3. Small sellers can leverage the deep co-creation model for their AI projects: proactively sharing your needs with the ecosystem will get you fast access to technical support to solve problems. Combined with the large library of open-source code and ready-to-use solutions already available in the community, this drastically lowers the barrier to development and compatibility, freeing small teams to focus on product development and go-to-market instead of filling ecosystem gaps on their own.

This article breaks down the latest development of China's domestic Kunpeng and Ascend AI computing ecosystem, with new opportunities and insights for factories pursuing AI-powered digital transformation:

1. The advancement of full-stack domestic AI has created new opportunities for manufacturing. As AI deployment accelerates across all industries, factories now have a mature, stable domestic computing foundation to build intelligent production and smart products. No longer constrained by the price and supply volatility of overseas computing, factories have far more autonomy in their transformation journey.

2. The Kunpeng-Ascend ecosystem provides a new, more efficient path for digital and AI transformation. Previously, factories had to devote massive human and capital resources to underlying infrastructure compatibility work. Today, the open-source Kunpeng-Ascend ecosystem offers abundant pre-built solutions, with very low code migration costs. This allows factories to focus their resources on AI optimization tailored to their specific production scenarios, rather than wasting resources on underlying compatibility work.

3. For factories with custom AI requirements for production, the domestic ecosystem's co-creation model enables direct access to technical experts to resolve issues. Compared with closed overseas ecosystems, this model is more flexible with faster response times, making it ideal for factories to develop AI innovation aligned with their unique production needs and build differentiated intelligent production capabilities.

This article summarizes the latest development trends of China's domestic Kunpeng and Ascend AI computing ecosystem, with key insights on industry trends, client pain points and solution directions for AI-related service providers:

1. The clear core trend in China's AI industry is the accelerating replacement of overseas ecosystems with domestic alternatives. The Kunpeng-Ascend ecosystem has passed its inflection point and entered a phase of rapid growth, with a growing number of developers and enterprises shifting their AI projects to the domestic ecosystem. Service providers that proactively lay out compatibility, development and consulting services for the domestic ecosystem will unlock significant room for market growth.

2. Core client pain points for AI deployment are now well-defined: in the past, clients commonly struggled with high costs from insufficient computing, the need to fix compatibility gaps in immature ecosystems, and no support for underlying technical problems. All of these pain points have been effectively addressed in the Kunpeng-Ascend ecosystem, and service providers can build their offerings around solving these pain points for clients via the domestic ecosystem.

3. Service providers can reduce their own service costs by leveraging Kunpeng-Ascend's open ecosystem, which already offers abundant open-source code and pre-built solutions that eliminate the need to build compatibility solutions from scratch. For custom underlying requirements from clients, service providers can collaborate with Huawei's expert technical team to resolve issues, effectively improving their own service capabilities to take on a wider range of client demands.

This article shares the ecosystem-building experience of Huawei's Kunpeng-Ascend ecosystem, with actionable insights on operations and ecosystem development for AI platform operators:

1. Developers' core demands for an AI computing platform can be summarized in two words: capable and accessible. Developers need both strong hardware performance and low costs for learning, migration and compatibility work. To attract more developers, platforms should handle complex underlying optimization work themselves, to deliver a simple, convenient user experience for developers.

2. A proven effective approach to ecosystem building is open-source openness paired with fast response to developer feedback. Through deep co-creation with developers, Kunpeng-Ascend quickly resolved developer pain points to fill ecosystem gaps. In just over a year, it increased CANN ecosystem coverage to 80-90%. After full open-sourcing, the number of open-source projects grew from zero to 65 in 5 months, with total code volume exceeding 12 million lines and total downloads surpassing 10 million. This rapid iteration model is well worth learning for platform operators.

3. A thriving platform ecosystem should cover multiple types of participants, attracting startups, research institutions, industry enterprises and other players to build a mutually prosperous ecosystem. Platforms should also keep up with emerging technology trends, complete architectural upgrades in advance for new directions such as Agentic AI, and build ecosystem advantages through early布局.

This article shares the latest industry developments in China's domestic Kunpeng and Ascend AI computing ecosystem, providing new research directions and materials for industry researchers:

1. A clear new shift is underway in China's domestic AI computing industry: the Kunpeng-Ascend ecosystem has passed its inflection point and entered a phase of rapid growth. It has not only delivered hardware technical innovation, but also rapidly improved its software ecosystem through open co-creation, and is now fully capable of supporting large model training and industry-wide AI deployment. A growing number of developers are migrating from imported ecosystems to domestic alternatives, making this a pivotal new development for China's AI industry.

2. China has developed a new business model for ecosystem building: deep co-creation between developers and ecosystem providers. Developers report real-world problems, and ecosystem providers iterate quickly to resolve them. This model is more flexible and delivers faster iteration than closed overseas ecosystems. Industry observers already consider the development speed and controllability of the CANN ecosystem to outperform the traditional CUDA ecosystem, making this new model a particularly worthy topic for in-depth research.

3. There are also new research-worthy questions emerging in the industry. While early pain points such as insufficient computing power and incomplete ecosystems have been resolved through open co-creation, the full end-to-end link for full-stack independent innovation still needs further improvement. Key emerging research directions for the AI industry include how to sustain fast ecosystem iteration and deliver inclusive AI for all.

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.

曾经的盐碱地,已经变成了肥沃的黑土地。

AI正在落地。

这几年,人们都看到了AI很厉害,也不满足于它只是“看上去很厉害”。

“打工人”希望AI替自己干活,企业主想叫AI帮自己赚钱。就连习惯了在AI中“放飞自我”的极客们,也觉得该着陆了。

“这几年,AI就像是我的‘飞天扫帚’,刚开始骑着它冲冲冲很爽,但现在回头看又没什么拿得出手的成果,感觉有点空虚。我也想做出‘龙虾OpenClaw’这种人们日常也想用的东西。”一位开发者在为自己的新项目找投资时表示。

类似的渴求与尝试,正在快速涌现。

但落地的路,并不轻松。

算力是AI的土壤,而国内的算力本就紧张。

不少2024-2025年创业的极客开发者,都曾吐槽过有限的算力优先流向大厂,生态不够完善、开放,团队要花费大量时间写算子、填充生态缺口等问题。

但情况已经发生改变。

开发者们渐渐发现,自己能和大厂用上同等级的算力了,以鲲鹏昇腾生态为代表的中国算力生态变得更加好用、易用,也加速走向开放、开源。

曾经的盐碱地,成了肥沃的黑土。在这里落地AI看起来已经不是难事,而那些“死磕”过来的极客们,要动手搞出点真正的大事情了。

与华为“极限拉扯”的极客“小团队”

AIGCode(蔻町智能)团队最初找到的,只是一片“荒漠”。

这家公司想把vibe coding(氛围编程)这种前沿的AI编程范式做成实用产品,让不懂代码的人也能仅凭文字描述,就生成自己的网站和应用。

2024年1月公司成立,核心团队是来自微软、华为以及国内互联网大厂的一群极客。

“我们这二、三十人的小团队中,很多人在传统大厂的评价体系中并不以综合能力见长,但技术能力都非常强,80、90分以上那种。”公司联合创始人兼CTO陈秋武介绍称。

这群人一上来就想“搞件大事情”,打破Anthropic模型的能力垄断,用自研基础模型做出“一句话生成网站和应用”的vibe coding产品。

但落地挑战巨大,尤其是在算力上。

彼时的国内AI圈,缺卡焦虑正浓。

用国产卡训练大模型,还缺少足够的成功经验,更关键的是,当时的国产算力生态很难与被英伟达视作“最强护城河”的CUDA生态等量齐观。

按业内人士所言,芯片硬件代表着算力的理论能力上限,软件生态才决定着算力的实际能力上限。换言之,没有生态的算力,只是徒有其表的空中楼阁。

“说实话,一开始用国产是因为‘穷’,我们初创小公司,没那么多钱买N卡。”陈秋武回忆称,AIGCode当时选的是华为的昇腾芯片。

但团队很快发现,这片土地比想象中更“荒”。

华为对标英伟达CUDA的是CANN生态,用陈秋武的话说,2024年初的感觉有点像“荒漠”,很多东西都没有。团队做7B模型的预训练,20台机器跑起来,缺算子、缺工具链,常常“一边踩坑,一边填坑”。

极客们向来“爱较劲”,这支小团队也由此开启了跟华为的“极限拉扯”,常常提出各类问题和需求,让华为拿出解决方案。

陈秋武称,自己曾很严厉地向华为团队指出CANN生态的差距,甚至向负责的高层表示,必须要直面问题,不要讲漂亮话,否则永远都赶不上CUDA。

“质变”比想象中来得更快。

大半年过去后,CANN生态已经“越用越顺手”。到2025年,陈秋武称看到CANN的覆盖率到百分之八九十,连他自己“都吓了一跳”。

回顾那段“走出荒漠”的经历,陈秋武觉得CANN生态之所以进步如此之快,一方面得益于昇腾硬件很扎实,擅长创新,另一方面,沟通和人的因素也很重要。

他举例称,团队开发涉及到汇编层面的底层优化,如果是在英伟达的CUDA生态中,类似的深度定制根本找不到人,查资料也查不到。但在跟华为团队的沟通中,能很方便地链接到人,还有专家级人员来跟团队密切合作解决问题。

其实,所谓“拉扯”,就是有来有回。对于开发者与算力产业来说,出现问题能有反馈,是生态建立的基础;收到反馈能解决问题,则是生态发展的要义。

尤其在AI时代,个性化的真实数据、反馈正是难得的资产和养料,也是AIGCode看到“荒漠”变为“黑土地”的关键。

如今,这支极客“小团队”已经成为昇腾CANN生态的深度共创者,交付了多个项目。他们还把MoE(混合专家模型)预训练的MFU(算力利用率)做到了65%,是行业平均水平的两倍,让昇腾上“一张卡可以当两张卡用”。

陈秋武认为,从可掌控性和发展速率来看,CANN已有胜过CUDA之处。业内不少声音也认为,鲲鹏昇腾已跨过生态拐点,正在加速发展。陈秋武还表示,自己现在仍会参加很多次华为的技术闭门会,也经常提出各种问题和建议,而华为的反馈也一如既往地快。

看起来,“拉扯”还远未结束。

AI落地何所依?

“死磕”华为生态的,不止AIGCode。

清华大学团队联合多方合作单位,基于鲲鹏服务器构建“地球系统模拟器”。这些科学家极客将AI引入传统地球系统模拟中部分依赖经验公式估算的环节,把全球模拟分辨率提升至公里级;只需计算一天,就能完成约一年模式时间的天气-气候演化模拟。

中科大团队基于鲲鹏进行算法创新,优化求解器,“榨干”鲲鹏张量运算部件和高带宽内存的优势,相比传统方法提升40倍左右。

更受关注的,是DeepSeek今年4月末发布的V4,这让中国的算力生态和中国大模型的深度耦合站到全球聚光灯下。DeepSeek当时预告,“预计下半年昇腾950超节点批量上市后,V4 Pro的价格会大幅下调”。就在5月22日晚,DeepSeek宣布6月起Pro版正式降至原价的四分之一,不少声音猜测中国算力在其中发挥了关键作用。

金融、能源、教育、交通……越来越多的行业在鲲鹏昇腾生态上落地AI能力,有的管证券交易,有的管水力发电,有的管反诈拦截。

极客开发者们,似乎找到了AI着陆的所依之地。

有“戒断英伟达,用上国产卡”的开发者,视这片新土地为“意外惊喜”;也有率先迁移到基于中国的生态的团队,将之作为“秘密武器”。

有业内人士称,在DeepSeek与昇腾进行深度适配后,全球很多巨头、机构都在“深扒其中的门道”。

其中,自然有硬核的技术突破。

比如昇腾950芯片在HiF8新型数据格式、底层架构等方面的创新,鲲鹏CPU面向Agentic AI时代在多核数、高性能、低时延等方面的针对性设计,尤其是让芯片能“集群作战”的超节点,为中国的算力提供了更多可能。

但另一方面,常被忽视的生态构建,同样关键。

生态的“门道”,听起来千头万绪,其实核心无非两点:好用,易用。

想要“好用”,除了做好硬件、软件、软硬结合,持续提升性能,更要考虑开发者的需求、偏好和实际感受。

与华为有深入合作的中科大先进计算机系统结构实验室副研究员陈俊仕评价,鲲鹏在高性能计算领域没有选择CPU+GPU异构路线,而是“另辟蹊径”走出一条独特的鲲鹏架构路线,本质上就是“把复杂留给自己,把简单留给开发者”。

而“易用”,则是降低学习成本和迁移门槛,提高兼容性和便捷度。

华为强调“硬件开放,软件开源”,CANN生态2025年底全面开源开放后,社区开源项目5个月从零增至65个,开源代码从827万行激增到1244万行,平均每天新增3万行,下载量突破千万。这意味着开发者遇到的大多数问题都能找到现成方案,不必从“造轮子”开始。清华大学团队的王一鸣提到,鲲鹏昇腾生态的代码迁移成本低,社区反馈丰富,“能让科研人员的精力回到科研本身,不用在适配和优化上反复踩坑”。

好用、易用,就会吸引更多人来用。而用的人越多,AI就会越好用。

技术革新、大国争锋、商业竞赛都会带来很多不确定性,对于致力于在变动的时代落地自身热情、创意与价值的开发者来说,一个能汇聚众人、持续滋养创造力的生态,就是一种确定性。

让智能抵达更多人

生态的意义,还在于成就超乎个体能力和预料的事。

DeepSeek的“高性价比”让更多不那么富裕的团队也能接入AI;四处涌现的Agent,将数字员工、AI助理带进现实。AIGCode的vibe coding产品,让非技术人群能通过AI快速落地自己的创意;清华大学团队的“地球系统模拟器”,则致力于为普通人预警气候灾害。

鲲鹏、昇腾的生态总会提到“使能”这个词,让合作伙伴具备能力。而在AI时代,快速跃进的智能已经引发了“能力垄断”“能力鸿沟”等焦虑。只有解决好智能的普惠,才能实现技术的平权。

那些把“AI价格打下来”、让普通人也能coding出应用、为社区居民预警灾害的极客们,正把自己的能量和生态的“使能”聚在一起,传递给更多人。AI落地,智能扩散,最终,繁荣才能属于所有人。

寻找到土壤的极客们,对未来的底气更足了,他们都想要“搞点大事情”。

DeepSeek专注于实现AGI;清华大学团队则致力于攻克气候建模领域的国际前沿难题;AIGCode则想把AI Coding做到L5,在搞定“一句话前端、后端、数据库全部生成”后,把更高复杂度的软件一句话生成自动化完成。陈秋武还希望继续与华为“拉扯”,期待打通国内算力+自研模型完整链路,让千行百业都能用上全栈自主创新的大模型。

华为自身也在鲲鹏昇腾开发者大会2026上披露,将在Agentic AI时代继续深耕超节点架构,打造好用易用的算力底座,共建繁荣共生的智能体时代算力新生态。

相信在这片土地上,还会有更多的大事发生。

注:文/胡珈萌,文章来源:钛媒体(公众号ID:taimeiti),本文为作者独立观点,不代表亿邦动力立场。

文章来源:钛媒体

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