[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fQscL6bhGpmaNxJyoaS7mBwSpyPeDQ-eoqpJ14V5eB6k":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},1417,"AI硬件崛起：从算力到终端的系统性跨越","\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">从智能录音设备Plaud、系列AI玩具，再到正如火如荼拓展海外市场的Robotaxi，中国AI新创公司正强势占领全球市场心智。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这背后是\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">国内市场高效的智造供应链体系、AI快速迭代和敏锐产品嗅觉所累积的竞争底气。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“中国市场产业链的效率远远高于其他国家。”一名回国创业的3D视觉领域公司创始人对21世纪经济报道记者感慨。这不仅推动国内创新生态快速持续迭代，更能借力由此形成的上下游合力，更快突破海外市场落地。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">不止于此，在AI硬件终端的关键上游，国产算力产业链也正加速协同。从摩尔线程88天IPO过会，到以海光信息、运营商为代表推进的AI算力产业链加速聚合，以及华为昇腾超节点持续引领算力集群能力拓展——国内AI基础设施生态链正形成合力，以应对本轮AI大模型引发的第四次工业变革。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">从芯片到软件栈，从智算集群到智能终端，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">中国AI硬件正在完成一场系统性跨越。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">AI新硬件群雄并起\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">2024年，仅靠攻占海外市场心智就已获得1亿美元年化收入的智能录音产品品牌Plaud.AI，近日终于开始登陆国内市场发售。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这家总部在深圳的AI硬件初创公司，2021年才刚刚成立（企查查显示），截至今年7月，已经实现在全球突破百万台规模的销量记录。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">其是本轮AI新硬件创新浪潮的一种缩影，创立之初就面向全球市场。乘风破浪的远不止于该品类，AI玩具、AI机器人、XR、智能制造等细分赛道都是重要突破方向。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">今年8月，AI玩具新锐跃然创新（Haivivi）官宣完成2亿元A轮系列融资，由中金资本旗下基金、红杉中国、华山资本、愉悦资本领投，招银国际等共同参投。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这同样是一家2021年于深圳成立的公司。据悉，其推出的首款AI玩具Bubble Pal截至8月销量已突破20万台，已与国内外多个知名IP达成合作授权。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">XR眼镜是另一个本轮创新热点。一名智能眼镜公司创始人受访时也指出，其此前有手机和相关配件打磨经验，在梳理供应链时发现，配件外壳与XR眼镜镜片是相同供应商；而在手机行业的多年打磨让他发现，将能力和经验可以顺势迁移到打造XR眼镜。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">更遑论已经在海外市场大放异彩的无人扫地机、电动两轮车、自动驾驶等产品。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">云鲸智能高级副总裁Hower此前接受21世纪经济报道等媒体采访时就分析道：“现在的机器人出海不再是当年简单制造、卷价格的逻辑，而是我们带着更好的技术、更好的方案走出去，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这是‘中国制造’和‘中国创造’的区别。\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这也意味着，除了对产品本身的打造，在新技术浪潮下，中国科技公司也有望成为新硬件定义标准的关键角色。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">比如，国产方案供应商文远知行和小马智行，是一边用技术试水、一边探索业务外延。如今两家公司都在海外多个市场进行自动驾驶研发、测试和运营。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">沙利文大中华区分析师颜逸扬对21世纪经济报道记者分析，技术能力是中国智能驾驶方案提供商成功推进海外市场拓展的基础。“更进一步来说，在不同国家、法规、交通环境下实现Robotaxi的运营，让中国智能驾驶方案提供商有机会参与甚至主导部分区域性乃至全球性的自动驾驶技术标准、安全标准和运营规范的制定。”他指出，同时，通过与全球或区域性交通巨头合作，是中国智能驾驶方案提供商探索“技术赋能+本地化运营”商业模式的实践，为构建可复制、可扩展的全球化商业模式积累经验。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">从录音笔到机器人，从AI玩具到XR眼镜，中国AI新硬件的爆发，不仅是一场产品创新竞赛，更是一场产业范式的重塑。\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">过去，中国制造以“性价比”著称，如今则以“智能力”突围。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">深圳、上海等为代表的创新区域，正成为全球AI硬件生态的“实验室”和“策源地”。随着技术积淀、供应链协同与全球化布局的加速，AI硬件的新出海时代已经到来。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">AI和制造的深度耦合\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">支撑本轮AI新硬件创新的底层逻辑，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">是AI大模型浪潮必将催生众多以“原生AI”为出发点的AI硬件产品。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">而中国市场的特殊优势在于，此前借力手机为核心硬件的产业链思维和能力，让中国硬件产品无论是从研发到迭代的速度还是成品、渠道综合竞争力都不容小觑。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">中国市场产业链的效率远远高于其他国家。\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">”一名回国创业的3D视觉领域公司创始人对21世纪经济报道记者感慨，在他所在领域，早年间的学术贡献多集中在海外市场，但随着包括他在内的视觉行业人才学成归来，借力国内高效的供应链协同能力，在诸多视觉领域，中国的整体实力已经对海外形成超越。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“供应链的高效，让我们能快速迭代产品。可能上一代产品还不够好，但市场会逼着你以‘月’为单位持续进化。”该创始人续称，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这种快速迭代能力，使中国公司能在短时间内实现技术突破和成本优化。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">由此取得的沉淀，则支持国内AI智能硬件进一步走向海外，开疆拓土。“我们跟国内其他供应商一起，可以更快寻找到海外客户的业务结合点，并把需求转化为可以落地的方案。这种上下游抱团出海，会让在海外取得成果的速度快很多。”该人士进一步分析道。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">换言之，AI硬件的崛起，是AI与制造产业链深度耦合的产物，曾经的“世界工厂”正在加速进化为“智能创新的源头”。无论是智能终端、机器人，还是工业自动化设备，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">背后都体现出中国在软硬件融合、快速验证、成本控制、全球化供给等方面的独特竞争力。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这显示出，如今的中国AI终端正形成一个强大创新矩阵。它们共享同一种底气：源于国内激烈市场竞争所锤炼出的快速迭代能力、基于庞大用户场景打磨出的产品定义能力，以及一个成熟、高效且愿意拥抱创新的本土供应链体系。这背后绝不仅是单个产品的胜利，而是一个系统性能力的全球绽放。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">当AI的智能触角深入每一个场景，中国制造产业链的角色也将被重新定义。\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">不只是生产，更是定义、引领和输出智能化标准。\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">随着AI与智能制造的融合，AI终端正在成为全球科技产业角逐的新焦点。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">底座跃迁\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">值得注意的是，在AI终端的重要底层能力方面，国内算力产业链持续突围，为下游提供核心动能。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">今年，多家国产AI芯片头部公司正迎来上市辅导潮，或处于待审核、待过会、待挂牌状态。几家公司在2024年都实现营业收入大幅增长，虽然还处在亏损阶段，但根据回复问询函的内容显示，这些头部公司均已获得可观的智算业务订单。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">资本市场的正反馈，将为这些硬科技企业持续投入研发、扩大生态提供宝贵“弹药”。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这不是个例。2024年末开始，亏损多年的“寒王”（寒武纪）开始单季扭亏，到2025年上半年大幅盈利。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">产业龙头的业绩兑现则印证了国产AI算力从“可用”到“好用”的商业化闭环正在显现雏形。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">寒武纪公司高管在2025年上半年业绩说明会上透露，公司旗下产品持续在运营商、金融、互联网等多个重点行业规模化部署并通过了客户严苛环境的验证。同时，为满足大模型市场对算力的旺盛需求，公司基于对未来收入的乐观预期进行了战略性备货，存货规模相应增长。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这表明，国产AI芯片已不再是实验室里的样品，而是能够真正进入核心业务场景、产生持续现金流的商品。寒武纪的“翻身”，是国产AI基础设施生态走向成熟的一个缩影。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">同时，国内AI芯片公司分别采取GPU和ASIC等差异化芯片架构路线，快速迭代发展，并积极参与生态链协同。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">近期DeepSeek在官方文件中提到，DeepSeek-V3.1使用了UE8M0 FP8 Scale的参数精度，是针对即将发布的下一代国产芯片设计。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">对此，一名芯片公司高管对21世纪经济报道记者指出，“我们很高兴看到，UE8M0 FP8这样的路线会被DeepSeek投入到具体应用中，同时也乐于见到，有DeepSeek这类企业，告诉大家哪些是趋势。这可以让未来我们国产芯片厂商，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">有更多机会与模型厂商发挥协同作用，共同让计算效率提升到新的水平。\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">腾讯公司高管也提到，旗下异构计算平台已适配主流国产芯片，并以软硬件协同的全栈优化，整合不同类型芯片，对外提供高性价比的AI算力。而包括阿里、百度在内的互联网巨头，也在通过自研ASIC定制芯片的方式，积极推进内生基础设施能力的夯实。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这一趋势支撑下，国产AI芯片经历多年沉淀，正迎来阶段性关键节点。面对高端算力供给不足、国产加速卡性能短期内存在差距、软硬件生态不成熟等多重挑战，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">中国AI产业选择了一条难而正确的道路：开放架构+生态协同。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">生态合力\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这不再是个别企业的孤军奋战，而是一场由技术、产业、政策共同驱动的系统性突围战，这让我国在算力领域得以由“点”连成“面”。其合力正以前所未有的速度重构着中国AI基础设施的基座。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“通常来说，如果购买了100张GPU卡，每张卡的算力是100%来看，实际上在部署过程中，最开始能实现的算力效率可能只有30%-40%。问题就在于数据互联层面缺乏高效协同，限制了算力潜力的释放。”一名GPU行业从业者对21世纪经济报道记者打了个比方，让100个人在一个空间内沟通，对比让同样多人数分散在不同空间内通过打电话方式沟通，显然前者效率更高，这也是scale up逻辑下构建超节点的意义所在。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">尤其在当下AI推理需求正持续累积过程中，借力于此，可以让算力集群能力更好释放。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">超节点的构建背后，涉及从底层芯片，到中间软件栈和互联协议，再到智算集群等不同层面计算环节的能力协同。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">9月，中科曙光联合包括上述芯片企业在内的20多家产业链伙伴，发布国内首个AI计算开放架构，并推出支持百万卡扩展的AI超集群系统。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">中科曙光总裁助理、智能计算产品事业部总经理杜夏威表示，克服国内AI算力多维挑战，需汇聚芯片、计算系统、大模型等产业链多方力量，构建开放AI计算架构并拉通产业间跨层协作，以打破“技术墙”与“生态墙”。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">相比封闭系统，曙光AI超集群系统除了通过紧耦合设计，像一台计算机般高效工作外，还支持多品牌AI加速卡以及兼容CUDA等主流软件生态，大幅降低硬件成本和软件开发适配成本。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“表面上看，这是一场关于工具链的开放；但\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">在更深层的逻辑中，它折射出中国AI算力生态的走向：从追赶、模仿，到开始尝试制定自己的规则。\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">”杜夏威续称。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">作为中科曙光的重要伙伴，国产CPU\u002FDCU龙头海光信息也做出了开放系统互联总线协议（HSL）的战略决策。HSL的核心内容包括开放完整的总线协议、提供IP参考设计、开放指令集等。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">中国工程院院士郑纬民对此总结：CPU、GPU、NPU、加速卡存储与网络模块的高效协同，是释放异构系统整体性能的关键。“要实现这一点，必须依赖统一的互联总线协议。它既是算力单元之间沟通的语言，也是实现系统互通、算力融合的重要基础。”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">华为则端出了成熟的落地方案。今年上半年，华为昇腾384超节点首次亮相，被业内直接对标英伟达GB200 NVL72。9月18日，华为再度更新超节点产品Atlas 950 SuperPoD和Atlas 960 SuperPoD超节点，分别支持8192及15488张昇腾卡。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">华为还透露，明年要推出的Atlas 950 SuperPoD，其数量规模是英伟达GB200 NVL144的56.8倍，算力是6.7倍，内存容量是15倍，互联（内部）带宽更达62倍。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">由此，一条AI发展路径已清晰勾勒出来：资本助力技术单点突破，而开放生态则将这些突破凝聚成无可替代的系统性能力。这股自上而下与自下而上相结合的巨大合力，正在打破“技术墙”与“生态墙”，为中国在AI时代的竞争构筑起坚实的基座。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">从底层算力到终端落地，中国AI硬件崛起不仅是一场技术竞赛，\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">更是一场关于产业链协同、生态构建和全球市场重塑的系统性工程，也让中国有望从“跟跑者”向“领跑者”蜕变，为人工智能的普惠化提供创新方案。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(187, 187, 187);\">【新闻来源】21世纪经济报道记者 骆轶琪 \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845511183702469808&wfr=spider&for=pc\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845511183702469808&wfr=spider&for=pc\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F7fec82514c674d65880360e88cce9591\u002FAI领域.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fthumbs\u002F7fec82514c674d65880360e88cce9591\u002FAI领域.jpg",0,1,80,"2025-10-13 22:02",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3Acbbc6022-f78f-4951-8a71-290f4d8b4591%3A0.wav?Expires=1761210603&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=xKIdvq%2Fxmf5AXc7TfH97qpz59OE%3D",24240534,"cbbc6022-f78f-4951-8a71-290f4d8b4591","2025-10-13 21:57","AI Hardware Rises: A Systematic Leap from Computing Power to Terminals","\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">From the smart recording device Plaud, AI toy series, to the Robotaxi that is currently rapidly expanding its overseas market, Chinese AI startups are strongly occupying the global market's mindshare.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This is backed by the efficient manufacturing supply chain system of the domestic market, the rapid iteration of AI, and the keen product sense that have accumulated competitive confidence.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“The efficiency of China's industrial chain is far higher than that of other countries,” a founder of a 3D vision company who returned to China for entrepreneurship said to the reporter of 21st Century Economic Report. This not only drives the rapid and continuous iteration of the domestic innovation ecosystem but also allows leveraging the resulting upstream and downstream synergy to break through the overseas market more quickly.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Not only that, in the key upstream of AI hardware terminals, the domestic computing power industry chain is also accelerating collaboration. From Moer Thread's IPO being approved in 88 days, to the acceleration of the AI computing power industry chain driven by Higan Information and operators, as well as Huawei's Ascend super node continuously leading the expansion of computing power cluster capabilities - the domestic AI infrastructure ecosystem is forming a joint force to cope with the fourth industrial revolution triggered by this round of large AI models.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">From chips to software stacks, from intelligent computing clusters to smart terminals,\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">China's AI hardware is completing a systematic leap.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">The Rise of the New Generation of AI Hardware\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In 2024, the smart recording product brand Plaud.AI, which has achieved an annualized income of $100 million by capturing the overseas market's mindshare, has recently started selling in the domestic market.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This AI hardware startup company, headquartered in Shenzhen, was just established in 2021 (as shown by Qichacha), and by July of this year, it had already achieved sales records exceeding one million units globally.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">It is a microcosm of this round of AI new hardware innovation wave, and it was aimed at the global market from the beginning. It is not only this category that is riding the wave, but also AI toys, AI robots, XR, and intelligent manufacturing are important breakthrough directions.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In August this year, the AI toy newcomer Yueru Innovation (Haivivi) announced the completion of a Series A financing of 200 million yuan, led by funds under CICC Capital, Sequoia China, Huashan Capital, and Yulian Capital, with招商银行 International and others participating jointly.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This is also a company established in Shenzhen in 2021. According to the information, its first AI toy Bubble Pal has sold more than 200,000 units by August, and has reached cooperation agreements with multiple well-known IP brands both domestically and internationally.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">XR glasses are another hot spot in this round of innovation. When interviewed, a smart glasses company founder also pointed out that he had experience in mobile phones and related accessories. When sorting out the supply chain, he found that the casing of the accessories and the lens of the XR glasses were supplied by the same supplier; and after years of experience in the mobile phone industry, he found that he could transfer his capabilities and experience to build XR glasses.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Let alone the products that have already made a splash in the overseas market, such as autonomous sweeping robots, electric two-wheelers, and autonomous driving.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Hower, Senior Vice President of Yunjing Intelligence, previously told media outlets including 21st Century Economic Report, “Now, the export of robots is no longer based on the logic of simple manufacturing and price competition. Instead, we are going abroad with better technology and better solutions,\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">this is the difference between 'Made in China' and 'Created in China.'\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This means that apart from the development of the product itself, in the new technological wave, Chinese technology companies are also expected to become key players in defining new hardware standards.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">For example, domestic solution providers WeRide and Pony.ai are both using technology to test and exploring business extensions. Now both companies are conducting autonomous driving R&D, testing, and operations in multiple overseas markets.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Yan Yiyang, analyst of Frost & Sullivan Greater China, analyzed for 21st Century Economic Report, \"The technical capability is the foundation for Chinese intelligent driving solution providers to successfully expand their overseas markets.\" He further pointed out, \"In different countries, regulations, and traffic environments, achieving the operation of Robotaxi allows Chinese intelligent driving solution providers to participate in or even lead the formulation of some regional or even global autonomous driving technical standards, safety standards, and operational specifications.\" He added that through cooperation with global or regional transportation giants, Chinese intelligent driving solution providers are practicing the \"technology empowerment + localized operation\" business model, accumulating experience to build replicable and scalable global business models.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">From voice recorders to robots, from AI toys to XR glasses, the boom of Chinese AI new hardware is not only a competition of product innovation, but also a reshaping of the industrial paradigm.\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In the past, \"cost-effectiveness\" was known about Chinese manufacturing, now it is breaking through with \"intelligent capacity\".\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Innovative regions represented by Shenzhen and Shanghai are becoming the \"laboratory\" and \"source\" of the global AI hardware ecosystem. With the acceleration of technological accumulation, supply chain coordination, and globalization layout, the new era of AI hardware exports has arrived.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Deep Integration of AI and Manufacturing\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The underlying logic supporting this round of AI new hardware innovation is\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">that the AI large model wave will inevitably give rise to numerous AI hardware products starting from \"native AI\".\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">China's unique advantages lie in the fact that, by leveraging the industrial chain thinking and capabilities centered around mobile phones, Chinese hardware products, whether from R&D to iteration speed or finished product and channel comprehensive competitiveness, are formidable.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The efficiency of China's industrial chain is far higher than that of other countries.\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75>” A founder of a 3D vision company who returned to China for entrepreneurship said to the reporter of 21st Century Economic Report. In his field, early academic contributions were mainly concentrated in overseas markets, but with the return of visual industry talents including himself, leveraging the efficient supply chain collaboration capability in China, China's overall strength in many visual fields has already surpassed overseas.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">“The efficiency of the supply chain allows us to rapidly iterate products. Perhaps the previous generation of products is not good enough, but the market will force you to continuously evolve on a monthly basis.” The founder continued, \u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">this rapid iteration capability enables Chinese companies to achieve technological breakthroughs and cost optimization in a short time.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The resulting accumulation supports the further internationalization of China's AI smart hardware, opening up new territories. \"We can find the business intersection points with overseas customers faster together with other domestic suppliers, and convert needs into implementable solutions. This upstream and downstream teaming up to go abroad will make the speed of achieving results overseas much faster,\" the person further analyzed.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In other words, the rise of AI hardware is the result of the deep integration of AI and manufacturing industry chains. The former \"world factory\" is accelerating its evolution into the \"source of intelligent innovation.\" Whether it is smart terminals, robots, or industrial automation equipment,\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">they all reflect China's unique competitiveness in soft and hardware integration, rapid verification, cost control, and global supply.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This shows that today's Chinese AI terminals are forming a powerful innovation matrix. They share the same confidence: derived from the rapid iteration ability cultivated by intense domestic market competition, the product definition ability polished by a vast user scenario, and a mature, efficient, and innovative-friendly local supply chain system. Behind this is not just the victory of a single product, but the global bloom of a systematic capability.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">When AI's intelligent touch extends to every scene, the role of the Chinese manufacturing industry chain will be redefined.\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Not only production, but also definition, leadership, and output of intelligent standards.\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">With the integration of AI and smart manufacturing, AI terminals are becoming the new focus of global technology industry competition.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Base Upgrade\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">It is worth noting that in the important underlying capabilities of AI terminals, the domestic computing power industry chain continues to break through, providing core momentum for downstream industries.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This year, several leading domestic AI chip companies are facing a wave of listing guidance, or are in the process of being reviewed, waiting for approval, or waiting to be listed. Several companies have seen significant growth in revenue in 2024, although they are still in a loss-making stage, according to the replies to the inquiry letters, these leading companies have already obtained considerable intelligent computing business orders.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Positive feedback from the capital market will provide valuable \"ammunition\" for these hard tech enterprises to continue investing in R&D and expanding their ecosystems.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This is not an exception. Starting from the end of 2024, the \"Cold King\" (Cambrian) has begun to turn a profit in a single quarter, and by the first half of 2025, it has made a significant profit.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The performance realization of industry leaders confirms that the commercial closed loop of domestic AI computing power from \"usable\" to \"user-friendly\" is beginning to take shape.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">A senior executive of Cambrian Company revealed at the performance briefing in the first half of 2025 that the company's products have been continuously deployed in key industries such as operators, finance, and the internet, and have passed the verification of customer's strict environments. At the same time, in order to meet the strong demand for computing power in the large model market, the company has made strategic inventory preparations based on optimistic expectations of future income, and the inventory scale has accordingly increased.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This indicates that domestic AI chips are no longer samples in laboratories, but products that can truly enter core business scenarios and generate continuous cash flow. Cambrian's \"recovery\" is a microcosm of the maturing of the domestic AI infrastructure ecosystem.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">At the same time, domestic AI chip companies have taken differentiated chip architecture routes such as GPU and ASIC, rapidly iterated, and actively participated in ecological chain collaboration.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Recently, DeepSeek mentioned in an official document that DeepSeek-V3.1 uses UE8M0 FP8 Scale parameters, which is designed for the next-generation domestic chip that is about to be released.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In response, a chip company executive told 21st Century Economic Report journalist, \"We are very pleased to see that the UE8M0 FP8 approach will be applied in specific applications by DeepSeek, and we are also happy to see that there are companies like DeepSeek that tell people what the trends are. This can allow our domestic chip manufacturers to\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">have more opportunities to work with model manufacturers to jointly improve computing efficiency to a new level.\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Tencent executives also mentioned that their heterogeneous computing platform has been adapted to mainstream domestic chips and provides high-cost-performance AI computing power by integrating different types of chips through full-stack optimization of software and hardware. Meanwhile, internet giants including Alibaba and Baidu are also advancing the consolidation of their internal infrastructure capabilities through self-developed ASIC custom chips.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Under this trend, domestic AI chips, after years of accumulation, are now reaching a critical phase. Facing multiple challenges such as insufficient high-end computing power supply, performance gaps of domestic acceleration cards in the short term, and immature software and hardware ecosystems,\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">China's AI industry has chosen a difficult but correct path: open architecture + ecological collaboration.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Ecological Synergy\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This is no longer a lone battle of individual companies, but a systematic breakthrough driven by technology, industry, and policy, allowing China to connect \"points\" into \"surfaces\" in the computing power field. Its synergy is restructuring China's AI infrastructure base at an unprecedented speed.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\"Generally speaking, if you purchase 100 GPU cards, each card's computing power is 100% viewed, but in practice, the computing power efficiency achieved at the beginning may only be 30%-40%. The problem lies in the lack of efficient collaboration at the data interconnection level, which limits the release of computing power potential,\" said a GPU industry practitioner to the reporter of 21st Century Economic Report, giving an analogy: let 100 people communicate in a space, compared to having the same number of people dispersed in different spaces communicating by phone, obviously the former is more efficient. This is the significance of building super nodes under the scale-up logic.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Especially during the current period of continuous accumulation of AI inference demand, by leveraging this, computing cluster capabilities can be better released.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The construction of super nodes involves the collaborative capabilities of different levels of computing links, from the underlying chips, to the middle software stack and interconnection protocols, and then to the intelligent computing clusters.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">In September, Sustech joined over 20 industry chain partners, including the aforementioned chip companies, to launch the country's first AI computing open architecture and introduce an AI super cluster system supporting millions of cards expansion.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Du Xiawei, assistant president of Sustech and general manager of the intelligent computing product division, stated that overcoming the multidimensional challenges of domestic AI computing power requires gathering the forces of the chip, computing system, and large model industries to build an open AI computing architecture and coordinate cross-layer collaboration between industries, thus breaking down \"technical walls\" and \"ecological walls\".\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Compared to closed systems, the Sustech AI super cluster system, in addition to working efficiently like a computer through tight coupling design, also supports multi-brand AI acceleration cards and is compatible with mainstream software ecosystems such as CUDA, significantly reducing hardware costs and software development adaptation costs.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">\"On the surface, this is an open toolchain; but\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">in a deeper logic, it reflects the direction of China's AI computing power ecology: from following and imitating to beginning to try to set its own rules.\u003C\u002Fstrong>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">” Du Xiawei continued.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">As an important partner of Sustech, the leading domestic CPU\u002FDCU company Hanguang Information has also made the strategic decision to open the system interconnect bus protocol (HSL). The core content of HSL includes opening the complete bus protocol, providing IP reference designs, and opening the instruction set, among others.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Academician Zheng Weimin of the Chinese Academy of Engineering summarized: \"The efficient collaboration of CPUs, GPUs, NPUs, accelerator cards, storage, and network modules is the key to unleashing the overall performance of heterogeneous systems.\" \"To achieve this, it must rely on a unified interconnect bus protocol. It is the language for communication between computing units and an important foundation for realizing system interoperability and computing fusion.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Huawei has also put forward a mature implementation plan. In the first half of this year, the Huawei Ascend 384 super node made its debut, directly benchmarked against NVIDIA GB200 NVL72. On September 18, Huawei updated the super node products Atlas 950 SuperPoD and Atlas 960 SuperPoD, supporting 8192 and 15488 Ascend cards respectively.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Huawei also revealed that the Atlas 950 SuperPoD to be launched next year will have a quantity scale 56.8 times that of NVIDIA GB200 NVL144, computing power 6.7 times, memory capacity 15 times, and internal (interconnect) bandwidth 62 times.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Thus, a path for AI development has been clearly outlined: capital helps to break through technologically, while an open ecosystem consolidates these breakthroughs into irreplaceable systemic capabilities. This huge synergy combining top-down and bottom-up efforts is breaking down \"technical walls\" and \"ecological walls\", building a solid foundation for China's competition in the AI era.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">From the bottom computing power to terminal implementation, the rise of Chinese AI hardware is not only a technological competition,\u003C\u002Fspan>\u003Cstrong style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">but also a systematic project involving industrial chain collaboration, ecological construction, and global market reshaping, enabling China to potentially transform from a \"follower\" to a \"leader\", providing innovative solutions for the popularization of artificial intelligence.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(187, 187, 187);\">【News Source】Reporter Luo Yiqi of 21st Century Economic Report \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845511183702469808&wfr=spider&for=pc\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845511183702469808&wfr=spider&for=pc\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（This article is reposted by this website to provide readers with more information and news. The content does not constitute investment or consumption advice. If there are any questions about the facts of the article, please verify with the relevant parties. The views of the article are not the views of this website, and are for reference only.）\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A9c205253-d8ad-4673-81e7-c273827d856c%3A0.wav?Expires=1774838460&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=6IjFnVG%2F%2FlVF2D1PwBZn0FY%2BXZw%3D","9c205253-d8ad-4673-81e7-c273827d856c",17066410]