转自：新华财经 联想集团此次推出的"AI工厂"解决方案，旨在构建出一套可管理、可复制且支持持续运营的标准化体系，重塑AI应用的开发与部署流程，将原本复杂且孤立 的AI开发任务转变为一条标准、高效的现代化"AI生产线"。陈振宽表示，从场景定义到将客户数据采集作为原材料，经过智能体开发平台与AI训练引擎的 深度处理，"联想AI工厂"解决方案最终能够生产出可交付的智能体、垂域模型、推理服务等成熟"产品"。 要让"AI工厂"高效运转，离不开坚实可靠的底层基础设施作为承载。作为"AI工厂"的算力基石，发布会上发布的联想万全异构智算平台4.0，迎来四大技术 场景的全新升级。 首先，在大模型预训练场景，通过长序列并行优化，模型训练时间可缩短35%；其次，在大模型后训练场景，计算引擎优化、小样本强化学习、训练自动并 行可使训练时间缩短50%；第三，在推理场景，ROCE网络负载均衡优化，带宽利用率可提升60%，通信原语性能提升30%，大模型推理性能提升30%；第 四，在超智融合计算场景，可支持国际和国内硬件生态，完成从底层架构到算子的全面优化，同时新增超16个制造业应用的作业模板和脚本，实现超算、智 算高效融合。 新华财经北京 ...

坚实的算力基础设施是AI工厂高效运转的核心支撑。此次发布的联想万全异构智算平台4.0，通过九大差异化创新实现四大技术场景升级：大模型预训练时 间缩短35%，后训练时间缩短50%，推理场景带宽利用率提升60%、性能提升30%，超智融合计算场景新增16个制造业应用模板。该平台已在多个重点场景 落地见效，在国家级智算集群中实现千卡训练MFU从30%提升至60%，企业本地部署场景中DeepSeek R1模型极限吞吐量突破12000 Tokens/s，为高校科研提 供稳定高效的HPC/AI融合算力支持。 【环球网科技综合报道】以"异构智算 本地引擎"为主题的2025联想异构智算产业联盟高峰论坛暨AI算力基础设施新品发布会近日在北京举行。会上，联想 集团正式推出"AI工厂"解决方案，同步发布新一代高端大模型训练AI服务器、万全异构智算平台4.0，并联合产业伙伴发布业内首个模型训推服务标准验证 成果，多项重磅举措为我国人工智能产业高质量发展注入强劲动能。 当前，"人工智能+"行动连续两年写入政府工作报告，多部门密集出台配套政策，推动人工智能与各行业深度融合。在技术突破、政策支持、供应升级的多 重驱动下，中国AI市场正处于" ...

在2025年，中国AI产业正经历从狂热的"百模大战"向冷静的"应用落地"转型的关键周期。12月5日，在 北京举办的2025联想异构智算产业联盟高峰论坛上，联想集团试图通过发布"AI工厂"解决方案、核武级 服务器以及全新的行业标准。在算力不再稀缺的当下，如何解决效率与商业价值的错配？联想中国基础 设施业务群战略管理总监黄山与服务器事业部总经理周韬在接受媒体专访时，不仅详解了硬件架构的激 进演进，更直言不讳地指出了当前大模型商业化面临的挑战与机遇。 算力中心的终结与"AI工厂"的崛起 他进一步强调，这种差距将导致竞争层面的不对等："这些业务如果有生态形成了联动以后，我只做数 据中心，我就被形成降维打击了。怎么叫降维打击？当赋能用户的时候，给AI工厂只给计算中心。首 先方案的完整性不如别人。第二个，因为你没有跟其他的业务板块协同，你的效率肯定不如一个完整的 AI工厂，最终的效率一定是不行的。" 联想的商业创新在于"补足"客户的能力短板。黄山解释道："我们的客户有的有自己的数据，有自己的 设计能力，但是没有生产能力。有的客户也有生产能力，但没有生产管理能力，有的客户出来的应用如 何做迭代，他没有成周期升级维护的能力。 ...

新浪科技讯 12月5日晚间消息，"异构智算 本地引擎"2025联想异构智算产业联盟高峰论坛暨AI算力基础 设施新品发布会在北京举办。 据悉，联想万全异构智算平台4.0在大模型预训练场景，模型训练时间可缩短35%；在大模型后训练场 景，计算引擎优化、小样本强化学习、训练自动并行可使训练时间缩短50%；在推理场景，ROCE网络 负载均衡优化，带宽利用率可提升60%，通信原语性能提升30%，大模型推理性能提升30%；在超智融 会上，联想集团副总裁、中国基础设施业务群总经理陈振宽正式推出"联想AI工厂"解决方案，同时发布 新一代高端大模型训练AI服务器—基于英特尔®至强®6处理器的联想问天 WA8080a G5，以及全新升级 联想万全异构智算平台4.0。 另外，联想问天 WA8080a G5作为联想首款高端大模型训练AI服务器，采用英特尔®至强®6最新处理 器，支持本地化主流最新OAM2.0 GPU，同时兼顾风冷和液冷双模散热方案。 责任编辑：刘万里 SF014 新浪科技讯 12月5日晚间消息，"异构智算 本地引擎"2025联想异构智算产业联盟高峰论坛暨AI算力基础 设施新品发布会在北京举办。 会上，联想集团副总裁、 ...

Core Viewpoint - Lenovo is positioning itself at the forefront of a new growth cycle in AI infrastructure, emphasizing systematic innovation and responding to the accelerating demand for AI computing power in China [2][3]. Group 1: Product Launches and Innovations - Lenovo ISG China launched the "Lenovo AI Factory" solution, the high-end AI server Lenovo Wenti WA8080a G5, and the upgraded Wanquan Heterogeneous Computing Platform 4.0, along with the industry's first model training and inference service standard [2][9]. - The new AI Factory solution aims to standardize and streamline AI application development and deployment processes, transitioning AI from experimental to industrialized systems [9]. - The Wanquan Heterogeneous Computing Platform 4.0 has achieved significant upgrades in various scenarios, including a 35% reduction in model training time for pre-training and a 50% reduction for post-training [9][10]. Group 2: Market Demand and Growth - The AI server market in China is projected to reach $16 billion in the first half of 2025, doubling year-on-year, with expectations to exceed $140 billion by 2029 [5]. - The overall market for intelligent computing services in China is expected to grow from $5 billion in 2024 to $7.95 billion in 2025, with a compound annual growth rate of 57.3% from 2023 to 2028 [5]. - Lenovo ISG China's revenue has increased by 175% over the past two and a half years, establishing itself as one of the fastest-growing AI infrastructure providers in China [3][5]. Group 3: Strategic Positioning and Future Outlook - Lenovo's growth is driven by structural demand from large model implementations, indicating a shift in the AI computing industry towards deeper integration with various sectors [6][17]. - The company aims to enhance its product offerings while leveraging differentiated advantages in liquid cooling technology and heterogeneous computing scheduling [16]. - Lenovo's strategy focuses on localizing and AI-oriented solutions, integrating full-stack capabilities to provide comprehensive AI solutions to clients [16][17].

Core Insights - Lenovo officially launched the "Lenovo AI Factory" solution aimed at transforming traditional data centers into efficient AI factories, leveraging its advantages in computing products and solutions [1] - The new generation of high-end AI servers, Lenovo Wenti WA8080a G5, based on Intel Xeon 6 processors, was introduced alongside the upgraded Lenovo Wanquan Heterogeneous Computing Platform 4.0 to support enterprises in embracing the AI era [1] - The Chinese government has emphasized the integration of artificial intelligence in its work reports, leading to a series of policies aimed at deepening the industrial layout and promoting the integration of AI across various sectors [1] Industry Developments - The AI industry is entering a new phase characterized by the deep coupling of large models with specific industries, necessitating advancements in hardware innovation, software optimization, standardization, and application implementation [2] - The first industry standard for model training and inference services, "Information Technology Computing Power Service High-Performance Training and Inference Service Capability Requirements," was released to establish a unified capability benchmark for high-performance model training services [2] - The industry faces challenges such as fragmented heterogeneous chip scheduling, communication bottlenecks, and inefficient resource utilization, which hinder the effective release of computing power resources [2]

Major Companies - Chasing Technology has been selected for the 2025 "Brands That Matter" global list by Fast Company, being the only domestic brand on the list due to its localized innovation and user experience improvements [1] - Chasing Technology's products cover various categories including high-end smart home appliances, smart cars, smart kitchen appliances, smart cleaning devices, outdoor smart devices, and personal care products [1] - The company's sweeping robot products hold the number one market share in 22 countries and regions globally, including Germany, France, and Singapore [1] Investment and Financing - Hengxuan Technology's next-generation low-power, high-performance wearable chip, BES6000, is expected to enter the sample delivery phase in the first half of next year [7] - The BES2800 chip has been widely applied in low-power smart hardware markets such as smartwatches and smart glasses [7] - Zhongke Qiancheng has recently completed several million yuan in angel round financing, with funds primarily aimed at product improvement and technology research and development [8] - Aoluo Pharmaceutical has successfully completed a $100 million Series A financing round, with funds to accelerate its differentiated pipeline layout focusing on unmet clinical needs in immunology, inflammation, and oncology [9] New Products - CATL has launched the world's only "ship-shore-cloud" zero-carbon shipping and smart port integrated solution, achieving full-chain integration from onboard power systems to shore-based energy supply networks and cloud-based intelligent management [10] - Gaode has released the "AI Parking Radar," the first real-time perception function for road parking spaces in the country, providing users with a new smart parking experience [11] - Lenovo has officially launched the "Lenovo AI Factory" solution to help enterprises transition from traditional "computing centers" to efficient "AI factories," along with the release of a new generation of high-end AI server based on Intel® Xeon® 6 processors [12]

凤凰网科技讯12月5日，联想在北京举办"异构智算本地引擎"2025产业联盟高峰论坛暨AI算力基础设施 新品发布会。会上正式推出"联想AI工厂"解决方案，旨在帮助企业数据中心实现从传统算力中心向智能 化AI工厂的转型。 同期发布的新品包括新一代高端大模型训练AI服务器——基于英特尔至强6处理器的联想问天WA8080a G5，以及全新升级的联想万全异构智算平台4.0。该平台在大模型预训练、后训练、推理及超智融合计 算四大场景中实现性能提升，其中在部分训练场景中训练时间最高可缩短50%。 本次大会还联合多家产业伙伴发布了业内首个《信息技术算力服务高性能训推服务能力要求》标准验证 成果。该标准旨在建立统一的训推服务能力评价体系，推动算力服务从"资源交付"向标准化"能力交 付"演进。联想作为主编单位之一，参与了标准制定并率先完成相关验证。 联想集团副总裁、中国基础设施业务群总经理陈振宽表示，当前AI正进入与行业深度耦合的新阶段， 多元算力、芯片模型生态协同成为构建中国AI生态的重要力量。中国科学院院士陈润生指出，算力产 业需从"可用"向"好用"进阶，推动形成自主强大的AI算力生态。 据介绍，联想将继续围绕"AI导向"与 ...

会上，联想问天WA8080a G5正式发布。据悉，新品支持八颗最新OAM2.0 GPU和最高支持单颗1000W 散热、液冷超1000W，同时节点间互联速率高达3.2Tb/s。 在稳定性方面，12V和54V分离供电设计电源N+N冗余、双BIOS/BMC flash冗余以及部件级功耗实时监 控，进一步保障了新品的稳定可靠。此外，新品还拥有高密前后可扩展IO设计、兼容国内外主流GPU 等扩展能力。 新浪科技讯 12月5日下午消息，联想举办"2025联想异构智算产业联盟高峰论坛暨AI算力基础设施新品 发布会"。 在稳定性方面，12V和54V分离供电设计电源N+N冗余、双BIOS/BMC flash冗余以及部件级功耗实时监 控，进一步保障了新品的稳定可靠。此外，新品还拥有高密前后可扩展IO设计、兼容国内外主流GPU 等扩展能力。 责任编辑：刘万里 SF014 新浪科技讯 12月5日下午消息，联想举办"2025联想异构智算产业联盟高峰论坛暨AI算力基础设施新品 发布会"。 会上，联想问天WA8080a G5正式发布。据悉，新品支持八颗最新OAM2.0 GPU和最高支持单颗1000W 散热、液冷超1000W，同时节点间互联 ...