基流科技获近亿元A+轮融资，自研RoCE组网降本40%，服务智谱AI等大客户。 文 ｜ 施嘉翔 封面来源 ｜ unsplash 近日，全栈自主AI算力建筑商"基流科技"完成近亿元A+轮融资，由上海国投孚腾资本和老股东国方创新联合领投，克拉玛依城发 基金、张江燧芯基金跟投，老股东卓源亚洲连续四轮追加投资。融资资金将主要用于核心技术和产品研发，市场拓展与团队建 设。 基流科技成立于2023年2月，源自清华大学网络安全实验室。创始人胡效赫是清华大学博士、博士后，在高性能网络和分布式系 统领域有十余年研究经验。核心团队成员来自清华、北大等顶尖高校及阿里、百度、中兴等头部厂商。 基流科技定位为全栈自主的AI算力建筑商，团队专注于攻克大规模算力集群的分布式计算通信难题。作为国内极少拥有数个万卡 集群落地经验的AI基础设施厂商，公司已服务包括智谱AI、商汤科技、运营商、数据中心及地方国企在内的多个用户。 目前，基流科技打造了自研的高性能开源智算系统Galaxy HPAC（High-Performance AI Computing）。该系统主要包括三大核心 产品：高性能融合AI网络（Mercury-X）、人工智能平台（Ven ...

Group 1 - The company "Jiliu Technology" has recently completed nearly 100 million yuan in A+ round financing, led by Shanghai Guotou Futeng Capital and Guofang Innovation, with participation from other funds and existing shareholders [1] - Founded in February 2023, Jiliu Technology originated from Tsinghua University's cybersecurity laboratory, with a core team from top universities and leading companies [1] - The company focuses on building a high-performance open-source intelligent computing system, differentiating itself from competitors by providing a comprehensive solution that includes both software and hardware [1][2] Group 2 - With the explosion of large models and AIGC applications, computing power has become a new type of infrastructure, leading to challenges such as network communication bottlenecks and high costs due to vendor lock-in [2] - Jiliu Technology's self-developed RoCE networking solution can reduce customer costs by 40% and shorten delivery times from months to weeks [2] - The company has developed a high-performance open-source intelligent computing system called Galaxy HPAC, which includes three core products: Mercury-X, Venus-AICloud, and Venus-AIDOC [2] Group 3 - Jiliu Technology has delivered a total of 23 clusters, utilizing over 66,000 GPUs and 4,000 switches, and has successfully implemented long-distance training and inference clusters [3] - The demand for computing power is expected to shift from a training-to-inference ratio of 7:3 in 2023 to 3:7 by 2025, indicating a growing need for inference capabilities [3] - The company is actively participating in national strategies by engaging in computing power projects in regions like Xinjiang and Ningxia, while also investing in the design of a 100,000-card cluster [3]

成立至今，公司 累计组网调优交付的算力规模 已 达 23 个集群，超 6 万 6 千张 GPU ，超 4000 台交换机，超 32 万个光模块 ， 服务客户覆盖智谱 AI 、商汤科技、运营商 、 数据中心及地方国企 等。 AI Infra 赛道创业窗口期已过，不再是蓝海。 作者丨巴里 编辑丨关雎 图源丨 Midjourney 每一次技术变革驱动的 "掘金时代"开启时，率先实现价值的，往往是那些 铺路修桥的人。 这句 话正是 当下 AI 产业格局 的 生动诠释 ——大模型这座金矿刚刚开始挖掘，挖金子的 AI 企业大 多尚未盈利，而英伟达 这样的 底层算力基础设施 提供商已率先实现规模化收益 。 7 月 21 日， 作为国内极少拥有数个万卡集群落地经验的 AI 基础设施厂商， 基流科技完成 了近亿 元 A+ 轮融资， 由上海国投孚腾资本和国方创新联合领投，克拉玛依城发基金、张江燧芯基金跟投， 老股东卓源亚洲连续四轮追加投资 。 此次融资 主要用于加大研发投入、市场拓展和团队建设 。 这已经是基流科技成立两年多以来的第 7 轮融资，融资速度不可谓不快，投资方不乏北京、上海、新 疆等地国资，以及智谱 AI 、 招商局 ...