训推一体GPU板卡
Search documents
沐曦股份12月18日获融资买入4.60亿元,融资余额8.69亿元
Xin Lang Cai Jing· 2025-12-19 01:41
资料显示,沐曦集成电路(上海)股份有限公司位于上海市浦东新区海科路999弄C8栋,成立日期2020年9 月14日,上市日期2025年12月17日,公司主营业务涉及研发、设计和销售应用于人工智能训练和推理、 通用计算与图形渲染领域的全栈GPU产品,并围绕GPU芯片提供配套的软件栈与计算平台。主营业务收 入构成为:训推一体GPU板卡97.55%,智算推理GPU板卡1.25%,其他0.80%,训推一体GPU服务器 0.32%,IP授权0.08%。 截至12月17日,沐曦股份股东户数2.51万,较上期增加20138.71%;人均流通股722股,较上期增加 0.00%。2025年1月-9月,沐曦股份实现营业收入12.36亿元,同比增长453.52%;归母净利润-3.46亿元, 同比增长55.79%。 责任编辑:小浪快报 12月18日,沐曦股份跌5.16%,成交额54.08亿元。两融数据显示,当日沐曦股份获融资买入额4.60亿 元,融资偿还3.97亿元,融资净买入6329.75万元。截至12月18日,沐曦股份融资融券余额合计8.69亿 元。 融资方面,沐曦股份当日融资买入4.60亿元。当前融资余额8.69亿元,占流通市值 ...
沐曦股份上市前的财务表现如何,他们也会拿融资款去理财吗?
Sou Hu Cai Jing· 2025-12-18 10:14
沐曦集成电路(上海)股份有限公司(股票简称:沐曦股份)2020年成立,2025年12月在上交所科创板上市,总部位于上海,在北京、南京、成都、杭州、深 圳、武汉和长沙等地建立了全资子公司暨研发中心。 沐曦股份致力于为异构计算提供全栈GPU芯片及解决方案,可广泛应用于智算、智慧城市、云计算、自动驾驶、数字孪生、元宇宙等前沿领域,为数字经济 发展提供强大的算力支撑。 "训推一体GPU板卡"是沐曦股份的核心业务,还有少量的"智算推理GPU板卡"、"训推一体GPU服务器"和"IP授权"等业务;现在主要针对国内市场,也就是 所谓的国产替代。 沐曦股份是国内少数系统掌握高性能GPU芯片研发、设计和量产技术的企业,其核心业务聚焦于全栈高性能GPU芯片的自主设计,且已形成成熟的芯片设计 成果与技术体系。 具体体现为,从核心底层技术入手,具备完整自主的芯片设计能力;已构建起覆盖多场景的 GPU 芯片产品线矩阵,且多款设计产品实现量产落地;同时还 拥有专利和专有技术等成体系的技术储备。 2025年发布的旗舰产品曦云C600,采用12nm工艺制程,由中芯国际代工生产,已实现该工艺的规模化量产,其Chiplet良率已突破92%。曦云C ...
沐曦股份IPO上会:高性能GPU光环下的三大隐忧
Sou Hu Cai Jing· 2025-10-23 09:22
Core Insights - Muxi Co., Ltd. is facing significant challenges as it prepares for its IPO on the STAR Market, highlighted by a cumulative loss of 3.29 billion yuan over three years and a reliance on a single product for over 97% of its revenue [1][2]. Financial Performance - The company has exhibited a "high investment, high loss" financial profile, with cumulative losses reaching 3.29 billion yuan from 2022 to Q1 2025, peaking at 1.409 billion yuan in 2024 [2]. - Revenue surged from 426,400 yuan in 2022 to 743 million yuan in 2024, yet this revenue only accounted for 53% of the losses incurred during the same period [2]. - Operating cash flow has been persistently negative, with a total outflow of 4.51 billion yuan from 2022 to Q1 2025, including a single-year outflow of 2.148 billion yuan in 2024 [2]. Product Dependency and Pricing Power - Muxi's revenue structure is heavily dependent on the Xiyun C500 series, which constituted 97.87% of total revenue in Q1 2025, a sharp increase from 30.09% in 2023 [3]. - The average selling price of the GPU boards has declined from 56,900 yuan to 38,900 yuan from 2023 to Q1 2025, representing a 31.6% drop, indicating a lack of pricing power in a competitive market [3]. - The company faces significant product iteration pressures, with new chip designs taking 2-3 years to develop, while AI model technologies evolve every 3-6 months [3]. Competitive Landscape - Muxi operates in a highly competitive environment dominated by NVIDIA, which held a 70% market share in China's AI chip market in 2024 [4]. - Despite U.S. export controls creating opportunities for domestic GPU manufacturers, NVIDIA's tailored H20 chip for the Chinese market poses a threat to local firms [4]. - Domestic competition is fierce, with Muxi competing against other local players like Haiguang Information and Moore Threads, all vying for a limited market share [4]. Customer Concentration and Technology Sources - The company has a high customer concentration, with the top five customers accounting for 88.35% of sales in Q1 2025, raising concerns about revenue independence [5]. - The core technology team includes members with backgrounds at AMD, a direct competitor, which could pose potential risks in terms of intellectual property disputes [5]. Industry Challenges and Muxi's Path Forward - Muxi's struggles reflect broader challenges faced by domestic GPU companies, including how to achieve a commercial breakthrough amid technological lag and weak ecosystems [6]. - The company plans to leverage CUDA compatibility to reduce user migration costs, but building a robust software ecosystem is a long-term endeavor [6]. - The upcoming IPO aims to raise 3.9 billion yuan primarily for new chip development, but the high failure rate in chip R&D raises questions about the effectiveness of these investments [6].