Domestic AI Chips

Search documents
华为领衔,国产AI芯片撑起算力脊梁
21世纪经济报道· 2025-08-04 06:55
Core Insights - Huawei's Ascend 384 supernode has made its debut, achieving a computing power of 300 Pflops through self-developed Matrix-Link technology, positioning it as a competitor to NVIDIA's latest GB200 NVL72 [1] - Domestic AI chip manufacturers are collaborating to build a robust ecosystem, with significant advancements showcased at the World Artificial Intelligence Conference [2] - The performance of domestic AI chip companies has surged, with notable revenue increases reported across the sector [2][3] Group 1: Huawei and Its Innovations - Huawei's Ascend 384 supernode has been recognized as a "treasure of the exhibition" at the World Artificial Intelligence Conference, attracting significant attention [1] - The supernode's architecture allows 384 cards to operate as a single computer, showcasing its advanced capabilities [1] - The deployment of Huawei's technology is already in multiple data centers across the country, providing services to various sectors including internet and automotive [2] Group 2: Domestic AI Chip Developments - Companies like Nucypher and Suirun Technology are developing advanced chips, with the C600 model set for small-scale production in Q4 of this year [1] - The domestic AI chip industry is experiencing rapid growth, with companies like Moore Threads reporting a 256% increase in revenue, driven by AI computing business [2] - Cambrian Technology has turned a profit for the first time in years, with a 40-fold increase in revenue and over 2-fold profit growth in Q1 of this year [3] Group 3: Industry Trends and Collaborations - The domestic AI chip industry is moving towards a trend of adopting local chips, with significant collaborations among companies like Xizhi Technology and ZTE [2] - The introduction of light-interconnected GPU supernodes is expected to reduce power consumption and latency significantly [2] - The overall performance of domestic AI chips has led to a substantial increase in market presence, indicating a shift away from reliance on foreign technology [3]
华为领衔,国产AI芯片撑起算力脊梁
2 1 Shi Ji Jing Ji Bao Dao· 2025-08-04 03:33
Group 1 - Huawei's Ascend 384 super node has made its debut at the World Artificial Intelligence Conference, showcasing a computing power of 300 Pflops through its self-developed Matrix-Link network, positioning it against Nvidia's latest GB200 NVL72 [1] - Domestic AI chip manufacturers are collaborating to build a robust domestic computing power ecosystem, exemplified by products like the Xiyun C550 and C600 from Nuxi, which utilize self-developed architectures and are set for small-scale production in Q4 [1][2] - The domestic AI chip market has seen significant growth, with companies like Moore Threads reporting a 256% increase in revenue to 432 million yuan, and Nuxi's revenue surging over 1300% to 742 million yuan [2] Group 2 - The trend of adopting domestic AI chips is becoming irreversible, as evidenced by the increasing usage of these chips in various sectors, supporting China's computing backbone [3] - The collaboration among companies such as Xizhi Technology, Biyuan Technology, and ZTE has led to the development of the first domestic optical interconnect GPU super node, LightSphere X, which aims to reduce power consumption and latency [2]