（文章来源：每日经济新闻） 华泰证券认为，国产算力厂商产品迭代与融资进展均在加速，华为于2025全联接大会公布昇腾 Roadmap，2026年将推出昇腾950系列，同时微架构从达芬奇架构更换为 SIMD/SIMT架构，2028年计划 发布昇腾970系列，芯片的编程灵活性得以提升；沐曦与摩尔线程已提交IPO材料，9月26日摩尔线程 IPO已经过会，其他如燧原、天数智芯、壁仞等芯片产品也在政府、运营商端及行业垂类细分领域不断 落地。2026年国产算力卡国产化趋势将进一步加速，而产业链瓶颈在于上游的先进制造产能，在云厂数 据中心以及地方算力集群需求带动下，拥有先进产能的代工厂有望充分受益。 科创人工智能ETF华夏（589010）紧密跟踪上证科创板人工智能指数，覆盖全产业链优质企业，兼具高 研发投入与政策红利支持，20%涨跌幅与中小盘弹性助力捕捉AI产业"奇点时刻"。 截至2025年10月10日 13点31分，上证科创板人工智能指数下跌5.00%。成分股方面，芯原股份领跌 8.68%，恒玄科技下跌8.32%，虹软科技下跌8.15%，星环科技下跌7.53%，海天瑞声下跌7.51%。科创人 工智能ETF华夏（589010 ...

Group 1: AI Competition Landscape - The AI sector is characterized as an "infinite competition game," reflecting the intense actions of US and Chinese tech giants in the field [1] - Nvidia plans to invest up to $100 billion in OpenAI to develop a super AI data center with a power capacity of 10 gigawatts, equivalent to the electricity consumption of 8 million US households [1] - OpenAI's "Gateway to the Stars" project aims to establish multiple sites with a total power capacity of 10 gigawatts, targeting a computational power goal that matches China's current total AI computing capacity of 90 EFLOPS [1][2] Group 2: China's AI Infrastructure and Investments - Alibaba's CEO announced a three-year plan with an investment of 380 billion yuan in AI infrastructure, aiming to establish 30 global data centers [2] - Huawei is set to release its Ascend 950 series chips in 2026 and 2027, aiming to double computing power annually to compete with Nvidia's technology leadership [2] - China's total computing power stands at 280 EFLOPS, ranking second globally, but still trailing behind the US in terms of high-impact talent density and influential AI research [3] Group 3: Market Efficiency and Utilization Challenges - There is a significant supply-demand gap in GPU requirements in China, with estimates indicating a demand exceeding millions of units and a notable production capacity shortfall [3] - The average utilization rate of AI computing centers in China is only 30%, indicating inefficiencies in the application of computing power [3] - Despite leading in AI paper production and patent applications, China's high-impact research output lags behind that of the US [3] Group 4: Consumer Market and Adoption - Chinese consumers show a high acceptance of AI technologies, with over 80% awareness, compared to approximately 62% in the US [4][6] - The Chinese AI application market is characterized by a "universal" approach, with companies like ByteDance significantly reducing API costs to one-tenth to one-twentieth of US counterparts [6] - Manufacturing is a key sector for AI investment in China, with 43% of AI venture capital directed towards this industry, contrasting sharply with only 3% in the US [6] Group 5: Future Projections and Growth - By 2030, China's AI-related industry value is expected to exceed 10 trillion yuan, while the US has a head start with an AI adoption rate of 78% among enterprises [7] - China's intelligent computing center construction is projected to grow at an annual rate of 40%, with significant energy efficiency upgrades planned for data centers by 2032 [7]

"AI是一场无限竞争游戏。"英伟达CEO黄仁勋的这句话，完全是近期中美巨头在AI领域密集动作的真实 写照。随着全球芯片产业即将迈入万亿美元时代，这场关乎未来经济话语权的竞赛已进入决定格局的深 水区。 9月下旬，美国科技圈抛出重磅合作——英伟达计划向Open AI分阶段注资高达1000亿美元，联手打造电 力容量达10吉瓦的超级AI数据中心，相当于800万户美国家庭的用电量。英伟达CEO黄仁勋直言，这 是"为下一个智能时代提供动力"，而OpenAI CEO奥尔特曼更将其上升到战略高度："计算基础设施将成 为未来经济的基础"。 与此同时，OpenAI的"星际之门"计划正加速落地，9月24日官宣5个新站点已敲定4地，规划电力容量达 10吉瓦，支撑的远期算力目标可对标中国当前智能算力总和（90EFLOPS），首个站点预计2026年下半 年启用，未来3年还将追加4000亿美元投入。 中国同样动作频频。9月24日开幕的2025云栖大会上，阿里集团董事兼首席执行官吴泳铭表示，大模型 是下一代操作系统，而AI云是下一代计算机，也许未来全世界只会有五六个超级云计算平台。并披露 三年3800亿元AI基建计划的最新进展——今年已启用 ...

Core Insights - Huawei unexpectedly revealed its future chip roadmap during the Huawei Connect 2025 event, showcasing several new chips including the Ascend 950, 960, and 970 series for AI computing, as well as the Kunpeng 950 and 960 processors for general computing [1][3][10] Group 1: Chip Developments - The Ascend 950 series chips will support low-precision data formats and achieve computing power of 1P and 2P, enhancing training efficiency and inference throughput [3][10] - The Ascend 960 is planned to double the performance of the Ascend 950 and will support Huawei's self-developed HiF4 data format, set to launch in Q4 2027 [7] - The Ascend 970 will further enhance specifications compared to the Ascend 960, with plans for release in Q4 2028 [7] Group 2: Supernode and Cluster Innovations - Huawei introduced the Atlas 950 supernode, which will consist of 8192 Ascend 950DT chips, achieving FP8 computing power of 8E FLOPS and FP4 computing power of 16E FLOPS, set to launch in Q4 2026 [11][13] - The Atlas 960 supernode, planned for Q4 2027, will be based on 15488 Ascend 960 chips, with FP8 computing power reaching 30E FLOPS and FP4 computing power reaching 60E FLOPS [13] - The Atlas 950 SuperCluster will consist of 64 Atlas 950 supernodes, achieving FP8 computing power of 524 EFLOPS, making it the world's strongest computing cluster [18] Group 3: Software and Ecosystem Development - Huawei aims to develop a robust software ecosystem to complement its hardware, with the CANN deep learning framework and MindSpore framework serving as alternatives to NVIDIA's CUDA [21][22] - The company plans to open-source its CANN compiler and virtual instruction set interface by the end of 2025, along with the Mind series application tools [22][24] - Huawei's strategy emphasizes hardware evolution through existing chip technology while fostering an open-source ecosystem to address challenges posed by U.S. sanctions [24]