在商业化方面，阿里巴巴看好云业务具备庞大增长潜力，主要受强劲的AI原生需求、国际扩张以及传 统产业AI采用率提升所驱动。凭借其全栈式"云+AI"服务，美银证券预期阿里电动车、金融及具身智能 等垂直领域将快速增长。 美银证券发表报告，将阿里巴巴(09988,BABA.US)2026至2028财年每股ADS盈利预测上调0%至4%，并 将目标价由168美元上调至195美元，对应阿里巴巴港股目标价由164港元升至190港元。基于未来三年云 业务复合年增长率预估超过30%及稳固的电商前景，重申对阿里"买入"评级。 该行指阿里巴巴举行2025云栖大会，阿里云CEO吴泳铭强调，为迎接人工智能超级智能(ASI)时代的到 来，公司将在未来三年内，超越原定的3,800亿人民币预算，进一步增加对人工智能和云计算的投资。 管理层认为，大型模型将成为下一代作系统，而AI云将成为下一代电脑。管理层预期，长远来看，全 球可能仅会存在5至6个超级云计算平台。有鉴于新ASI时代的巨大机会，公司将自身定位为全球领先的 全栈AI服务提供商，提供顶级的大型模型、全球AI云端网络，以及开放且对开发者友善的生态系统。 ...

Core Viewpoint - The Vice Chairman of the China Semiconductor Industry Association, Wei Shaojun, suggests that Asian countries, including China, should abandon reliance on NVIDIA's GPUs for AI development to reduce dependence on U.S. technology [1] Group 1: Industry Trends - Asian countries are currently imitating the U.S. in developing algorithms and large models, primarily using NVIDIA or AMD GPUs for training large language models, which limits autonomy and increases dependency on U.S. technology [1] - The domestic GPU market in China is witnessing a significant shift, with the share of domestic computing power in the data center accelerator card market rising from 14% to 34.6% from 2023 to 2024 [2] - Predictions indicate that by 2027, the proportion of domestic AI chips in China will soar from 17% in 2023 to 55%, while the share of U.S. suppliers like NVIDIA and AMD will drop from 83% to 45% [2] Group 2: Company Developments - NVIDIA's H20 chip has faced significant market challenges following the exposure of serious security issues, with the company reportedly not selling any new H20 chips in the domestic market as of July, and only releasing $180 million worth of inventory to other regions in the second fiscal quarter [1] - The rise of companies like DeepSeek demonstrates that Chinese enterprises can achieve significant breakthroughs in algorithms even without the most advanced hardware [2] - Wei Shaojun emphasizes the need for domestic companies to focus on creating new types of chips specifically designed for large model development, rather than continuing to rely on NVIDIA's GPUs originally designed for gaming and industrial graphics [2]