Group 1 - OpenAI plans to deploy one million GPUs by the end of this year, significantly increasing its GPU capacity from the current 100,000 GPUs used for ChatGPT 4.5 [1][2] - The surge in GPU deployment indicates a corresponding increase in demand for AI servers, with major suppliers like Hon Hai, Quanta, and Inventec expected to benefit from this trend [1][2] - The "Stargate" initiative by the Trump administration is anticipated to further drive AI infrastructure demand, leading to a substantial influx of AI server orders [1] Group 2 - OpenAI's transition to a "multi-cluster architecture" is aimed at supporting the training needs of next-generation models, with future considerations for using up to 10 million GPUs simultaneously [2] - The Stargate project, with a budget of hundreds of billions, is progressing, with the first data center in Texas expected to house up to 400,000 NVIDIA AI chips, potentially becoming a leading AI computing center globally [2] - OpenAI's cloud service providers include Microsoft, Oracle, and CoreWeave, with major Taiwanese suppliers like Hon Hai and Quanta set to gain from increased server production [3]

图 点击 上方 硬AI 关注我们 超微电脑第三财季营收46亿美元，低于预期的54.2亿美元。第四季度指引为营收56-64亿美元，远低于市场预期的68.2亿 美元。CEO查尔斯认为经济不确定性和特朗普关税影响已开始显现。而由于关税相关不确定性，公司不会提供2026财年 指引。 硬·AI 作者 | 鲍弈龙 编辑 | 硬 AI 超微电脑第四财季收入预期大幅低于分析师预测，同时下调全年指引，反映出AI基础设施热潮可能 遭遇首个障碍。 周二美股盘后，AI服务器制造商超微电脑（Super Micro Computer）发布第四财季收入，业绩显著 低于预期，指引令人担忧： 财务表现不佳 ： 第三财季营收46亿美元，低于预期的54.2亿美元；调整后每股收益0.31美元，低于预期的0.50美 元 。净利润每股仅为0.17美元，较去年同期的0.66美元大幅下滑。 疲软指引引发关注 ： 第四季度指引为营收56-64亿美元，远低于市场预期的68.2亿美元 ；调整后每股收益0.40-0.50美 元，低于预期的0.69美元。 CEO查尔斯·梁(Charles Liang)在财报电话会议上指出： 我们看到客户在当前的Hopper和即将推出 ...