公众号记得加星标⭐️，第一时间看推送不会错过。 来源 ： 内容来自 编译自morethanmoore 。 AMD 和 OpenAI 今日宣布了一项计划，OpenAI 将通过关联公司购买并部署高达 6 千兆瓦的 AMD Instinct GPU，涵盖多代硬件，从 2026 年下半年的 MI450 系列开 始，潜在销售额可达 900 亿美元。作为交换，AMD 已向 OpenAI 发行认股权证，以每股 0.01 美元的价格认购最多 1.6 亿股 AMD 股票。随着 OpenAI 的部署规模从 1 千兆瓦扩大到 6 千兆瓦，以及 AMD 股价突破一系列业绩门槛（最高可达每股 600 美元），认股权证将逐步归属。 对于那些计算的人来说，如果 OpenAI 以每股 600 美元的价格出售所有股份，那么 1.6 亿股将价值 960 亿美元，大致相当于交易中硬件的价值。 AMD 宣布与 OpenAI 建立战略合作伙伴关系，这不仅代表着又一项 GPU 供应协议，更是对 AI 基础设施融资和配置方式的结构性重构。表面上，这只是一项供应协 议。但其本质可以被视为一种金融工具，将 AI 硬件销售转化为股权配置，将 AMD 的长期估值 ...

Investment Rating - The report provides a positive outlook for AMD, indicating an outperform rating based on expected revenue growth and strong performance in key segments [19]. Core Insights - AMD reported FY2Q25 revenue of $7.69 billion, exceeding market expectations of $7.43 billion, with a Non-GAAP EPS of $0.48, aligning with consensus [6][1]. - The growth was primarily driven by strong sales of EPYC and Ryzen processors, which offset the negative impact of export controls on Instinct sales [6][1]. - The company anticipates significant revenue contributions from the MI400 series and Helios platform starting in 2026, representing a multi-billion dollar opportunity [8][3]. Revenue Breakdown - For 2Q25, revenue by segment was as follows: - Data Center revenue: $3.2 billion, up 14% YoY, with an operating loss of $155 million due to inventory costs [6][1]. - Client and Gaming revenue: $3.6 billion, up 69% YoY, with an operating profit of $767 million and an operating profit margin (OPM) of 21% [6][1]. - Embedded revenue: $824 million, down 4% YoY, with an operating profit of $257 million and an OPM of 33% [6][1]. Future Outlook - For 3Q25, AMD expects revenue between $8.4 billion and $9.0 billion, with a midpoint of $8.7 billion, representing a 28% YoY increase, primarily driven by high double-digit growth in the Data Center segment [10][4]. - The Non-GAAP gross margin is projected to be 54%, with operating expenses estimated at $2.55 billion [10][4]. AI Business Growth - AMD's AI business continues to show strong growth, particularly with the MI350 series accelerators, which have demonstrated competitive performance in training and inference workloads [7][2]. - The MI355 model has entered mass production ahead of schedule and is expected to significantly contribute to Data Center growth in the second half of the year [7][2]. Product Development - The MI400 series is on track for development, with expected AI performance reaching 40 PFLOPs and leading in memory and bandwidth metrics [8][3]. - The Helios platform, designed for complex AI tasks, is anticipated to enhance performance significantly for advanced models [8][3]. Shipment Delays - The MI308 product is currently under review by the U.S. Department of Commerce, causing delays in shipments, with limited output expected in 3Q25 [9][3].