微软发布第三代自研AI芯片 Maia 300：计划明年部署逾百万颗，每兆瓦算力较领先GPU系统高40%
雷德蒙德（华盛顿州）10月13日——据路透社报道，微软本周发布第三代自研 AI 芯片 Maia 300，并表示计划明年部署可能超过100万颗该芯片，为其内部 AI 服务提供算力。Maia 300 采用芯片粒（chiplet）设计：核心计算部分使用台积电2纳米制程，其他芯片粒使用成本较低的3纳米制程负责数据搬运。
微软称，Maia 300 每兆瓦可完成的有效计算量比「领先的 GPU 系统」高约40%。搭载该芯片的机架每架含96颗芯片、功耗约167千瓦；对比之下，英伟达最新 GB300 AI 服务器机架含72颗 GPU、功耗约142千瓦。微软 Azure 硬件系统与基础设施总裁拉尼·博卡尔（Rani Borkar）对路透社表示，这一设计是在整个系统层面优化，以每一瓦、每一美元换取更多智能产出。
背景：Maia 300 的上一代 Maia 200 已在运行微软自有 AI 模型以及通过 Azure 云服务交付的 OpenAI 模型。云计算巨头与 AI 公司纷纷自研芯片，以松动英伟达在 AI 计算市场的主导地位。
来源：路透社2026年10月12日报道（微软 Maia 300 发布）。
Microsoft Unveils Maia 300 AI Chip, Plans Deployment of More Than a Million Units Next Year
REDMOND, Wash., Oct. 13 — Microsoft has announced Maia 300, the third generation of its custom AI chip, and says it plans to deploy potentially more than a million of them next year to back its internal AI services, Reuters reported. The chip's central computing processor is composed of "chiplets," using Taiwan Semiconductor Manufacturing Co's 2-nanometer technology for key computing functions and cheaper 3-nanometer technology for moving data.
Microsoft said the Maia 300 will deliver about 40 percent more useful computing per megawatt than "leading GPU-based systems." Racks built around the chip will contain 96 processors and use about 167 kilowatts of power, compared with roughly 142 kilowatts for Nvidia's latest GB300 AI server racks with 72 GPUs. "It's basically delivering more intelligence for every watt, every dollar, and every byte of memory by optimizing across the entire system," Rani Borkar, president of Azure hardware systems and infrastructure, told Reuters.
The predecessor Maia 200 is already running Microsoft's in-house AI models as well as models from OpenAI delivered on the Azure cloud service. The new chip arrives amid a flurry of custom silicon from cloud rivals and AI firms aimed at loosening Nvidia's hold on the AI computing market.
Source: Reuters, October 12, 2026.
图片来源：半导体晶圆在洁净室加工设备中取出，CC BY 4.0（Wikimedia Commons，摄影/作者：Purdue Engineering）
