内容来源：https://www.geekwire.com/2026/microsoft-makes-biggest-bet-yet-on-its-own-silicon-unveils-new-ai-chip-to-reduce-reliance-on-nvidia/

内容总结：

微软计划将大部分人工智能服务运行在自主研发的芯片上，这一转变有望为公司节省数十亿美元，并减少对行业巨头英伟达的依赖。

微软于周一发布了Maia 300，这是其自研AI芯片的第三代产品，并称将于2027年开始大规模部署。Azure负责人斯科特·格思里在博客文章中表示，Maia将成为“微软产品和服务的主要AI推理系统”。微软CEO萨提亚·纳德拉在X平台上发帖称，这是“我们在自研芯片上迄今最大的赌注”。

据微软介绍，在相同功耗下，Maia 300的AI输出量比以英伟达为主导的GPU系统高出40%。公司称，一整机架的新芯片AI输出量可达当前世代的七倍。该芯片专为增长最快的任务类型设计，例如可处理多步骤工作的AI智能体，以及处理长文档或长对话的模型。

这家总部位于雷德蒙德的公司负责芯片设计，并委托台积电代工制造，而非从英伟达购买成品芯片。更高效的芯片还能让微软以更少的电力运行AI服务，而当前电力供应已成为AI扩张的最大瓶颈之一。

微软与英伟达仍是重要合作伙伴。纳德拉与英伟达CEO黄仁勋上周同台发布了基于英伟达芯片的Windows PC。英伟达仍是微软数据中心的关键供应商，纳德拉在7月财报电话会上表示，微软将是最早部署基于英伟达Vera Rubin和AMD Helios下一代系统的云服务商之一。

微软此举是科技巨头自研AI芯片大趋势的一部分。谷歌已开始直接向部分客户销售其张量处理单元，亚马逊凭借Trainium芯片赢得了包括Anthropic在内的AI公司青睐。Meta已推出自研AI芯片，OpenAI也在6月发布了与博通合作打造的首款定制芯片。

微软一直在追赶。与谷歌和亚马逊的芯片不同，Maia并不对外开放租用。微软首款Maia芯片于2023年发布，经历延期后，第二代Maia 200于今年1月推出。据The Information八月报道，截至今年夏季，Maia 200仅在美国两个数据中心运行。

当前一代Maia 200已为Microsoft 365 Copilot提供算力，并运行OpenAI及微软自研模型。据微软介绍，该芯片目前正在美国、欧洲和澳大利亚部署。GitHub Copilot和Microsoft Foundry计划于年底前迁移至该平台。

这一转变需要时间。据CNBC报道，Evercore ISI分析师估计，到2027年初，Maia将占微软AI芯片支出的约10%。

微软表示将在下月于旧金山举行的Ignite大会上分享更多细节。

中文翻译：

微软计划将大部分人工智能服务运行在自家设计的芯片上，这一转变有望为公司节省数十亿美元，并减少其对行业巨头英伟达的依赖。

该公司周一发布了Maia 300，这是其自研人工智能芯片的第三代产品，并称将于2027年开始大规模部署。Azure负责人斯科特·格思里在博客文章中表示，Maia将成为“微软产品和服务的主要人工智能推理系统”。

微软首席执行官萨提亚·纳德拉在X平台上发帖称，这是“我们在自研芯片上迄今最大的赌注”。

据该公司称，在相同功耗下，Maia 300的人工智能输出量比英伟达主导的领先GPU系统高出40%。

公司表示，一整机架的新芯片的人工智能输出量将达到当前世代的七倍。这些芯片专为增长最快的任务类型而设计，例如执行多步骤任务的AI智能体，以及处理长文档或长对话的模型。

这家雷德蒙德公司负责设计芯片，并依赖台积电代工制造，而非从英伟达购买成品芯片。更高效的芯片也让微软能以更少电力运行人工智能服务，而当前电力供应已成为人工智能扩张的最大制约因素之一。

微软与英伟达仍是重要合作伙伴。纳德拉和英伟达首席执行官黄仁勋上周同台发布了搭载英伟达芯片的Windows个人电脑。英伟达也仍是微软数据中心的关键供应商，纳德拉在7月的财报电话会上表示，微软将是最早部署基于英伟达Vera Rubin和AMD Helios下一代系统的云服务商之一。

微软此举是科技巨头自研人工智能芯片大趋势的一部分。谷歌已开始直接向部分客户销售其张量处理单元，亚马逊则凭借Trainium芯片赢得了包括Anthropic在内的人工智能公司的青睐。Meta已推出自研人工智能芯片，OpenAI也在6月发布了与博通合作打造的首款定制芯片。

微软一直在追赶。与谷歌和亚马逊的芯片不同，Maia并不提供给外部客户租用。微软于2023年发布首款Maia芯片，经历延期后于今年1月推出第二代Maia 200。据The Information八月报道，截至今年夏天，Maia 200仅在两个美国数据中心运行。

当前一代Maia 200已为Microsoft 365 Copilot提供算力，并运行OpenAI及微软自研的模型。据微软称，该芯片目前正在美国、欧洲和澳大利亚部署。GitHub Copilot和Microsoft Foundry将于年底前迁移至该平台。

这一转变需要时间。据CNBC报道，Evercore ISI分析师估计，到2027年初，Maia将占微软人工智能芯片支出的约10%。

微软表示，将于下月在旧金山举行的Ignite大会上分享更多细节。

英文来源：

Microsoft plans to run most of its AI services on chips it designs itself, a shift that could save the company billions of dollars and reduce its dependence on industry juggernaut Nvidia.
The company on Monday unveiled Maia 300, the third generation of its custom AI chip, saying it will begin large-scale deployments in 2027. In a blog post, Azure chief Scott Guthrie said Maia will become “the primary AI inference system for Microsoft products and services.”
Microsoft CEO Satya Nadella called it “our biggest bet yet on our own silicon” in a post on X.
Maia 300 will produce 40% more AI output for the same amount of power as leading GPU-based systems, which Nvidia dominates, according to the company.
A full rack of the new chips will produce up to seven times as much AI output as the current generation, the company says. They’re designed for the kinds of tasks that are growing fastest, such as AI agents that work through multistep jobs and models that handle long documents or conversations.
The Redmond company designs the chips and relies on Taiwan Semiconductor Manufacturing Co. (TSMC) to make them, rather than buying finished chips from Nvidia. More-efficient chips also let Microsoft run AI services using less electricity, at a time when power supply has become one of the biggest constraints on AI expansion.
Microsoft and Nvidia remain big partners. Nadella and Nvidia CEO Jensen Huang shared the stage last week to launch Windows PCs built on Nvidia chips. Nvidia also remains a key supplier for Microsoft’s data centers, and Nadella said on the company’s July earnings call that Microsoft will be among the first cloud providers to deploy next-generation systems based on Nvidia’s Vera Rubin and AMD’s Helios.
Microsoft’s move is part of a larger trend among tech giants to design their own AI chips. Google has started selling its Tensor Processing Units directly to some customers, and Amazon has won over AI companies including Anthropic with its Trainium chips. Meta has rolled out its own AI chips, and OpenAI unveiled its first custom chip, built with Broadcom, in June.
Microsoft has been playing catch-up. Unlike Google’s and Amazon’s chips, Maia isn’t available for outside customers to rent. Microsoft’s first Maia chip, announced in 2023, was followed by a second-generation Maia 200 this January after delays. As of this summer, Maia 200 was running in just two U.S. data centers, The Information reported in August.
Maia 200, the current generation, already powers Microsoft 365 Copilot and runs models from OpenAI as well as Microsoft’s own. It’s now being deployed across the U.S., Europe and Australia, according to Microsoft. GitHub Copilot and Microsoft Foundry are set to move onto it by the end of the year.
The shift will take time. Evercore ISI analysts estimate that Maia will account for about 10% of Microsoft’s spending on AI chips by early 2027, CNBC reported.
Microsoft says it will share more details next month at its Ignite conference in San Francisco.