New Microsoft AI Chip Reduces Reliance on Nvidia GPUs
Microsoft today unveils the third generation of its Maia AI chip, designed as the 'primary AI inference system' for Microsoft's products and services. Maia 300 not only improves the cost and performance of running AI workloads in the cloud, but also handles the latest reasoning models and agentic tasks more capably. Maia 300 [...]
Posts in this thread will be added to your daily email digest and homepage feed.
The new Maia 300 chip will debut in 2027 to power AI workloads.
Posts from this author will be added to your daily email digest and home page feed.
Microsoft Maia 300 chip. Image: Microsoft
is a senior reporter and writer for Notepad, covering all things Microsoft, PC and technology for over 20 years.
Microsoft today announced the third generation of its Maia artificial intelligence chip, which is designed as the "primary artificial intelligence inference system" for Microsoft products and services. Maia 300 not only improves the cost and efficiency of running AI workloads in the cloud, but is also more capable of handling the latest reasoning models and agent tasks.
The Maia 300 will go into production in 2027 and is designed with performance in mind. "We expect the Maia 300 to deliver 40 percent more tokens per MW compared to leading GPU-based systems," said Scott Guthrie, executive vice president of Microsoft's cloud group. And in a full 96-accelerator configuration, it delivers up to 7x the rack-scale token throughput of the Maia 200.
These performance improvements will soon allow Microsoft to rely primarily on its own in-house AI chips instead of Nvidia and AMD chips. "We're going to grow our Azure Maia family into a multi-gigawatt deployment starting in 2027, and we're going to make Maia the primary AI inference system for Microsoft products and services," Guthrie says.
Maia 300 chips will be used for production loads in 2027. Image: Microsoft
The Maia 300 is the successor to the 200 which was introduced only in January. Maia 200 is currently running on production AI workloads across Microsoft, including Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and Microsoft's own AI models.
"The Maia 200 is the best AI infrastructure in our fleet as measured by real-world benchmarks for AI services," Guthrie said. Maia 300 is a complete AI system designed for the new types of AI workloads emerging today: long-context reasoning, latency-sensitive conversational experiences, and persistent multi-stage agent tasks.
Microsoft is finally making custom chips — and they're all about artificial intelligence
Microsoft's latest AI chip takes on Amazon and Google
Microsoft is using a new chiplet architecture to achieve its significant gains in productivity. Built on TSMC's N2P process node, the latest Maia 300 architecture separates compute, memory controller, and I/O functions. "This increases the silicon area dedicated to computing performance and allows each function to use the right technology for its role," says the Azure Hardware and Infrastructure team.
Maia 300 is also an important part of Microsoft's AI model work and could help the company reach its goal of catching up with Google DeepMind, OpenAI and Anthropic. We need to work across the whole stack, to the end,” says Microsoft AI CEO Mostafa Suleiman: “At this scale we need a platform that we own and design with our own models. That's why running Microsoft AI models on top of the Maia chip is a critical focus for our team.
A college in Illinois tried to cancel his contact with Flock and was denied
The GTA VI reveal continues with a lengthy (and very nude) gameplay video
Learning to use local AI is exciting, overwhelming, and frustrating
What to expect from Apple's "Welcome Home" event
A free daily digest of the news that matters most.
By providing your information, you agree to our Terms of Use and Privacy Policy. We use vendors who may also process your information to help provide our services. This site is protected by reCAPTCHA and Google's Privacy Policy and Terms of Service apply.