Microsoft unveils Maia 300 chip to lower reliance on NVIDIA
What's the story
Microsoft has unveiled its third-generation Maia in-house AI chip, the Maia 300. The new chip is designed to be the "primary AI inference system" for Microsoft products and services. It promises improved cost-effectiveness and efficiency for running cloud-based AI workloads, while being more capable of handling advanced reasoning models and agentic tasks.
Deployment
It will be in production starting 2027
The Maia 300 chip is set to be deployed in production starting 2027.
Scott Guthrie, Microsoft's executive vice president of cloud group, said, "We expect Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems."
In a full 96-accelerator configuration, it can provide up to seven times the rack-scale token throughput of its predecessor, Maia 200.
AI integration
Microsoft's own AI model work will benefit from Maia 300
The Maia 300 chip will play a key role in Microsoft's own AI model work.
Mustafa Suleyman, CEO of Microsoft AI, said, "We need a platform that we own, and that we co-design with our models."
He stressed the importance of running their Microsoft AI models on top of the Maia chip as a critical focus for their team.
Chip design
Chiplet architecture helps Maia 300 achieve efficiency gains
The Maia 300 uses a new chiplet architecture to achieve its efficiency gains.
Built on TSMC's N2P process node, the latest design separates compute, memory controller, and I/O functions.
"This increases the silicon area dedicated to compute performance and allows each function to use the technology best suited to its role," according to the Azure hardware and infrastructure team.
Service enhancement
The chip will also enhance Microsoft's Azure cloud
The Maia 300 will also enhance Microsoft's Azure cloud, which runs a series of models for clients.
It will improve the performance of Microsoft 365 Copilot and GitHub Copilot, both offering AI-powered services to software engineers and office workers.
Suleyman said, "Maia 300 is a complete AI system designed for the new types of AI workloads emerging today."