Microsoft to launch Maia 300 inference chip in 2027

Microsoft said it will launch its Azure Maia 300 in 2027 in multi-gigawatt data centers. When it scales, Microsoft's Azure Maia family of chips will be the go-to inference processor for Microsoft's products.

In a blog post, Microsoft said Azure Maia 200 is in production for Microsoft Copilot with GitHub Copilot and Microsoft Foundry n deck. Maia 200 also runs OpenAI and Microsoft AI models.

According to Microsoft, Maia 300 is designed for long-context reasoning, low latency and multi-step AI agent tasks. Like other hyperscalers, Microsoft has developed its own processors to lower AI costs.

• Microsoft launches Maia 200 as custom AI silicon accelerates

"We expect Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems. And in a full 96-accelerator configuration, it delivers up to 7x the rack-scale token throughput of Maia 200," said Scott Guthrie, executive vice president of cloud and AI at Microsoft.

Microsoft Maia 300 includes the following:

• A software-defined dataflow architecture and AI-assisted kernel optimization to reduce data movement.
• Higher memory bandwidth.
• Chiplet architecture on TSMC's N2P process node.
• Lower memory and bandwidth requirements.
• A design that integrates networking as part of the software-defined dataflow architecture.