Microsoft will scale its in-house Azure Maia artificial intelligence chips to multi-gigawatt deployments from 2027, as the technology giant looks to make its own silicon the primary infrastructure for running AI workloads across its products and services.
The company on Monday announced its third-generation AI accelerator, Maia 300, and said it will begin production deployments next year. The move marks a significant expansion of Microsoft’s custom silicon programme as demand for AI inference—running trained models to generate responses—continues to drive computing and power requirements.
“As we move into the era of Super Intelligence, Microsoft is entering a new phase of infrastructure,” Scott Guthrie, executive vice president, Cloud + AI at Microsoft, said in a blog post. “We will scale our Azure Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for Microsoft products and services.”
Microsoft Maia 300: What the new AI chip is designed for
Microsoft said its current-generation Maia 200 is already serving production workloads, including Microsoft Copilot. The chip is also being used by Microsoft’s MAI team to develop next-generation models, while GitHub Copilot and Microsoft Foundry are expected to follow by the end of the year.
Maia 200 is deployed across the US, Europe and Australia and runs OpenAI models, Microsoft’s own MAI models, and leading open-source model families. Microsoft said the chip delivers up to 1.7 times more tokens per watt for built-in AI capabilities in Teams, Copilot and Cowork compared with leading GPU infrastructure.
The new Maia 300 is designed primarily for inference workloads, with Microsoft targeting long-context reasoning, latency-sensitive conversational applications and sustained, multi-step agentic tasks. “Maia 300 is a complete intelligent system designed for the new types of AI workloads emerging today,” Guthrie said.
Microsoft targets higher AI performance per megawatt
Microsoft expects Maia 300 to deliver 40% more tokens per megawatt than leading GPU-based systems. In a full 96-accelerator configuration, the company said the system can deliver up to seven times the rack-scale token throughput of Maia 200.
The chip uses a multi-chiplet architecture that separates compute, memory-controller, and input/output functions. Its compute chiplets are built using TSMC’s N2P process technology. The system also includes eight HBM4 memory stacks capable of 18.0 TB/s of peak memory bandwidth within a 950-watt thermal design power envelope.
Microsoft said the design is intended to reduce data movement, which it described as an increasingly important bottleneck in large-scale AI inference systems. Its software-defined dataflow architecture and AI-assisted kernel optimisation are designed to improve utilisation across compute, memory, and networking resources.
Maia 300 to power multi-gigawatt AI infrastructure
At rack scale, 96 Maia 300 accelerators operate within a 167-kilowatt power envelope, providing 27.6 TB of HBM capacity and 1.7 petabytes per second of aggregate HBM bandwidth across the scale-up domain.
Microsoft said Maia 300 will allow it to serve more simultaneous AI interactions and larger workloads from each rack, improving datacentre capacity and the economics of expanding AI services.
“AI is becoming a foundational layer of every application, workflow and business process,” Microsoft said, adding that enterprises are moving beyond chat-based experiences towards systems of agents capable of reasoning, acting and collaborating.
Why Microsoft is building its own AI chips
Microsoft’s push into custom silicon comes as hyperscalers increasingly develop their own AI accelerators to improve control over performance, power consumption and infrastructure costs.
However, the company has not indicated that Maia will replace GPUs across its entire AI infrastructure. Instead, it describes its strategy as an open, heterogeneous and model-diverse platform in which model and silicon choice remain important.
Maia 300 production deployments are scheduled to begin in 2027. Microsoft is expected to provide further details on its AI infrastructure strategy at its Ignite conference in November.