Microsoft to Scale In-House AI Chips to Multi-Gigawatt Deployments from 2027
Microsoft CEO Satya Nadella announces massive scale-up of in-house Azure Maia AI chips to boost efficiency and performance by 2027.
Microsoft is making a massive move to dominate the artificial intelligence landscape by scaling its proprietary Azure Maia AI chip family. The technology giant aims to reach multi-gigawatt deployments by 2027, positioning these in-house silicon solutions as the primary engine for running AI models across its entire ecosystem of products and services.
Microsoft CEO Satya Nadella described this strategic shift as the company’s “biggest bet yet on our own silicon,” signaling a deep commitment to reducing reliance on external hardware providers while maximizing energy efficiency and performance.
The Leap to Maia 300: Efficiency Reimagined
As part of this ambitious roadmap, Microsoft has introduced the Maia 300. This next-generation chip is engineered to tackle the increasingly heavy demands of modern AI, including long-context reasoning, complex conversational applications, and multi-step tasks handled by autonomous AI agents.
The performance leap is significant. Microsoft expects the Maia 300 to deliver 40 per cent more tokens per watt than current leading GPU-based systems. In a full 96-accelerator configuration, the Maia 300 is projected to provide up to seven times the rack-scale token throughput compared to its predecessor, the Maia 200.
Technical Powerhouse: Memory and Architecture
To support these massive workloads, the Maia 300 systems are designed with cutting-edge specifications:
- Power Management: 96 accelerators contained within a 167-kilowatt power envelope.
- Memory Capacity: 27.6 terabytes of high-bandwidth memory.
- Bandwidth: An aggregate memory bandwidth of 1.7 petabytes per second.
- Manufacturing: A multi-chiplet design utilizing Taiwan Semiconductor Manufacturing Company's (TSMC) N2P process technology.
Current Progress: From Copilot to Global Deployment
While the massive scale-up is set for 2027, the foundation is already being laid. The Maia 200 is currently being deployed across the United States, Europe, and Australia. According to Scott Guthrie, Microsoft Executive Vice President for Cloud and AI, the Maia 200 is already serving production workloads, including Microsoft Copilot.
The real-world impact is already visible in Microsoft’s existing tools. Nadella noted that the Maia chip architecture is delivering up to 1.7 times more tokens per watt across user experiences in Microsoft Teams and Copilot. Looking ahead, GitHub Copilot and Microsoft Foundry are expected to integrate these advancements by the end of this year.
By coordinating hardware, software, memory, and networking, Microsoft is building an integrated AI infrastructure stack. As the global demand for AI services continues to surge, this in-house silicon strategy is designed to support larger workloads and more frequent AI interactions within the constraints of available power capacity.