Microsoft's 'biggest bet' on its own silicon: Azure Maia AI chips

Microsoft plans to scale its Azure Maia AI chip family to multi-gigawatt deployments from 2027. CEO Satya Nadella introduced the Maia 300 chip, expected to deliver 40% more tokens per watt than leading GPU-based systems, calling it a major bet.

Microsoft will scale its in-house Azure Maia AI chip family to multi-gigawatt deployments from 2027, aiming to make it the primary system for running artificial intelligence models across its products and services, as the technology major steps up efforts to improve AI performance and energy efficiency.

Microsoft's 'Biggest Bet' on In-House Silicon

Microsoft Chief Executive Officer Satya Nadella announced the plan in a post on X, describing it as the company’s “biggest bet yet on our own silicon”. He also introduced Maia 300, which the company expects to deliver 40 per cent more tokens per watt than leading GPU-based systems.

The expansion builds on Maia 200, which is already being deployed across the United States, Europe and Australia. Nadella said the chip was delivering up to 1.7 times more tokens per watt across experiences in Microsoft Teams and Copilot.

Current Production and Workloads

In a separate blog post, Microsoft Executive Vice President for Cloud and AI Scott Guthrie said Maia 200 was serving production workloads, including Microsoft Copilot. GitHub Copilot and Microsoft Foundry are expected to follow by the end of the year.

Maia 300: Designed for Demanding AI Workloads

The chips are designed to run AI models more efficiently by coordinating hardware, software, memory and networking. Microsoft said experience from operating its AI services on Maia 200 had informed the development of Maia 300. The new system is designed for increasingly demanding AI workloads, including long-context reasoning, conversational applications and multi-step tasks performed by AI agents.

Technical Specifications and Performance

Microsoft expects Maia 300 to deliver up to seven times the rack-scale token throughput of Maia 200 in a full 96-accelerator configuration. The company said Maia 300 systems would combine 96 accelerators within a 167-kilowatt power envelope, with 27.6 terabytes of high-bandwidth memory capacity and aggregate memory bandwidth of 1.7 petabytes per second.

Maia 300 is expected to begin production deployments in 2027. Microsoft said the system uses a multi-chiplet design, with its compute chiplets built on Taiwan Semiconductor Manufacturing Company's N2P process technology.

Broader AI Infrastructure Strategy

The company said its broader strategy was to improve efficiency across the entire AI infrastructure stack while continuing to incorporate innovations from the wider technology ecosystem. As demand for AI services expands, Microsoft is positioning its in-house chips as a central part of its infrastructure strategy, with the planned rollout aimed at supporting larger workloads and more AI interactions within available power capacity. (ANI)

(Except for the headline, this story has not been edited by Asianetnews Editorial staff and is published from a syndicated feed.)