Microsoft's Maia 300 cuts cloud AI power use, targeting 2027 rollout to reduce Nvidia reliance

Microsoft will transition its primary AI inference workloads to its in-house Maia 300 chip starting in 2027, significantly reducing dependence on Nvidia and AMD graphics processing units for cloud services. The third-generation accelerator is designed to handle emerging reasoning models and agentic tasks while improving cost efficiency across the Azure infrastructure.

How it unfolded

• Microsoft announced the Maia 300 chip today as its next-generation AI hardware solution.
• The predecessor, Maia 200, was announced in January and is already running production workloads.
• Production deployment of the new Maia 300 architecture is scheduled to begin in 2027.

Who is affected

Organisations using Microsoft Azure for AI inference will eventually see their workloads run on this custom silicon rather than third-party GPUs. Scott Guthrie, executive vice president of Microsoft’s cloud group, stated that Maia 300 will deliver 40 percent more tokens per megawatt compared to leading GPU-based systems. This efficiency gain applies specifically to cloud AI workloads.

The chip is engineered for specific modern AI demands, including long-context reasoning, latency-sensitive conversational experiences, and sustained multi-step agentic tasks. According to The Verge, the hardware is built on TSMC’s N2P process node and uses a chiplet architecture that separates compute, memory controller, and I/O functions. This design increases the silicon area dedicated to compute performance.

Maia 200, which was announced in January, is currently described by Guthrie as the best-performing AI infrastructure in Microsoft’s fleet based on real-world service metrics. It already supports Microsoft Copilot, GitHub Copilot, Microsoft Foundry, and internal AI models. The new Maia 300 aims to expand these capabilities with up to seven times the rack-scale token throughput of its predecessor in a full 96-accelerator configuration.

Our take on the response

Microsoft’s move to scale the Azure Maia family to multi-gigawatt deployments indicates a strategic shift toward hardware independence for AI inference. By positioning Maia as the primary system for its products, the company seeks to compete more directly with rivals like Google DeepMind, OpenAI, and Anthropic. This suggests that managed service providers should monitor Azure hardware updates closely, as the underlying infrastructure for client AI workloads will change significantly over the next few years.

What to do and how to stay safe: Microsoft

• Review current Azure AI workload configurations to understand which services might migrate to Maia-based infrastructure.
• Monitor Microsoft documentation for performance benchmarks specific to Maia 300 to adjust cost projections for 2027 deployments.
• Assess whether client applications relying on specific GPU features need architectural updates for the new chiplet design.
• Track announcements regarding the transition timeline to plan capacity management for the shift from Maia 200 to 300.

Step-by-step guide: How Smartphone Security Updates Work: The Hidden Chain

General security guidance from the Patch Observer newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

When will the Maia 300 chip be available for production use?

Microsoft states that the Maia 300 chip will start rolling out to production AI workloads in 2027.

How does Maia 300 compare to GPU-based systems in efficiency?

Microsoft expects Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems.

What AI tasks is the Maia 300 chip designed to handle?

The chip is designed for long-context reasoning, latency-sensitive conversational experiences, and sustained multi-step agentic tasks.