Microsoft Unveils Maia 300 AI Chip to Reduce Reliance on Nvidia GPUs

Image: Illustrative image · prashanth dara · CC BY-SA 3.0 · Source

Microsoft announces its third-generation Maia 300 AI chip, set to become the primary AI inference system by 2027, offering significant efficiency improvements over GPU-based systems.

Microsoft has introduced its third-generation in-house AI chip, Maia 300, designed to serve as the primary AI inference system powering the company's products and services. Scheduled for production rollout in 2027, Maia 300 focuses on enhancing cost-efficiency while handling advanced AI workloads, including long-context reasoning, latency-sensitive conversational tasks, and complex, multi-step agentic processes.

According to Scott Guthrie, executive vice president of Microsoft's cloud group, Maia 300 aims to deliver a 40 percent increase in tokens processed per megawatt compared to leading GPU-based systems, achieving up to seven times the rack-scale token throughput of its predecessor, Maia 200, in a full 96-accelerator configuration. These gains are attributed in part to a new chiplet architecture implemented with TSMC's N2P process node, which separates compute, memory controller, and I/O functions to optimize silicon area dedicated to computation and allow each function to utilize the most suitable technology.

Maia 200, launched earlier in the same year, currently supports production AI workloads for Microsoft offerings like Microsoft Copilot and GitHub Copilot and has been recognized as the company's best performing AI infrastructure. Maia 300 is expected to extend these capabilities further, enabling Microsoft to scale multi-gigawatt Azure Maia deployments and reduce dependence on GPU providers like Nvidia and AMD.

Mustafa Suleyman, CEO of Microsoft AI, highlighted the strategic importance of co-designing hardware with AI models, stating that owning the platform down to the silicon level is critical for the company’s goal to compete with AI leaders such as Google DeepMind, OpenAI, and Anthropic. Maia 300 will play a vital role in running Microsoft’s custom AI models efficiently, signaling a significant shift towards proprietary AI hardware within the company’s ecosystem.

With these advancements, Microsoft positions itself to control more of its AI infrastructure stack, aiming for greater efficiency and performance in cloud AI services by the middle of the decade. While exact application details and broader deployment strategies remain limited, the Maia 300 represents a substantial step in Microsoft’s in-house AI chip development roadmap.