Microsoft announced the Maia 300, its latest custom-built artificial intelligence processor, positioning it as the company’s primary inference chip for handling AI workloads across its product portfolio. The announcement represents Microsoft’s continued effort to develop internal alternatives to graphics processing units from Nvidia and AMD.
The new chip delivers significant performance improvements over its predecessor. According to Scott Guthrie, executive vice president of Microsoft’s cloud division, Maia 300 achieves 40 percent greater token throughput per megawatt compared to leading GPU systems. In a full configuration of 96 accelerators, the chip provides up to 7 times the processing throughput of the earlier Maia 200 model. The company plans to begin rolling out Maia 300 to production operations in 2027 and scale deployments to multi-gigawatt levels starting that same year.
The chip incorporates a new chiplet design built on TSMC’s N2P process technology. This architecture separates computing, memory control, and input-output functions to optimize efficiency. By isolating compute functions into dedicated silicon, Microsoft says each component can employ technology specifically tailored to its purpose, maximizing overall performance and power efficiency.
Maia 300 is engineered to handle contemporary AI workloads including long-context reasoning, low-latency conversational experiences, and multi-step agentic tasks. The Maia 200, which was introduced earlier this year, already handles production workloads across Microsoft services including Copilot, GitHub Copilot, and Microsoft’s internal AI models. Mustafa Suleyman, Microsoft’s AI chief, emphasized that developing proprietary silicon allows the company to co-design hardware and software, which he described as essential for competing in AI development against rivals like Google DeepMind, OpenAI, and Anthropic.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI