Microsoft Unveils Maia 300 AI Chip to Boost Cloud Inference Efficiency
Microsoft has revealed its third-generation Maia AI chip, marking a significant step in the company's efforts to build custom silicon for its cloud infrastructure. The Maia 300 is designed to function as Microsoft's "primary AI inference system" across its products and services.
The chip addresses both performance and efficiency concerns in AI inference workloads. According to Scott Guthrie, executive vice president of Microsoft's cloud group, the company expects Maia 300 to deliver 40 percent more tokens per megawatt compared to leading GPU-based systems. This efficiency gain could translate into meaningful cost savings at scale.
Beyond efficiency improvements, the Maia 300 has been specifically engineered to handle newer AI workloads, including reasoning models and agentic tasks. These capabilities position the chip to support more sophisticated AI applications that require multi-step reasoning rather than simple pattern matching.
The rollout is planned for 2027, indicating Microsoft's long-term commitment to developing proprietary AI hardware. By expanding its custom silicon portfolio, Microsoft aims to diversify its supply chain and reduce its reliance on third-party GPU vendors for inference workloads.