Microsoft Launches Maia 200 AI Chip to Cut Nvidia Reliance

On January 26, 2026, Microsoft unveiled the Maia 200, its second-generation in-house AI inference accelerator. Fabricated by TSMC using a 3-nanometer process, the chip features over 140 billion transistors and integrated SRAM to improve throughput and memory efficiency for large-scale reasoning models. The hardware is designed to reduce the company's reliance on Nvidia and lower the cost of AI token generation, offering a 30% improvement in performance per dollar compared to Microsoft's existing fleet.

Deployment began at a data center in Des Moines, Iowa, with subsequent expansion planned for the US West 3 region near Phoenix, Arizona. The Maia 200 will power Microsoft 365 Copilot, Microsoft Foundry, and OpenAI's GPT-5.2 models. To support the hardware, Microsoft introduced a new software development kit and programming tools, including the open-source Triton tool developed with contributions from OpenAI.

Microsoft executives claim the chip outperforms competitors, delivering three times the FP4 performance of Amazon's third-generation Trainium and exceeding the FP8 performance of Google's seventh-generation TPU. While the company has already begun designing a Maia 300 successor, CEO Satya Nadella stated that Microsoft will maintain a dual-track strategy, continuing to purchase hardware from Nvidia and AMD to meet growing computational demands. The launch coincides with a 40% year-on-year increase in Azure and other cloud services revenue for the first quarter of fiscal year 2026.