Microsoft launches Maia 200 AI chip for faster model inference

Microsoft Corp. (NASDAQ: MSFT) announced the launch of Maia 200, an AI accelerator chip designed for inference workloads. The chip is built on TSMC's 3-nanometer process and contains over 140 billion transistors.

The Maia 200 delivers over 10 petaFLOPS in 4-bit precision and over 5 petaFLOPS of 8-bit performance within a 750W power envelope. The chip features 216GB HBM3e memory at 7 TB/s bandwidth and 272MB of on-chip SRAM.

"Maia 200 is the most efficient inference system Microsoft has ever deployed, with 30% better performance per dollar than the latest generation hardware in our fleet today," said Scott Guthrie, Executive Vice President of Cloud + AI at Microsoft.

The chip will run multiple AI models including GPT-5.2 models from OpenAI across Microsoft Foundry and Microsoft 365 Copilot services. Microsoft's Superintelligence team will use the chip for synthetic data generation and reinforcement learning applications.

Microsoft has deployed Maia 200 in its US Central datacenter region near Des Moines, Iowa. The company plans to expand deployment to the US West 3 datacenter region near Phoenix, Arizona, with additional regions to follow.

The company introduced the Maia SDK preview, which includes PyTorch integration, a Triton compiler, and access to Maia's programming language. The development kit provides tools for building and optimizing models for the new chip.

Each Maia 200 system includes 2.8 TB/s of bidirectional bandwidth and supports clusters of up to 6,144 accelerators. The architecture uses standard Ethernet with a custom transport layer and integrated network interface card.

Microsoft stated that AI models were running on Maia 200 silicon within days of the first packaged parts arriving, with deployment time reduced by half compared to similar AI infrastructure programs.

You May Also Be Interested In