Microsoft has announced that its Azure Maia 200 chip is now in production, with its successor, Maia 300, slated for deployment in 2027.

In a blog post discussing the company’s in-house AI chips, Scott Guthrie, EVP, cloud and AI, said Microsoft was planning to scale its Azure Maia family to “multi-gigawatt deployments” starting next year, making the hardware the “primary inference system” for its products and services.

Maia 200 is built using TSMC's 3nm technology and delivers around ten petaflops of FP4 and five petaflops of FP8 compute performance within a 750W SoC (System-on-Chip) TDP (Thermal Design Power) envelope. The chip has a “redesigned memory system” comprising 216GB HBM3e at 7Tbps and 272MB of on-chip SRAM, alongside data movement engines.

At present, Maia 200 has been deployed across the US, Europe, and Australia and is supporting workloads including Microsoft Copilot, OpenAI models, and Microsoft’s own MAI models, Guthrie said.

Meanwhile, he described Maia 300 as a “complete intelligent system,” designed for “new types of AI workloads,” such as long-context reasoning, latency-sensitive conversational experiences and sustained, multi-step agentic tasks.

Maia 300 racks are comprised of 96 accelerators with a 167kW power envelope, delivering 27.6TB of high-bandwidth memory (HBM) capacity and 1.7Pbps of aggregate HBM bandwidth across high-speed scale-up domains. Built using TSMC’s N2P process node, a new multi-chiplet architecture separates compute, memory-controller, and I/O functions to improve efficiency, scalability, and cost structure.

According to Microsoft, Maia 300 is expected to deliver up to 40 percent more tokens per megawatt compared to leading GPU-based systems and, in a full 96-accelerator configuration, up to seven times the rack-scale token throughput of Maia 200.

“Experience running our AI models and services on Maia 200 directly informed the design of Maia 300,” Guthrie said. “Together, the two generations are designed to support a broad set of additional Microsoft AI services.”