Microsoft Maia 200

Microsoft's second-generation in-house inference chip for Azure

N/A Datacenter
Visit Product Page →
216GB HBM3e
7000 GB/s bandwidth
3nm process

Microsoft unveiled Maia 200 in January 2026 as the successor to its first in-house Maia 100 chip, moving from TSMC’s 5nm node to 3nm and packing 144 billion transistors. It carries 216GB of HBM3e memory at 7 TB/s of bandwidth plus 272MB of on-chip SRAM, and delivers around 10 petaflops of FP4 and 5 petaflops of FP8 compute within a 750W envelope, with native FP8/FP4 tensor cores tuned for inference rather than training. Microsoft says it offers roughly 30 percent better performance per dollar than the prior generation. At Hot Chips 2026 in August, Microsoft gave its deepest public look yet at the chip’s architecture as it prepares to deploy Maia 200 across Azure data centers.