Microsoft to make Maia its primary AI inference chip from 2027
Microsoft says Azure Maia accelerators will reach multi-gigawatt deployments starting in 2027; the Maia 300 packs 288GB of HBM4 at 18TB/s.
Microsoft said it will scale its Azure Maia accelerators to multi-gigawatt deployments starting in 2027 and make Maia its primary inference system for its products and services, according to TechRadar. The company also detailed the Maia 300, the accelerator at the centre of that plan, in an architecture deep dive.
Maia 300 moves from a single die to a multi-chiplet package. Compute sits on TSMC's N2P process, while memory-controller and I/O functions are placed on separate dies. The accelerator offers eight HBM4 stacks delivering 288GB at 18TB/s, compared with 216GB at 7.1TB/s for Maia 200.
According to TechRadar, Maia 300 delivers roughly a 36% compute gain and 2.5x the memory bandwidth of Maia 200, and Microsoft claims a 40% efficiency bump. The chip uses enhanced MXFP4 and sub-3-bit storage formats with hardware conversion between them, plus software-managed on-die SRAM and a non-uniform memory layout across chiplets.
Production of Maia 300 is slated for 2027. Microsoft's announcement sets out an inference-focused roadmap that stretches from the current Maia 200 generation to multi-gigawatt scale.
Industry context
The move comes as cloud providers and AI companies expand their own silicon efforts. In the second quarter, Google began recognising revenue from TPU system sales delivered to customer data centres, CFO Anat Ashkenazi said on Alphabet's July earnings call. In April 2026, Amazon CEO Andy Jassy said Amazon's chip business spanning Trainium, Graviton and Nitro had passed a $20 billion annual revenue run rate. In the quarter to July 2026, Nvidia reported $96.2 billion in revenue, $89 billion of it from data centres, at a 75% gross margin. This week, Gartner forecast that AI providers will spend nearly $1 trillion on infrastructure in 2026 alone.
Quick answers
When will Microsoft's Maia 300 enter production?
Microsoft says production of the Maia 300 is slated for 2027, the same year it plans to begin scaling Azure Maia to multi-gigawatt deployments.
How does Maia 300 compare with Maia 200?
Maia 300 offers 288GB of HBM4 memory at 18TB/s, versus 216GB at 7.1TB/s for Maia 200, with roughly a 36% compute gain, 2.5x memory bandwidth and a claimed 40% efficiency bump.
What process does Maia 300 use?
The compute chiplets are built on TSMC's N2P process, with memory-controller and I/O functions on separate dies in a multi-chiplet package.