Overview
- Microsoft unveiled Maia 300 on Monday, October 12, and said it will scale the Azure Maia family to multi‑gigawatt deployments starting in 2027 to run most of its inference workloads.
- The Maia 300 uses a multi‑chiplet design with compute chiplets built on TSMC’s 2nm N2P process and eight HBM4 stacks, and Microsoft says a 96‑accelerator rack will draw about 167 kilowatts and deliver roughly 40% more tokens per megawatt than leading GPU‑based systems.
- Microsoft estimates rack‑level throughput up to seven times that of Maia 200 and plans large volume sourcing talks with TSMC that could involve hundreds of thousands to more than a million chips, according to reporting.
- Company performance claims are based on internal benchmarks and await independent verification as real‑world deployments begin, and Microsoft will continue to rely on Nvidia and AMD hardware for model training and other workloads.
- If Maia 300 hits its targets, Microsoft could cut data‑center operating costs and speed services such as Copilot and GitHub Copilot, but the actual customer impact will depend on manufacturing capacity, deployment pace, and validation of the efficiency gains.