Overview
- Citing Monday's reporting, Microsoft is preparing to unveil the next-generation Maia 300 as soon as September, with the company still treating the timing as unconfirmed.
- Microsoft has held talks with Taiwan Semiconductor Manufacturing Company to secure roughly 300,000 Maia 300 units for delivery in 2027 and says it ultimately wants capacity for more than 1 million chips.
- The Maia 200, built on TSMC's 3‑nanometer process, is reported to be deployed inside Microsoft data centers in Iowa and Arizona and is running internal services including Microsoft 365 Copilot and OpenAI models.
- Microsoft has publicly claimed strong Maia 200 specs such as 216GB of HBM3e memory and support for FP8 and FP4 precision but independent benchmarks and broad Azure availability remain unconfirmed.
- The Maia program fits a wider trend of cloud providers building custom AI silicon to cut costs and control supply which could help Microsoft and its suppliers if production scales but will face limits from advanced-node, chip packaging and component shortages.