Microsoft Announces Maia 300 AI Chip with TSMC N2P and HBM4
Microsoft has officially announced the Maia 300 AI chip, built on TSMC's N2P 2nm process with 288GB of HBM4 memory, delivering 18TB/s bandwidth and up to 13.8 PFLOPS of FP4 compute. This announcement marks a major hardware advancement for hyperscalers building custom silicon to reduce reliance on merchant hardware providers like NVIDIA. The chip's advanced memory architecture and massive bandwidth specifically target demanding AI inference workloads. The Maia 300 features a multi-chiplet design with a 950W SoC power rating, supporting bidirectional scaling interconnect bandwidth of 3.2TB/s and clustering up to 12,000 accelerators.
## BACKGROUND
TSMC's N2P is an advanced semiconductor manufacturing node extending their 2-nanometer family, while HBM4 represents the latest high-bandwidth memory standard designed to resolve critical memory bottlenecks in high-performance AI processors.