Microsoft Unveils Maia 300 AI Chip Built on TSMC's 2nm Process
Microsoft introduced its third-generation custom AI chip, the Maia 300, which utilizes a chiplet architecture and TSMC's 2nm process. The company plans to deploy over a million of these chips next year to support its internal AI infrastructure and Azure cloud services. This development highlights major cloud providers' ongoing efforts to reduce their heavy reliance on NVIDIA and lower the skyrocketing costs of AI compute infrastructure. By optimizing full-stack hardware systems, Microsoft aims to deliver higher energy efficiency and cost-effectiveness for heavy workloads like code generation and complex enterprise tasks. The Maia 300 features a chiplet design where the core computing logic is built on TSMC's 2nm nodes, while data transfer components use cost-effective 3nm nodes. A single server rack contains 96 chips with a power consumption of approximately 167 kilowatts, yielding roughly 40% higher effective computation per megawatt than mainstream GPU systems according to Microsoft.
## BACKGROUND
Chiplet architecture is an advanced semiconductor design approach that breaks a large monolithic processor into smaller, specialized modular dies to overcome manufacturing yield and cost constraints. TSMC's 2nm process marks a major milestone by transitioning from older FinFET transistors to Gate-All-Around Field-Effect Transistor (GAAFET) technology, significantly boosting energy efficiency and performance.