Microsoft Launches Maia 200 AI Chip on 3-nm, Deployed in US Data Centres

Microsoft launches its second-generation in-house AI chip, Maia 200, along with a new software stack. The chip goes live in a Microsoft data centre in Iowa and is deployed in US data centres to compete with Nvidia’s AI chips and CUDA platform.

Maia 200 AI Chip (Microsoft):

Dimension	Key Details
Developer	Maia 200 is developed by Microsoft as an in-house AI chip.
Generation	Maia 200 is Microsoft’s second-generation custom-built AI chip; original Maia is introduced in 2023.
Manufacturing	Taiwan Semiconductor Manufacturing Company (TSMC).
Process Technology	Built using 3-nanometer chipmaking technology.
Deployment	Live in a Microsoft data centre in Iowa; deployed in US data centres.
Deployment-Future	Deployment provides for rollout in Arizona.
Target Workloads	Designed for large-scale AI workloads such as chatbots and generative AI systems.
Rationale	Launch provides for reduced reliance on Nvidia and direct competition with Nvidia’s AI hardware and software ecosystem.
Memory Features	High-bandwidth memory (HBM) from an older generation than Nvidia’s latest designs and a large SRAM (Static Random Access Memory) capacity.
Software Stack	Microsoft bundles open-source software tools with Maia 200, including Triton; Triton is a programming framework with significant contributions from OpenAI and is designed to perform similar tasks to Nvidia’s CUDA.
Industry Trend	In-house AI chips are developed by major cloud players including Google and Amazon Web Services to compete with Nvidia.
Collaboration	Google has attracted interest from Meta Platforms, which is collaborating to narrow the software gap with Nvidia.