Product attributes
Other attributes
The Microsoft Azure Maia 100 AI Accelerator is a custom chip designed and built by Microsoft to run cloud-based training and inferencing for artificial intelligence (AI) workloads. The Maia 100 is the first generation of the series and contains 105 billion transistors. At the time of release, the Maia 100 was one of the largest chips built on 5nm process technology.
The Maia 100 is designed specifically for the Azure hardware stack, aligning chips with the larger AI infrastructure for optimized performance and efficiency. Microsoft announced new innovations across the software, network, racks, and cooling to meet the needs of new AI workloads. The Maia 100 runs on custom server boards and tailor-made racks that fit into existing Microsoft data centers and uses a liquid cooling system to prevent overheating while running intensive AI workloads.
Custom-built rack with liquid cooling for the Maia 100 AI accelerator.
Microsoft announced the Maia 100 AI Accelerator series on November 15, 2023, at Ignite 2023 in Seattle, the company's annual conference for developers and IT professionals. Microsoft has history in silicon development, collaborating on silicon for the Xbox and co-engineering chips for its Surface devices. Rani Borkar (head of Azure hardware systems and infrastructure at Microsoft) stated that the process behind these new chips began in 2017 when the company started architecting its cloud hardware stack.
The Maia 100 AI chip is built in-house at Microsoft but was designed with input from OpenAI (Microsoft's partner) to tailor the chips for running their large language models. Of the launch of the new chip, OpenAI CEO Sam Altman said:
Since first partnering with Microsoft, we’ve collaborated to co-design Azure’s AI infrastructure at every layer for our models and unprecedented training needs... We were excited when Microsoft first shared their designs for the Maia chip, and we’ve worked together to refine and test it with our models. Azure’s end-to-end AI architecture, now optimized down to the silicon with Maia, paves the way for training more capable models and making those models cheaper for our customers.
During the announcement, Microsoft stated the Maia 100 will roll out to its data centers in early 2024. Initially, the chips will power services such as Microsoft Copilot or Azure OpenAI Service. They will join third-party products Microsoft offers from industry partners including the Nvidia H100 and H200 Tensor Core GPUs and the AMD MI300X. Microsoft has stated the addition of first-party silicon will allow the company to a greater variety of products to customers, in terms of both pricing and performance.
During the Maia 100 announcement, Microsoft stated it is designing the second-generation Maia AI Accelerator chip.
