To put an end to rumors that Microsoft was working on a proprietary chip, the Redmond-based tech giant unveiled its own AI chip at the annual Ignite Conference. The Maia AI Accelerator, which competes with Nvidia’s AI graphics processing units, and the Cobalt CPU are both custom-designed processors and integrated systems.
The Microsoft Azure Maia AI Accelerator is designed for AI tasks and generative AI, whereas the Microsoft Azure Cobalt CPU is an Arm-based processor designed to execute general-purpose compute workloads on the Microsoft Cloud.
Microsoft claims that the chips are the final component needed for the company to deliver infrastructure systems, which range from servers, racks, and cooling systems to silicon selections, software, and entire systems that have been meticulously designed and optimized with internal and external workloads in mind.
Executive vice president of Microsoft’s Cloud + AI Group Scott Guthrie stated, “Microsoft is reimagining every aspect of our datacenters to meet our customers’ needs, and we are building the infrastructure to support AI innovation.” “To improve performance, diversify our supply chain, and give customers infrastructure option, it’s critical for us to optimize and integrate every layer of the infrastructure stack at the scale we operate.”
The chips will initially power Microsoft’s services, like Microsoft Copilot and Azure OpenAI Service, and will be deployed to the company’s datacenters. In order to help fulfill the rapidly growing need for effective, scalable, and sustainable computational capacity as well as the needs of customers ready to take advantage of the most recent advancements in cloud and artificial intelligence, Microsoft said they will complement an expanding selection of products from industry partners.
Microsoft has been working on the chip’s development for years, albeit not so discreetly as the rumor mill was aflutter with the revelation. A lab full of machines was at the company’s Redmond location, painstakingly testing and working on the silicon.
A tech behemoth has previously announced a customized chip. Google revealed their AI Tensor Processing Unit in 2016. With the release of its Inferentia AI processor and Graviton Arm-based chip in 2018, as well as the announcement of Trainium for model training in 2020, Amazon Web Services entered the fray.