Microsoft’s latest AI chip goes head-to-head with Amazon and Google

Microsoft is announcing a successor to its first in-house AI spot today, the Maia 200. Built connected TSMC’s 3nm process, Microsoft says its Maia 200 AI accelerator “delivers 3 times the FP4 capacity of the 3rd procreation Amazon Trainium, and FP8 capacity supra Google’s seventh procreation TPU.”

Each Maia 200 spot has much than 100 cardinal transistors, which are each designed to grip large-scale AI workloads. “Maia 200 tin effortlessly tally today’s largest models, pinch plentifulness of headroom for moreover bigger models successful the future,” says Scott Guthrie, executive vice president of Microsoft’s Cloud and AI division.

Microsoft will usage Maia 200 to big OpenAI’s GPT-5.2 exemplary and others for Microsoft Foundry and Microsoft 365 Copilot. “Maia 200 is besides the astir businesslike conclusion strategy Microsoft has ever deployed, pinch 30 percent amended capacity per dollar than the latest procreation hardware successful our fleet today,” says Guthrie.

Microsoft’s capacity flex complete its adjacent Big Tech competitors is different to erstwhile it first launched the Maia 100 successful 2023 and didn’t want to beryllium drawn into nonstop comparisons pinch Amazon’s and Google’s AI unreality capabilities. Both Google and Amazon are moving connected next-generation AI chips, though. Amazon is moreover working pinch Nvidia to merge its upcoming Trainium4 spot pinch NVLink 6 and Nvidia’s MGX rack architecture.

Microsoft’s Superintelligence squad will beryllium the first to usage its Maia 200 chips, and the institution is besides inviting academics, developers, AI labs, and open-source exemplary task contributors to an early preview of the Maia 200 package improvement kit. Microsoft is starting to deploy these caller chips coming successful its Azure US Central information halfway region, pinch further regions to follow.