Microsoft to unveil Maia 300 this Fall, a next-generation AI chip to challenge Nvidia as it targets 1 million units

Microsoft is preparing its biggest push yet into homegrown AI silicon. The company plans to unveil its next-generation Maia 300 AI chip this fall, potentially as soon as September, and is seeking manufacturing capacity that could eventually exceed 1 million chips, according to a report from The Information.

The scale of the plan signals how serious Microsoft has become about reducing its dependence on Nvidia, whose GPUs sit at the center of the AI boom and represent a major cost for companies building and operating large AI systems.

Microsoft introduced its first Maia AI accelerator in November 2023, joining a growing group of cloud giants building custom chips rather than relying entirely on outside suppliers. Its internal chip program has moved more slowly than similar efforts at Alphabet’s Google and Amazon, both of which have spent years developing their own AI processors.

Microsoft now appears ready to close some of that gap.

“Microsoft is planning to significantly increase production of its internally designed next-generation AI chips next year in hopes of persuading big cloud customers like Anthropic to use them,” The Information reported, citing two people with direct knowledge of the plans.

Landing Anthropic would carry particular significance. The AI startup is one of the industry’s largest buyers of computing infrastructure and already has deep ties to Amazon, which has invested billions of dollars in the company. Anthropic has been using Amazon’s Trainium chips alongside other computing hardware to train and run its Claude AI models.

Microsoft’s reported pitch suggests Maia is moving beyond an internal cost-saving project. The company wants its custom silicon to become part of the computing infrastructure it sells to major Azure customers.

Microsoft is chasing Google and Amazon in the custom AI chip race

Microsoft has plenty of catching up to do.

Google has spent more than a decade developing its Tensor Processing Units, or TPUs, and began recognizing revenue from direct sales of the chips in the quarter ended in June. Amazon has gained adoption for its Trainium processors as it tries to give AWS customers another option beyond Nvidia GPUs.

Microsoft unveiled Maia 200, its second-generation AI accelerator, in January. Manufactured by TSMC using a 3-nanometer process, Maia 200 packs a large amount of SRAM, a high-speed form of memory that can improve performance for AI workloads processing large volumes of user requests, Reuters reported.

Maia 300 would push Microsoft’s chip strategy into a much larger production phase.

The company has been talking with TSMC about securing manufacturing capacity for more than 300,000 Maia 300 chips scheduled for delivery in 2027, The Information reported. Microsoft’s longer-term goal is capacity for more than 1 million units, though component availability and continuing negotiations with TSMC could limit those plans.

The timing matters. Nvidia remains the dominant supplier of advanced AI accelerators, giving it an enviable position as Microsoft, Meta, Amazon, Google and scores of AI companies spend heavily on computing infrastructure. That dominance has created a powerful incentive for the largest cloud providers to develop alternatives.

Microsoft has an added reason to move faster. It operates one of the world’s largest cloud platforms and provides infrastructure to OpenAI, making the economics of AI computing increasingly important to its margins.

Maia does not need to replace Nvidia across Microsoft’s data centers for the strategy to work. Every workload Microsoft can economically shift onto its own silicon gives the company another option for managing costs, supply and infrastructure planning.

That makes the reported 1 million-chip ambition more revealing than the Maia 300 launch itself.

Microsoft entered the custom AI chip race later than Google and Amazon. With Maia 300, it appears ready to find out whether it can turn that late start into a chip business large enough for some of the biggest AI companies to take seriously.