Microsoft is preparing to publicly unveil its next-generation Maia 300 artificial-intelligence chip in fall 2026, potentially as early as September, according to The Information in a report cited by Reuters. The planned launch would mark another significant step in Microsoft’s effort to build more of the computing infrastructure behind Azure and its AI services with silicon designed in-house.
The strategy is closely tied to Microsoft’s attempt to reduce its dependence on Nvidia’s costly AI processors. Demand for generative AI has driven enormous spending on data centers, accelerators, networking and power, prompting the largest cloud providers to develop custom chips that can lower operating costs, improve efficiency and give them greater control over their infrastructure.
Maia 300 also arrives at a time when Microsoft is under pressure to catch up with Alphabet and Amazon in custom AI silicon. Google has already begun recognizing revenue from direct sales of its Tensor Processing Units, while Amazon continues to expand adoption of its Trainium processors.
Maia 300 Could Be Unveiled as Early as September
Reuters reported that Microsoft plans to unveil Maia 300 this fall, potentially as soon as September. The timing is based on people with direct knowledge of the plans cited by The Information, so it should still be treated as reported information until Microsoft makes a formal product announcement.
Microsoft said it continues to invest in custom silicon as part of its long-term AI infrastructure strategy. Andrew Wall, general manager for Microsoft’s Azure Maia business, told Reuters that the production figures in the report do not reflect the scale of Microsoft’s program, while the company did not disclose its own production-volume targets.
Microsoft Is Discussing More Than 300,000 Maia 300 Chips With TSMC
According to the report, Microsoft has been in talks with TSMC to secure manufacturing capacity for more than 300,000 Maia 300 chips for delivery in 2027. Reaching that level would represent a substantial expansion of Microsoft’s internal AI accelerator effort compared with the early stages of the Maia program.
Read More
Microsoft’s longer-term ambition is even larger. The company ultimately wants to secure capacity for more than one million Maia 300 chips, although component supplies and continuing negotiations over available TSMC manufacturing capacity could constrain those plans. TSMC could not be reached by Reuters for comment outside regular business hours.
Anthropic Is Among the Major Cloud Customers Microsoft Wants to Win
Microsoft is also reportedly looking to increase production significantly and persuade major cloud customers, including Anthropic, to adopt Maia 300. Anthropic has previously been reported to be in talks about renting servers powered by Microsoft-designed Maia chips as it seeks more computing capacity for its AI services.
If Microsoft can attract external customers at meaningful scale, Maia would become more than a chip used primarily inside Azure. It could develop into an alternative cloud-computing platform for AI companies that currently depend heavily on Nvidia GPUs or other hyperscalers’ proprietary accelerators.
Google and Amazon Have Moved Faster With Custom AI Chips
Microsoft has lagged Alphabet and Amazon in scaling its in-house AI chip efforts. Google’s Tensor Processing Units have been developed across multiple generations and are now being sold directly to customers, while Amazon’s Trainium family is gaining adoption as AWS expands its own AI infrastructure.
That competitive gap makes Maia 300 strategically important. Microsoft needs enough performance, software support, manufacturing capacity and customer demand to justify its custom-silicon investments while also lowering exposure to the high cost and limited supply of leading Nvidia accelerators.
Maia 300 Builds on Maia 200 and Microsoft’s 3nm Chip Roadmap
Microsoft introduced its first Maia AI accelerator in November 2023 and unveiled the second-generation Maia 200 in January 2026. Maia 200 is manufactured by TSMC using 3-nanometer technology and contains a significant amount of SRAM, a type of on-chip memory that can deliver speed advantages when AI systems handle large numbers of user requests.
Maia 300 is expected to extend that roadmap, but Microsoft has not yet formally disclosed the chip’s complete specifications, performance targets, pricing or general availability. Until the company officially announces the product, the September timing, production quantities and customer-adoption plans remain reported plans rather than final commitments.
