Reuters reports that Microsoft is planning to unveil its next-generation Maia 300 AI chip this fall, potentially as soon as next month, citing The Information, which spoke to people with direct knowledge of the plans. The move is the latest step in Microsoft’s long-running effort to reduce its reliance on Nvidia’s expensive data center processors.
The report says Microsoft has been in talks with chipmaker TSMC to secure manufacturing capacity for more than 300,000 units of the chip for delivery in 2027, and ultimately aims to secure capacity for more than one million Maia 300 chips, though component supplies and ongoing capacity negotiations could constrain those plans. Microsoft is also looking to significantly ramp up production and persuade major cloud customers such as Anthropic to adopt the chip.
A chip that has lagged its rivals
Microsoft introduced its first Maia AI chip in November 2023, but has trailed rivals such as Alphabet and Amazon in scaling up its in-house silicon. Alphabet began recognizing revenue from direct sales of its custom Tensor Processing Units in the quarter ended June, while Amazon has seen growing adoption of its Trainium processors. Microsoft’s second-generation Maia 200 arrived in January 2026, built by TSMC using 3-nanometer technology and packed with a large amount of SRAM, a memory type that can offer speed advantages for AI systems handling many user requests.
The Maia 300 would build on that design as Microsoft pushes its custom-silicon program further. The company’s position on the reported volumes is more cautious. “Microsoft continues to invest in custom silicon as part of our long-term AI infrastructure strategy. While we don’t share production volumes, the figures reported don’t reflect the scale of our program,” Andrew Wall, general manager for Microsoft’s Azure Maia, told Reuters. TSMC could not be reached for comment outside regular business hours.
Why in-house silicon matters
For Microsoft, the stakes are partly about cost and partly about control. Nvidia’s accelerators dominate AI training and inference, and demand has pushed prices and lead times up as data center builders race to add capacity. A competitive Maia line gives Microsoft leverage in those negotiations and a cheaper path for powering its own Copilot and Azure AI workloads. Persuading outside customers such as Anthropic to run on Maia would be a bigger validation, and the report’s emphasis on production ramp suggests that is exactly what Microsoft is aiming for.
An unveiling in September would put the Maia 300 in the spotlight at the height of the fall product season. Whether Microsoft can hit the reported capacity targets will depend on TSMC’s availability and on component supply, both of which have constrained AI hardware plans across the industry this year. For now, the chip’s exact specifications and pricing remain unconfirmed, and Reuters notes the plans are not yet public.
The fuller picture should come when Microsoft makes the chip official. If the timeline holds, the Maia 300 becomes the clearest sign yet that Microsoft intends its custom silicon to be more than an internal experiment.
