Microsoft plans to introduce its next generation Maia 300 artificial intelligence chip in the fall, possibly in September, to reduce its reliance on Nvidia. The company is stepping up its custom chip efforts to manage artificial intelligence infrastructure costs.
Microsoft began developing custom artificial intelligence processors in November 2023 with the launch of its first Maia chip. However, the company has lagged behind competitors such as Google, which generates direct revenue from its TPU processors, and Amazon, which is expanding the use of its Trainium chips.
Custom chip expansion and TSMC talks
Microsoft is conducting talks with TSMC to secure manufacturing capacity for more than 300,000 Maia 300 chips for delivery in 2027. Over the long term, the company aims to secure production capacity for over 1 million Maia 300 processors.
The company intends to increase production significantly and convince major cloud clients, including Anthropic, to use Maia processors. Andrew Wall, general manager of Azure Maia at Microsoft, stated that the reported figures did not reflect the actual scale of the program. Wall added that custom chip investments formed part of a long-term artificial intelligence infrastructure strategy.
Features of the Maia 200 processor
Microsoft introduced the second generation processor in the family, the Maia 200, in January. Built by TSMC using a 3-nanometer manufacturing process, the chip was designed to handle high volumes of simultaneous user requests in artificial intelligence systems.
The Maia 200 features a large amount of SRAM memory, which operates close to the processor to deliver fast access speeds and performance benefits during heavy workloads. Microsoft is expected to expand its production capacity and chip applications further with the Maia 300.
