Microsoft is preparing to unveil its next-generation custom artificial intelligence (AI) processor, the Maia 300, as early as next month, a report has said. This comes as the software giant accelerates efforts to scale in-house silicon and reduce its heavy reliance on Nvidia hardware. Citing The Information, news agency Reuters reported that the upcoming chip rollout comes as major cloud rivals, including Google-parent company Alphabet and Amazon, report growing commercial adoption and revenue from their own custom silicon offerings, such as Google’s Tensor Processing Units (TPUs) and Amazon’s Trainium processors.
Expanding custom silicon manufacturing with TSMC
To support its long-term AI infrastructure demands, Microsoft is said to be actively negotiating manufacturing capacity with semiconductor foundry TSMC. Reports indicate that Microsoft is seeking capacity to produce more than 300,000 units of the Maia 300 for delivery in 2027, with a broader long-term target of securing capacity for over 1 million chips.However, final production volumes remain subject to component availability and ongoing manufacturing allocations at TSMC, the Reuters report said. Alongside internal deployment across its Azure data centres, Microsoft is working to persuade major enterprise cloud clients, including AI research firm Anthropic to adopt the new chips for model training and inference workloads.
Accelerating silicon development cycles
The reported planned September reveal suggests a rapid cycle for Microsoft’s custom hardware division, following the introduction of its original Maia 100 processor in November 2023 and the second-generation Maia 200 in January.The previous-generation Maia 200 has been manufactured using TSMC’s advanced 3-nanometer process node and featured high-density Static Random-Access Memory (SRAM), which is a specialised memory architecture designed to speed up processing times for high-volume conversational AI requests.