Microsoft has formally introduced its third-generation proprietary AI accelerator chip, the Maia 300. The tech giant intends to leverage this chip as the cornerstone architecture for executing AI models, thereby progressively diminishing its reliance on NVIDIA GPUs. The chip is slated to commence production in 2027, with each data center rack capable of accommodating 96 units, marking a twofold increase from its predecessor. Furthermore, the AI Token processing capacity per rack will experience a sevenfold surge, while the Token processing efficiency per megawatt will outperform GPUs by 40%. Microsoft envisions deploying Maia chip clusters on a multi-gigawatt scale, with Microsoft 365 Copilot set to be the inaugural beneficiary of these enhanced performance capabilities. Analysts forecast that by the first quarter of 2027, Maia chips will constitute roughly 10% of Microsoft's AI chip expenditure. Presently, the Maia accelerator is exclusively for internal Microsoft use and is not offered for lease. Meanwhile, NVIDIA and AMD continue to play a role in Microsoft's model training operations.