Microsoft announced it will increasingly use its own AI chips, starting with the third-generation Maia AI accelerator (Maia 300) unveiled on Monday, to handle incoming requests. This move aims to expand margins and bolster its AI stack. The Maia 300, set for production in 2027, will offer significant performance improvements, including up to seven times more AI tokens per rack compared to GPUs. While Nvidia's GPUs have been standard, Microsoft's finance chief Amy Hood noted that using its own chips has contributed to margin improvement. Nvidia will remain a key partner for model training.