Microsoft Launches MAI Models, Cuts Voice AI GPU Costs 89%

Microsoft is expanding its in-house generative AI portfolio as it seeks greater control over inference costs, product development and enterprise deployment. The strategy could gradually reduce its reliance on frontier models supplied by partners including OpenAI, while preserving that commercial relationship. Microsoft also plans to turn lessons from training and optimizing its own systems into new enterprise AI offerings delivered through its Azure cloud platform.

The company unveiled MAI-Image-2.5-Pro for image generation and MAI-Voice-2-Flash for real-time voice applications. Microsoft said optimization of the voice model could cut GPU computing costs for customer-service workloads by 89%, positioning the system as a lower-cost alternative to GPT-based tools. The information provided did not specify an investment amount or an exact launch date for the two models.