Microsoft is testing its own AI models to potentially reduce reliance on OpenAI for its Copilot service. The company has unveiled MAI-Voice-1, a speech model, and MAI-1-preview, a text-based model, for future Copilot versions. MAI-Voice-1 is designed for efficiency, generating a minute of audio in under a second on a single GPU.
The MAI-1-preview model was trained using 15,000 Nvidia H-100 GPUs. Microsoft aims for cost-effectiveness and in-house expertise to create leading AI models. This move could intensify competition between Microsoft and OpenAI, despite their existing partnership. Microsoft intends to deepen its collaboration with OpenAI.
