Microsoft announced it has built two powerful homegrown AI foundation models after years of relying on OpenAI. The company introduced MAI-Voice-1, a lightning-fast speech generator, and MAI-1-preview, a text model for its Copilot assistant semafor.com semafor.com. Mustafa Suleyman, now CEO of the Microsoft AI division, said this marks a strategic shift: “We have to be able to have the in-house expertise to create the strongest models in the world” semafor.com. The move pits Microsoft directly against OpenAI – its close partner turned rival – in the race for cutting-edge AI models semafor.com semafor.com. Microsoft claims MAI-Voice-1 is one of the most efficient speech systems available, capable of generating a minute of audio in under a second on a single GPU semafor.com. The text model MAI-1-preview was trained on 15,000 Nvidia H100 chips and optimized with novel techniques from the open-source community to “punch above its weight” in performance semafor.com semafor.com. Both models focus on cost-effective AI – aiming to deliver high-quality outputs with far less compute than competitors. Analysts see this as Microsoft asserting control over its AI destiny, reducing dependence on OpenAI’s models and potentially accelerating AI features across Windows and Office. Suleyman emphasized that Microsoft will still use the “very best models” from partners and open source, but developing its own gives it flexibility and competitive edge microsoft.ai microsoft.ai.
