Microsoft introduces a pair of in-house AI models

Microsoft has unveiled two new in-house AI models: MAI-Voice-1, its first natural speech generation model, and MAI-1-preview, its first end-to-end trained foundation model. The models were developed with a focus on efficiency and cost-effectiveness, with MAI-Voice-1 running on a single GPU and MAI-1-preview trained on about 15,000 Nvidia H-100 GPUs, a significantly lower number than other models like xAI's Grok, which required more than 100,000 chips. The decision to build its own AI models, despite having investments in OpenAI's GPT technology, indicates that Microsoft wants to be an independent competitor in the AI space. Microsoft's AI division leader, Mustafa Suleyman, said the company has an "enormous five-year roadmap" for its AI development, and the timeline will need to be aggressive to ensure that the independent path is worthwhile, especially with concerns about a potential AI bubble-pop.