Microsoft launches in-house AI models MAI Voice-1, MAI-1-preview to reduce OpenAI reliance

Microsoft has unveiled its first fully homegrown AI models, marking a major milestone in its artificial intelligence strategy. The two models, MAI-Voice-1 and MAI-1-preview, are designed to enhance the company’s consumer-facing AI tools while reducing its reliance on OpenAI’s technology.

MAI-Voice-1, a natural speech generation model, can produce a minute of audio in under a second on a single GPU. It already powers features such as Copilot Daily, where an AI host recites news stories, and generates podcast-style discussions to explain complex topics. Users can also experiment with the model through Copilot Labs, customising its voice and speaking style.

Alongside it, Microsoft introduced MAI-1-preview, a text-based foundation model trained on around 15,000 Nvidia H100 GPUs. While significantly smaller in scale than rival models such as xAI’s Grok, which used over 100,000 GPUs, Microsoft says its efficiency-driven approach prioritises the right data to minimise wasted compute. The model is currently available for public testing on the LMArena benchmarking platform and will soon be integrated into select Copilot scenarios.

Despite its continued dependence on OpenAI’s GPT models, Microsoft is positioning these launches as the start of a long-term plan to establish greater independence. Mustafa Suleyman, head of Microsoft AI, said the focus is on building systems optimised for consumer use rather than enterprise. “We have vast amounts of very predictive and useful data on the ad side and consumer telemetry,” he noted during an episode of the Decoder podcast with The Verge, adding that the goal is to create models that serve as effective digital companions.

Suleyman, who has previously spoken about a “five-year roadmap” for Microsoft AI, also stressed that model training needs to be efficient and cost-effective to remain sustainable. This comes at a time when concerns are growing about an AI investment bubble and the strain of escalating development costs.

Looking ahead, Microsoft says it has “big ambitions” beyond simply making bigger models. Instead, it plans to orchestrate a suite of specialised systems tailored to different user needs and contexts. This multi-model strategy, the company believes, will unlock long-term value and strengthen its role in the next phase of AI evolution.