Microsoft has unveiled its first proprietary foundational AI models, signalling a strategic shift in its approach to generative artificial intelligence. While the company has invested billions in OpenAI, it is now betting on developing its own technologies.
This move is aimed not only at becoming less dependent on a key partner, but also at optimising costs and efficiency in a highly competitive sector.
The new solutions are MAI-Voice-1, a model for speech generation, and MAI-1-preview, a large language model. A key differentiator for both projects, overseen by Microsoft AI chief Mustafa Suleyman, is efficiency.
MAI-Voice-1 is capable of generating a minute of natural-sounding speech in less than a second, running on a single graphics processing unit (GPU). This model is already used in some Copilot assistant functions.
The MAI-1-preview, on the other hand, is the first such advanced text technology trained entirely in-house at Microsoft. Here, the company relied on an optimised training process, using some 15,000 Nvidia H-100 GPUs.
This is significantly less than some competing models, which required more than 100,000 GPUs. As Suleyman points out, the art of AI training is increasingly about choosing data intelligently and avoiding wasting resources.
The introduction of proprietary models inevitably redefines Microsoft’s relationship with OpenAI. The previous arrangement, in which Microsoft was primarily an infrastructure provider and customer, is evolving into a complex partnership combined with competition.
By developing its own AI foundations, Microsoft is gaining more control over products and costs, reducing its reliance on OpenAI technology, which to date has driven much of Copilot’s functionality.
The company admits that it has several years of catching up to do with market leaders such as Google and OpenAI itself. However, public testing of the MAI-1-preview on the LMArena platform shows that Microsoft is getting into the game with a competitive solution.
This dual track strategy – investing in OpenAI and developing its own high-performance models at the same time – positions Microsoft as a flexible player, ready for a long-term battle for dominance in the artificial intelligence era.
