Microsoft Unveils First In-House AI Models to Challenge ChatGPT and Gemini The company has already invested billions into OpenAI but is now laying out its own five-year roadmap to ensure it remains competitive in the rapidly evolving AI landscape

Microsoft has taken a major step into the artificial intelligence race by unveiling its first fully in-house AI models, marking a clear attempt to compete directly with OpenAI's ChatGPT and Google's Gemini.

The tech giant introduced two new systems: MAI-Voice-1, a natural speech generation model already powering Copilot Daily and Podcast features, and MAI-1-preview, a large-scale text-based foundation model currently in public testing on LMArena and set to be integrated into select Copilot applications in the coming weeks.

Mustafa Suleyman, who heads Microsoft's AI division, emphasised that these models were developed with a strong focus on efficiency and cost management. Unlike rival models that require more than 100,000 Nvidia H-100 GPUs for training, MAI-1-preview was built using around 15,000 GPUs, reflecting Microsoft's efforts to use computational resources more intelligently.

While Microsoft Copilot still relies heavily on GPT models from OpenAI, the launch of MAI-Voice-1 and MAI-1-preview signals a deliberate move toward becoming a more independent AI player.

The company has already invested billions into OpenAI but is now laying out its own five-year roadmap to ensure it remains competitive in the rapidly evolving AI landscape.

The release comes amid growing debate over whether the sector is entering a speculative bubble. However, Microsoft remains bullish, stating that its long-term vision is to deliver a range of specialised AI models tailored to different user needs, thereby unlocking broader value. By advancing its own AI ecosystem, Microsoft is positioning itself not only as a partner to OpenAI but also as a rival shaping the next phase of AI innovation.