Microsoft is pushing its own AI model and will begin developing its own AI technology beyond its OpenAI partnership.
While continuing to develop Copilot as a core AI service, Microsoft also announced the launch of two AI models fully trained by its own team, signifying that it is no longer solely reliant on OpenAI technology but is also gradually developing more competitive proprietary AI technologies. The two models launched by Microsoft are MAI-Voice-1 and MAI-1-preview. The former is Microsoft's first natural language generation model, already used in its Copilot Daily service for Samsung smart TVs and services such as Podcasts. The latter is Microsoft's first completely self-built foundational model, currently undergoing public testing on LMArena and will be gradually integrated into specific Copilot scenarios in the coming weeks. Mustafa Suleyman, head of Microsoft's AI division, pointed out in an interview with Semafor that the core of the development of these two models lies in "efficiency and cost-effectiveness." For example, MAI-Voice-1 only requires a single GPU to run, while MAI-1-preview was trained on approximately 1.5 NVIDIA H100 GPUs. In contrast, xAI's Grok used over 10 of the same GPUs for training. Mustafa Suleyman emphasized, "The key to training models is no longer simply stacking hardware, but selecting the right data and avoiding wasting computing resources on tokens that don't significantly benefit the model." Nevertheless, Copilot's core remains built on OpenAI's GPT model technology, and Microsoft's investment in training its own models is not intended to immediately replace existing external collaborations, but rather to demonstrate a longer-term strategy. It is widely believed that after investing billions of dollars in OpenAI, Microsoft's continued focus on developing its own technology indicates its intention not to continue relying on external suppliers in the AI market. Mustafa Suleyman further revealed that Microsoft has planned a "major five-year investment blueprint," with continued quarterly increases to drive the evolution of its own models. This proactive approach may also be a response to recent market concerns about a potential AI industry bubble; if the competitive landscape changes in the future, Microsoft can immediately establish a foothold with its own technology. In the rapidly advancing AI race, Microsoft's strategic choices will profoundly impact the overall market. While OpenAI will remain the core of its Copilot in the short term, Microsoft has the opportunity to have greater autonomy in the field of generative AI as its own MAI series models are gradually expanded.