Microsoft Claims In-House MAI-Base-2 Model Outperforms OpenAI Products Using Less Compute
Microsoft AI CEO Mustafa Suleyman announced that the company's self-developed MAI-Base-2 model outperforms comparable OpenAI models while utilizing only a quarter of the computing power. This development highlights Microsoft's push toward AI self-sufficiency and efficiency, potentially shifting the internal dynamics between Microsoft and its key partner OpenAI. MAI-Base-2 is a 2.8-trillion parameter Mixture-of-Experts (MoE) model pre-trained on 16,000 GB200 GPUs, which Suleyman claims also achieves significantly higher computational efficiency compared to other industry models.
## BACKGROUND
Pre-training is the initial phase of AI model development where a model learns patterns from massive datasets, which traditionally requires immense amounts of compute and financial investment. Microsoft and OpenAI have maintained a close partnership where Microsoft provides the massive cloud and computing infrastructure needed to train OpenAI's frontier models.