Microsoft is reportedly shifting its reliance toward in-house models to mitigate the escalating costs of using third-party AI tokens. According to Bloomberg, tens of thousands of weekly prompts in Excel and Outlook are now being processed by Microsoft’s own MAI models rather than relying heavily on OpenAI or Anthropic. This strategic move follows the recent announcement of seven new in-house models, including the efficient MAI-Thinking-1, which is designed for high performance at a low-token cost. While Microsoft enjoys a partnership discount with OpenAI, the company faces massive expenses, with Microsoft AI CEO Mustafa Suleyman noting that internal teams spend millions on tokens. As competitors like China’s DeepSeek offer significantly cheaper alternatives, Microsoft aims to reduce and eventually eliminate its heavy financial dependence on expensive external providers like Anthropic to maintain profitability.