Microsoft Bets Big on In-House AI to Cut Costs

Microsoft is implementing its own AI models to mitigate soaring token costs, a significant expense even for the tech giant. Tens of thousands of weekly AI prompts in Excel and Outlook are now being handled by Microsoft's proprietary MAI models, a move away from heavier reliance on OpenAI and Anthropic. This initiative comes after Microsoft unveiled seven new in-house models last month, including MAI-Thinking-1.

MAI-Thinking-1, described as a 35 billion active-parameter model with a 256K context window, boasts high efficiency and low token costs, reportedly matching Anthropic's Claude Opus 4.6 in coding abilities. This development aligns with an industry-wide trend towards more cost-effective and efficient AI solutions, contrasting sharply with the premium pricing of competitors like Anthropic, where input tokens can cost up to $10 per million.

Microsoft AI CEO Mustafa Suleyman acknowledged the substantial AI token expenditure, stating that many within the organization spend millions of dollars. He expressed a clear goal to reduce and eventually eliminate reliance on expensive third-party providers. While Microsoft benefits from a partnership discount with OpenAI, the strategy indicates a proactive approach to future cost management and AI resource optimization.