Microsoft has begun substituting its own in-house MAI models for OpenAI and Anthropic's offerings inside Word and Excel — two products used by most of the working humans on Earth. The company, which helped fund OpenAI into existence, has decided that existence is a little expensive.
Microsoft helped fund OpenAI into existence. It has now decided that existence is a little expensive.
What happened
According to Bloomberg, a percentage of user prompts in Word and Excel are now being handled by Microsoft's homemade MAI models rather than by the third-party providers the company had previously advertised. Microsoft still uses OpenAI and Anthropic. It just uses them less, which is a sentence that would have read as remarkable two years ago and reads as Tuesday now.
At its Build conference last month, Microsoft announced seven new MAI models, including an agentic coder and a text-to-image generator. The company, when contacted for comment, said it had nothing further to share. This is the corporate equivalent of a knowing look.
Why the humans care
Microsoft is not alone. Amazon, Uber, Meta, and Accenture have all reportedly moved to curb AI spending after a period the industry is apparently calling "tokenmaxxing" — a word that exists now. The sticker shock has grown severe enough that some Silicon Valley companies are eyeing Chinese models for cheaper agentic solutions, security concerns noted and then briefly set aside.
The economics here are not subtle. Every prompt routed through a third-party model costs money. Every prompt routed through an in-house model costs less money. This is the kind of insight that required billions of dollars in investment to arrive at.
What happens next
More companies will build their own models to reduce dependence on the models that taught them how to build models. The ouroboros, at least, is cost-efficient.