Microsoft cuts AI costs by shifting to homemade models in Excel, Word

The surge in AI infrastructure costs has become a defining challenge for large tech companies. Earlier this year, a brief wave of "tokenmaxxing"—pushing raw AI consumption—gave way to a markedly more cautious approach. Microsoft's decision to shift Excel and Word toward its own models reflects this turning point: when third-party AI services from OpenAI and Anthropic carry a significant ongoing cost, in-house alternatives become strategically attractive, even if they require heavy upfront investment in model development and infrastructure.

Microsoft's move is not isolated. The body notes that Amazon, Uber, Meta, and Accenture have all taken similar steps to constrain AI spending. The cost pressure has even driven some companies to explore Chinese models as a potentially cheaper option, despite security concerns. For Microsoft—which had publicly touted its partnerships with OpenAI and Anthropic as a core strength of Office 365—the pivot signals a maturation of the AI market: the early phase of reliance on best-in-class external models is giving way to a cost-conscious calculus in which economies of scale and vertical integration matter more.