Microsoft Quietly Swaps OpenAI and Anthropic for Homegrown MAI Models in Word and Excel

Microsoft is pulling back on its dependence on third-party AI providers, deploying its own MAI models to handle a share of user requests inside two of its most ubiquitous productivity tools — Word and Excel. According to a Bloomberg report published Tuesday, the company has quietly begun substituting OpenAI and Anthropic outputs with responses generated by its homegrown models, marking a notable strategic pivot for a company that had previously marketed those third-party integrations as a core selling point of its Office 365 suite. Microsoft declined to provide additional detail when contacted by TechCrunch.

The MAI model family has been under development internally for some time, but received its most public showcase last month at Microsoft's annual Build developer conference. There, the company unveiled seven new MAI models spanning a range of capabilities, including an agentic coding assistant and a text-to-image generator. The announcements signaled that Microsoft's ambitions extend well beyond cost savings — the company appears to be positioning its own AI stack as a long-term competitive asset rather than a stopgap measure.

Microsoft's move fits neatly into a broader industry realignment that has gathered pace over the past several months. Following a period observers dubbed 'tokenmaxxing' — in which companies integrated AI features aggressively with little concern for expense — the pendulum has swung sharply toward fiscal discipline. Amazon, Uber, Meta, and Accenture have all reportedly taken steps to reduce AI-related spending, suggesting the correction is sector-wide rather than company-specific.

The financial pressure is real. Providing and consuming AI inference at scale carries enormous infrastructure costs, and the bill has grown uncomfortable enough that some Silicon Valley firms are reportedly exploring Chinese AI models as a cheaper alternative for agentic workflows — a workaround that carries its own set of security and regulatory concerns. Microsoft, with its substantial Azure compute base and deepening in-house research capabilities, may be better positioned than most to absorb that pressure without resorting to such trade-offs.

For everyday users of Word and Excel, the immediate impact may be imperceptible — the underlying outputs should feel functionally similar regardless of which model generates them. But the strategic implications are significant. Every prompt Microsoft routes through a MAI model instead of an OpenAI or Anthropic endpoint represents a dollar not spent on a partner and a step toward greater independence in the AI value chain, a dynamic that is likely to reshape partnership agreements across the enterprise software landscape.