Microsoft cuts AI costs by using its own MAI models in Office

Microsoft has begun routing a portion of Office 365 prompts in Word and Excel to its in-house MAI models, shifting some traffic away from software supplied by OpenAI and Anthropic. Bloomberg first flagged the change, and TechCrunch published the account on July 7, 2026, saying Microsoft now uses its homemade MAI models to respond to "a certain percentage of user prompts."

What changed inside Microsoft?

Microsoft now answers a share of Office prompts with its own MAI models rather than relying entirely on third-party models, and it still uses OpenAI and Anthropic in parts of Office 365. The company also recently unveiled seven new MAI models at its annual Build conference, including an agentic coder and a text-to-image generator, and has been standing up more in-house AI agents alongside continued third-party use. When asked for comment, Microsoft said, "it had nothing further to share."

This shift is presented inside Microsoft as a cost-savings measure. TechCrunch says the company has "begun to deploy a cost-savings strategy" by relying less on OpenAI and Anthropic and instead deploying internal models. The report pauses short of giving a concrete percentage of traffic rerouted, using the phrase "a certain percentage of user prompts" to describe how Excel and Word are being served.

How does this fit into the wider industry trend?

The move follows a broader wave of tech companies pulling back on AI spending in recent months, with Amazon, Uber, Meta and Accenture among firms that have reportedly made cuts. After an earlier period this year that TechCrunch calls a brief blitz of "tokenmaxxing," the news cycle has shifted toward thriftier deployments and cost controls in AI operations.

The sticker shock of providing and buying AI services is driving companies to explore alternatives. TechCrunch notes some firms are even looking to Chinese models for cheaper agentic solutions, despite concerns around potential security issues. Microsoft’s decision to increase use of in-house MAI models sits alongside those cost pressures and the industry’s search for cheaper inference options.

Why it matters

This is a practical cost-management move that affects both Microsoft’s vendor relationships and how Office features behave for users. Routing a share of prompts to MAI models reduces direct spend on OpenAI and Anthropic services, and it signals Microsoft will treat internal models as a meaningful part of its product stack. For enterprises and developers, it changes the calculus around which models power productivity features inside Office 365.

The company’s Build announcement of seven new MAI models adds technical depth to the strategy: Microsoft is not only shifting traffic but expanding its internal model set with capabilities such as agentic coding and text-to-image generation. That combination — product routing plus new models — makes the cost story also a product-architecture story.

What to watch

Look for Microsoft to disclose the actual share of Office prompts handled by MAI models or to provide more detail on the MAI models it launched at Build. A clear, quantified statement about the percentage of prompts moved on to in-house models would confirm whether this is a marginal saving or a larger platform shift. Also watch whether other large vendors follow with comparable moves away from third-party model providers.