Microsoft Shifts to In-House AI Models to Cut Costs in Office

Microsoft has begun routing a portion of AI requests in Excel and Word through its own internally developed models rather than relying exclusively on third-party systems from OpenAI and Anthropic, according to a report from Bloomberg.

The shift represents a notable change in strategy for the company, which has previously highlighted its partnerships with both AI providers as core components of Office 365's intelligence features. While Microsoft continues to use external models, the integration of its proprietary MAI systems marks a deliberate effort to manage the escalating costs associated with AI inference at scale.

Microsoft declined to provide additional details when contacted by TechCrunch, which first reported on the Bloomberg findings.

Expanding the MAI Model Portfolio

The cost-cutting approach aligns with Microsoft's recent expansion of its in-house AI capabilities. At its Build conference last month, the company unveiled seven new MAI models spanning multiple use cases, including an agentic coding assistant and a text-to-image generation system. These tools give Microsoft greater flexibility to handle different workloads without incurring per-token charges from external vendors.

By developing models tailored to specific Office tasks, Microsoft can optimize for both performance and cost efficiency. Simple queries that don't require the full capabilities of frontier models like GPT-4 or Claude can be handled by lighter-weight alternatives, preserving margins on its AI-enhanced productivity suite.

Why It Matters

This move signals that even the largest AI investors are feeling pressure to rationalize spending as inference costs compound across millions of users. For enterprises evaluating AI deployment strategies, Microsoft's approach validates a hybrid model where proprietary systems handle routine tasks while reserving premium third-party models for complex requests. The shift also suggests that the initial phase of AI adoption—where companies paid premium prices for cutting-edge capabilities—is giving way to a more cost-conscious operational phase focused on sustainable economics.

Industry-Wide Retrenchment

Microsoft's cost management efforts reflect a broader recalibration across the technology sector. Following an early-year period of aggressive AI spending that some observers dubbed "tokenmaxxing," major companies including Amazon, Uber, Meta, and Accenture have all reportedly implemented measures to control AI-related expenses in recent months.

The financial burden of AI services has become particularly acute for companies operating at scale. Some organizations have even begun exploring Chinese AI models as lower-cost alternatives for agentic applications, despite potential security considerations that accompany such decisions.

The details were first reported by Bloomberg.