Microsoft has started redirecting some AI-driven prompts within its Microsoft 365 suite away from third-party models such as OpenAI's ChatGPT and Anthropic's Claude toward its own proprietary AI systems. This shift impacts features embedded in popular apps like Excel and Outlook, signaling a strategic move to better manage escalating operational expenses related to AI usage.
The transition affects a relatively small portion of total AI requests but represents a clear commitment to rely more on Microsoft's internal AI infrastructure. This move aligns with statements from Microsoft’s AI chief, who outlined the company’s intention to reduce—and eventually stop—paying external providers for AI services. The newly promoted Microsoft AI (MAI) models have already been integrated into Copilot for Business and Enterprise, with specialized versions such as MAI-Code-1-Flash designed to rival external coding assistants like Claude Code and OpenAI’s Codex.
Microsoft’s efforts to bring AI workloads in-house also include plans for proprietary transcription models tailored for Teams, pushing the company toward tighter control over AI functionality and related costs. The decision follows a broader pattern among major tech firms reining in external AI expenditures amid soaring costs for AI "tokens"—units of computation or language processing. Earlier in the year, Microsoft limited employee access to some outside AI tools due to steep expenses and data security concerns.
This strategic pullback from third-party AI vendors mirrors wider industry trends. For instance, Uber exhausted its entire annual AI budget within months, while some companies have reported enormous monthly expenditures on AI tokens, fueling renewed focus on cost efficiency. By centralizing AI operations around its own models, Microsoft expects to streamline billing and better control its investment—though the financial returns and long-term impact of this approach remain difficult to measure.