Microsoft replaces OpenAI with its own AI in several applications.

      Microsoft has started to replace OpenAI and Anthropic models with its own in-house MAI models for certain application features where cost or data residency is more favorable, according to Bloomberg. This transition is gradual; OpenAI and Anthropic still manage the majority of Copilot's operations, but it indicates a continued effort by Microsoft to reduce reliance on these partners, a dependence it spent years and substantial resources building. This change was made possible by the renegotiation in 2025, which allowed Microsoft to develop competing models.

      The shift involves directing specific tasks to Microsoft's MAI models when it is economically advantageous. While this is an incremental adjustment, OpenAI and Anthropic continue to manage most production traffic in Copilot, with MAI models integrated where financially beneficial.

      At its Build conference, Microsoft introduced seven MAI models, including its first reasoning model, image, voice, and transcription systems. The MAI-Transcribe-1 model is currently being tested within Teams and Copilot, while MAI-Image-2 is being integrated into Bing and PowerPoint.

      The primary focus is on efficiency, as Microsoft can operate these models on its own Azure infrastructure, eliminating third-party costs. It claims one MAI model, optimized for consulting firm McKinsey, outperformed OpenAI’s GPT-5.5 on cost efficiency by tenfold.

      However, this does not signify a complete separation. Microsoft launched the MAI series as a direct competitor to OpenAI while still utilizing its partners’ models where they excel.

      For several years, Microsoft's initial contract with OpenAI prevented it from independently pursuing advanced AI. This changed after the two companies renegotiated, lifting Microsoft's exclusivity and allowing it to create competing models while retaining a license for OpenAI’s technology until 2032.

      This agreement was mutually beneficial, permitting OpenAI to sell through competitors, allowing AWS to offer its models as well. Microsoft’s strategy has involved diversifying its investments: holding a significant stake in OpenAI, incorporating Anthropic’s Claude into Copilot, and now rolling out its own MAI models.

      Satya Nadella has hinted at this strategic rationale, expressing concerns that Microsoft might become “the next IBM” if it overly depended on one partner. By owning its models rather than just leasing them, Microsoft seeks to avoid that scenario.

      Microsoft does not need MAI to lead every performance ranking since its broad access to hundreds of millions of Office and Teams users generates substantial engagement. Even transferring a small portion of that usage to in-house models has significant financial implications.

      However, there is an ongoing challenge, as Microsoft delves deeper into AI while the question of paying customers remains uncertain. The use of more cost-effective, internal models can enhance profit margins regardless of user payments.

      The costs associated with external AI are considerable, and bringing features in-house helps reduce that expenditure. The partnership that previously shaped Microsoft’s AI narrative is being subtly adjusted, one replaced model at a time.