Microsoft Swaps Its Own AI for OpenAI in Some Apps
Microsoft has begun swapping OpenAI and Anthropic models out for its own in-house MAI (Microsoft Artificial Intelligence) models in select app features where cost or data residency is advantageous, according to Bloomberg.
A Shift in Dependence
The move marks a gradual reduction in Microsoft’s dependence on third-party AI providers, which the company spent years and billions building. This transition is facilitated by a 2025 renegotiation that freed Microsoft to develop competing models while retaining a license for OpenAI’s technology through 2032.
Incremental Change
While Bloomberg reports a replacement of models, the shift is incremental. OpenAI and Anthropic still handle most production traffic within Copilot. MAI models are being implemented where their economics stack up, such as in transcription and image systems.
Benefits and Strategy
Microsoft’s strategic logic centers around efficiency. By running these models on its own Azure infrastructure, Microsoft can avoid third-party costs. One MAI model, tailored for McKinsey, reportedly outperforms OpenAI’s GPT-5.5 in cost efficiency by a factor of ten.
This approach aligns with Satya Nadella’s concerns about avoiding becoming "the next IBM" by leaning too heavily on a single partner. Microsoft wants to own the models it uses, not just rent them.
The Future of Partnerships
Although Microsoft is developing its own AI, it doesn’t intend to cut ties entirely. The company launched MAI as a challenge to OpenAI while still offering partners’ models for specialized tasks. This distribution game focuses on leveraging Microsoft’s vast user base rather than dominating benchmarks.