Microsoft swaps in its own AI models for OpenAI's in parts of Copilot

In July 2026, Microsoft began using its own AI models instead of OpenAI and Anthropic solutions in a number of Copilot features, Bloomberg reports.

What MAI is and why Microsoft is rolling it out

MAI (Microsoft AI) is the company's in-house family of language models, developed in parallel with its use of partner solutions. Until recently they were used mainly for research; now Microsoft is starting to move real product workloads onto them.

Two factors drive the shift. Economics: for certain classes of tasks, running inference on its own infrastructure is cheaper than paying partners for API calls. As Copilot scales, the cost gap widens. Data residency: enterprise customers in regulated industries such as finance, healthcare and the public sector often require guarantees that their data is processed and stored in a specific region. Running MAI inside Microsoft's own infrastructure makes it easier to meet those requirements than routing data through third-party AI services.

Key details of the shift, according to Bloomberg:

• Microsoft routes selected tasks to MAI where it wins on cost or data residency
• OpenAI and Anthropic keep most of the traffic in Copilot
• The changes affect individual product features, not the whole ecosystem
• The transition is happening without public announcements

Does this mean a break with OpenAI and Anthropic?

Bloomberg explicitly describes the changes as incremental: OpenAI and Anthropic still process most of the production traffic in Copilot. MAI does not displace the partners across the board. It takes specific niches where it offers an advantage on price or on compliance with enterprise requirements.

The change is incremental rather than a complete break:

OpenAI and Anthropic still handle most of the production traffic inside Copilot — Bloomberg

This strategy is typical of mature technology companies. They start with external AI vendors, gradually accumulate data and expertise, and then build their own capabilities. They do not end partnerships; they manage a portfolio of models more flexibly. That reduces dependency risk and gives them leverage when negotiating terms with partners.

What this means

Microsoft is methodically building an AI stack that depends less on a single external supplier. Even a targeted replacement sends the market an unmistakable signal: the company is ready to compete with those it also works with.

For OpenAI and Anthropic, this is the start of a new reality. Their largest enterprise partner and key distributor is developing its own models and gradually moving real traffic onto them. In the long run, the line between "strategic partners" and "competitors" in the AI industry is getting blurrier: the trend toward in-sourcing models is gaining strength among tech giants, and Microsoft is not alone.

What is Microsoft's MAI?

MAI (Microsoft AI) is the company's in-house family of language models, developed in parallel with its use of partner solutions. Until recently they were used mainly for research; now Microsoft is starting to move real product workloads onto them.