Microsoft Replaces OpenAI and Anthropic Models with Its Own AI in Several Copilot Functions

Microsoft in July 2026 began gradually replacing OpenAI and Anthropic models with its own MAI developments in various Copilot product functions — according to Bloomberg.

What is MAI and why Microsoft is implementing them

MAI (Microsoft AI) — Microsoft's internal language models, developed in parallel with the use of partner solutions. Until recently, they were used primarily for research purposes; now Microsoft is beginning to transition them to real tasks in the production environment.

The transition logic is determined by two factors. Economics — inference on its own infrastructure for certain classes of tasks is cheaper than paying partners for API calls. As Copilot scales, the cost difference grows. Data residency requirements — corporate clients from regulated industries: finance, healthcare, government sector — often require guarantees that data is processed and stored in a specific region. Using MAI within Microsoft's infrastructure simplifies compliance with these requirements compared to routing through third-party AI services.

Key transition parameters according to Bloomberg:

• Microsoft directs selected tasks to MAI where it wins on cost or data storage
• OpenAI and Anthropic retain the main volume of traffic in Copilot
• Changes affect individual product functions, not the entire ecosystem
• The transition occurs without public announcements

Does this mean a break with OpenAI and Anthropic?

Bloomberg directly characterizes the changes as incremental: OpenAI and Anthropic continue to process the majority of productive traffic in Copilot. MAI does not displace partners globally — they occupy specific niches where they provide advantages in price or compliance with corporate requirements.

The change is incremental in nature, rather than a complete break:

OpenAI and Anthropic continue to process the majority of productive traffic within Copilot — Bloomberg

Such a strategy is typical for mature technology companies: they start with external AI suppliers, gradually accumulate data and expertise, and then develop their own competencies — not breaking partnerships, but managing a portfolio of models more flexibly. This simultaneously reduces the risks of dependency and provides leverage in negotiations on working conditions with partners.

What this means

Microsoft is systematically building an AI stack less dependent on a single external supplier. Even targeted replacement sends an unambiguous signal to the market: the company is ready to compete with those it simultaneously cooperates with.

For OpenAI and Anthropic, this is the beginning of a new reality. Their largest corporate partner and key distributor is developing its own models — and gradually transitioning real traffic to them. In the long term, the boundary between "strategic partners" and "competitors" in the AI industry is becoming increasingly conditional: the trend toward model insourcing among technology giants is gaining momentum, and Microsoft is not alone in this.