Microsoft bets on in-house AI models to cut costs

Microsoft on July 7, 2026 became another Silicon Valley technology giant to announce a shift toward reducing AI spending: the company intends to significantly increase the share of its own language models in its products and services.

Why Microsoft is reducing dependence on external models

The scale of AI features in Microsoft products has grown so large that spending on third-party models has become a significant line item in the operating budget. With hundreds of millions of users across Microsoft 365 Copilot, GitHub Copilot, and other AI-integrated services, even marginal differences in the cost per request ultimately add up to hundreds of millions in annual savings.

The company is not starting from scratch: the Phi series of small language models—developed by Microsoft Research—has demonstrated that compact, carefully trained models can match large solutions on many practical tasks while operating at significantly lower cost. Switching some workloads from external APIs to proprietary models allows the company to control inference costs, accelerate response times, and fine-tune behavior more precisely for each product.

This does not mean a complete break with OpenAI: the partnership with the world's leading AI lab remains unchanged. Microsoft is apparently building a hybrid approach—the most resource-intensive and complex requests are handled by GPT models, while routine tasks fall to internal developments.

Who is already following this path

Microsoft is joining a trend that has gained momentum among all major Silicon Valley technology players.

• Meta is systematically transitioning products—from Facebook to WhatsApp—to its proprietary Llama family of models instead of buying external AI APIs
• Apple has bet on on-device models for iOS and macOS, minimizing dependence on cloud providers
• Google balances between its commercial Gemini lineup and specialized internal models for specific services
• Amazon is developing Titan models and a range of internal solutions for AWS and its own products

It is symptomatic that Microsoft itself—the largest corporate investor in OpenAI—has become one of the flagships of this movement. This is not a rupture, but pragmatic diversification: use partner models where they are irreplaceable, and switch to proprietary solutions where it is economically justified.

What is changing for the AI market

The exodus of large corporations from external APIs is shifting the balance of power. AI model suppliers, betting on sustained corporate demand growth, are forced to rethink their strategy: competition now must be on quality, price efficiency, and customization capabilities.

For the entire industry, the trend drives further reduction in inference costs and the spread of specialized, lighter models. "Getting the same quality for less money" is becoming the main technical priority for the coming years.

What this means

Microsoft's joining the wave of AI savings means: the first era of "buy capacity from the leader" is ending. Large corporations are building their own AI stack and vertically integrating model expertise. For OpenAI, Anthropic and other suppliers, this signals the need to seek value where internal corporate teams have not yet achieved parity—in advanced reasoning, multimodality, and specialized scientific tasks.

*Meta is recognized as an extremist organization and is banned in the RF.