Microsoft CEO warns enterprises about proprietary AI model risks, urges data control and open alternatives

Microsoft CEO Satya Nadella warned in a blog post that companies using proprietary AI models are "paying for intelligence twice"—once in token usage fees and again by surrendering proprietary business knowledge that model providers can use to train competitors. He argues that the more sensitive data enterprises feed into models to improve performance, the more they risk teaching those models about their business nuances in ways that could be exploited by rivals.

Nadella specifically criticized the practice of model providers reserving the right to learn from customer usage and interaction data, calling it hypocritical for providers to freely train on public data while restricting others from studying their models. He proposed that enterprises should "retain ownership" of their data, including prompts and feedback, and build proprietary learning environments—ideally on cloud platforms like Microsoft Azure—to maintain control.

He also advocated for "orchestration layers" that let companies switch between AI models from different providers, framing this as a way to avoid lock-in. Tools that enable multi-provider model switching, often described as AI "gateways," have grown in popularity as enterprises seek flexibility.

The shift toward open source and on-premises models is already underway among large enterprises, according to Idit Levine, CEO of Solo.io, whose technology powers the Linux Foundation’s Agentgateway project. Levine said customers are increasingly asking whether they can run open source models on their own premises, noting that such models can deliver "almost 90%" of the performance of proprietary models at lower cost while giving companies full control over their data.

Traffic data from model-switching platforms supports this trend. Vercel, which added AI model-switching tools to its hosting platform, reported that open models accounted for 29% of all traffic routed through its gateway last month. OpenRouter, a company that helps developers route requests across different AI models, has also seen a surge in traffic to open source models.

Nadella’s stance—coming from the CEO of a company invested in both OpenAI and Anthropic—signals a potential inflection point in enterprise AI strategy. His argument centers on the idea that "in consuming intelligence, you are creating intelligence. And what you create should belong to you." This framing challenges the default assumption that proprietary model providers should have unfettered access to customer data for model improvement.