Nadella Warns Enterprises Pay Twice for Proprietary AI Models

Satya Nadella published a blog post on July 13, 2026, arguing that companies using proprietary AI models face a hidden cost: their prompts, corrections, and tool usage—what he calls "model exhaust"—flow into vendor training pipelines, handing competitors access to proprietary institutional knowledge. The practice mirrors an asymmetry: model makers train freely on public data under fair use while forbidding customers from fine-tuning their outputs. Nadella urges enterprises to retain data ownership and build orchestration layers enabling model switching.