Microsoft CEO Satya Nadella warned on Monday that enterprises buying AI services are paying twice – first with money for token usage, then by handing over proprietary knowledge that powers the models. He explained that every prompt, correction, and tool interaction feeds the provider’s systems, turning a company’s internal know‑how into a competitive asset.
Nadella argued that if AI labs can freely scrape public data to improve their models, it is unfair for them to restrict others from “distilling” those models. He urged businesses to retain ownership of their data—including prompts and feedback—by building proprietary learning environments on the cloud, preferably Azure, and by adding orchestration layers that let them switch between providers without lock‑in.
The warning aligns with a growing shift toward on‑premise open‑source models, which many firms see as a cheaper, controllable alternative. Companies such as T‑Mobile, ADP and SAP are already experimenting with self‑hosted AI, and platforms like Vercel report rising traffic to open‑source models. Nadella’s call may accelerate that trend as enterprises seek to protect their data while still leveraging AI capabilities.
