Microsoft CEO Satya Nadella warned that companies using proprietary AI models may be giving away valuable data while paying for AI services. He said customers pay twice — first with money for AI token usage and then with data that helps model providers refine their systems. He noted that model providers can learn from prompts, agent tools and corrections, which may become institutional knowledge. He argued that while model providers have fair‑use rights to train on public data, they should not restrict enterprises from using the outputs of their own models to build cheaper alternatives. He recommended that companies retain ownership of their data, including prompts and feedback, and build proprietary learning environments on cloud platforms, preferably Azure. He also urged the implementation of orchestration layers that allow switching among multiple model providers. Companies are increasingly moving to open‑source models run on‑premises to reduce costs and maintain control. Solo.io CEO Idit Levine said many customers are experimenting with open‑source models that perform most functions of proprietary ones at lower cost. Her firm supports enterprises such as T‑Mobile, ADP and SAP. Vercel and OpenRouter have reported a rise in traffic to open‑source models, with open models representing about 29 % of traffic on Vercel’s gateway last month. Nadella's remarks underscore a debate over data ownership and model distillation as enterprises seek greater control over AI usage.