[DigitalToday reporter Chi-gyu Hwang] Microsoft CEO Satya Nadella (사티아 나델라) warned that companies could ultimately struggle to survive if they leave all aspects of AI use to a specific AI model provider.

TechCrunch reported on July 27 that Nadella stressed companies must directly control their data and prompts, as well as metadata accumulated during the process of using models.

Appearing on CNN program "Fareed Zakaria GPS", Nadella said how much information a company shares with an AI model provider matters. He said companies should hold relevant metadata themselves each time they use a model so they can later use it to train their own weights or their own open models.

The key is a structure that is not dependent on a specific model. Nadella said companies without their own models, or without infrastructure such as an AI gateway that manages prompts separately from models, could become at risk. He also argued companies should reduce reliance on embedded coding tools provided by AI companies. He cited Anthropic's Claude Code and OpenAI's ChatGPT Codex as such examples.

He said separating the harness from the model, and separating context and memory from the model, allows companies to use multiple models according to each one's strengths. He added that companies can retain control even if a single specific model disappears.