Recently, Satya Nadella, CEO of Microsoft, stated in a CNN podcast program that companies should not rely on a single AI model, otherwise they may "be unable to continue surviving." He believes that the knowledge, prompts, interaction records, and metadata generated by using AI belong to important assets of the enterprise, and should be controlled by the enterprise itself, rather than being tied to a single AI supplier or model.

Nadella said that unlike consumer scenarios, enterprises need to ensure their own knowledge assets remain within the company. As AI usage becomes more deeply integrated into business processes, enterprises will not only have "human capital," but also accumulate more and more "token capital," which refers to prompts, context, knowledge, and metadata formed through interactions between employees and AI. This information can further be used to train the enterprise's own models or model weights.

Copilot

He also emphasized that AI models themselves may change or even disappear, so enterprises must maintain model selection rights and technological autonomy. "Any model may disappear, but you can still control your own fate," Nadella said. He believes that if enterprises cannot control AI foundational capabilities, it actually means outsourcing their thinking capabilities.

This view also aligns with Microsoft's current product strategy. Microsoft 365 Copilot allows enterprises to choose different AI models, including OpenAI and Claude, according to their needs, rather than being bound to a single model. As enterprises accelerate AI deployment, model replaceability, data control, and knowledge asset ownership are becoming key issues in AI infrastructure development.

Regarding user data, Microsoft stated that it will not use personal data to train AI models without consent, but Copilot provides optional switches such as "conversation activity training" and "voice conversation training," which are turned off by default. Additionally, the "optional diagnostic data sharing" in Windows Copilot may collect browser usage, visited websites, and enhanced error reports to improve Microsoft products.

From model selection to data governance, the competition in enterprise AI is shifting from merely pursuing model capabilities to competing for long-term control over data, knowledge, and AI infrastructure.