Companies using AI may face a cost beyond the tokens they purchase. Microsoft CEO Satya Nadella says they can also pay with proprietary knowledge needed to make an AI model genuinely useful.
The warning centers on what happens after an organization starts using a model in daily work. Prompts, corrections and business information can provide the model with new knowledge about how that company operates.
Nadella described this as a double payment in a blog post cited by CNBC Indonesia from Tech Crunch. The first payment is monetary, while the second involves exclusive knowledge supplied by the user.
“Essentially, you pay twice, once with money and once with something more valuable: proprietary knowledge to make the intelligence useful,” Nadella said. He added that better model performance requires users to provide more knowledge.
This creates a direct connection between the output a company expects and the information it may need to share. A more capable result, in Nadella’s view, calls for a deeper supply of business context.
| Area | What the User Provides | Effect Highlighted by Nadella |
|---|---|---|
| Tokens | Payment for AI use | Direct monetary cost |
| Data and knowledge | Prompts, corrections and business details | The model gains new knowledge |
| Model performance | Demand for better results | More knowledge must be shared |
Who Controls the Learning Process
Nadella also raised concerns about the balance of power in the AI ecosystem. Model owners can train their systems using data distributed around the world, while others may not have equivalent room to learn from the owners’ models.
He expressed concern about model makers having the right to study usage data and customer interactions. That arrangement makes control over company data a central issue for organizations adopting AI.
The concern is not limited to information entered in a single request. Feedback given when an AI response is corrected can also become part of the knowledge accumulated through use.
For companies, this means proprietary data must be considered alongside the immediate cost of access. The information that helps a system understand a business in greater detail may itself carry strategic value.
Keeping Data Under Company Ownership
Nadella’s proposed approach is for companies to retain ownership of their data, including prompts and feedback. He also emphasized building an exclusive learning environment in the cloud.
He further pointed to an orchestration layer that would allow companies to move between AI models from different providers. Such an arrangement is intended to preserve flexibility in model use while keeping control over the data generated during that use.
The broader point is that AI adoption involves more than selecting a model and paying for tokens. Organizations also need to consider how the knowledge created through interactions is handled, owned and controlled.
