Microsoft CEO Satya Nadella has once again publicly addressed the risks of enterprise AI use. Foreign media reports that his statement this time was more direct than earlier in the month: if companies rely on a single model provider for their AI capabilities in the long term, they may gradually lose autonomy and ultimately weaken their competitiveness.
Advocating for data preservation and call control
In an interview with CNN, he stated that companies need to handle the content they provide to model providers carefully, including not only training data but also cue words, context, and metadata generated during usage. According to him, this information should be kept in the company's own hands as much as possible.
Nadella believes that companies don't necessarily need to build their own large models now, but they shouldn't completely hand over key control layers to external platforms. If they can retain call logs and related metadata, companies will have the opportunity to train their own models in the future, or at least retain the ability to switch vendors.
The model and memory layer should be separated.
He suggested that the model itself should ideally be separated from the calling layer, context, and memory layer. This way, enterprises can access multiple models simultaneously and allocate their use according to different tasks; even if a model vendor withdraws, business operations will not be forced to stop.
Nadella also specifically mentioned the coding tools provided by the model provider, which are essentially a complete development layer encapsulated around the model. The article cites Anthropic's Claude Code and OpenAI's ChatGPT Codex as examples of such tools.
Worried that suppliers will become competitors
Foreign media believe that this reminder is not without Microsoft's own interests. Microsoft is both a large cloud service provider and sells enterprise-grade AI infrastructure. Nadella's advocacy of multi-model management and call isolation is precisely the service direction that cloud platforms can undertake.
However, the article also points out that this assessment is not unfounded. As enterprises introduce more AI applications, issues such as cost, performance, and vendor lock-in are becoming more prominent. More and more companies are beginning to focus on open weighted models that can be fine-tuned and deployed on their own hardware.
Nadella's core concern extends beyond budget. He believes that once companies outsource their "thinking processes" to model providers, those providers could launch directly competing services based on these use cases. This risk will increase further, especially as AI agents begin to integrate into internal company processes.
Additional information:The article notes that Nadella's approach to ordinary consumers differs from his approach to enterprise clients. He stated that in the consumer internet, users exchanging data for free services is a common practice, therefore this warning is primarily aimed at enterprise customers.
