Satya Nadella, Microsoft’s CEO, has warned companies against building their entire artificial intelligence strategy around a single provider. In an interview with CNN broadcast on 26 July, he argued that companies should retain in-house the prompts, metadata, context, memory and tools used by AI systems. This would allow them to modify models without halting processes, and in future also to train their own solutions.
“In my view, any company that lacks this control will not survive for long, because by relinquishing it, you have essentially outsourced your own thinking.”
This is an important argument, but it comes from the perspective of a player with a vested interest in this market direction. Microsoft remains a major shareholder in OpenAI, has announced an investment of up to $5 billion in Anthropic, and is simultaneously developing the Foundry platform, which enables the use of thousands of models from various companies. The company therefore also profits from the infrastructure required to support a multi-model environment.
Nadella’s statement reflects a shift in corporate priorities. AI is moving from the testing phase to everyday processes, and usage-based billing is increasing the pressure to control costs. Selecting different models for specific tasks can reduce expenditure, the risk of downtime and reliance on a single partner during negotiations.
A likely consequence will be increased investment in AI gateways, monitoring, data management and open-weight models deployed on in-house infrastructure. However, this will increase the complexity of systems and the demand for specialists. Competition may shift from the question of the ‘best model’ to a company’s ability to securely combine models, protect its know-how and switch suppliers quickly. A multi-model strategy alone does not eliminate technological dependence, but it can give a company greater control.
