Microsoft Chief Executive Satya Nadella has cautioned businesses against unintentionally handing over valuable intellectual property when using proprietary artificial intelligence models, arguing that companies may be “paying twice” for AI services.
In a blog post, Nadella said organisations not only pay for AI through usage fees but also contribute proprietary knowledge by sharing prompts, feedback and business-specific information that improve the models over time.
According to Nadella, enterprises effectively train AI systems through their interactions, with every prompt, correction and workflow helping models better understand how businesses operate. He argued that this accumulated knowledge represents valuable institutional expertise that competitors would otherwise struggle to obtain.
The comments echo concerns previously raised by several technology executives and investors, who have warned that AI providers such as OpenAI and Anthropic could gain access to commercially sensitive information from customers and potentially use those insights to strengthen their own competitive position.
Nadella also criticised what he described as an inconsistency in the AI industry’s approach to data. While many model developers defend the use of publicly available information to train AI systems under fair use principles, they often restrict others from using model outputs to develop competing systems through a process known as model distillation.
He argued that if AI companies benefit from training on publicly available data, enterprises should similarly be able to retain and benefit from the intelligence generated through their own use of AI models.
To address these concerns, Nadella encouraged organisations to maintain ownership of their data, including prompts and user feedback, by developing proprietary AI learning environments. He also advocated deploying orchestration layers that allow businesses to switch easily between different AI models, reducing dependence on a single provider.
Although Nadella did not explicitly endorse open-source AI, his remarks align with a growing trend among large enterprises towards deploying open-source models on their own infrastructure to improve control over data, reduce costs and minimise reliance on proprietary providers.
Industry executives say that shift is already gaining momentum. Idit Levine, founder and chief executive of Solo.io, said many enterprise customers that initially adopted proprietary AI models are now exploring open-source alternatives that can be deployed on-premises, offering lower costs while giving businesses greater control over their data and operations.
The trend is also reflected in growing demand for platforms that enable organisations to switch between multiple AI models. Companies including Vercel and OpenRouter have reported increasing adoption of open-source models, with Vercel saying open models accounted for 29% of all traffic routed through its AI gateway last month.
Nadella concluded that businesses should retain ownership of the intelligence they generate through AI, arguing that the knowledge created through enterprise use of these systems should remain with the organisations that produce it rather than the model providers.
