Microsoft CEO Satya Nadella has published a blog post that has shaken Silicon Valley. His argument is clear: companies using proprietary AI models, such as those from OpenAI or Anthropic, risk giving away their most sensitive data to the providers. Nadella states that these companies pay twice: once with money for token consumption, and a second time with valuable corporate knowledge used to train the models further.
The double payment companies make for artificial intelligence
According to Nadella, every interaction with a proprietary AI model leaves precious traces: the prompts written, the tools used by agents, and especially the corrections users make when the model provides wrong answers. Each correction is distilled into institutional knowledge that the provider can exploit to improve its service but also to gain a competitive advantage. The fear is that big AI labs act like Trojan horses, infiltrating business processes and then becoming competitors to their own customers.
Sponsored Protocol
Nadella's solution: orchestrate models and retain data ownership
Nadella proposes two strategies for enterprises. The first is to create "proprietary learning environments" in the cloud, where corporate data remains under direct company control. The second is to build "orchestration layers" that allow easy switching between different AI models, avoiding vendor lock-in. Tools like AI gateways are becoming increasingly popular for this purpose. Nadella emphasizes that while model providers claim fair use rights to train on public web data, they then impose restrictions on distillation by third parties, creating a double standard.
Sponsored Protocol
The trend toward open source models deployed on-premise
Many companies are already following this path. Idit Levine, CEO of Solo.io, observes that after experimenting with proprietary models, clients start asking whether they can use an open source model on-premise, achieving about 90% of the performance at a much lower cost and with full data control. Vercel and OpenRouter also report increased traffic to open models, which accounted for 29% of all traffic routed through Vercel's gateway in one month. This movement confirms that companies seek solutions preserving data sovereignty.
To explore how to integrate AI into business processes without compromising security, read our article on AI for Italian E-commerce. The shift to open source models is also supported by growing attention to data center infrastructure: learn how transformer delivery delays are affecting the industry. For a broader perspective, see the Wikipedia page on open-source model.
Sponsored Protocol
In conclusion, Nadella's warning marks a turning point for companies adopting AI. As the CEO writes: "In consuming intelligence, you are creating intelligence. And what you create should belong to you." Enterprises must now choose whether to entrust their secrets to external providers or invest in solutions that guarantee data ownership and operational flexibility.
Source: https://techcrunch.com/2026/07/13/satya-nadella-has-issued-a-shocking-warning-to-companies-using-ai
