Microsoft Chairman and CEO Satya Nadella issued a stark warning to enterprises adopting artificial intelligence: businesses are quietly risking their most valuable asset—institutional know-how—simply by using the technology. Nadella introduced the concept of the “Reverse Information Paradox,” a direct inversion of Nobel Prize-winning economist Kenneth Arrow’s 1962 “Information Paradox.” While Arrow’s original theory stated that a seller of information risks losing its value the moment they reveal it to a prospective buyer, Nadella argued that the AI age flips this dynamic entirely onto the buyer. Today, enterprises essentially pay for AI twice: once in cash, and a second, more expensive time in the proprietary knowledge they must expose to make the models useful.
Nadella emphasized that this risk goes far beyond traditional data leaks, occurring through what he termed “intelligence exhaust.” This includes the daily prompts employees write, the tools AI agents call, and most critically, the manual corrections humans make when a model gets something wrong. Every single correction is institutional expertise distilled into data—a hidden asset that leaks trace by trace, eval by eval. As a result, an intense information asymmetry develops: the AI infrastructure vendor learns an immense amount about the buyer’s business operations, while the buyer learns virtually nothing about what the vendor is absorbing in return. Nadella criticized this “ironic” industry status quo, pointing out that while AI labs aggressively rely on fair-use rights to train on the public web, they simultaneously restrict customers from model distillation while quietly feeding off their private interaction data.
