Microsoft's CEO just warned enterprises that AI labs could become their competitors

In a blog post published Sunday, July 13, Microsoft chief executive Satya Nadella told companies using frontier AI models that they are paying for the technology twice. "You essentially pay for intelligence twice," he wrote, "once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful."

The mechanism he describes is specific: models learn from user interactions, and especially from corrections — the moments a person tells a model it got something wrong. Nadella argues each of those corrections becomes distilled into institutional know-how, captured by whichever lab trained the model, describing it as knowledge "a competitor could never buy." The implication is direct — the lab hosting your AI usage is, in the process, absorbing a compressed education in how your business actually runs.

The reciprocity argument

Nadella did not stop at the warning; he drew a fairness argument from it. AI labs, he noted, built their models by scraping the open internet — everyone's writing, code, and data — without asking permission. If that is acceptable, he argued, it is hypocritical for those same labs to then restrict enterprises from studying the resulting models as closely as the models studied them. He named the dynamic himself: the "reverse information paradox."

You essentially pay for intelligence twice: once with money, and again with something even more valuable — the proprietary knowledge you must reveal to make that intelligence useful.

The tension the post doesn't address

The obvious question sits underneath the whole post and goes unmentioned in it: Microsoft is a major investor in OpenAI and, more recently, Anthropic — two of the labs whose customer relationships this warning implicates. Reporting on the post found no official Microsoft statement reconciling that position, and outlets covering it noted the irony without resolution. Nadella's own company also sells Copilot, an enterprise AI product built on models from the labs he is describing as risks. Whatever the merits of the underlying argument, it arrives from someone positioned on more than one side of the trade he is warning about.

What is not established

This is an argument and a framing from one executive, not a documented case study. Nadella's post names no specific company whose proprietary knowledge became a competing product, and no lab has been shown extracting institutional knowledge from customer corrections in the way he describes. "Reverse information paradox" is his own coinage, not an established term with independent measurement behind it. The critical read from ProMarket — that the warning doubles as a pitch for Microsoft's own AI offerings — deserves equal weight to the warning itself.