Nadella criticizes OpenAI, Anthropic for banning model distillation

Nadella's criticism zeroes in on what he sees as a fundamental inconsistency in how major AI labs operate. These providers prohibit distillation in their terms of service—ostensibly to protect their intellectual property—yet they themselves train on public data under fair use doctrine and extract value from customer interactions. This asymmetry, Nadella contends, allows infrastructure operators to accumulate economic gains while the companies generating actual knowledge through their use and corrections receive no reciprocal benefit.

The "reverse information paradox" framing reflects a deeper concern about value capture in AI markets. As companies increasingly rely on large language models for business operations, every interaction—every correction, rating, and refinement—becomes training data that AI providers can potentially leverage to improve their own systems or build competing offerings. Nadella positions this as a structural flaw rather than a mere commercial disagreement, suggesting that whoever controls the infrastructure gains disproportionate leverage over the entire knowledge cycle. His mention of Microsoft's own infrastructure capability signals that he sees a market opportunity for companies seeking alternatives where they can retain greater control over how their own data is used and learned from.