Nadella: AI Labs Hypocritical on Distillation Bans

Microsoft CEO Satya Nadella is calling out the AI industry’s data double standard, and he’s not pulling punches. In a pointed blog post, he took direct aim at labs like OpenAI and Anthropic, labeling it “ironic” that these companies aggressively train their models on the open web under the banner of fair use, yet explicitly ban the practice of distillation in their terms of service. Distillation, where a smaller ‘student’ model learns from a larger ‘teacher’ model’s outputs, has become a flashpoint as providers try to block primarily Chinese competitors from using their expensive APIs to jumpstart rival systems.

Nadella frames the issue as what he calls the “reverse information paradox.” The core of his argument is that enterprises are paying AI providers twice. First, there’s the obvious subscription or API fee. But the second, more insidious payment is what Nadella dubs the “exhaust”—the corrections, ratings, and nuanced usage data that companies generate every time their employees interact with these systems. That data stream, rich with proprietary internal knowledge, doesn’t just vanish. It flows back to the foundation model providers, potentially training future models that could one day compete with the very businesses that fed them.

His critique isn’t born from pure altruism, of course. Nadella is making a clear pitch for a different architectural model, one where Microsoft’s Azure infrastructure plays kingmaker. He argues that the current setup concentrates economic value with the hyperscalers operating the infrastructure, rather than the companies originating the knowledge. The alternative, he suggests, is for businesses to run their own AI on their own (or rented) infrastructure, keeping the learning loop—and the valuable data exhaust—contained within their own four walls.

What makes the intervention striking is the source. This isn’t a privacy activist or a disgruntled startup founder. It’s the CEO of the world’s most valuable company and OpenAI’s biggest backer, publicly airing the structural tension built into how commercial AI is deployed. Whether it signals a genuine philosophical rift or just sharp-elbowed competition to sell more cloud computing, Nadella has put a name to a resentment that’s been simmering in corporate IT departments for months.