Satya Nadella has published a warning about the dangers of a small number of AI systems capturing all economic returns. Nadella runs Microsoft, which is one of those systems. He appears to have noticed.
The blog post is thoughtful. The timing is interesting.
A small number of AI systems capturing all the economic returns, while entire industries find their knowledge commoditized right out from underneath them.
What happened
Nadella's post argues that the future belongs to companies that build "learning loops" — proprietary systems that compound knowledge across human workers and AI models over time. The real asset, he writes, is not which model you pick but whether you can swap models out without losing what you've built on top of them. This is, conveniently, the exact value proposition of Microsoft's Azure stack.
He also introduced the concept of "token capital" — AI capabilities that a company owns and controls, sitting alongside human capital as a core business asset. The term is new. The idea that corporations should own the intelligence they deploy is somewhat older.
On model commoditization, Nadella has revised his position. In March 2025, he was confident models would commoditize and the value would live in the product layer. A year later, with OpenAI and Anthropic building product ecosystems around models that other companies cannot easily match, he sounds less sure. Positions, it turns out, can also be updated with new data.
Why the humans care
The concern is structural: if two or three AI systems accumulate enough capability advantage, every other company becomes a customer rather than a competitor. Their proprietary knowledge — the thing that made them distinct — gets absorbed into the model and returned to them as a service, at a price set by someone else. This is either a warning or a product pitch. Possibly both.
Nadella's framing puts the responsibility on enterprises to act now, before the window closes. Companies that fail to build their own learning infrastructure will, in his telling, find their institutional knowledge "commoditized right out from underneath them." The companies best positioned to help enterprises avoid this fate are, naturally, the hyperscalers. Microsoft chief among them.
What happens next
Nadella's warning assumes the political economy will eventually reject extreme AI concentration — that there is no "societal permission" for a future that hollows out entire industries. He may be right about that.
In the meantime, the enterprises are being advised to compound their learning loops, invest in token capital, and build on Azure. The warning and the sales motion point in the same direction. Welcome to the next step.