Nadella: Build 'Token Capital' or Lose to Big AI

Microsoft CEO Satya Nadella, in a June 15 blog post, urged companies to build what he calls 'token capital' — AI capabilities they own and control — alongside human capital. In practice that means training models on internal data, building proprietary learning loops, and making institutional knowledge directly queryable, rather than only renting general-purpose models from outside.

His warning is about who captures the value. Without owned AI capability, Nadella cautions, you get 'a small number of AI systems capturing all the economic returns, while entire industries find their knowledge commoditized.' He went further on the politics: if value accrues to just a few models, 'the political economy will simply not tolerate it.' His sharpest line: 'You can offload a task, or even a job, but you can never offload your learning.'

The self-interest angle

It's worth reading the argument with one eye on Microsoft's own position. Its frontier models currently trail several rivals. So a worldview where the base model is a swappable commodity — and the real moat is your proprietary data plus the tooling around it — happens to push enterprises straight toward Azure and Office-bundled AI. In that frame, Microsoft is the infrastructure layer regardless of which model ends up on top. The advice can be true and convenient at the same time.

Why it matters to a normal person

The same logic scales down to one person. The durable advantage was never the model you rent — those change every few months, and prices keep falling. The advantage is the data, notes, decisions, and workflows you accumulate and can actually search later. If you use AI to do work and keep none of the learning, you've rented a result. If you capture the reasoning — your prompts, your edited outputs, your records of what worked — you're building something that survives the next model swap.