Breaking

Nadella urges firms to build their own AI 'token capital'

June 16, 2026 at 00:16 EDT

Sales & Business
AI Agents
Foundation Models

Microsoft CEO Satya Nadella laid out a corporate blueprint on June 14, 2026, arguing that companies should accumulate two forms of capital — "token capital" they build and own, and the "human capital" of their employees — rather than depending entirely on AI models. The argument warns against a future in which a few frontier models absorb everything, and was detailed by Indian Express the following day.

AI-Era Strategy · Microsoft

"Token Capital": Don't Let a Few Models Eat Everything

Satya Nadella's blueprint for the AI-era firm: stop competing on whose general model is strongest, and instead build a continuously improving "learning loop" from the AI capability you own plus the judgment of your people.

$700B+

Global AI CapEx projected for 2026 — raising the question of how returns get shared

Pillars of implementation: private evals, private RL environments, queryable knowledge base

Forms of capital that compound together: human capital × token capital

Two strategies, two outcomes

As model power rises and prices fall, where does accumulated expertise end up?

One "best model"
does everything

Commoditization

Expertise gets absorbed by a few model providers — firms reduced to "data pipes."

Owned token capital

Human capital

Learning loop

Proprietary moat

Frontier ecosystem

Value distributed across firms, industries and nations — and a compounding edge.

The learning loop: a "hill-climbing machine"

Every improved workflow generates a training signal and accumulates tacit knowledge.

HUMAN CAPITAL

Judgment, relationships, decisions

→

TRAINING SIGNAL

Private evals + RL on internal data

→

TOKEN CAPITAL

A stronger, owned "AI muscle"

Litmus test: swap in a new general model — does it still preserve expertise equivalent to a "company veteran"?

Praise — practical

Lets firms build a proprietary moat from their own data, with a compounding effect between human and token capital.

Critique — idealistic

Thin on implementation detail and the cost of private evals; large-scale deployment cases remain limited so far.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.