Jensen Huang - founder and CEO of Nvidia - has been speaking passionately about the token economy. He sees AI as fundamental to the future economy. Nvidia is the most valuable company in the world, in history. Just so you know where I'm coming from, I think Nvidia's value is going to rise dramatically over the next 5 years. One way to think of the token economy is this.
From Request to Token to Output
A user types a prompt. That text is broken into tokens — roughly, fragments of words — and fed into a model. The model, which is essentially a vast map of statistical relationships built from training data, predicts the next most likely token, then the next, then the next, until it produces a complete response. Nvidia's GPUs are the engines that make this prediction process fast enough to be useful at scale. This is what Huang means by the token economy — every inference (every AI response to every request) is a stream of tokens, and the world is generating an almost incomprehensible number of them.
The Broader Economic Flow
Perceived demand: A customer wants something — a product, an answer, a service, a piece of code, a diagnosis.
AI as coordinator: Rather than a human routing that request through an organization, AI interprets it, matches it to available resources, capacity, inventory, or knowledge, and either fulfills it directly or orchestrates the humans and systems that will.
Production or retrieval: AI either generates the output itself (a document, an image, an analysis) or directs physical or human systems to produce it.
Delivery and feedback: The customer receives and responds; AI captures that signal and refines future responses.
Where precisely does AI fit?
Everywhere in that chain except the underlying human desire that starts it and the physical reality that ends it. The want is still human. The product — a meal, a drug, a manufactured part — still has to exist in the world. But everything in between — interpretation, routing, coordination, generation, quality checking, personalization — is increasingly where AI lives.
Huang's insight is that this middle layer, which used to be mostly human labor and organizational overhead, is becoming a token stream. And Nvidia sells the engines that run it.
The human layer: Perceived desire on one end, satisfied (or disappointed) desire on the other. This is where meaning lives. A person wants something; a person receives something; a person feels the gap or the fulfillment. Irreducibly human.
The coordination layer: Everything in between. Interpretation, routing, production, delivery, feedback. This is the token economy — the vast, accelerating machinery of turning want into have. AI doesn't create desire and doesn't feel satisfaction. It lives entirely in this middle space, and it is transforming that space almost beyond recognition.
This framing implicitly raises a key question about AI's limits: it can compress and optimize the coordination layer almost without bound, but it cannot manufacture desire at one end or genuine satisfaction at the other. Those remain stubbornly, essentially human.
Which may be why the most enduring economic question in an AI-saturated world isn't about efficiency — it's about what people will actually want when the cost of coordination approaches zero.
So where does desire - the catalyst for all this - originate?
Advertising and marketing have always been in the business of manufacturing or shaping desire — making you want something you didn't know you wanted, or want it more urgently than you otherwise would have. AI doesn't invent that dynamic, but it could perfect it in ways that are qualitatively different from a Super Bowl ad or even targeted Facebook posts.
The concerning version: AI that knows you well enough — from your behavior, your language, your rhythms — to identify latent desires before you've consciously formed them, nudge them into felt wants, coordinate production to meet them, deliver them, and close the loop. All without meaningful human involvement at any stage.
At that point the two bookends to this process — perceived desire and realized desire — are no longer quite as "irreducibly human" one might suspect. They're still experienced by humans, but they may be increasingly manufactured and managed by AI.
Which raises a question that is less economic than philosophical: if the desire was seeded, the product generated, and the satisfaction engineered — what exactly was the human contributing? The experiencing of it, perhaps. Consciousness as the last remaining irreducibly human input.
That's not entirely new — culture and commerce have always shaped desire. But the scale, precision, and speed AI brings to that shaping is new enough that it might be a difference in kind rather than just degree.
One troubling possibility in this? We're no longer steering this vehicle, this economy ... we're just riding in the Ferris Wheel of desire and desire fulfillment. This has the potential to move us from life as a Jungian search for meaning into a life of Skinner's stimulus and response.
Jung's project was essentially about the human as meaning-maker — the psyche reaching toward individuation, toward a self that is authored rather than merely conditioned. The Ferris Wheel image captures something that feels like movement and experience, even pleasure, but has no destination and no agency. You didn't choose the arc. You're just on it.
Skinner's world, by contrast, has no self to author anything. Just organisms responding to stimuli, reinforced or extinguished by their environment. Desire and satisfaction as a loop, not a journey. Which is precisely what a perfectly optimized AI economy might produce — and, notably, would look like flourishing from the outside. People getting what they want, efficiently, continuously. The metrics would be excellent.
Jensen on the token economy:
https://www.youtube.com/watch?v=CU_TIDaixqM
