Follow the Value

I predict sales of Nvidia chips will go up.

That’s not a particularly impressive prediction. A child of three could make it. Nvidia is helping finance infrastructure that will require Nvidia chips. The financing creates the conditions for the thing that makes the financing valuable.^1,2

»But did anyone’s back actually itch?«

When critics raised the possibility of circular financing, Nvidia CEO Jensen Huang rejected it: »Is this circular financing? No. OpenAI will pay the lease.«³

There’s something strange about this. Normally, investment and demand are separate things. I invest in a restaurant because I believe people will want to eat there. I do not normally give the customers money to buy the food and then point to their purchases as evidence that I made a good investment.

Imagine I manufacture back-scratchers. I give you £100 to buy £100 worth of my back-scratchers. You buy them. I now have evidence that there is demand for back-scratchers.

But did anyone’s back actually itch?

The answer might be yes. The transaction might create real value. But the signal is no longer independent of the thing it measures. It has become part of the mechanism that produces the demand.

This is an old problem. Sociologist Robert K. Merton called one version of it the »self-fulfilling prophecy« – a belief can become true because people act on it.⁴ Investor George Soros later called a related phenomenon »reflexivity« – participants’ perceptions alter the conditions they are trying to perceive.⁵ Philosopher and physicist Mario Bunge approached the problem through causation itself – systems act on environments that can, in turn, alter the systems.⁶

A weather forecast does not normally cause the weather. But suppose a forecast says Nvidia’s sales will rise, Nvidia finances the infrastructure that makes those sales possible, and the resulting sales are then used as evidence that the forecast was correct. At that point, asking whether the forecast was, in fact, »correct« becomes strange. It didn’t merely describe the future. It participated in making it.

Social media works through a similar loop. A platform learns what people want, shows them things, measures their reactions, and uses those reactions to decide what to show them next.

Eventually the system is no longer simply learning what people want. The system begins to teach itself what it wants people to want.

Markets have always contained feedback. Expectations influence investment; investment changes production; production changes prices; prices change expectations. But a sufficiently reflexive system can begin to inflate itself. This is where the »bubble« metaphor becomes interesting. The question isn’t so much whether expectations are optimistic. It’s whether anything of value enters the loop from outside.

If I give you money to buy my back-scratchers, the transaction tells me very little about whether anybody’s back actually itched. But if you then sell them to someone who genuinely wants them, value has entered the system from outside the loop. If nobody does, apparent demand can grow indefinitely without anyone needing a back-scratcher.

This is where the Trustor affair makes an almost comic appearance. In 1997, Joachim Posener and his associates acquired the Swedish investment company Trustor using Trustor’s own money. Posener describes the affair himself in the recent Netflix documentary Inside the Trustor Scandal.⁷ The company financed its own acquisition. At the extreme, you don’t need demand at all. All you need is a mechanism capable of producing the appearance of value.

That’s not necessarily where Nvidia and OpenAI are. The chips are real. The data centre is planned. The customers may be real. But the mechanism is visible enough to ask a larger question: how much of the apparent demand for AI-driven technologies is being generated by the AI economy itself?

If the answer were »a lot«, the economy would not merely be responding to a future. It would be participating in the production of the conditions that make that future valuable.

Which, by the way, need not be pathological. The same mechanism could be used deliberately. A climate NGO could fund heat pumps, which creates a market for heat pumps, which makes them cheaper, which creates more demand, which attracts more manufacturing. A future that helps finance itself isn’t necessarily a bubble. It may be a highly efficient way to fund the building of the futures we want.

The problem begins when we mistake the loop for the value, and grows when the futures we fund aren’t the ones we want.

The future can finance itself. Follow the value.

References