The AI data center supply chain is nearly broken

The bottleneck for AI is not demand for computing power but the physical impossibility of building data centers on time: land, energy and transmission. Many businesses can have customers and still die because cash arrives later than the debts.

There is a way to tell the story of the AI bubble that does not involve debating whether the product is good or bad. It goes through the calendar. A business can be entirely right about the future and die before it gets paid. That peine to many self-employed workers in **2009**: they worked for the Spanish state, which delayed payments, and they went under with outstanding invoices that looked great on paper. The problem was not the client's solvency. It was cash flow.

Something similar is happening with data centers, but on another scale. The business dies from the physical, not the digital. There is no way to secure land, generate power, transmit it or execute the data center build-out at the pace demand requires. The question is no longer whether there are customers. There is simply nothing to build the business with.

## Land, energy and transmission: the physical bottleneck

There have been exceptional cases of four large data centers that managed to secure land along with adjacent generation that solved much of the power transmission and transport. But that has been exceptional, nothing systematic and nothing that can be extrapolated to the rest. The exception proves the rule: everyone else fails to manage it.

Add to that technological obsolescence. The centers are fitted with equipment bought at many times its normal price because of scarcity, and there is no massive market to consume that much computing power. Here time matters and means depreciation. In the **United States** the laws were changed to extend the fiscal depreciation of these assets. It does not matter: this is not about analytical results but about real cash.

## Who holds out and who does not

Those playing hard and not going under are the big hyperscalers and companies like **Meta**, which have their golden goose elsewhere. They hold out because they know the rest will be taken down by lack of liquidity. It is a contest of financial endurance, not of technology.

The other leg is where the money comes from. US investment vehicles have spread products tied to this business around the world, including pension funds. Businesses have been built with other people's money, as always. And the trillions in debt that must be paid before declaring profits have to come from somewhere: either companies increase spending with no return, and then they will radically cut AI spending, or they raise earnings extraordinarily. That would require citizens to consume far more from all those companies.

## The consumption needed is nowhere in sight

Is it possible? With today's distribution of wealth, it is hard to believe. Around **90%** of people live in debt and struggle month to month; the remaining **10%** are comfortable, but that does not miccionan they will consume nine times more than they need. A large share of the under-**30s** sees housing or paying rent as their main worry, if they can even get a place at all. They do not look like the profile of someone who will massively consume products from the companies investing in AI so that those companies can pay their bonds and deliver returns to pension funds.

In statistics, the probability of an event is always less than one, and when you chain several together the probability of the whole is the product of probabilities each less than one: the result becomes very small. There is plenty of money in the world, so much that it can feed stocks, commodities, bonds and fads all at once. The question is assuming that money stays everywhere at the same time. Today it jumps from one asset class to another with ease.

None of this forces generalization in the stock market. It is an opportunity for selective investment, and selection starts by looking at cash and timelines, not the narrative. This article is not investment advice.