Markets begin to price in data center risk

GPUs used as collateral lose value with every generation, and circular financing between chipmaker and customer inflates balance sheets, according to circulating criticism.

English · Original discussion in Spanish · Published

Markets begin to price in data center risk
AI's circular financing is starting to sustancia ilegal

Data center debt is partly backed by the very GPUs it buys, according to criticism circulating among analysts. The problem is that those cards depreciate over five or six years, and each new generation, roughly every one or two years, cuts into their resale value. When the collateral is worth far less than the balance sheets claim, the loan is worse covered than it looks. That is the crux of the matter.

The second piece of the machinery is circular financing: the chipmaker invests in the computing rental company or the AI lab, and that money flows back to it in the form of orders. A closed loop that inflates profits on paper and, when it cools down, leaves everyone staring at each other.

Who pays AI's electricity bill

The other front is energy. A large share of the energy costs of AI companies in the United States ends up falling on the taxpayer, according to circulating criticism. All very free-market, sure, right up until it's time to open the public tap. The new electrical infrastructure is being designed for data centers before businesses or households, and that reshuffles priorities without asking too many questions.

The ACS slump nobody wants to look at

The market has already started pricing some of this in. The construction firm owned by Florentino Pérez went from 140 euros to 90 euros per share in a few months, a slash of around 35% from its peak. And yet it is still up 27% from a year ago, because the earlier rise had been that of a promising startup. For a construction and services company, the precedent is hard to find.

Why patents don't sustain this business

Some argue that a company's financing should rest on patents for material things and processes, not on debt or the end user. In this sector it weighs less than it seems: almost all the foundational techniques were published in open papers; what separates one lab from another is data, computing capacity and internal know-how, which is protected as a trade secret rather than by patent. Software patents are hard to defend, especially in Europe, and expire after twenty years, an eternity in a field that changes every few months.

The loan that ends up with the taxpayer

The question hanging in the air is who absorbs the risk when the loop breaks. If the debt is worse covered than the balance sheets claim and energy is paid by the public purse, the uncomfortable scenario is the usual one: socializing the losses. The blowup, as they say on the street, is not paid for by those who inflated the profits.

What if the slump is already priced in and what comes next is another round of public money to compete with China?

Also available in: Deutsch

Summary of a discussion on Burbuja.info - Foro de economía, actualidad y política., translated from Spanish and reviewed before publication. Read the full discussion (19 replies).

More summaries

All summaries in English →

-
Back