And finally, in the last Daily Planet on hyperscaling: a podcast on whether we are in a bubble. Before we get into the substance of the podcast, let's first ask:
what is a bubble?
Among other things, it's a situation where expectations run far ahead of reality, where investments are unlikely to pan out. Nevertheless, a bubble must be based on the possibility of big returns - we don't (yet) have a "Real Estate on Mars" bubble, for it's not possible to imagine humans settling on Mars in large enough numbers to justify real estate investments.
From that perspective, there are at least three bubbles floating around:
The metaphysical-technical bubble: will AI be as good or better than humans at everything? Will it be a better mathematician? A better painter? When we talk about AGI or ASI, we mean this bubble.
The Adoption Bubble: even if AI approaches AGI levels, will companies adopt it? Will it change productivity? Will it help them create new, better products?
The Investment Bubble: even if AGI is at hand and it transforms how we make things, will the vast amounts of money poured into AI - especially Data Centers - ever produce a profit?
This week's bubble accusations (including today's) have all been about the third. If it bursts, it will impact the other two bubbles for sure, but that doesn't mean they will disappear.
The AI industry is experiencing an unprecedented surge in spending, with U.S. tech companies investing $300-400 billion annually in AI infrastructure, primarily on data centers and GPUs. This massive capital influx has sparked concerns about an economic bubble akin to historical booms in railroads, telegraphs, and broadband. Unlike those long-lasting infrastructures, AI hardware like GPUs depreciates rapidly, becoming obsolete within two to three years, creating a challenging environment for companies to recoup their investments. This rapid obsolescence, combined with the high costs of AI infrastructure—where GPUs can represent over half the data center expenses—raises the risk of a bubble burst.
The AI buildout is also distorting the broader economy by diverting capital away from traditional manufacturing and other sectors, reminiscent of the telecom boom in the 1990s. Private equity firms prefer large investments in data centers over smaller manufacturing ventures, exacerbating this capital shift. Additionally, the growing energy demands of data centers contribute to energy inflation, potentially causing consumer backlash and prompting some data center construction to move offshore.
Financially, companies use complex financing structures to obscure AI spending, increasing systemic risk. The bubble could burst within 2-2.5 years as operating costs and declining rental prices make data centers economically unsustainable. Despite these risks, AI’s transformative potential remains significant, promising profound long-term impacts even if a short-term crash occurs.