It’s annoying that the journalists and academics who, not long ago, were nattering on about ‘AI’ as an ethereal force, untethered from the Earth, are now talking about data centers with the fervor of new converts. Some of us have talked for years, and from the start, about the dirty, industrial nature of what’s called ‘artificial intelligence’ (and indeed, the entire tech industry).
Better late than never, I suppose.
More than one thing can be true at a time, people say; usually while trying to appear subtle. It’s a true statement but insufficient. Not only is more than one thing true at a time but many things are true, simultaneously and in relation to each other, moving through time. This is the heart of a dialectical view which can be applied to understanding the mania of data center construction, a mad dash to a landscape, dotted with ruins.
…
Recently, as if from nowhere (but of course, not as suddenly as it seems), data center projects have been spreading across the United States. On social media, the rough consensus is that these warehouses are evidence of a plan for mass surveillance and population control that requires hyperscale computing. This is plausible, Palantir exists, after all, but not the entire, or even main mover: it is a factor among factors moving dialectically with other elements.
It’s common for US people (who vaingloriously claim the title of ‘American’ as if the rest of the hemisphere doesn’t exist or arrived from the 12th dimension) to fret about things worth fretting about while failing to see political economy. In the case of data center mania, the circular economics and failing quest for super-profits motivating the tech industry and its camp followers (the ‘Neocloud’ firms and venture capitalists waiting for investment returns).
David Gerard recently said:
“The story of the AI bubble, is setting real dollars on fire to put imaginary valuation dollars on the books”
To set those ‘real dollars’ on fire requires having something physical to show for investors’ capital; thus, a plague of data centers. Capitalism is not rational and as Ali Kadri informs us, is based on the accumulation of waste.
Another factor to consider, from Ed Zitron:
In reality, OpenAI and Anthropic are the only meaningful companies in the AI industry. They are the majority of revenue, the majority of capacity and the majority of demand. Microsoft, Google and Amazon have exploited the desperation in a tech industry that’s run out of hypergrowth ideas, and created a near-imaginary industry by propping up both companies.
The mistake that most make in measuring the circularity of OpenAI and Anthropic is to focus entirely on the money raised — $13 billion from Microsoft and up to $50 billion from Amazon for OpenAI, and as much as $80 billion from Amazon and Google for Anthropic.
The correct analysis starts with measuring infrastructure. Based on discussions with sources and analysis of multiple years of reporting, I estimate that of the roughly $700 billion in capex spent by Google, Meta and Microsoft since 2023, at least 5.5GW of capacity costing at least $300 billion has been built entirely for two companies. This has in turn inflated sales through multiple counterparties involving NVIDIA, ODMs like Quanta, Foxconn, Supermicro and Dell, and created a form of market-driven AI psychosis that inspired Meta to burn over $158 billion in three years and the entire world to convince itself that AI was the biggest thing ever.
Zitron calls what’s occurring “AI psychosis” – an apt description (though one based on an idea of the lost sanity of an inherently insane system) for a massive build-out of infrastructure for a project that has not, and will not produce profit or useful work for the organizations in the grip of this mania.
This financial ‘psychosis’ is what the (mostly liberal) doomsayers who predict a world of perfect surveillance and control are missing. Of course, elites want this but they also want to make money and, having failed to accomplish this are doubling down in a way never before seen on this scale.
Exhibit C, a story from Tom’s Hardware:
New AI data center in Utah will generate and consume more than twice the amount of power the entire state uses — Kevin O’Leary’s 9 Gigawatt Utah data center campus approved
These stories are breathlessly reported, as if, by waving a magic wand of billionaire wish fulfillment, a gigantic data center will spring from the Earth. But there are other factors (remember, think dialectically) such as the impact of hydrocarbon supply constriction on the PCB supply chain and the lack of transformers.
…
The data center ‘boom’ or, more accurately, to borrow from Zitron, ‘psychosis’ is the final act in the inflation of the greatest of all bubbles, rivaling mighty Jupiter in size. Because the US is little more than a nuclear armed used car sales lot, commanded at every level by people looking for an angle, there is no one to stop this process.
It will run its course, leaving the population to pick up the pieces and scavenge for parts when the power and water run out. One can hope that at long last, as children of the not too distant future wander the ruins of data centers, lessons will have been learned.
Computational Impacts 