Pascal's AI Wager
- Karl Johansson
- 1 sep.
- 3 min läsning
Uppdaterat: 7 sep.
Can you afford not to build AI data centres? Silicon Valley's AI tycoons say no.
How many data centres does the world really need? The common view now seems to be more based on a misguided belief in the transformational power of AI, but there seems to be no further developed answer than simply more. As the supposed use cases for AI are largely hypothetical it is difficult to estimate how much AI compute and buy extension data centres the economy will need. The logical way to find out would be gradually, building more as the current capacity is used up. But the media hype around AI is so strong that companies are throwing caution to the wind and blindly putting their faith in Sam Altman and the other AI tycoons. If the glorious AI future does indeed come to pass you have so much to lose that you should behave as though it will; the capex version of Pascal’s wager.
The main worry about any large scale infrastructure build out is that big fixed investments could turn out to be underutilised and thus a giant waste of money. I think for the reasons I go into greater depth in my essay No Acoustic Guitars in Silicon Valley that AI is a dead end economically if not technically. As such I think almost all the new data centres being built now are destined to be underutilised. Of course, the beauty of capitalism is that eventually someone will find a profitable and productive use of these new facilities. The problem with creative destruction however is that it does involve a lot of destruction.
And believe me when I say that there is a lot of destruction if the AI utopia fails to materialise. The Economist gives some statistics on the power usage in AI and normal data centres: “A non-AI data-centre computing unit, or rack, needs around 12 kilo-watts (kW) of power to run. An equivalent AI module requires 80kW when training large language models like the one behind ChatGPT, then 40kW when responding to users’ prompts.” What use are GPU-intensive compute if AI remains unprofitable or simply bad at replacing our current ways of doing things? Can we find some other use for power hungry chips or will old AI data centres be turned into badminton halls?
John Maynard Keynes wrote in the General Theory of Employment, Interest and Money: “When the capital development of a country becomes a by-product of the activities of a casino, the job is likely to be ill-done.” Which sums up the data centre craze well.
As much as the data centre folly is the fault of a media unwilling to ask tough questions to people like Altman and Amodei, it is but a symptom of deeper issues in the American economy. The power financiers and tech types have to set incentives and fashions coupled with an economy where high valuations are more profitable than producing goods result in a system where the casino really does dictate capital development of a country.
If you liked this post you can read a previous post about why the EU wasn't invited to the Alaska summit here or the rest of my writings here. I also have a section for longer reads I call essays here, I particularly recommend my essay on Silicon Valley and AI called 'No Acoustic Guitars in Silicon Valley'. It'd mean a lot to me if you recommended the blog to a friend or coworker. Come back next Monday for a new post!
I've always been interested in politics, economics, and the interplay between. The blog is a place for me to explore different ideas and concepts relating to economics or politics, be that national or international. The goal for the blog is to make you think; to provide new perspectives.
Written by Karl Johansson
Sources:
Kommentarer