Blog homepage RSS feed Mastodon Bluesky About me
This blog uses a new RSS feed. Please update the old QuirksBlog feed you used to follow.
October 2026 1 September 2026 3 August 2026 2 May 2026 6 April 2026 9 2004-2021 blog
At Fronteers Dark Mode last Friday I had AI conversations with three or four people. At the end of each of them I asked: “Yeah, but what would those tokens really cost?” They didn’t know either, but agreed that it was the right question to ask. Is AI truly cheaper than the workers it replaces?
I encourage you to do the same. After each conversation about AI, ask what the true cost of the tokens would be. Nobody knows, but getting used to asking the question is a good idea.
AI is being very heavily subsidised right now. I still don’t know what the true cost of a token is. Nobody does. (This analysis says “hyperscalers have spent about $1.2 trillion on AI investments while only making $277 billion in revenues.” That’s $4.33 spent for every dollar earned.)
On the consumer side, this is the first time the tech giants release a service that is not “free” — and that after years of training us to expect free goodies everywhere. They can fund their free social networks and such by productizing the consumer, but there’s no way they can port the ad model to AI. The money just isn’t there. (The total global ad market in 2025 was about $300 billion. This analysis estimates that the hyperscalers need at least $1.3 trillion per year over the next ten years. That’s four global ad markets and a bit.)
Also, this is the first time the tech giants release a service that is cordially disliked by a large percentage of the world — maybe even to the extent that it influences sentiment about the tech giants themselves.
But consumer opinions don’t matter, it’s the corporate market that counts. So the question becomes: can AI efficiency save the world about 25% of its total IT costs? (The 2020 global IT market — untouched by AI — was estimated to be $5.2 trillion, and as we saw we need at least $1.3 trillion per year to keep AI running. The savings would not only come from IT, but IT would be expected to deliver the lion’s share.)
Quick history lesson: the industrial revolution, which replaced tradesmen by huge factories, didn’t take place because people liked huge factories, but because they produced stuff cheaper.
Right now I’m not seeing any evidence that AI is cheaper than the workers it replaces, especially not when taking into account the amount of checking that AI errors and bugs still require. I could be wrong, but it’s unprovable either way, and the numbers I quoted above seem to support my viewpoint.
The tech giants are creating something with the appeal of a 19th-century factory, complete with oppressed and estranged workers, without delivering the sole benefit: cheaper IT and other services.
Do they create their factories because they like them? Regardless of the costs? Did that make them bite off more than they can chew?
Hey, I’m just asking questions.
For this article I read a few overview stories about the cost of AI — and not in my local anti-AI socialist rag website, but in the sort of unsuspect capitalist publications that Very Serious Investors read.
I found that, while the authors still sprinkle sentences throughout their articles about “obvious benefits,” and company X saving $50K, and the general wonders tech brought to the world, they start to be pretty critical of the huge expenditures.
To me, this sounds as if the writers are preparing the world for a shifting viewpoint; for the moment that money will run out, unlikely as that might appear to anyone within the bubble. Oh, yes, we could be wrong, mr AI True Believer who pays for a subscription to the magazine we write these words in ... except that we’re not.
To me, it sounds as if at least part of the capitalist class is waking up to the fact that AI is just a bubble. And once enough people say “AI is just a bubble” it becomes reality.
So help it come about. After each conversation about AI, ask what the true cost of those tokens would be. Sow doubt. Hint at bubbles.
Let’s hope the bubble bursts soon. It’s only after the inevitable crash that we can assess the true cost — and benefits — of AI.
