KI ist zu teuer: KI ist derzeit für niemanden wirtschaftlich rentabel, außer für die Baufirmen NVIDIA und die umliegenden Hardwarefirmen, die von der irrationalen Überschwänglichkeit eines Rechenzentrumsausbaus profitieren, der offenbar nicht in der von uns angenommenen Geschwindigkeit erfolgt
https://www.wheresyoured.at/ai-is-too-expensive/
48 Kommentare
To date, ed zitron’s claims are more accurate than those of scam altman or slop musk.
Ecologically it would be an absolute disaster if the speed at which we are informed this expansion is taking place. It amazes me that plans can be made to build without the required resources in place.
nvidia gets paid whether AI works out or not. that tells you everything about who’s actually winning here
This is a long, but excellent read.
No matter if you agree or disagree with the article’s content, this goes notably deeper and is better researched than the typical AI hype-good or AI hype-bad articles.
Recommended to anyone interested in GenAI technology.
Too expensive for the US’ electricity infrastructure but China, on the other hand, is perfectly positioned to have a thriving AI industry. Especially since US entertainment copyright law isn’t enforced and the stigma and resistance from the public and unions doesn’t exist.
Yeah that’s why they’re faking literally everything to make sure most people have no idea how bad things are until it’s too late.
Imagine what people would do if they knew they would have no rights to anything in less than ten years.
Fake it til you make it.
There is no business model nor some killer app that is going to make them money without them inventing a need. Its currently the most incestuous and ouroboros-like business I have seen since the Fashion Industry (making a bunch of crap none of us will ever wear etc).
The AI companies need some kind of regulated society switch to force people to use them for „something“. A manufactured need that makes them the only game in town for exactly what they have already built. I think they will try to make „compute“ the commodity and somehow ensure none of us can have powerful computers at home. I wouldn’t be surprised if we end up with an actual „token“ economy. A token can let you play a game with top graphics for a few hours, or write your essays. Pricing us out of owning anything but a terminal, or just a smart phone and a TV. It has already begun, if only due to the centers themselves sucking up supply and jacking hardware prices.
You don’t make money in a gold rush digging in the ground, you make money selling shovels.
It’s almost as if the demand is artificial….
They are also late to the party. The fad is over, yet they are building and pissing money on it
Something like a year or two ago, you had the slop boys saying stuff like „we’re going to need dozens of new nuclear plants by 2030 to keep pace with AI expansion“, and that’s when I knew for certain Zitron was worth paying attention to. The US and Europe have built like 2 such power stations in the last 20 years, so that was just not going to happen.
Forgetting that A.I will pay high dividends via revealing the level of incompetence behind extraction cults masquerading as technology companies.
I think it is primarily a problem of use-case. I’ve used some AI and AI-Assistants recently to do some market research and build small tools that help me collect data. If I had given that task to a coding firm, it would have cost me thousands, but AI did it for me at the cost of around $50.
It is definitely not replacing highly skilled tasks and is not yet ready to perform critical roles, but especially in the bottom-segment of code, where you need small tools that serve one purpose for a short or mediate timeframe, they do have their advantages.
But I primarily use them as human->machine translators. That’s what they are doing well.
Understand what I want and write a standardized script that does just that.
The same applies to the dot-com bubble of 2000.
It was commonly said during the gold rush(es) that the most profitable position was to be the man selling shovels….
“AI is a perfect storm of failed concepts and organizations, and the apex of the Era of the Business Idiot, an epoch where we’re ruled by people so thoroughly disconnected from the actual workforce that it was inevitable that a technology would be created specifically to grift them.
LLMs are dangerous for many, many reasons, but the under-discussed one is how well they play to a certain kind of executive imbecile. Generative AI is — to quote Mo Bitar — really good at doing an impression of work, much like most managers and c-suite executives, and even if it’s completely incapable of doing something, it’ll absolutely say it can and tell you you’re amazing for suggesting it.
And that’s why Business Idiots love it.”
This quote is so apt. I love the term “Business Idiot.” Perfectly describes the group of executives that are so completely out of touch when it comes to their own businesses
Ed Zitron has a firm grasp of the financials involved but I don’t think he has the right take on the reason for construction delays and cancelations.
These data centers are massive and require both heavy utility support and the approval through zoning and permitting. One strategy to deal with local pushback on data centers is to announce and plan them in more locations than you actually plan to build so you still meet your target. The results is cancelations being announced.
The delays also make sense in that there is only so much manufacturing capacity for long lead items like transformers and generators. Data centers have to compete with everyone else on these items in a country with neglected infrastructure.
Things like permitting and utility permission to operate can take months-to-years depending on the size of what you are building.
These data centers were never going to appear overnight even if society wanted them. People also have no desire to foot the bill for the data centers through higher utility rates because we know these are not being built for the public good, but to profit the very few.
But buuuuut buuuuuuuuut its not going to pop…
It’s a great big grifting circle jerk, and when their funds dry up they’ll take the whole world economy with them.
NVIDIA is like the Pied Piper of Hamelin, and everyone is following blindly until it crashes and it will.
Who wants to start a salvage company with me? Once this whole fad passes over, so many data centers are going to end up derelict and ripe with sweet, sweet ram! We could be kagillionaires!
‚AI‘ if you insist on calling it that, is good for pulling up random facts, searching recipes, and other reddit post level problem solving that already has a huge body of online data.
For the average person it’s a neat way to turbo browse the internet but It’s not worth Trillions of dollars.
AI token costs need to be 5x – 10x more than they are now, just for the AI companies to break even and START paying back the trillions they „borrowed“, let alone actual profit.
They’re just trying to get you hooked so it’s essential so you „can’t possibly live without“ it, which is why they’re shoving it into everything to make you dependent.
Then, when you’re unable to back out, they’ll start to increase costs to reflect the true cost of those tokens you’re using, and it’ll be too late.
Honestly, it’s a beartrap and you’re all falling for it… again… (e.g. cloud, etc…. remember when that used to be cheaper than on-prem?)
yeah the compute costs are still brutal and a lot of the revenue math only works if you assume insane scale or future price cuts
right now it feels like the infrastructure is ahead of the actual willingness to pay and a lot of companies are basically betting on efficiency gains and cheaper inference later to fix the gap
And the impact it will have on our bills and groundwater. The grift of all grifts, really.
The breakdown of the finances in this article is truly eye opening. This horrendous expenditure is NEVER going to be recouped.
Narrow scope AI is profitable and just fine, it’s the big dumb LLM wanna be AGIAI that winds up being very expensive for what it does.
Narrow scope, AI crunching out new drug candidate so we’re doing the pet detection in your camera. We’re running your robot. Vacuum are all just fine for being affordable and practical.
Nearest scope is also generally a lot faster because it’s specialized so you get more of that sweet sweet AI rapid data parson capacity out of the narrow scope then you ever get out of the LLM style AI.
Like if you try to get a LLMAI to crunch new drug candidates and compare its performance to a narrow scope, AI designed specifically to crunch drug candidates, the narrow scope AI will completely blow it out of the water while the LLM mostly just hallucinates. Narrow scope will post results in thousands of percent faster than an LLM..
Tell us something we don’t know
All of the big LLM companies are putting their „reasoning models“ behind paywalls at a time when the average person is having trouble just paying for groceries. That does not seem like a recipe for profit.
Nobody wants this outside of the profiteers.
I can’t believe a machine that just predicts most human sounding next word possible isn’t gonna revolutionise our entire society
It’s Tech Bubble 2.0, and you’d have to be completely ignorant to not see it. Which, unfortunately, most people are.
NVIDIA are the guys selling the pick axes during the gold rush
The big investors are shoving money into this to fund it to the point where companies fire a significant amount of their workforce due to AI only to get bamboozled by the price increase that they now can’t afford not to agree to.
They will expect government subsidies and they might get it.
Down with BIG DATA
Nobody wants this except for people that stand to make money from it.
while the article is fairly cogent in outlining the snake oil that’s being drunk and the reasons why its being drunk, the question remains: what’s next?
these companies can’t keep building castles in the cloud-y air. something is going to break sooner or later. regardless of the fact that a large segment of the population is unable to visualise that upcoming crash, what happens when it does arrive?
are we seeing the seeds being sown of big tech pyramid collapsing into a heap of rubble? and what then? what does one do to prepare from being burned alive?
What… do you mean spending massive amounts of compute to multiply giant matrices to achieve the same fucking outcomes of a few simple if-statements and maybe some loops for every damned problem under the sun isn’t a sustainable way of doing things?
Having an AI company nowadays has to be like running a steam-train with money for coal
Can’t AI figure out how to make AI cheaper?
It’s all a giant Ponzi scheme,
>If capex never stops being invested, you need revenues to explode dramatically — to the tune of effectively doubling [..] **tripling** **Amazon Web Services’ annual revenue** ([$128 billion](https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Fourth-Quarter-Results/?ref=wheresyoured.at))
This is such a good line and puts things into perspective. AWS literally runs the web at this point. To make this cap expend worthwhile it would have to triple AWS. Wow.
No wonder the new pricing on Github Copilot is literally 10-20x.
These subscriptions are like 90% subsidized.
Michael Moon, Milton Bradley representative, 1977:
> video entertainment is just a passing fad
https://gamerant.com/1977-article-predicted-video-games-fad/
Literally anyone on the planet can build their own standing wave memory tech for free and embody it in an android. There is no bottleneck. The slavery companies are trying to create an industry out of selling air.
Why can they just ask AI to make it cheaper? /s
*This* AI is. They unveiled economic AI in China last year. the whole reason this is so inefficient is so people can profit off all the waste.
AI is the current metaverse, except taxpayers will be left holding the bag when the bubble bursts
I remember back when you paid for something you were using and you got value out of it, that mattered. A company sold you a product and that was it. You didn’t have to worry about whether other bigger companies are also paying for the same product. Nowadays, I talk about how ChatGPT helped me get a thyroid cancer diagnosis and I get told that literally doesn’t matter.