I have previously written some bits about the AI-bubble:
- ‘Artificial Intelligence’ Is (Mostly) Glorified Pattern Recognition – June 2 2023
- Is The AI Bubble Ready To Pop? – Oct 3 2025
- Those ‘Scary’ AI Models Are Still Only Slop – Apr 24 2026
- It’s The Math – How China Is Avoiding The AI Bubble – Aug 4 2026
I lamented that the much hyped Large Language Models (LLMs), as proffered by OpenAI and Anthropic, are mostly unreliable search engines. There are only a few real use cases for these. That is why I argue that these do not justify the current AI investments.
But the AI companies (and those who promote them) want lots of money. As their is no real use for their very expensive models they always point to future cases. When asked about current use gains, they divert and point to science fiction.
A master of this tactic is Scam Altman, the head of OpenAI.
In a recent interview of him he was asked how he personally is using his company’s product. That is of course a softball question. It is like asking the CEO of Mercedes why he loves to drive his company’s cars. Any reasonable salesperson would catch the softball and laud his products.
Altman, however, veers off (@20:25):
Q: How are you using ChatGPT? How are you using it every day and what’s that look like?
Altman: Ahem- let me answer a more interesting question than that. How I would like to use it.
I think we are close to a world where you can have like a descendant of ChatGPT watch your computer screen all the time, watch every meeting you are in, like record every call, everything like that, have perfect context of your whole life, everything you see …
When I watched that blubber I had two questions:
- Who the hell would want such a thing?
- Why doesn’t Altman say how he is using his product?
A second softball interview with Altman, published three days ago, answers the second question.
Sam Altman is not using ChatGPT (@5:58) or Codecs, his company’s other product.
Altman: I have for 20 years been using computers the same way. I now have a magic thing called codecs. So do you. So does everybody.
That means I should completely be using my computer in a different way. I should not be clicking around uh you know pasting from one messaging app to another. I should not be scrolling mindlessly through my emails and trying to figure out which one is like least painful for me to open and respond when I don’t want to be dealing with it. I should not be like keeping a to-do list and doing sort of this like these wrote computer tasks in the same way that I have for so long.
And yet there’s like something in my mind that is encoded that like doing this kind of stuff is what it means to work and what it means to be productive.
What Altman is really saying here is that his company’s product is not useful for what he does every day. It could be useful, he say, if …, if … and if …. But it currently isn’t. At least not for him.
But still he is out there to hustle for more money to be thrown into ever bigger OpenAI models which will add nothing of value.
It is no wonder then that the rats are leaving his ship:
OpenAI’s Head of Data Centers Has Left the Company (archived) – WSJ
A key executive overseeing OpenAI’s data-center build-out, Chris Malone, left the company last week, according to people familiar with the matter, joining a wave of leadership departures ahead of a planned public offering.
…
His departure comes amid a broader exodus at the firm, which is working toward an IPO that is expected in 2027 and racing to catch up to rival Anthropic in sales to business customers. In recent weeks, Chief Revenue Officer Denise Dresser, Chief Operating Officer Brad Lightcap, and Fidji Simo, who served as second-in-command to Chief Executive Sam Altman, have all left.
OpenAI is on its death bed. Its demise will likely cause a larger crash. Then Microsoft will swallow its carcass.
Its main competitor, Anthropic, is just as lunatic (archived) and will also go down:
SpaceX’s record-breaking IPO tested the limits of an obscure financial metric. Anthropic’s could push it even further.
The maker of Claude is likely to tell investors its potential revenue opportunities are above $30 trillion, topping SpaceX’s $28.5 trillion estimate, according to people familiar with the matter.
Tech startups or other growing companies going public often estimate their “total addressable markets,” or TAMs, to show investors they have ample room to grow. Such figures estimate the amount of annual revenue a company could theoretically capture if it achieved 100% market-share using inputs ranging from industry data to bankers’ models.
…
To put its more than $30 trillion vision in context, the 191 technology companies in the S&P 1500 brought in $2.4 trillion in revenue last year …
The evaluations thrown around by these companies, and the hundreds of billions being invested into data centers to keep them going, are absurd when compared to the potential revenue these companies might make.
Imagine a product which is so good that every household in North America and Europe is willing to spend some $20 per month to use it. Those are 400 million households for a total revenue of $48 billion per year. That is a laughable sum when compared to the ~$1.1 trillion contracted by OpenAI and Anthropic with Microsoft, Amazon and Oracle for current and future compute.
There is no AI product in sight (but hype) and no revenue, that can justify such investments.
BCA Research (via Ed Zitron) agrees:
AI companies may need to ultimately generate $10 trillion per year in revenue to justify their capex. Barring a massive increase in productivity growth, this will be very difficult to achieve.
The capital expenditure that is currently put into LLMs/AI has no relation to its potential future revenue. It is wasted money.
There has yet to be any ‘productive growth’ due to AI. Within companies AI projects notoriously fail. On a personal use level just ask Scam Altman, who hasn’t seen any personal productivity gain yet during his work day.
The scare mongering about AI leading to job losses is also nonsense. It is a marketing device to argue for more capital investment. In the real world AI job losses are a myth.
Large Language Models are reasonably good, though still defective, semantic search engines. With some functionality added on top of them they can be used for text, code, picture or video generation. But as these models are inherently unreliable all of their outputs need human supervision.
There is only so much money people are willing to pay for what LLMs do. The costs of training new models exceeds their potential revenue by a wide margin. The underlying algorithms of current LLMs (transformer models) are compute intensive and make their use expensive.
There is nothing that justifies enormous investments in them.