News AI bubble is worse than the dot-com crash that erased trillions, economist warns — overvaluations could lead to catastrophic consequences

[USAFRet]

I sure wish there was a way to down vote here.

Just post your opposing viewpoint.

Of course, be 'nice' about it.

 

[atomicWAR]

Just post your opposing viewpoint.

Of course, be 'nice' about it.

I oppose me opposing your very correct viewpoint. Have a wonderful day!!! (takes myself out back and jumps myself...jk)

 

[Eximo]

I oppose me opposing your very correct viewpoint. Have a wonderful day!!! (takes myself out back and jumps myself...jk)

Oh great, the nuclear option...

 

[Amdlova]

Crypto and AI will die when the Earth flip poles... Get ready your abacus...

 

[jlake3]

My last employer was an engineering and manufacturing firm where mid-level management wanted to roll out AI to impress higher-level management, and everything we tried felt like a solution looking for a problem.

We tried AI document summaries! ..."what do you mean 'What AI services do our NDAs allow us to upload to? Uh... You're exempt from this initiative.'"

We tried using AI to research technologies we wanted to implement! "...I've actually done some research on this, and I recognize some of what the AI is presenting as facts as being marketing copy from someone selling this technology. And in the marketing copy it's hyped up against hand work, they don't compare it to the tech we're currently doing, but the AI summary makes it sound like advantages such as 'more repeatable' and 'less labor hours' are relative to ALL competing options."

We contemplated using AI for calculations! "...so I've gotta take the measurements from the CAD, type the measurements into the AI and explain what I want it to calculate like I'm writing a word problem... or I can put the numbers into highlighted boxes on the Excel sheet we made for this, and get an answer with only 12 keystrokes?"

I tried just looking up instructions for how to do a certain thing on the office 3D printer, and Gemini leapt in to hallucinate a walkthrough that involved menus and settings that don't exist in the very thorough documentation that exists on the manufacturer's website.

Machine vision for detecting part orientation and/or spotting defects has gotten both better and faster with new models and NPUs that are cheaper and perform better per watt, but that's a field that existed before the current AI boom and has been retroactively rolled into it.

The idea that we skipped 10 years ahead and it's undervalued by trillions of dollars because it's gonna invent new processes that revolutionize every industry and discover new science... That's not the experience I'm having. And right now, it feels like the massive valuations depend on someone making an AI that is capable of all the things that people have hyped up AI eventually being able to do... but whether that is possible with the current approaches is still an open question.

 

[DougMcC]

My last employer was an engineering and manufacturing firm where mid-level management wanted to roll out AI to impress higher-level management, and everything we tried felt like a solution looking for a problem.

We tried AI document summaries! ..."what do you mean 'What AI services do our NDAs allow us to upload to? Uh... You're exempt from this initiative.'"

We tried using AI to research technologies we wanted to implement! "...I've actually done some research on this, and I recognize some of what the AI is presenting as facts as being marketing copy from someone selling this technology. And in the marketing copy it's hyped up against hand work, they don't compare it to the tech we're currently doing, but the AI summary makes it sound like advantages such as 'more repeatable' and 'less labor hours' are relative to ALL competing options."

We contemplated using AI for calculations! "...so I've gotta take the measurements from the CAD, type the measurements into the AI and explain what I want it to calculate like I'm writing a word problem... or I can put the numbers into highlighted boxes on the Excel sheet we made for this, and get an answer with only 12 keystrokes?"

I tried just looking up instructions for how to do a certain thing on the office 3D printer, and Gemini leapt in to hallucinate a walkthrough that involved menus and settings that don't exist in the very thorough documentation that exists on the manufacturer's website.

Machine vision for detecting part orientation and/or spotting defects has gotten both better and faster with new models and NPUs that are cheaper and perform better per watt, but that's a field that existed before the current AI boom and has been retroactively rolled into it.

The idea that we skipped 10 years ahead and it's undervalued by trillions of dollars because it's gonna invent new processes that revolutionize every industry and discover new science... That's not the experience I'm having. And right now, it feels like the massive valuations depend on someone making an AI that is capable of all the things that people have hyped up AI eventually being able to do... but whether that is possible with the current approaches is still an open question.

Let me worry you a little:
We're finding AI extremely effective, and it is generating significant revenue for us. I know people in various adjacent industries, and their experience is the same. There's good reason people are pretty hyped about the potential that we've only begun to tap.

Anyway, the worry you should have is: maybe your company is just going to be on the losing end of this as your competitors who figure out how to leverage AI for a significant jump in productivity leave you in the dust.

 

[Alex/AT]

Yes. This bubble will burst.

 

[Alex/AT]

Just think about it... every government of the world will need to leverage their own AI infrastructure...

The power industry won't be able to cope. And given the 'hallucinated' results, it won't allow to profit much in any precise & practical area, is a sheer loss against power already (only the rising hype keeps it afloat). One of the reasons this bubble will burst.

 

[Persister]

I sure wish there was a way to down vote here.

It forces you to say what you mean, which is not a bad thing.

 

[Persister]

There are far worse things in history and life than bursting financial bubbles, but everyone must be aware and stay clear of large collapsing structures.

AI is definitely going to change our lives (and it already shows in some areas), even more so than the Internet did. But the stock valuations and infrastructure spendings are way ahead of the results/returns. It's like we skipped 5 or 10 years ahead. And just like in the case of the Internet companies, there will be just a few winners and hordes of losers. Hopefully, people's retirement accounts are not going to get devastated too much from this.

 

[tamalero]

The all-time high for Amazon's stock (AMZN) was $242.06, which was reached on February 04, 2025. Sure, some dot-com stocks crashed — but Amazon didn’t just survive, it became a pillar of the global economy. If you bought in the 90s, you’re not licking wounds, you’re sitting on a fortune!

AI isn’t just hype —it changed my daily routine and replaced internet search. It continues to be integrated everywhere. It's a huge deal. The P/E ratios may be high but weren't not talking Tesla (180) level projections for the big 7. The real question isn’t whether AI is transformative — it’s whether investors are pricing in that transformation too early.

Amazon wasn't dependent on a single technology, they evolved. Plus they were before then a multi billion store.
Not every Tech company have a sidebusiness that can take the hit.
Take Disney. Many of their business are floundering and kept afloat only by their parks and cruises sectors.
If they were not that expansive, they would have died.
And many many many companies are 100% AI or united to AI.
That includes a lot of tech companies right now.

Skynet will become real. 😛

I think weŕe closer to Idiocracy to be honest.

Let me worry you a little:
We're finding AI extremely effective, and it is generating significant revenue for us. I know people in various adjacent industries, and their experience is the same. There's good reason people are pretty hyped about the potential that we've only begun to tap.

Anyway, the worry you should have is: maybe your company is just going to be on the losing end of this as your competitors who figure out how to leverage AI for a significant jump in productivity leave you in the dust.

Sure, until your numbers, data, etc.. are all cooked because AI s hallucinated variables.
And then you will have to hire people (which you promised the investors you would never hire again because of AI) to fix the AI issues.

 

[jlake3]

Let me worry you a little:
We're finding AI extremely effective, and it is generating significant revenue for us. I know people in various adjacent industries, and their experience is the same. There's good reason people are pretty hyped about the potential that we've only begun to tap.

Anyway, the worry you should have is: maybe your company is just going to be on the losing end of this as your competitors who figure out how to leverage AI for a significant jump in productivity leave you in the dust.

• I said "my last employer", past tense, so I no longer particularly care if that company is on the losing end of anything. But the implied threats are very cool, and definitely make me more open to your side.
• What's your industry and what problems are you solving? I had four specific examples of attempts to use AI in a small engineering office with attached machine shop where every job was custom, and how it failed to deliver on the hype each time. You brushed it off as basically "Works at my job, you must be doing it wrong" with zero specifics or suggestions.
  • We actually tried AI-generated meeting notes too, to make five attempts in my department alone. Between the things they got wrong and the things it included that didn’t really need to be included (like diversions off topic, speculation, etc.), they couldn’t be trusted blindly and needed a human review and correction that took half as long as the meeting itself did.
  • Our IT guy used AI to generate a new policy document outlining how they planned to use extremely invasive monitoring software that employees were all forced to sign, and it turned into an absolute dumpster fire as they got deluged in questions about privacy and data retention where they either couldn’t answer, or the human answer contradicted the AI-written policy. And we learned it was AI because the IT guy accidentally outed himself by typing his prompt asking ChatGPT to respond to employee concerns into MS Teams instead of ChatGPT by accident and people saw it before he deleted it and replaced it with more self-contradictory slop. We burned days worth of man-hours on internal discussions, and came out of it with morale and trust in the toilet.
• There's big blockchain vibes to everyone trying to sell AI in this thread: It's gonna change "everything" and the valuation is gonna go to an unfathomable number, so you've gotta get in now even if it's got a lot of problems at the moment, because otherwise you'll get left behind (and be poor)… and don’t mind the energy use, it’s worth it and also we have a plan to at some point in the future make it sustainable.

 

[Arkitekt78]

The all-time high for Amazon's stock (AMZN) was $242.06, which was reached on February 04, 2025. Sure, some dot-com stocks crashed — but Amazon didn’t just survive, it became a pillar of the global economy. If you bought in the 90s, you’re not licking wounds, you’re sitting on a fortune!

AI isn’t just hype —it changed my daily routine and replaced internet search. It continues to be integrated everywhere. It's a huge deal. The P/E ratios may be high but weren't not talking Tesla (180) level projections for the big 7. The real question isn’t whether AI is transformative — it’s whether investors are pricing in that transformation too early.

It's hype... if you are that invested in is use, you will get what's coming to you. Do the research... its not good.

 

[Arkitekt78]

Problem is that this will also leads to massive layoffs while CEOs happily scurry away with golden parachutes while blaming anything else but their own work.

Then they should find new jobs now.

 

[JJAbery]

Every tech nerd may THINK that, but they don't know. Keep waiting and get left behind.

 

[CyberFlash]

Regardless of whether the suspected AI bubble bursts or not, end result is that commoners will get screwed while the elites chill. Annoying never ending trend of our world sadly.

 

[tamalero]

Let me worry you a little:
We're finding AI extremely effective, and it is generating significant revenue for us. I know people in various adjacent industries, and their experience is the same. There's good reason people are pretty hyped about the potential that we've only begun to tap.

Anyway, the worry you should have is: maybe your company is just going to be on the losing end of this as your competitors who figure out how to leverage AI for a significant jump in productivity leave you in the dust.

Then they should find new jobs now.

New jobs where? everyone is cutting to ride the wave of AI.

 

[hwertz]

Good article, people seem to be riding the AI hype train hard and a voice of reason is good to have.

Incidentally is the post moderation AI? I had a post listing issues with AI and like 'that's not worth billions? No kidding' (with the 'Good article' bit after that) and the moderation refused to let me post it even after several edits. Like the AI didn't like itself being criticized LOL.

 

[USAFRet]

Incidentally is the post moderation AI?

You mean the moderation here at Tom's Hardware?
I assure you, good sir, I am not an AI bot...😉
(of course, that is what an AI would say)

 

[DougMcC]

• I said "my last employer", past tense, so I no longer particularly care if that company is on the losing end of anything. But the implied threats are very cool, and definitely make me more open to your side.
• What's your industry and what problems are you solving? I had four specificexamples of attempts to use AI in a small engineering office with attached machine shop where every job was custom, and how it failed to deliver on the hype each time. You brushed it off as basically "Works at my job, you must be doing it wrong" with zero specifics or suggestions.
  • We actually tried AI-generated meeting notes too, to make five attempts in my department alone. Between the things they got wrong and the things it included that didn’t really need to be included (like diversions off topic, speculation, etc.), they couldn’t be trusted blindly and needed a human review and correction that took half as long as the meeting itself did.
  • Our IT guy used AI to generate a new policy document outlining how they planned to use extremely invasive monitoring software that employees were all forced to sign, and it turned into an absolute dumpster fire as they got deluged in questions about privacy and data retention where they either couldn’t answer, or the human answer contradicted the AI-written policy. And we learned it was AI because the IT guy accidentally outed himself by typing his prompt asking ChatGPT to respond to employee concerns into MS Teams instead of ChatGPT by accident and people saw it before he deleted it and replaced it with more self-contradictory slop. We burned days worth of man-hours on internal discussions, and came out of it with morale and trust in the toilet.
• There's big blockchain vibes to everyone trying to sell AI in this thread: It's gonna change "everything" and the valuation is gonna go to an unfathomable number, so you've gotta get in now even if it's got a lot of problems at the moment, because otherwise you'll get left behind (and be poor)… and don’t mind the energy use, it’s worth it and also we have a plan to at some point in the future make it sustainable.

I didn't mean to imply anything. AI is a very real threat to people's jobs, and if you hate on AI for that reason I'll happily clap along.

I'm in Insurance Software so we have:
AI helping us produce more/better software faster, which because we've been fairly aggressive in adoption has us now creating a gap with our competitors who were slower to take advantage. We are selling more software faster.

We are also using AI in those features, creating workflows that omit previously necessary humans and saving our customers a LOT of money. People usually worry about hallucinations in this context but those are largely solved now, and the result is the output from AIs is consistently better than the humans they replaced. Not perfect, but already better than human, and getting better with time. If your job is digital cog (that is, you can and do do your job from a computer), then AI is coming for your job. Adjacent industries are seeing this too: logistics, banking, defense, travel, realty .... lots of places where people are doing jobs that involve forms and very basic domain expertise where it is just not proving very hard at all to automate now where that automation was previously much too complex for anyone to solve.

 

[TerryLaze]

Sure, until your numbers, data, etc.. are all cooked because AI s hallucinated variables.
And then you will have to hire people (which you promised the investors you would never hire again because of AI) to fix the AI issues.

A.I. only hallucinates if the data that it's being fed is wrong, basically only in public A.I. that uses other public A.I. that is fed by internet trolls and so on.
As the saying goes "garbage in, garbage out"

Big corpos are building their own A.I.s that they only feed with their own data to do the thing they want it to.

 

[hwertz]

You mean the moderation here at Tom's Hardware?
I assure you, good sir, I am not an AI bot...😉
(of course, that is what an AI would say)

Oh no I didn't post and then have it not show up... I know the moderators are human. I got an automated message saying the message was rejected because it resembled spam or some such thing, and to try editing and try again. I rephrased things about 5 times and it was rejected, so I started over LOL. I had a list of AI falilings (hallucinations, people with 6 fingers, customer support AIs that can't answer basic questions or make up a response, etc.) and snarked that "of course that's worth 100s of billions of dollars", something in my snark triggered the filter.

 

[USAFRet]

Oh no I didn't post and then have it not show up... I know the moderators are human. I got an automated message saying the message was rejected because it resembled spam or some such thing, and to try editing and try again. I rephrased things about 5 times and it was rejected, so I started over LOL.

Some posts with specific keywords are automatically rejected. Due to being associated with SPAM.

 

[hwertz]

Some posts with specific keywords are automatically rejected. Due to being associated with SPAM.

It's all good I was mainly joking that the filter might be an AI that might not like posts bad mouthing AI. I figured I probably had some word or phrase in there that triggered a filter.

 

[hwertz]

A.I. only hallucinates if the data that it's being fed is wrong, basically only in public A.I. that uses other public A.I. that is fed by internet trolls and so on.
As the saying goes "garbage in, garbage out"

I can assure you that's not true. If you ask something that it doesn't know the answer to, rather than saying 'I don't know' they tend to just come up with a plausible sounding but incorrect answer. I asked if a particular engine used aluminum or steel heads (I meant to say cast iron) and it assured me it was steel (I don't know if any car used steel heads). Then I asked again if it was aluminum or cast iron and it said it was cast iron (it's not, it's aluminum.) That's not from internet trolls, that's from it not knowing the answer and just making one up.

Of course, if you use AI as an 'expert system'... it's used for specific purproses so it's trained on the data it needs for it's purpose, and is used to get particular results or answers from that data, you can have one that will not hallucinate.

And, I assume this is a solveable problem, there's some way to have the AI realize the 'connections' it's going through are not strong* (as they would be for something it actually saw in it's training data), and therefore it's not confident in it's answer. But current ones don't seem to do so.

*What I mean by 'connections not being as strong', the model has weights in it, so you have a 'neuron' with several connections and weight on each one saying which it is more likely to pick; data it's confident on will have weight strongly pointing in one direction, like (if there were only 2 choices) 0.8 toward the 'correct' answer and 0.2 toward the other. If it's not confident it'll be like 0.52 and 0.48.