Ask HN: When will the AI bubble burst?
16 comments
It won't die. AI is to stay here. All the companies that are building the infrasturcture and the ones that are giving the APIs like Anthropic and OpenAI won't die probably because the world needs them right now. The ones to die would be the startups who are using AI to build something.
Bubble bursting doesn't mean death. It means bringing down the value significantly. When the dot com bubble burst, the Internet didn't die.
AI is here to stay, but the valuations are likely way, way too high. OpenAI may survive, but their investors may not. Nvidia's stock will likely tumble, which will bring down the S&P 500, etc.
AI is here to stay, but the valuations are likely way, way too high. OpenAI may survive, but their investors may not. Nvidia's stock will likely tumble, which will bring down the S&P 500, etc.
I used to think the OpenAI valuation was a bit steep and if you base it purely on revenue today, it's an absurd valuation.
But when you think Google did ~$265B in ad revenue last year (a large portion of their ~$350B in total revenue) - I think $500B is probably about right for OpenAI.
ChatGPT has 800 million users today. Not crazy to bet on that being 1B within the next couple years. If they turned on ads for that product alone the advertising revenue would be immense. They also have so many other avenues to find and generate revenue from (partnerships, FDEs, their PaaS).
I can see why investors are still piling in at $500B. The bets are 1) The costs will come down 2) they will start to capture a lot of revenue from their products and services 3) their consumer products will continue to grow.
None of these seem crazy to me. OpenAI could well be a legit $1T+ company (based on revenue + growth) in the not too distant future.
But when you think Google did ~$265B in ad revenue last year (a large portion of their ~$350B in total revenue) - I think $500B is probably about right for OpenAI.
ChatGPT has 800 million users today. Not crazy to bet on that being 1B within the next couple years. If they turned on ads for that product alone the advertising revenue would be immense. They also have so many other avenues to find and generate revenue from (partnerships, FDEs, their PaaS).
I can see why investors are still piling in at $500B. The bets are 1) The costs will come down 2) they will start to capture a lot of revenue from their products and services 3) their consumer products will continue to grow.
None of these seem crazy to me. OpenAI could well be a legit $1T+ company (based on revenue + growth) in the not too distant future.
If cryptocurrency is any frame of reference, AI hype will die down once an even more lucrative digital gimmick is invented. I don't know if it will "burst" besides all the wildly overvalued value-add corps like OpenAI and Xai. Unless something changes, they're dead-in-the-water even now.
Keep in mind, even the rapid abandonment of the cryptocurrency market didn't kill Nvidia. I don't think things will ever go back to normal for people who desire a 2012-esque tech outlook again.
Keep in mind, even the rapid abandonment of the cryptocurrency market didn't kill Nvidia. I don't think things will ever go back to normal for people who desire a 2012-esque tech outlook again.
If cryptocurrency is any frame of reference, it means AI hype will never die. Cryptocurrency is still alive and well today (and I'm saying this as someone who is NOT a fan of it)
The only thing that could pop the bubble is an alternative architecture for inference that doesn't need GPU clusters and datacenters to compete within the ecosystem.
AI itself isn't going away anytime soon.
Do you mean the bubble in the financial markets, or disillusionment of capital assuming they get to lay off half the employees for the same output?
Sry don't have expertise to answer your question but it got be wondering if there is question about this on polymarket.
The way I see it, world has too many GPUs, and they're too expensive, and it doesn't make sense by any measure.
It started with crypto, which inflated the value GPUs actually have. Now, LLMs are doing that. It's still the same thing, and I think it will transform again.
In a similar post in the past (3mo ago) I mentioned that this (ultra fancy GPUs) either evolves towards AR (augmented reality) or war (cyberwarfare), which are the most obvious next steps for "something-that-uses-fancy-GPUs".
Let's see some recent developments:
Apple dropping their AR glasses. This points towards a disinterest in augmented reality. It seems that's not a viable way to scale silicon supply chains.
Sora stays right in the middle of these two possibilities, being a proto-holodeck of some sorts ("hey, help create a new simulated world with you in the middle"), and also possibly a cyberwarfare platform ("hey, look, we could produce misinformation in levels previously unseen!").
It could be that these are not exactly organic market bubbles, but instead, stealthy cyberwarfare spending, dating back to the crypto days (possibly earlier!).
A burst, in this scenario, is just a way of moving money to these more shady goals without being too obvious about it.
In this scenario, the burst is also somehow a controlled burst. It means performance numbers are likely to be not that predictive of what happens.
I could also be wrong, but I don't think I am (it all circles back to silicon and geopolitical movements).
It started with crypto, which inflated the value GPUs actually have. Now, LLMs are doing that. It's still the same thing, and I think it will transform again.
In a similar post in the past (3mo ago) I mentioned that this (ultra fancy GPUs) either evolves towards AR (augmented reality) or war (cyberwarfare), which are the most obvious next steps for "something-that-uses-fancy-GPUs".
Let's see some recent developments:
Apple dropping their AR glasses. This points towards a disinterest in augmented reality. It seems that's not a viable way to scale silicon supply chains.
Sora stays right in the middle of these two possibilities, being a proto-holodeck of some sorts ("hey, help create a new simulated world with you in the middle"), and also possibly a cyberwarfare platform ("hey, look, we could produce misinformation in levels previously unseen!").
It could be that these are not exactly organic market bubbles, but instead, stealthy cyberwarfare spending, dating back to the crypto days (possibly earlier!).
A burst, in this scenario, is just a way of moving money to these more shady goals without being too obvious about it.
In this scenario, the burst is also somehow a controlled burst. It means performance numbers are likely to be not that predictive of what happens.
I could also be wrong, but I don't think I am (it all circles back to silicon and geopolitical movements).
That's like trying to time the market.
A bit its like asking when mankind will go burst, looking like it is destroying our environment like there is no end, but hoping that something new magically will fix everything before that point.
Will an AI saving killer app, technology or whatever happens before the bubble burst, or no? It might, a lot of companies with a lot of talented people and deep pockets are working on that, as their near term existence depend on it. But also you can hit rock bottom before finish inventing the parachutes.
Will an AI saving killer app, technology or whatever happens before the bubble burst, or no? It might, a lot of companies with a lot of talented people and deep pockets are working on that, as their near term existence depend on it. But also you can hit rock bottom before finish inventing the parachutes.
There was an interview on Fortune where a Wall Street executive believes it probably won't happen for at least 9 months, but will likely happen within the next 24.
https://fortune.com/2025/10/07/ai-bubble-cisco-moment-dotcom...
https://fortune.com/2025/10/07/ai-bubble-cisco-moment-dotcom...
C'mon, AI is a fraud. It should be called NN (Neural Networks). No size of data centers will fix AI's "forgetfulness" of the previous context. AI coding tools are not working well through existing code bases. If you let it loose over an existing code base, it will shit your code. This reduces AI to a merely coding helper.
Yup, AI is actually great for vibe coding yourself up a prototype in a technology you're unsure in but we all know that most of the work on a given systems is in scaling and maintaining the original idea.