back
147 comments
Outside a relatively small world of circular investment and FOMO feeding FOMO the general consensus seems to be “let it burn.”

It appears very unlikely we will ever see an IPO of OpenAI. Anthropic appears less doomed, but still iffy at best. Tons of other large, but little discussed, AI startups are just dead-companies-walking at this point.

The likes of AWS are showing good headline numbers but are taking out massive debt to build infrastructure that looks increasingly unneeded. Those with capacity are looking to offload it, quickly. Yes AWS has “committed contracts” for this capacity but if those commitments are with shaky AI startups then it’s mostly just fluff PR and these hyperscalers will get left holding the bag on all this debt.

Anthropic is almost purely a model company. They own close to no data centers.

If models are becoming commodities, and the main bottleneck is actually serving them at scale, Anthropic does not appear particularly well positioned.

If AWS had a ~60-80% margin for decades, I see no reason why inference can't have a ~60-80% margin for quite some time.

The problem is, if costs continue to drop ~90% for the same level of quality every 18 months, demand is unlikely to grow 10x to keep the revenue stable.

Who knows. Jevon's paradox. But the cost/quality is dropping too fast that it's hard for me to imagine demand keeps up long term to keep revenues (and profits) GROWING.

> infrastructure that looks increasingly unneeded

This is the crux that needs to be substantiated. Without substantiation none of your other arguments hold up.

It's holding up the entire US stock market and therefore the entire US economy and the dollar value. You probably don't want this particular Atlas to shrug.
"increasingly unneeded"?

In what way/shape/form is the infrastructure showing signs that are contrary to growth in need/demand? I'd really like to know if there's something I'm missing that I should keep my eye on, thanks

I mean, talk about circular investment and FOMO all you want, but if you use Fable every day you know what's coming. This is no bubble, we're just getting started.
More and more companies are signing up with OpenAI and Anthropic at near exponential growth to automate everyday tasks. If anything, these two should manage to IPO just fine. The rest of the downstream startups probably won't make it.
I would argue we still have not even really gotten started.

What do we have in the decade ahead? Robotics in every household, models 10x+ faster and more intelligent than today.

Really no significant impact in life sciences, R&D, and 'offline' world / robotics today as of yet, which is where most of the value will live.

Even if we agree with this take (and I do think it's a likely take that you're right on the long term), it doesn't change that it seems likely we're in a bubble, and it probably will pop.

We see a similar paradigm with lots of revolutionary technology. The initial promise is high, people get very excited, lots of money pours in, and.... 15-30 years go by before we start seeing real impact across the economy at large.

It's just a real slog to actually implement and roll out new tech.

So take robots: I can promise you that you won't see robotics in every household in the next decade (especially so if we exclude the current market of robot vacuums). Even if a company makes an incredibly capable robot "today" (and to be clear - they are not) it won't have time to scale out production, reduce costs, generate a used market that's accessible to less wealthy consumers, deal with regulatory hurdles and quality problems that only pop up in real-world usage, etc...

It's just slower than you're implying.

The change very well will happen (I'm inclined to agree that things are going to shift). That doesn't mean that the current investment is sane and will pay off.

So many historical examples of this, just two here real quick:

- Ford built his first automobile in 1896, founded a company in 1901, went out of business, got sued by ALAM, didn't build more than 10k Model T's until 1910, then only finally hit real scale (of low hundred of thousands of units) in 1913: More than a decade to "basic scale". Household ownership didn't hit 60% until 1929... 30+ years later.

- The initial web enthusiasm, followed by the dot-com crash in early 2000s...

Robotics in the household has been the realm of science fiction for decades and it still hasn't happened. We get dedicated, compact machines like dishwashers and washing machine/dryers, that's it. Most recent innovation has been the automatic vacuum.

If you want someone else to do household chores, hire someone. You can pay someone to do your house chores for years for the cost that these things will have initially.

The absolutely massive scale and influence of the Web happened after the dot com burst.
Yep, that is what the bet is. The build out will continue, even if it looks different.

The Internet was a 'bubble' at one point, and after it crashed in 2000, it didn't go away, it continued to build out. We're still using the Internet after the Internet bubble popped.

Of course we know it. It’s been obvious since at least 2023. Everyone in AI oversells, except a few companies that built an actual business with revenue, like Midjourney. There is no AGI coming anytime soon no matter how much hype is being thrown around. We are not in the singularity. However, peak bullshit is NEAR.
The trick the frontier labs have done is define AGI as "better than humans at the vast majority of valuable knowledge work" which is definitely not what most people think it means.
There is something I have been pondering recently. If we compare the cost of AI subscriptions (let's say Claude's 100/month) to a median developer salary (let's say 100k/year to 200k/year), the difference is orders of magnitude. This fills like a gap that needs to close. I suspect llms are too cheap right now but will raise their prices to a point where only big companies will be able to afford subscriptions to use them. I think soon we will see models that are only sold at very high prices.
This is something I have been pondering recently. If I compare the cost of a plunger ($23.99 on Amazon) to a median plumber salary ($62,970 per year per BLS), the difference is orders of magnitude. This feels like a gap that needs to close. I suspect plungers are too cheap right now but will raise their prices to a point where only big companies will be able to afford subscriptions to use them. I think soon we will seen plungers that are only sold at very high prices.
Why is this a gap that needs to close? Can you elaborate more than feelings? Right behind these big models are local inference with AMD/Apple having a great hardware start and a lot of local sized models making great progress.
Open weight models tell a different story. Inference is not that much more expensive than what a 100$ plan would allow and will only get cheaper (for current capability models of course, frontier not so much).
Also, something else to add. At first I thought no one wants to build data centers in hotter areas in the middle of deserts (many places in the American continent). So nobody would spend money building a data center in the Chihuahuan Desert for instance. However, a game of latencies will either require cover llm access from these areas, or make people move closer to the other data centers. In the former, llm prices will go up; in the latter, there will be a migration towards data centers that increase the price of the areas around.
LLMs are not as cheap on enterprise plans.
> I suspect llms are too cheap right now but will raise their prices to a point where only big companies will be able to afford subscriptions to use them.

"A.I." == "Artificially Inexpensive"

I don't understand this argument.

Go to OpenRouter and look at all of the unsubsidized providers.

> 100k/year to 200k/year

Is this range just Silicon Valley or what is this? Even including just Europe, you're looking at a lower bracket of 10k. If you expand to the rest of the world... Or do you think rich cities in the USA, where developers make 100k+ per year, can alone sustain this industry?

It's obvious. Most people were expecting this.
"The line separating investment and speculation, which is never bright and clear, becomes blurred still further when most market participants have recently enjoyed triumphs. Nothing sedates rationality like large doses of effortless money. After a heady experience of that kind, normally sensible people drift into behavior akin to that of Cinderella at the ball. They know that overstaying the festivities ¾ that is, continuing to speculate in companies that have gigantic valuations relative to the cash they are likely to generate in the future ¾ will eventually bring on pumpkins and mice. But they nevertheless hate to miss a single minute of what is one helluva party. Therefore, the giddy participants all plan to leave just seconds before midnight. There’s a problem, though: They are dancing in a room in which the clocks have no hands." Warren Buffett 2000 https://www.berkshirehathaway.com/2000ar/2000letter.html
> The same thing happened with their massive investment in Anthropic. They accounted for $53.4 billion due to deals with Anthropic last quarter. He said if you follow one Anthropic dollar through the earnings release, it's counted in AI business revenue, chips business, and AWS segment revenue.

That is insane if that is true, is that even legal?

There's a bit of Chinese whispers happening here. If you read the original piece they're talking about -- https://www.theregister.com/paas-and-iaas/2026/07/31/amazon-... (definitely worth a read, it's hilarious) -- you'll realize it means that that Anthropic dollar is "counted" multiple times in the marketing of three different Amazon businesses.

As far as accounting of revenue is concerned, it would have been counted as appropriate. Else, like you, I would guess it's not very legal.

It is. However remember that when the bubble pops it all works in the reverse direction too. Suddenly you have to mark down investment losses, missed revenue, and written off commitments.

Companies can go from looking really good to a complete financial mess almost overnight when all that leverage and self-reinforcing stuff unwinds. See last weeks headlines for one such scenario.

> And the fundamentals here are OpenAI and Anthropic, which are massively valued companies. They have humongous commitments and are generating real revenue on the order of twenty billion a year.

I think the size of their commitments is predicated on demand. Anthropic's annualized revenue run rate is now close to $50 billion, a fivefold increase from a year before [1]. They are making big investments, like $200 billion on Google's TPUs over the next five years [2], but those numbers seem justified by their expected revenue this year alone. If Anthropic cannot capture that revenue, someone else will.

Stock market valuations are a different beast, I personally think we have been due for a correction for ages now. But criticism of AI investment and particularly betting that it will all come crashing soon appears misguided to me. I can see a future where AI expenditures shifts around, not a future where everyone simply stops spending in AI all of a sudden.

[1] https://www.marketscale.com/industries/software-and-technolo...

[2] https://www.resultsense.com/news/2026-05-06-anthropic-200bn-...

I mean we are due our 6-8 year financial crash that we won't learn from. Once again it'll be coming from the USA's feral financial investments all to be bailed out by the tax player whilst the rest of the world picks up the pieces. Maybe its time we moved away from the petro-dollar if the USA can't be trusted to keep its finances in order.
Financial markets are already diversifying away from dollar as the sole reserve currency. Think swiss francs are doing pretty good.
There's a bubble around data centers, mainly. That's fueled by projected demand of AI and assumptions companies make about how the pie for that revenue is going to be divided up.

What's very real is the rapidly growing amount of revenue for both OpenAI and Anthropic. That's already tens of billions per year and growing quite rapidly. Investments against that kind of revenue aren't completely horrible. To a point. But at the multi trillion dollar valuation level, of course there are going to be issues with living up to those expectations.

In my view some of the base assumptions are looking not so solid currently. It's not a given that OpenAI and Anthropic will end up with most of the revenue. The Chinese trust Silicon Valley just about as much as vice versa. Which is why they are doing their own models, chips, and data centers. This is driving a rapid commoditization for things like frontier models, open model weights, and chips. This in turn gives countries outside the US a lot of options to stay independent. Which burst the bubble that all that global revenue was going to flow towards Silicon Valley. Some of that still might. But that will have to happen based on cost and merit.

There are also geopolitical circumstances that cause most data center plans to be bottle necked on permitting, chip shortages, grid connectivity, availability of gas turbines, gas, solar panels, inverters, batteries, water, and other resources. As it turns out, you can't just willy nilly plan for hundreds of GW of data centers and expect those to pop into existence overnight along with all the needed infrastructure. Most of the announced/planned capacity for this will likely not be realized. Certainly not this decade. 5-10% by 2035 would be a lot given all the constraints and scarcity. No amount of reality distortion can change the physical constraints on this topic.

The good news is that most of the money needed for this hasn't been spent yet. And what has been spent won't be going to waste. Up and running data centers are a hot commodity right now. They won't be running idle if a bubble bursts. But probably investors dreaming of multi trillion dollar IPOs might be a bit more cautious now that SpaceX stock is trading well below its IPO value.

Wow I hope this is true. This can only be good for the world.

I'm sure AI will still come but it is too disruptive now. It needs a slowdown. As usual all the greedy investors are to blame.

The sooner it pops, the better. Later on it will be even worse.
I'm looking forward to prices for RAM and SSDs returning to sane levels.
Just before coming to read this article and thread, I read how Hyatt got rid of something like 30% of their call centre staff, and how the industry is gearing for replacing human work with AI.

There is sadly ample fiscal headroom in mundane drone-like work that was being outsourced (still cheaply, I might add) that AI can replace and even do a marginally better job of. I suspect AI prices can even increase and it will still be profitable for enterprises.

Companies like this will certainly keep expanding their AI use, and once they commit to that, there's little stopping them from moving to open models or local inference if need be.

My concern is less over the bubble and more over the social cost of AI. Call centres and the like provide a tremendous number of jobs. As AI moves into enterprises more and more, where are all these people supposed to work? Become baristas? They certainly won't be "learning to code"... What sorts of social and other unrests will this cause?

These forces I think will muddy the waters and make predictions difficult. Say what you want about the AI bubble, but if it pops, it will be different than previous ones. The bubble doesn't even need to burst because of the insane economic model, if enough people are economically devastated by it, it will cause ripple effects of its own.

The thing is that companies that are investing in AI crap to replace mundane drone-like workers is that the AI will get better but more importantly we will get more used to the way behaves towards us. Right now we some janky-as-fuck implementations but we're currently being trained on how to react to AI just as much it is trained on us responding to it.
> If you look at most big tech earnings this quarter, with the exception of Amazon, almost all others lost significant value after reporting. Microsoft, Alphabet, Meta, and Apple did.

Microsoft spiked on earnings, and is now actually about ~24% above it's pre-earning level.

Alphabet did lose about 7% the day of the earnings, but it recovered and now, after MSFT earnings, is actually 9% above the pre-earnings level.

Meta dropped 10% on earnings but is now back to its pre-earnings level.

Apple dropped ~10% and has not recovered (yet) but it's also famously "sitting out the AI bubble", so not sure why it's included here other than a "tech stock that went down."

Oracle has been dropping forever but after Microsoft's earnings it's climbing again.

If you zoom out, the stories change, and as you keep zooming out, they keep changing all over again.

My point is, 1) reading stocks in isolation is like reading tea leaves, and 2) if you want to point to any stocks, you should make sure they support your narrative.

The title's clickbait (from The Register, of all people - colour me shocked!).

There's nothing really groundbreaking at all in there, just "chips are expensive, and open weights models hosted locally in enterprise could displace Claude/GPT"

> I caution anyone looking at API prices: they dropped the price, but is it actually less expensive?

> Yeah, they dropped the price, but count the number of tokens you're tossing into it and see if it's actually cheaper. It's good for marketing, but the rub is how much you're actually using.

The level of discourse is so horrible now, I don't have words. Are these the ones making predictions on AI bubble?

This is poorly stated in the podcast, but the underlying point is correct: while cost-per-token is going down, overall token use is way up. This is causing the cost of using LLMs in corporations to skyrocket.
Of course, the price of tokens going down makes it cheaper. You can now afford to throw millions of tokens at solving a problem that would have been impossible for AI to solve at all a year ago for any price.

People completely lack imagination about this stuff. The main problem right now with AI isn't even AI successfully producing code at a reasonable cost, it's human coordination and review that is the bottleneck.

The bubble in the US pops maybe. China is only limited by chips, not costs.
well if bubble pops, everyone will die EXCEPT google

UNLESS openAI/etc actually succeeds in making AGI that never hallucinates and goes over the current LLM limitations

as for google... well they own the web

(+google has plenty of other revenue sources, so it can just pay out its AI survival)

if you're a website owner, would you welcome chatgpt/etc's data-collection bots?

but... as for google's bots... you need your website to be on the Google search results...

Note that the “AI Bubble” term used here is only defined in the context of market speculation. If you are not an investor of AI companies then there is nothing to worry about for you. If you are an investor, then you should know that people can’t really predict when bubbles burst. Every prediction in the markets is a speculation and some investors can simply bets against the popular expectations to make good money.

Movements of AI stocks shouldn’t be confused with “AI as a technology” and “AI as a business”. Market valuation is a different game.