Everybody knew it.
But some investors don't have the patience for these "sure things" to mature into their commercial destiny and others know they can siphon off a killing by taking advantage of them as they speculate.
These bubbles aren't about potential, they're about timing.
And today's AI, for as revolutionary as it may be, is taking just as long to mature as anything in the past.
Most of the top players involved in AI right now are going to be rich. And in 20 years, the world is going to involve a lot of AI in soooo many things. But a lot of shirts are going to be lost below those top players and a lot of companies are going to fold because infrastructure, refinement, and integration take time while the FOMO speculation starts as soon as the big discovery is recognized.
Canals... where a repeating pattern was the company underestimating difficulty/costs, and going unrecoverably into debt. Another company, which then bought the largely finished canal, cheap, they had potential profit... until tech change (railroads).
Absolutely right. Somebody may buy the DCs. The tech changes very very fast, the DCs may be entirely worthless. US has high energy costs, it can never be cost competitive.
Why Data Centres Will Change the World and Bankrupt Their Builders — John Hempton https://www.youtube.com/watch?v=GY5YyPLfqp0
Useful (like solar, airlines, fiber), but it won't be profitable. High capital cost, low marginal cost, subject to competition and with technical improvement --> these four factors guarantee businesses will never be profitable, and they go bust.
There a lot of parallels with AI…it doesn’t take a genius to see the long-term potential, but the risk is in implementation, so it’s a mad rush to the finish line by a wide variety of less scrupulous early players. Rockefeller got his big start by sweeping up failing producers who were all capital weak, and I’m sure something similar will happen here during the inevitable bust.
I can do the hard, risky work of drilling wells / deploying AI, or I can sell you a vision of that future, today. The most dangerous technologies in terms of grifting are the ones that everyone can understand.
Almost all the "top players" in the dot-com crash died or lived on in relative poverty; only Google and Amazon turned out "very rich".
Another reason is that the way we find out what works and what not is through experimentation. Especially in a transformation like this it’s impossible to predict which configurations will work and which won’t.
Yes. And valuation. I think the leading frontier labs will be fine in the long-run and NVidia will be fine. Some of their investors who bought at too high of a valuation, expecting to flip it in too short a time frame will lose a chunk of their money when they have to sell for less.
The hyper-scalers in the middle are the most vulnerable. They borrowed billions at increasingly high rates to lock up GPU and power capacity. They are 'ground zero' of a lot of the 'circular money' deals. Big banks who financed hyper-scalers initially, expecting they could package that debt up and sell it on to diversify, are already struggling to find buyers for the debt, causing its value to drop, which then forces the banks to cover with more collateral, and that's how the snowball starts down the mountain.
And do these not all have established business models that are ridiculously profitable, and huge cash reserves?
How the heck are THEY vulnerable, and the "bare" frontier labs like Anthropic and OpenAI with no preexisting business models or profits, but equally ridiculous borrowing are supposedly not?
Google, Meta, Amazon, MS have huge cash-producing businesses they can just fall back on. Of course the stocks will get hammered in the short term, but they are still sitting on cash volcanos.
Pure AI plays are what are most at risk. In the bad case, all the CoreWeave type companies have their assets acquired on the cheap by the existing big players. In the armageddon case, companies like Anthropic and OpenAI also end up in the discount bin.
I don't think so.
The reality is that the adoption is incredibly fast and that revenues of the AI labs are like doubling by the quarter.
Further, there is the expectation that breakthroughs in applying AI to science, robotics, or medicine may unlock trillions in value.
That this will take 20 years, rather than happen before 2030, is just speculation. None of us can see the future.
I believe AI is here to stay as general concept turned into practically working implementation.
What that practicality means is we all who consume AI are learning but we're still in the early adopters phase of all that.
Of course AI companies invest a lot and try to stay on top of the traction but I also believe not all of them will survive test of time.
I personally believe that a lot of over investment in data centers and supply of energy will be burnt and will have to be reinvested somewhere else when technology advances enough to "miniaturize" AI enough so it won't require couple of football fields to provide what it does today.
I do believe cloud based AI will be always attractive to small business if price is right and to big enterprises so they don't need to invest inhouse for the scale they need, but that will be sort of alliance based enterprise B2B market.
anything between (including me for the scale I represent business wise), I can't wait until I will be able to buy appliance from 12 to 48U maximum that can do all what me and my company needs AI wise as local AI.
moment subscriptions become equally expensive as pay as you go models, I switch to mac studios with 96GB and reasonable weighted models that will be specialized in something instead of one LLM/AI to prompt everything to it and hoping it can address all sort of problems as one solution.
AI one day must fly to space, be called Mother and discover aliens. it can't be 6 football fields as physics won't let it take off. It must be one day small enough so we all can have ChatGPT 6.1 medium model at home without subscription, just as fixed price with TCO over lifecycle of the whole appliance (software + hardware together)
that may become a drama to many companies that apparently may not be worth trillions unless they diversify enough its AI offering.
I don't see all these things as being comparable to AI... At the bottom, they were "100%" technologies. Trains were just able to transport commodities, and the internet was able to reliably connect computers. So their path to success was about gaining market.
Since they were 100% technologies, they disappeared. One could build solutions using them and forget they are building blocks, because they just worked.
LLM AI is different at the fundamental level, because it's non-deterministic. It needs human-in-the-loop. It won't ever disappear. Gaining market share doesn't fix anything.
> And today's AI, for as revolutionary as it may be, is taking just as long to mature as anything in the past.
They may be as expensive to mature (in terms of equivalent duration of total economic output), but by the wall-clock we've gone from "invent Transformers to make translation less bad" to "solves Millennium maths problems" in 9 years.
After 9 years of railroads… well, those were originally an Ancient Greek way to drag ships over an otherwise annoying bit of land, and the more modern attempt had wooden rails and a rope for humans and animals to pull for around 250 years so let's say after 9 years of the steam locomotive: 1784 for the prototype, 1802 for the "first full-scale working railway steam locomotive" (according to Wikipedia), so 9 years past the invention and it was still a prototype.
Semiconductors (well, transistors, i.e. us understanding the process and doing it deliberately, rather than getting lucky with cat's whiskers before we had any clue about quantum mechanics) did at least get commercialised in less than 9 years, but AFAICT in the form of "now our radios are better, sturdier, and smaller" which is a rather smaller change than what's already happened since Transformers.
(Of course, why Transformers, rather than the other attempts at AI? Transformers are when people cared this time around, but there have been previous AI bubbles. I'd argue those are like pointing at 1515 (or the Ancient Greeks) for the railway, or cat's whiskers for semiconductors).
Nor would it have been thrust blindly into the hands of your manager, the marketing department, the CEO, and countless other folks who have no understanding of how it works, its limitations, or its risks.
Speculative bubbles, I.E. risk is where much of the capital comes from to develop these ideas to start. I’m not sure you can have one without the other.
Look to Europe to see how that has worked out.
If that was true it is strange that you would find yourself saying as much for a while now, given it sounds most appropriate in response to someone denying the utility of slop generators.
and, apparently, our brains https://www.media.mit.edu/publications/your-brain-on-chatgpt...
Is it?
Every single vibe-coded software I've ever used looks pretty on the outside but has massive, glaring holes in functionality that make it unusable.
It's like having food faster but then you find out it's actually full of silicone and microplastics.
That's a bubble. It happens every time because it has to happen every time. If you stay out of the game, the people who stay in the game win and your holdings in redundant technology are worthless.
AI models can be deployed on the same datacenter as the old one and be either orders of magnitude cheaper to run, or orders of magnitude more capable.
The only thing that can pop this bubble is if we're not smart enough.
This hasn't historically been true. The clusters have been getting bigger along with the models over time. And now we may even want to have these computers have rapid access to extra compute or caches of the internet or other data stores.
Not if the datacenters are in space, you can upgrade the models for a little while but once you need to upgrade the silicon you need to launch new hardware.
It might be dated but not too long ago a report showed the effect AI had on the SAP 500 and 490 companies couldn't even find any substantial improvement. Mind you, this is during the craze, this is happening while every manager has all the incentives to use as much of the technology as possible, lest they be the one that missed out on the new steam engine. All of these companies invested heavily in cars because none of them wanted to be stuck using horses while everyone else drives off into the sunset... yet most of them didn't manage to measurably outrun the people on horses. That's not the prediction I would have made if I were switching from horses to cars.
> And today's AI, for as revolutionary as it may be, is taking just as long to mature as anything in the past.
I suspect this opinion may come across as naive for the myriad software engineers whose job has become pressing the enter key. Your point about timing is based on solid principles, but I don't think your comment contains a satisfactory argument that the timing is incorrect.
If no one is willing to pay increasingly high costs of training frontier models, we'll still have the models that we have now that can be fine tuned for incremental improvements. Still tons of utility in that, but little investor upside.
FYI Claude Code’s default is Auto Mode these days
In every bubble some say this. And it could be the exact same herd behavior (mentality) that creates a bubble.
Oh, I thought they were about unchecked greed and insufficient regulation, but I suppose that works, too
As we all know, GE then went bankrupt and we went back to whale oil.
Expect more differentiation in the future and not less.
This kind of thing amounts to a LoRa applied to your model. There's no reason (apart from potential IP infringement) that someone else couldn't produce a "Claude LoRa" for their model (or, more likely, given the dislike for Claude's writing style, Anthropic could offer a "non-Claude LoRa" for their own models)
Assuming you meant this for personality, I’d still disagree but not that extensively. I think personality is nuanced and complex - it can’t be simply lifted and shifted. It’s grown using a meticulous process.
Fundamentally, people underestimate the marginal increase in utility for marginal increase in “intelligence” as measured in benchmarks. Like I’d pay 5x for a model that has 2 units higher in AA intelligence.
GPT has the same performance. And performance is definitely a commodity. Why wouldn't it be?
I wonder why so few people make the comparison to the recent VR hype. We heard the same things there about its unlimited potential, how it's inevitable that we will soon resolve all its shortcomings and that it will become ubiquitous.
I still believe VR will eventually live up to its potential, but no one can deny that bubble has popped after Meta took a huge loss on it. It took too long to resolve the hard problems so the timing didn't work for them.
AI is already a gigantically transformative technology, with a staggering amount of paid consumption, and virtually everyone I know uses it every day, where I don't know anybody who regularly uses VR in any form.
The ship has absolutely sailed on "is AI a world-changing technology", where "are the leading AI companies worth what's been paid for them" is a completely open question.
VR is a toy with a few very narrow areas of application in industry. That does not describe the current status of AI.
That's a remnant of the past, where only a nation state could afford massive projects such as creating a railway line crisscrossing throughout a massive country.
Since the last couple of decades we see private companies investing in railway lines, including high-speed railway corridors. We also see a major liberalization push in Europe to decouple railway operators from infrastructure maintainers, and now there are plenty of private companies operating railway services across the world. In Spain alone there are 2 or 4 private companies providing high-speed railway connections between major urban centers.
The reasons for this are complicated and nuanced though.
Governments think they push the financial burden of transport maintenance on the people (and its a good tool to register and control people too, but thats not a strong consideration I wouldn’t wager), so they fervently cut spending on public transport and curtail cycling infrastructure in favour of cars.
But roads themselves operate at a crazy financial loss and are one of the reasons governments are in crazy debt (which nobody wants to acknowledge).
This is because road maintenance is significantly more expensive than building new roads. Which means that road use becomes heavily subsidised (much more than trains cost per user) but the cost of vehicle ownership, despite that heavy subsidy, is still enormous and one of the largest expenses an average person can have.
Is it? People love to have personal sovereignty over their transportation, given the choice.
"The fourth power law (also known as the fourth power rule) is a rule of thumb which states that the damage to the road caused by a motor vehicle increases in proportion to the fourth power of its axle load. This law was discovered in the course of a series of scientific experiments in the United States in the late 1950s and was decisive for the development of standard construction methods in road construction."
"Example: Car (total weight 2 tonnes, 2 axles): load per axle: 1 tonnes vs Truck (total weight 30 tonnes, 3 axles): load per axle: 10 tonnes. [...] The resulting damage differential between truck and car is 15,000 to 1"
PS: this is something electric car proponents don't talk much about
- Train capacity is much, much higher. You couldn't have the density of France, the Netherlands or even NYC without most people travelling by train. LA struggles with 1/10th of the density.
Corporations demanding payment for everything at whatever margin they want is little different than taxation without representation and local cops shaking people down for money at random; just different legal semantics
Specifically because of massive government investment in road infrastructure in preference of railways. This would be about like the government deciding to funds DCs directly...
AI is bigger than the Industrial Revolution, the Internet, than computers, than literally any innovation or invention that has ever come before, combined.
Today you see AI slop and vibe coding.
Tomorrow, all of humanity will be affected, forever.
Everything about the way that we learn, that we think, and that we invent will be changed forever.
I think civilization will have time to react and adapt, panic leads to haphazard action, we need deliberate consistent action.
How do you think we need honest deliberate action if 95% of people believe AI is a scam and a bubble and mostly just hype marketing? Just take HN's comments for example.
Their monopolies got broken and the profits disappear in a commodity market.
If nvidia or tsmc got split up i doubt its stock is going to stay at as a high a p/e.
Imagine a world without those. Oh, wait: tomorrow people won’t know how to use language, will they? And they won’t have water to farm with, so there you go.
Humans living forever is inevitable. Galactic colonization, inevitable. Over a long-enough time scale.
AI is inevitable on the same time scale, sure. But can it actually survive as a sustainable business given what is needed to keep it relevant—will people pay what it costs to support said business—not clear. Or, at least in the next five years.
The second generation AI companies will be built on different assumptions and will likely be slower-growing, and that’s when you’ll see it truly become ubiquitous. That could be a decade though.
is anyone else weirded out by folks chanting this like a mantra? imagine saying this for any other emergent tech. "NFTs are inevitable", "crypto is inevitable", "excel is inevitable", etc.
inevitability != usefulness/profitability