Apollo calls AI a 'bubble' worse than even the dotcom era
fortune.com
fortune.com
However, ML/AI has always been a cornerstone of most big tech companies, but it was usually in understated products or used in ways that might not be directly b2c or b2b like ChatGPT. For example, meta/google/netflix ad/video recommendation systems are likely mostly ML based, and have been for a few years. Most of the photography on your phone is using some form of CV/ML, Amazon probably uses ML for logistics planning, AV companies have been using ML models in perception and behvaior for years, generative models have existed for a while, etc.
ChatGPT just brought the idea of AI and ML to the forefront of investors and the general public as a "tangible" and fairly practical use of "AI." So I don't think AI is necessarily a "bubble" since theres potential for pretty high utility products, but I do think the market might have an overly optimistic view on the success of AI (specifically generative AI) monetization, which only a few companies have arguably successfuly done (mostly the AI "suppliers" like nvidia, OpenAI, etc). Its yet to be seen how Chatbots/generative models will actually impact the bottom line or streamline processes at businesses without necessarily "solving" AGI.
ML/AI existed before this "bubble" and will continue to exist after. In my opinion it was only a matter of time before we got here, though I'm not sure if this time if its over or under priced. I disagree with the claim that AI itself is a bubble, since it has actual utility.
Maybe everyone doesn't need an LLM assistant, but every business would love it if AI can optimize their sales pipeline 5%, reduce their manufacturing defects 3%, increase their utilization 10%, etc.
There are not enough customers with cash to milk. So to keep up appearance of growth they cultivate a subprime market propped up on credit thats never going to be paid back. So the illusions and delusions stay alive.
Efficiency is always a double edged sword. The more efficient the predator or the parasite gets, the more unstable the whole system gets.
https://longportapp.com/en/news/100635792
Let’s let Google’s new AI results tell us the signs of a bubble and I see all of these with the current AI grift.
Rapid price increases
High valuations
Speculative behavior
Increased media attention
Lack of fundamentals
Prices rise regardless of news
Other asset prices are soaring, too
New traders say that old investors "don't get it"
Stock valuations in the top percentiles
The stock market was on fumes before the current AI hype hit and we all crowded into 7 stocks. Wall Street needs this to keep the cycle going, but I don’t think it’s enough.
NVIDIA have high revenue right now because everyone is suddenly chasing the AI bubble. There hasn’t really been a product shift on their end to cause this, rather the bubble formed coincidentally around their compute libs prevalence that’s been laying the groundwork for almost two decades .
their high profit is well deserved but the sudden ramp is the bubble.
When will it pop is the question? I think that’s when a lot of these smaller companies fail to produce a viable product and the bigger ones get fed up of the high margins.
But I think NV will come out on top regardless, because a pop will still place them pretty high.
A savvy shovel seller will be aware of that, but it sounds a lot like a bubble to me.
That they both eventually exhaust doesn't make them the same; one is fumes, the other involves a real expansion in the stock of usable resources.
People's cynicism here has them ignoring real enduring positive shocks in the economy – even though that's been the nonstop history of our tech industry. A few suppliers transient stock valuations may be, at times, a 'bubble'. But the fundamental (& irreversible advances) in AI technology aren't the bubble.
And certainly you have parallel shovel-sellers in the dot-com bubble (notably Sun Microsystems).
ChatGPT and generative models being a panacea for every company, sure I can concede that its a bubble.
Like I said, it has yet to be seen how AI "consumer" companies will make more money and/or increase margins with generative AI products.
The big issue is: what is the product at the end of the tunnel? The really cool stuff is still too unreliable for day to day use because it hallucinates things. The cost to scale is also really high.
The bubble just refers to the speculative growth. The pop is often just the number of companies failing to capitalize on the hype, and the following pullback on investor funding.
I think AI is a bubble , but I think some companies will come out the other side with huge winning products. Many will burn their runways out.
The biggest parallel it has is that the world of AI is like the web was in the 90s. Merely a glimpse of what is to come. VCs and angel investors are likely to get killed, not so much public markets.
.com craze had a ton of public companies being valued at outrageous numbers. Not so today, even nVidia's P/E ratio is not so crazy. During the .com craze a ton of fiber was put down which at the time could not get used. But it was used eventually. Something like that is going on with GPUs.
This effectively makes NVidia a leveraged investment on the AI space, however it's unclear whether all investors recognize the leveraged nature of the investment - or the portion which is funded directly out of firms like Meta ramping up their GPU buys.
All the more reason to assume the new run up is indeed a bubble.
> actual utility
What utility?
Also denying it’s real world use is very ignorant as well. I have used generative ai very little yet I’ve gotten very real world use out of it. Image generation for a lecture to conceptualize something as pictures. Help in programming something fast in something unfamiliar. (Needed to use an azure sdk in java which I hadnt used before - instead of reading documentation chatgpt gave me a very usable example and I did the integration part of the work in minutes. I did check some things from docs afterwards, but the example sped the progress by a lot.)
Gen ai saves too much time / money / effort on several tasks not to keep on existing. I don’t deny a possible bubble, but a bubble and real world use are not mutually exclusive in the least.
Chatbots are not what's going to make AI great. Am sure there's some commercial utility out there for AI to improve efficiency in certain specific industries, that's what am more interested in hearing about instead.
We shouldn't use the same word to describe the tulip bubble and the dot com bubble.
The tulip bubble to me is what we really think of as a bubble. A collective mania causing price to go up in a completely irrational way.
If we can agree the valuation of a company should be all future discounted cash flows of a company then it doesn't strike me as irrational for the price of stocks during dot com or today to go through the roof. Valuation is really hard and really uncertain and then all the harder for something new and highly disruptive. The problem with using the word bubble in the tulip bubble sense with AI is there is surely no guarantee the price of NVDA is going to fall sharply from here as if it was a tulip bulb selling for the price of a house.
Do you use gmail, youtube, or google search? All 3 products use AI for several features, big and small. It might not register immediately since its not as overt as ChatGPT, but recommending you a new video to watch, a new restaurant to go to, or some product to buy is likely driven by some form of AI. Do you drive a car with lane keep assist? Same deal.
Where do you think the line should be drawn?
What reason is self awareness needed for an agent to be intelligent? What do you mean by 'self-determination'. Figuring out the rest of it's life?, or just being given a goal and choosing it's own path to get there.
Because some of you daydreamers are convinced AI will replace human intelligence one day. It is important to define and contextualize intelligence. I gave some of the qualities of what AI is up against (human intelligence).
That's why it is important for AI to be capable of achieving at least some of these qualities before claiming the world as we know it is coming to an end.
Do you mean AI as in (proto) AGI? I think we might have slightly different notions about the definition.
A market being a bubble does not imply that there is no utility for the products in that market.
> but I do think the market might have an overly optimistic view on the success of AI
This is exactly the definition of a "bubble".
But it's hard to argue with the results of the progress in AI over the last few years. I have no idea how that will translate for investors, but I wouldn't be surprised if AI fundamentally changes the stock market into something fully automatic within the next decade.
If it's reliably better at trading than people, then whoever has the most hardware is going to be able to make the best predictions, allowing them to buy more hardware and make better predictions. In 10 years, what's the purpose of a human trader?
Automated trading already exists, but automated trading firms value explainability and reproducibility. Most common models in the quant trading space are linear regressions - they are simple and explainable. If you made an AI trade, it better be explainable, but explainable AI seems pretty far away at this point and it doesn't seem like something many companies care about right now.
And despite all the automation in the trading space, there are still traders. Humans are decent at analyzing idiocyncratic events (0-shot learning), but machines are not (yet).
Again, I'm no expert in trading, and I'm sure you understand the processes better than I do. I know that automation has been a big part of trading for a long time. But is fully automated and fully explainable AI really that far off? As in unattainable within the next decade? We're all guessing here, but judging on recent progress, I would bet that AI will be able to do everything a human trader could do, including providing the rationalization in a comprehensible and convincing way to managers, within 10 years at the very least, probably sooner. I would bet that AI can even do a better job of hitting the specific points that will be convincing to target the decision makers before that.
The focus on AI in the stock market is already evidence of that. Right now, it's the people boosting AI that are convincing investors to put their money into it. And once they're invested, they're going to be more likely to be convinced by confirmation bias, therefore more likely to listen to the advice generated by the product that they already invested in. That's a cycle that will be hard to break.
There is no such thing as bubble definition. Everyone has one depending on their point of view. You have one.
> the market might have an overly optimistic view on the success of AI
What is overly optimistic? How much overly is overly? Nuance matters, as with everything. Also timing. Most if not all of the outrageous ideas during the dotcom era have been realized in the 20 years since.
My litmus test has been companies that call it machine learning are less hype'y and likely more real. Companies slapping AI onto anything and everything are more likely to be.
https://www.nvidia.com/en-gb/geforce/news/nvidia-ace-for-gam...
https://www.nvidia.com/en-gb/research/ai-playground/
And features for games like AI upscaling.
It's all to sell Nvidia hardware in the end of course.
I think your last sentence says it all.
Elon Musk isn't a flamethrower salesman. These are flamethrowers.
Dotcom s/stock/GPU/g. Although I do think NVIDIA will come out of this one unscathed.
And just to ward off any naysayers, I'm a GPT subscriber who finds it very useful in a variety of ways and uses it almost daily. But also anyone who's used this tech for a minute and approaches it from a viewpoint of cautious optimism rather than rose-tinted glasses knows that its limitations are quickly understood.
Someone who invested their money in Amazon on top of the "dotcom bubble" is today 50x times better off than someone who held USD.
If the company achieves a 4-5X growth (15% per year) and 30% margins (instead of 55%) that would place the share price at $176 - 1/4th of what it is today.
Let's look at the addressable market:
The sum of wages in the USA alone is about $100T per year. How much is it worldwide? Let's say $400T.
That is the current TAM when we expect computers to replace humans like cars replaced horses.
Capturing only 1% of that TAM would mean $4T in yearly revenues.
Nvidia's $2T market cap seems tame in face of the size of the opportunity.
But there is another point: A new technology not just replaces an old technology. It 10xes the usage. There are way more cars on the road now than there were horses back then.
The size of the market will grow beyond those $400T. We will build way more stuff when robots carry out all the work. Most of planet earth is unused at the moment. That will change.
But at the end of the day values of money are made up bullshit that attempt to assign a number to scarcity. If we make a world where computers can cheaply (energy/material wise) make all the items humanity needs, that future world is one that is a lot better off then the one we currently exist in. At least if we consider human needs being met valuable.
If science fiction tropes have any connection to reality, us humans may sit around all day in hedonistic gay space communism playing games while robotic/AI systems produce and trade needed resources and energy between each other producing 'value' while humanity itself is just a value sink.
Nvidia is a key player in this. So I would be surprised if they capture less than 1% of that value.
For example, in 1999 I bought food & other goods by delivery to my office and worked with companies selling online, using my handheld computer with a fast wireless connection. Clearly those were ultimately successful ideas, but that didn’t mean that investors in Kozmo, Pets.com, Ricochet, or Handspring got rich.
correct, it needs used for what its true purpose is: paper clip feedstock.
But seriously, what does that comment event mean. The earth isn’t being used? More use would invariably be mean more dystopian, despoiled and privatized, why is that good?
Mind you my personal perspective orients towards the opinion that tech companies do lead the markets more than others through metrics I can't personally rationalise but it's something we all seem to agree with (and I'm ok with if my superannuation/401k is always growing)
Lots of cash doing nothing on the sidelines is risky in the long term because inflation. A "risky asset" is risky in the short term because of volatility.
Not everyone can wait X years to get it back
If you imagine a person, say 30 years, who thought in 1980 "Hey, I'm going to hold my savings in tech companies because tech is where the music is" and started accumulating throughout the 80s, 90s, 00s, 10s ... They would have a great time all along. And would be a happy 70 year old now.
Every tech boom in history was kicked off by wild speculation and ridiculous prices. Some of them turn out to be worth it, but practicing hindsight like you're doing is less than worthless. The dot-com boom was followed by the dot-com crash, and this will be no different.
“lol what bubble? Stocks are much higher than they were in 2000.”
“But they still took 15 years to recover”.
Whether it was smart or not to go all in on NASDAQ is not really material.
To address your point anyway, plenty of investment and pension funds were badly hit. Because even if you diversify, when a whole sector of the industry loses 20% in a couple of months, it hurts even people who did not invest themselves. I lived through these years and the ripple effects went much, much further than just some reckless venture capitalists. And broadening the outlook a bit, not everyone can afford to stay illiquid for 20 years, particularly in the US where the social safety net is not very effective.
Also, what about the not-boomers who were not in a state to invest in the 1980s? Plenty of 20- to 30-somethings were wiped out as well. Knowing that the market will be up 20 years from now (and not the stocks you might hold, mind) is cold comfort.
If you don’t remember it or did not live through it, I can tell you that it hurt quite a lot and no amount of hindsight now about what people should have done is going to change that.
Just because something may eventually be a profitable investment that doesn't mean that it is the best place for you to currently invest your money.
They didn't just sell books and CDs either. They sold tax free books and CDs. It is easy to forget now that no one paid tax for anything sold online in 1999.
Of course, no one brags about how they bought Borders stock in 1995 today. I am sure if you did you were up huge in 1999 but not so great if you held.
yes, there were lots and lots of companies started with bad ideas, ideas way ahead of their time, and even good ideas that just didn’t get traction for whatever reason. but the thing underlying all of it was basically just the availability of this amorphous thing called the internet that had been around for a while, but not generally accessible to the average person. Now all of a sudden anybody and everybody had an email address and could get on the web and order pet food from the comfort of their own home. There was a lot of time and money spent trying to figure out what to do with all of those new possibilities. but I don’t think anyone argues that the Internet in general or the web specifically were revolutionary innovations.
AI is much the same. Everyone still figuring out what to do with it how it’s going to affect the economy, jobs, the tools we use, even the kinds of entertainment we seek. Is there a bubble around it? sure, almost certainly. does that make it itself worthless or transient? Absolutely not. and just as we had some companies come out of the Internet bubble like Amazon, they are going to be some long lived companies that come out of the AI bubble. I think trying to pick exactly which ones are going to be winners and losers is a very tricky game though.
Crediting the internet for the dot com bubble is like crediting the city streets for the success of the taxi industry. It's a necessary condition but not the innovation that made the money fly.
For example, Nvidia’s PE ratio has actually improved. The stock is technically cheaper now than it was before. Assuming of course that it is able to keep performing at the same level. Something I am hugely skeptical about that.
I have a working theory (unfinished and likely has faults): The automation of decision making in trading is the reason behind this. Which is also why I feel a repeat of 2000 will be unlikely as fewer humans are involved directly in the loop. Things are going to seem more plausible now, as the algorithms ensure that by looking at hundreds of signals before making buy/sell decisions. Our guts may be picking up on something uncanny that has not been accounted for in the algorithms yet. Only time will tell what that is.
Thats usually flows from hedge funds, family funds, pensions and large banks. And more retail presence in trading is playing a part in that too. Typically most of the things I described are not automated, though there might be more data science involved nowadays than say 30 years ago. You can probably confirm this yourself by looking at nvidias largest shareholders. I don't think Citadel or 2sigma is just sitting on tons of nvidia.
No shits are given for actual usefulness or if it even makes any sense for the product at hand to have some sort of LLM functionality built into it. Why would they care? Their investors want AI, so AI we build.
An absolute circus
Looking backwards saying "but this guy I know did 100x" is not informing of how people can know now, that Amazon and not Alexa by Digital is the one to pick. (note: they are just labels, alexa was not an amazon competitor. the point is digital was acquired, and died. Amazon didn't die. At the time digital was big, you would think it was a sure fire bet. It wasn't)
The narrower view in the context of artificial neural networks ("AI" is more of a bubble term. Even artificial neural networks might be a bit of a bubble term, as if we have completely recreated the human nervous system in the computer). Have they made advances in recent years, will they have profitable products with some development? Probably. They're a new technology which will almost certainly be profitable, just like the technologies in 1994 that made the current Internet profitable. It is of a certain size though.
Then there's the macroeconomic view. There is a lot of capital sloshing around looking for a decent return. The common sense thing in a sense would be to slowly finance "AI" and as it develops, put more and more money in. But there's few decent things to get a good return on, "AI" is obviously one of them, so a ton of money rushes in. More money than is needed at the moment, and ahead of schedule as more things are needed in the development of all of this (including more powerful Nvidia cards and Nvidia networks, more robust frameworks for Pytorch and such, more Phd's who know what neural network is needed for a problem and data engineers to move data around).
Fed Chairman Greenspan talked about "irrational exuberance" in 1996, and yet the dot-com bull run (if Netscape's IPO was the sign of it really starting) had barely even started yet, and ran on until early 2000.
It's true if people are handing money out left and right, carnies will show up to pull in that money. This really doesn't have much to do with the people who were developing the technology of the Internet or "AI" now. There's a ton of capital sloshing around looking for good returns, it's obvious "AI" will be a future source of profitable returns, so money comes pouring in, even if it is disruptive to a point. It's not just the carnies pulling in money, it's the situation of a massive amount of capital desperately looking for better returns then it has been getting.
Paying with coins for EU hosting and US goods while in Russia is quite convenient these days.
It's a good thing that crypto exists to serve such a need for ordinary people who are doing something that's illegal only because some blanket political game classified it as such. Mechanisms for escape from certain laws and rules are in general good because not all laws and rules are automatically right simply because they exist.
Except, AI is much easier for new players to enter and manufacture than the Model T ever was. We've already seen this with the fearsome market pressure people are putting on the current market leader, OpenAI, despite their sizeable first mover advantage.
In that sense, and only in that sense, do I think AI is a "bubble". Prices are going to fall very quickly as this technology weaves its way into every firm imaginable. Everyone else who finds a way to put AI to use productively moving atoms around, meanwhile, really will see their stock valuations explode.
What are you talking about?
There's exactly one company that makes the machines that makes the AI chips.
There's exactly one company that operates the machines that makes the AI chips.
There's exactly one company (or two?) that makes the AI chips (which are feasible for training).
And the most powerful governments in the world are worried about these monopolies causing geopolitical/national security problems, and they pretty much can't do anything about the situation in the short term.
And to be eligible as a player you'll need hundreds of millions of dollars of funding to buy the hardware.
I don't have the numbers, but I'm pretty sure opening a car manufacturing factory 100 years ago didn't involve that much upfront capital, even accounting for inflation.
Surf the wave, don’t drown.
The article of course is focused on nVidia which is making money (unlike the dot com businesses) but the 'bubbleness' will happen if AI startups that aren't making money go public anyway. If there isn't a way to get retail investors involved then just the rich people will get fleeced.
That said, it was interesting to watch the nVidia recovery from the bitcoin miner demand dropoff. One hopes they would react quickly and effectively when the bubble bursts and demand for tensor chips drops by 16 to 20dB.
Many people don't understand AI, but it is much more clear that it is a useful tool. Companies can make money with it by replacing workers and increasing the productivity of workers who know how to use it. The future of AI is much clearer as well. We know it's going to keep getting better. There is no limit that we know of on how good it can get. Feed it a script and make full length movies? That might happen. AI companions and care-givers for the elderly? Definitely. An AGI companion that serves as your assistant? That's not far off. AI psychotherapists? We're almost there. The possibilities are endless.
This is the difference as I see it.
Meanwhile most people are sitting on the side line paying rent and hoping to be able to pay it next month…
But that's only matters to investors.
Something could break one or two major assumptions undergirding the current form of capitalism, but if it caused the returns to evaporate, that would still look like a bubble to an investor. But capitalism would have changed, and it would still have been hugely impactful for regular people
And honestly, the 90s revolution of internet/webpages felt bigger to me than AI. A lot of the 90s hype was true and relevant, even if it was ahead of its time and overexaggerated.
These are part of the same revolution. At least I believe when historians look at this from a distant future they will not see these as two profoundly different things, but a continuation of the same effect.
I also feel the same way about Apple Vision Pro.
Likewise, I don’t think this report refers to “AI” being a fad as much as just AI companies being overvalued.
It seems possible to believe that AI will swallow the world and also many AI companies being overvalued.
Instacart was founded about a decade later, and its margins are pretty slim by tech company standards (net income ~$100m on $2.5bn revenue) so I think it's pretty clear that in the year 2000, there's no possible way that Webvan could have succeeded even if it had done everything right.
Extrapolate this logic to AI, and the parallel is clear: The current generation of AI over-promises and under-delivers at enormous cost, and yet billions are invested under the premise that quality will go up and costs will go down enough in the next few years to build viable businesses.
If those investors are correct and we're just a few years away from useful, cost-effective AI, they'll make their money back. If they're wrong and we're actually a decade away, not a single AI product that hits the market within the next few years will survive.
Webvan was a good idea with bad VCs. They were forced by their funders to expand to too many locations, for the "first mover advantage". So they had 3% market share in 30 cities, instead of 30% market share in 3 cities. Which means a lot of trucks. Which Webvan bought and owned. Webvan drivers were direct employees. The sub-minimum wage gig economy hadn't been worked out yet.
The people behind Webvan later did Kiva Systems and helped Amazon get warehouse automation going.
Look at NASDAQ between 1995 and 2005. I’ve never seen anyone claiming that it was not a bubble. Sure, some companies survived (as some always do, a bubble is not the end of the world), but a lot of people took a bath.
> 3D TV seems to be a true bubble so far, I don't know if AI is
That is not what a bubble is. A bubble is over-valuation of a bunch of stocks, when all investors want to hop on the same train. The fad is the companies’ stocks, not (necessarily) their products.
Nobody was claiming that the Internet was a fad in ~2000, just that investors were so eager that ridiculous startups with dubious business models got tons of money dumped on them for no reason, and that the expectations were out of line with reality, even for successful big companies.