Episode 125

The 'AI Bubble'

Watch on YouTube

About this episode

In this week’s episode of the Untangling Web3 podcast, we tackle one of the biggest narratives dominating tech and global markets right now: the so-called AI bubble.

With headlines predicting an imminent crash in artificial intelligence stocks and skepticism mounting around LLMs, the hosts unpack whether we’re witnessing genuine over-exuberance, or simply the growing pains of a transformative technology.

Drawing on real market signals, user adoption, and infrastructure realities, this episode cuts through the hype to examine what’s really going on beneath the surface

Key Points Discussed

  • Why People Are Calling It an AI Bubble. The episode explores the core arguments behind the bubble narrative, including sky-high valuations, heavy reliance on a handful of AI-driven tech stocks, and the dominance of companies like Nvidia in propping up broader markets. Alec and Jack discuss concerns around circular investment models, where capital flows between AI companies and hardware providers, raising questions about whether real value is being created or simply recycled on balance sheets.
  • The Reality of AI Adoption and LLM Progress. Despite investor anxiety, the hosts highlight a stark contrast at the product level: LLMs are improving rapidly and becoming deeply embedded in everyday workflows. From replacing search to enhancing productivity, AI tools are delivering tangible value to consumers and businesses alike. The conversation emphasizes that while enterprise-level ROI may take longer to materialize, dependence on AI systems is already a reality—and unlikely to reverse.
  • Hidden Risks. Chips, Energy, and Global Dependencies: Beyond market sentiment, the episode dives into structural risks that could act as real pressure points for the AI boom. These include reliance on a single chip manufacturer (TSMC), rising energy and water demands from data centers, and geopolitical tensions between the U.S. and China. Unlike valuation debates, these are physical constraints that could trigger disruption regardless of how strong AI demand remains.

Is there an AI bubble? Possibly—but not in the simplistic way the headlines suggest. While a short-term correction in tech markets may be inevitable, Alec and Jack argue that artificial intelligence itself is here to stay.

The real story is one of scale, infrastructure, and timing: massive long-term bets paired with very real near-term constraints. Whether markets cool or surge next, AI and LLMs are already too deeply woven into the digital economy to disappear.

Transcript

Auto-generated from the episode audio.

Welcome back to Untangling Web 3. >> We've got another great episode for you today. But before that, we'd love it if you could drop a like on the video and subscribe to the channel. >> It's the best way you can support the two of us and help the channel continue to grow. >> And with that said, on with the show. Welcome back to Untangling Web 3. It's been a minute, Alec. We haven't we haven't had an episode in a little while. We've been taking some stock uh spending a little bit of time, you know, sorting our shop for the new year.

Um, but today we I think we had to talk about this because it's something that's on everyone's minds. You have to have been living under a rock to have not heard the term the AI bubble, the impending bubble that everyone is supposedly worried about. So yeah, there's a lot of doomongering and fear-mongering out there on on on AI at the minute even though, you know, the models are still getting better all the time. I I find I'm using them more and more. Um, but yeah, so this is this is something that I think it's worth covering because there are there are some really interesting aspects to the story and there are arguments kind of on either side even though it can feel like everyone's everyone's just kind of doomsaying at the minute.

But yeah, I think it's a good time to uh to to end the year by covering a slightly uh more more worrying story. What do you think, >> Yeah, I think you hit the nail on the head. Taking some stock stocks and shares, all that. That's what we're talking about today. Because if you're just a normal everyday user who's not worried about, you know, the global economy, you're using GPT, you're using Gemini, you're using Claude, and you're thinking, my god, that these have come so far even in the two two and a half years that we've really been starting to experience LLMs at the product level.

Like they have come on bounds and leaps. Like they are absolutely incredible. It seems like every single week or every single month there is a new model that can do something that previously couldn't be done. And I think that's the that's the really interesting thing about this stuff is that there's a lot of fear-mongering coming from investors, VCs, you know, the big money makers of the world. But if you're just an everyday user, consumer, you don't see that. You're just seeing some of the most incredible advancements and innovation that in my mind has ever happened and such a fast rate, such a level of competition that we haven't seen in an industry before.

And I think, yeah, it's a really interesting time to be talking about this because we have a lot of users, a lot of people paying for these products, using these products, experiencing things that have never been possible before, but then there's still this kind of fear-mongering and worry that, oh, has the price been locked in? Are they making enough revenue? What does the future look like? So, yeah, I think it's a great time to talk about this stuff. >> Yeah. And I think for all the for all the good and like like you said, it is an incredible time just from a purely tech perspective.

>> um you know there's this doomongering around what this means for the everyday person right obviously and in terms of jobs that's something we've covered quite a lot before I think this term AI bubble is very much centered around the financial story and the what's happening in the markets broadly because obviously this year 2025 you know we've seen many many all-time highs of the stock market S&P 500 has been going gang busters for for most of the year um you despite some of the the early tribulations we had in the kind of early months of the year.

But obviously it's it's clear that that the market is heavily dominated by the what is it the magnificent seven now or eight I forget how many it is. Um but the the major tech companies obviously Nvidia currently the largest largest company in the world is you know driving a lot of this growth. uh is one of the f it's been one of the fastest growing companies for a little while now and they obviously are responsible for designing a lots of the the AI chips the GPUs that are powering this whole revolution.

So broadly in a nutshell what is the argument uh that we're in a bubble? Well, it's essentially that these AI stocks, these AI companies are in theory overvalued, right? So, there's a bunch of different arguments around why, you know, a lot they're trading trading at very high price to earnings ratios, which can generally signal that things are overpriced. That's one of the arguments. Um, that was kind of a that's a narrative that's been there all year. And then around two months ago, there was this story I think it was in Bloomberg where they started showing there's actually a lot of kind of circular uh revenue sharing and investment models going around in the industry, right?

Which is I think that's where a lot of people started using this term AI bubble in a slightly different sense that it's not just okay, these companies are worth too much, but actually maybe there's some inflation of revenue going on. Maybe the story isn't quite as clear. So I think that's that's I think a good starting point to dig into the story. >> Yeah. I know I know we said we weren't going to talk about stats because there's so many stats and numbers that can kind of be thrown around in this that can overwhelm the listeners and definitely overwhelm us as well just looking at them.

But I think like you talked about historic highs in the stock market and I was like yeah 2025 has been a great year for the stock market. It's been incredible. You know all I can think of is Trump speaking about that you're going to be so tired of losing. You're going to be like please stop stop winning. Let's stop we need to stop winning and start losing that meme. But 80% of all the gains have been AI related firms in 2025 which is bonkers. So it's like all of these historic highs in the last year have been almost entirely purely because of AI focused firms which is a bit bonkers.

So you kind of think then okay there's an absolute reliance right now on AI related firms propping up the stock market and all the gains we've seen in the last year. And then you're kind of thinking okay well I think a lot of tech is like this. It's often a bet on the future. Like we live in a society in which you kind of the promise of tomorrow [snorts] outweighs like what's possible today. So you're often kind of investing in the future and expecting gains ROI.

That's the kind of how the entire monetary system is based in the whole point of inflation and whatnot. So I think the point that people are trying to say here is that actually we've kind of over overinvested maybe like the price is locked in and are the basic economics of these firms do they make sense at the end of the day? And like you're seeing really big names now like the CEO even people heavily invested in the space like Sam Alman obviously the CEO and founder of GPT and open AAI um you've got Demis uh Hazabis the deep mind CEO he's also saying that he's starting to think that even their systems there is going to be a massive price correction because they're over inflated right now.

So yeah, I guess it's really makes sense to dive into some of the economics of like why that is. And you mentioned the idea of okay, Nvidia is obviously a big player in this space, right? I think their market cap in the last four years has quadrupled, which is a bit bonkers. They're effectively building the infrastructure and the hardware that kind of power a lot of these US-based AI models. Um, one of the big things we're going to talk about is the fact that they will invest in a company, that company will use that investment to buy the hardware off Nvidia.

And basically money is being passed around. But that's just capitalism and like basic economics, right? If I have£10 and I give10 to Jack and Jack gives£10 to his mom and his mom gives£10 to me, like it's still the same£10, but on the balance sheet it says, okay, our economy is worth 30 pound40 pound. So yeah, it's really kind of it's a bit complicated to dive into. Is this new? Is this novel? Is this something specific to AI? Or are people the bears of the world just trying to create some fear to maybe make some >> Yeah, it's difficult.

And I think we should obviously preface this whole conversation by saying we are very much not economists. We're not financial professionals. No, obviously no financial advice here anyway. Um but yeah, I think okay, there's a couple of things to to unpick and what you said, right? you touched on the kind of basic economic models of these companies. I think OpenAI is a good one as an example. So they're they're now topping revenues of around annualized 19 billion. So that sounds pretty good on paper, right? 19 billion in revenue.

They're obviously also not making profit yet, right? Which is classically a sign of a healthy company. Um but as you say, a lot of a lot of the valuations of these kind of companies and the interest in investing seems to have been driven by the fact that they're pricing in future revenues, future profits, right? So people are people are making bets that these companies will be you know the biggest companies in the world in 5 10 15 years time based on the fact that this you know the user base is going to grow significantly for these services that these guys offer.

Um now there's been some kind of detractors around that as well. So the the rate of expenditure of these companies you know so they're not making profit. they got some decent revenue. Um, but there's there's some evidence that the actual customer growth is slowing down. I think I saw that for OpenAI in particular, you know, the growth in new users or actual use of the product is not actually increasing as as as fast as you might expect for those future valuations, what they're kind of expecting in the future.

Um, and obviously we've seen a similar model of, you know, massive tech companies, as you said, it's it's not a new thing necessarily for AI. Um the whole blitz scaling model for tech companies has been around for ages, right? Where you just burn capital early on to be the winner in the category defining company for for a product category, right? And obviously there there's so much competition in that the AI and a [snorts] few kind of different verticals in it, but it seems like the blitz scaling model again for AI where these companies don't need to be profitable in the short term.

They just need to be the category winner in 5 10 15 years. Um, yeah. So, but on the that's kind of just a pure like base base level. How are they making money? What's the revenue model? What's the long-term goal? Then that circular the circular financing thing is really interesting because again it doesn't seem like it's a new thing. Um, I've seen lots of comparisons to I don't I don't remember if it was the dotcom bubble or a different bubble previously where you had you know lots of companies but but and effectively what we're talking about here is one company invests in another on the proviso that the that company then spends their money back to them right so it's like you're passing around the money birectionally and it can go in circles as we're seeing with the kind of Bloomberg reporting but essentially you're you're you might Nvidia for example is investing in open AI Okay, you're giving them capital that they can then spend on Nvidia chips, right?

So, it's not necessarily created new value in the world. It's just passing it around, but it does allow the game to continue for a bit longer. The question is how sustainable is this passing around of money. And how much is that feeding into overvaluation potentially of the companies? Because you'd think sophisticated investors, this is, you know, they're not going to they're not going to this isn't going to surprise them, right? they're going to have access they have access to this information. Maybe this is this is priced in and it's not so worrying.

But um on the surface it definitely looks like a worrying signal to me at least. Um but yeah, what do you think? >> Yeah, I I agree. I think but again I think it's basic capitalism and the way we measure like productivity is quite strange that you can just pass the same money around between 10 people and all of a sudden you've got 10 times the amount of money when really that's not the case. you've got the same money that's just being passed around. Um, I think you are right to focus on I think we're kind of both working from first principles here when we talk about how it should work and it should be is there a problem?

Are you solving a problem? Is the monetary value associated with the problem you're solving that you can then, you know, take and actually make money from. And I think that one of the promises of AI and a lot of the focus is typically on the retail use case. I think the real boom that we've seen is normal everyday users just playing around with chachi team making silly images using it to help respond to emails using it to help kind of I don't know draft things create content all this kind of stuff lots on social media FUD and all this kind of stuff and the question there is like how much productivity is that actually creating like a normal everyday user might not spend that much on it the real promise in my mind was deploying this at the enterprise level deploying this at the governmental level replacing maybe in replacing too wrong word, but at least enhancing everyday jobs to create new value and maybe reduce the reliance on more people.

So we saw like two years ago, I think it was, they were immediately saying, okay, we're not going to need developers in a year, two years time. Is that the case? No, it's obviously not the case. Like people can do more. Developers can do more. Maybe there's less need for junior developers or super junior developers in the world now, but we still need senior developers. We still need architects. We still need all these roles. like we I don't think we've seen the real kind of displacement of jobs that we were expecting at this stage which in my mind is probably is probably a good thing but that was going to be the real value ad in a lot of people's minds is okay an enterprise level what jobs can this fully replace so I don't have to spend 100k a year on a developer for example that was the real value ad and we haven't seen that yet are we going to see it I think we probably will I think like the one is

the technical angle like can these models actually outperform a developer in the day-to-day that's the one The other part is like culturally are we ready for that? Like are people ready to start replacing developers having you know instead of a person accountable having a model accountable for delivering a project? I don't think culturally we're ready for that especially when we talk in about normal companies that aren't at the frontier of this stuff. I think we're still at the experimental stage where people are like I said replacing junior developers but not the mid to senior levels that are actually going to run and own a project.

>> Yeah, I I I agree. I think like the ultimately when I mentioned the kind of slowdown of growth in in you know new customers you have to remember that OpenAI had the fastest adopted product in human history right it was like what it's something crazy like a million >> daily active users in some it was like under a month or something something crazy like that and that was three and a half years ago so it's natural to kind of reach some kind of saturation of using a product like that when you've had such high growth and that is all kind of retail customer end of you know day-to-day users like us that enterprise growth is is something that is going to take obviously longer to meet out and and actually understand how how much demand there is for it.

I I like you I think fully expect there will be huge demand for it you know for forever into the future right so long as this game continues um because of you know the tools are getting better all the time the models are getting better um and and they ultimately obviously just can replace a lot of a lot of you know teams of people to do things um so yeah parking the the kind of human cost and implications aside on that I think it's too it's probably too early to tell you know are are the valuations you know, if they are based on an assumption about huge enterprise adoption in the future and and kind of mass replacement of the workforce, it's [snorts] obviously way too early to know, you know, has this been priced incorrectly or not.

Um, but yeah, so that's kind of the the main I think the main narrative around the circular financing right now and people saying there's an AI bubble. So that's the second dimension, right? Number one, are these companies just too overvalued? Have we overvalued the impact of the tech? I think we're you and I are probably both on the side that, you know, I doubt it. >> I think that that the value is probably there in the long term. The circular funding and financing and investment thing I think uh maybe looks quite >> quite unsustainable in the short term, but also I I get the argument as well that there's so much there's so much benefit already that's been brought by these models.

Um and I think we'd all be quite angry as consumers if that all went away tomorrow, right? So there's already been comparisons with like open AI being too big to fail, right? I think they there's been stories around them saying, well, you know, maybe we need the government to step in and help fund some of this, right? Because ultimately we are going to become very dependent on it. And if something like open AI, you know, ceases to to to exist in the future, that will have a big negative knock on for for the economy probably straight away.

So, I think yeah, there's there's lots of uh there's there's lots to say about that funding argument that maybe this is part of just keeping the game going for long enough that things can, you know, can mature. So, all this expenditure and and working without profits. Um maybe that's just part of the cost of building this new insane industry that we we've never seen before. Um but yeah, that's that's the main kind of second dimension to the bubble. Um I don't know where we should go next, Alec.

I think we should probably talk about the hardware side of this because there's an interesting kind of dimension around the GPUs and the relationship between Nvidia and maybe TSMC as well. But I don't know if if you want to cover anything before that because I definitely are definitely interested in that side. >> Yeah, I think we should jump onto that. I think just before that just because you mentioned the idea of governments and we talked about Nvidia and Nvidia probably being the major player in this whole space.

I mean they're providing both the hardware and a lot of the investment into these AI companies like you said creating this kind of circular dependency. I think one of the most interesting things for me that's happened in recent weeks and months is around the kind of the US export controls of the Nvidia chips to China. Like there's two big players in the AI race. It's China and America and America is leading in a lot of ways and Nvidia has a monopoly in a lot of people's minds on the kind of the advanced chips that power these AI models.

And one of the interesting things that the current administration in the US did was actually starting to limit these advanced chips to China. And obviously that's not very good for business if China's the second biggest player in this space. And then they opened up the doors. The administration realized, okay, we shouldn't have done this. That's a massive market for us. We do want to dominate the space still. Um, and we want to create external revenue just, you know, outside the US, not just in the US.

And as soon as they open up the doors, Beijing were like, "No, we're going to make our own chips." like they don't want to be dependent on these kind of external US companies. And again, I think this is the point here that I'm trying to make is we're kind of assuming a monopoly and a dominance from these US tech giants, the Magnificent 7, Nvidia, all these people. But there are competitors coming out like that's only in the US domain. China is kind of offering a in my opinion like a much more efficient economical model for these systems that potentially isn't overinflated.

And the reason they can do that is because they are so governmentbacked like a lot of these are sub subsidized by the government. And it's just something we should start to think about if we're starting to work out, okay, are these companies overinflated? Are they going to dominate the world? Or can China create like a competitive, more economical, cheaper version of these? Like a good example of this is the Deep Seek model. When that came out, everyone assumed that Chachi PT was going to be the model that dominated absolutely everything.

You pay per usage. They have a black box complete control over that service. So effectively if they dominate the landscape they can charge whatever price they want and then all of a sudden Deep Seek came out and you know what is it like I think most of the AI companies dropped by like 5 to 10% overnight because all of a sudden they've got this much cheaper open free model that anyone could potentially run on their own hardware. That destroys the entire business model of a lot of these AI companies.

And I think they're just they're two dynamics that I think we should really take into consideration is what are people doing outside the US, specifically China. They're trying to build models that are much more much much cheaper. They're less compute focused, more efficiency focused, which again is very appealing to businesses, very appealing to everyday users potentially as well. And then the other one is free models versus premium AI models. like are is that going to completely disrupt and undermine the business models of these top seven AI companies around the world?

Like they're things that I think we should really consider when we're looking at the future of are these overinflated? Are they overpriced? Who's going to win? >> Yeah, that's actually a really interesting way of looking at it, right? because you're completely right in the sense that I think China is the way they're approaching AI um you know with deepseek looking at more local LMS as well um it's almost like America has and and arguably the rest of the world has assumed that this will follow the same path and trajectory as like the cloud industry right so that everyone accesses AI and the capabilities of LLM and you know whatever it is in the future robotics and via the giant cloud providers right giant through cloud services in the same way that we do most of the internet now it's not an unreasonable bet you know because ultimately convenience is it tends to be king um >> so I think there'll be huge role of these cloud providers and when you look at that that circular

kind of funding bubble uh and the structures there you know some of the other big names in that are Microsoft are Oracle right some classic cloud providers in there so a lot of that the industry is quite heavily tied to that And as you say, rightly so, America is incredibly compute focused right now. Um whereas China is offering I I guess you know a slightly a slightly alternative model as you said like lighter weight um maybe not so tied to these giant cloud providers. Um but yeah, it's it's interesting to see that it's so government-ledd and most countries in the world, especially the want to be major players are now obviously putting out their own AI strategy having to do, you know, major investment into the infrastructure as well behind it.

Um which I think dovetales very nicely into the conversation around the chips themselves because that that's an arguably a third dimension to the AI bubble. Maybe not not so much in how the bubble has grown or if there is a bubble, you know, the size of it, but it's more like a pin that could pop the bubble if it exists, right? Or even if it doesn't >> because there's this kind of single point of failure that we should discuss. So yeah, I think one point that is not necessarily that well known uh you know I guess with you know with um the average person like us but it's really interesting is the fact that Nvidia is primarily a chip designer right so they are they're designing the GPUs which is you know incredibly difficult and they are the world leaders in it for running these AI systems but they don't really fabricate the chips themselves like actually producing these wafers and these nanometer scale transistors.

That's actually not a job they do, right? This is this is a lot of it is being done by the Taiwanese semiconductor manufacturing company TSMC, which is very much arguably this very very acute single point of failure that if something went wrong or you know they suddenly stopped working then maybe the Jenga of of the whole industry could fall over quite quickly. >> Yeah. I think that it's it's quite interesting, right, that we're talking about US versus China and in this global AI race and then all of a sudden you mentioned Taiwan and the Taiwan semiconductor manufacturing company.

So there is obviously a potential issue there. You know, I don't think we're going to go too much into the politics of it, but >> what happens if say all of a sudden TSMC goes out of operation or can no longer be accessed by the West? And I think they are starting to I think a lot of the west is starting to wake up to this fact and trying to develop like trying to develop alternatives. But like I think a way we can kind of think about this is what happened when there was export controls from China on critical minerals and um precious metals and all this kind of stuff right the west lost its mind like all of a sudden you're thinking okay we can't develop the most advanced you know platforms what's it called like US defense systems F3 fighter system what are they called F-35 fighter jets all this kind of stuff and computers if all of a sudden you start to bottleneck and have a complete reliance on one company or

one supply and then that no longer becomes accessible which is a real possibility right it could be that that place doesn't become accessible in the next 2 to 3 years like we're not going to go too much into that but what does that mean for the AI dreams of the US and the west if that does happen well it doesn't happen right like if that was to be switched off today there wouldn't be a future and I think like what all of these the magnificent seven would collapse over a course of a week >> yeah and and I think it's worth mentioning like even geopolitical issues aside I the fact that again TSMC to to my understanding is one of these category defining companies in the sense that there is essentially no one else on the planet that has the chip fabrication capacity that that they do and to do because we're we're talking about transistors.

There's a really really good video I think I'll I'll link it in the description that shows you what the size of these nanometer scale transistors are actually like and it's it's mindboggling like it's smaller than like DNA or something like that. um the level you're getting to here and they are so they're they're manufacturing these with incredibly high precision. The facilities themselves again you can find videos of them like you have to control the air supply. You you you're having people in hazmat suits because any kind of contamination will massively you know increase the failure rate because there is a failure rate when you're when you're when you're manufacturing these kind of things.

uh it can be very easy to to to go wrong and actually I think Samsung is one of the the competitors to TSMC and they have you know one of the best they're one of the better uh chip fabricators in the world after them and their failure rates are so high you know I think Nvidia actually tried to move to using them a couple of years ago and they had to move back away because the failure rate was too high they couldn't supply you know something more like 30 or 40% right failure rate whereas TSMC are much And so just having a single point of failure like that in the industry which for whatever reason if it started failing or you know anything went wrong in the plant you know for you know if it burned down right or because someone left the lights on or something then that that's a huge issue and it's going to I think we are seeing as you said like other countries like America is now investing in chip

fabrication sites like it started these you know multi-billion dollar maybe even you know hundred billion dollars I can't the exact numbers but they've started these projects to build their own local chip fab uh in America lots of Texan based ones but they are going to take a long long time to do like could be 10 years before these are operational and there's no guarantee that they'll even be you know competitive with TSMC at that point so it is a really big it's a really big bottleneck like in terms of the the geography of of the AI landscape that everything is just based on on fab chip chips that come from essentially one >> it's interesting because like the other dynamics you've talked about in this kind of bubble setup are much more I guess vague and ambiguous like the idea of kind business models like is there enough value are people overinflating things like the secular and circular kind of investment cycles and fabricated money and whatnot but then TSMC is that's a very real

dependency right if that all of a sudden switches off and is can no longer be accessed by these big firms that is an overnight switch from okay go to no more like do you think that's the biggest potential disruptor or the biggest potential kind of critical point of failure for this bubble potential bubble or the AI industry rather than bubble. I think it probably is the biggest trigger, potential trigger of of a collapse in the industry, right? Because as you said, like after Deep Seek, you know, the market reacted um very quickly.

And I think the uh something happening with TSMC, you know, not being able to access it for whatever reason would be a real sharp acute cause for panic in the industry because you know all that money that's in going into these um you know those assumptions about AI growth in the future and who's going to be the market leader well all of a sudden if access to chips you know goes goes away then those assumptions get flipped on the head very quickly that the that the competition between different countries in the world can change very very quickly as well.

So yeah, I think it's it's not necessarily a c it's not like a bubble dynamic in that sense, but it is the pin that could that could you know pop the bubble um most quickly as far as I see it. I don't think there's any other single thing to my knowledge that that could do that. So yeah, it's definitely worth keeping an eye on. But yeah, as I said, I don't I don't I I also don't see that that that kind of issue will go away anytime soon because getting par with TSMC in other countries is just going to take so much investment and so much time um that yeah, it's hard to see hard to see how that goes away quickly.

>> That makes sense. I think another one like another risk factor that I'd like to bring up is you talked about western reliance on compute and like AI's growth. It's not just logical, financial, all this kind of stuff. It's quite physical like we've been talking about with the hardware dependencies and one thing we haven't mentioned is that hardware needs power, it needs cooling, it needs water and those costs especially in the west are like rising and quite real. Like I think I've seen some stats that AI workloads account for around 10% of global data center power right now and they think that's going to rise to about 30 to 50% at current rates by 2030.

And then you've got like so they're saying that by I think maybe this year or by maybe it's in the next couple of years they think that the AI power use could exceed that of Bitcoin mining right at its peak which obviously that caused an absolute kind of storm of people mal content with this stuff. So I think that there's the energy usage as well and this isn't something that a private company can necessarily solve instantly right energy usage is a grid dependent. It's obviously sustainability compliance dependent way actually sourcing that.

It tends to be quite governmentbased these big infrastructure projects. So it's not something that say an open AI of the world could just you know click their fingers and all of a sudden you know be able to generate 23 gawatt of energy right these are multi-year huge projects that typically need government financial backing. Like we look at ourselves in the UK right now our energy bills are skyrocketing and we're also building data centers. There's obviously a lot of push back to be like, okay, well, if we're going to up our energy usage at data centers by 10%, that's going to mean more costs for everyday consumers and users in their homes.

So, I see there being like a wider problem of the compute leading to more demand of energy and also water usage. Like, it's absolutely it's unbelievable the amount of water these places use as well. And I think there's going to be more and more not just environmental push back, but everyday, you know, retail user push back as well to be like my energy bills, my water bills are going through the roof. How can you still be building more energy centers that make it a bit less reliable, a bit less easy to use, and obviously a lot more costly?

>> Yeah, I'm glad you also mentioned the water usage because that's something that there's been a lot of criticism around recently is how these companies are actually accounting for water usage and you know, I think the creating metrics like uh water used per GPT call, right? That kind of thing, per prompt you send. Um it's it's very hard to estimate and you know you question do you include all of the water that went into cooling the the systems while they were being trained right and all all that compute time before or are you doing on an ongoing basis you know actually this call what is it taking >> um yeah and and again it goes back into a geopolitical issue in you know energy dependency water dependency in in different countries right we we don't live in a world where a lot of countries are self-sufficient and this and that are incredibly reliant on energy sources elsewhere.

Um, who knows? Maybe maybe this we can reuse the argument like I always say for Bitcoin in that it should it should in theory create this push to have uh you know minimizing energy dependence using renewable resources locally to power these kind of things and data centers. But again as you say like the pace of this industry is always going to cause problems here. Like did you I think there was a a story around Meta was adding um adding data center capability kind of just under tents like under Tarpoolin near their facilities right because they just had to expand.

They didn't have buildings yet so they were just like putting them up inside tents these these servers right so the growth is is is outstripping natural resource demand pretty pretty quickly at the minute. I think one of the the trends that I've been seeing as well is that I think that it seems to be getting cheaper over time like right now like it's expensive now but it seems like a lot of the models as we're getting more intelligent like the previous models like saying a good example is GPT40 like the mini version it is as advanced if not more advanced than you know the current Frontier model 2 years ago but at a fraction of the price I think that like Frontier models training obviously costs a lot it's like hundreds of millions to a billion per run or something like that and that's compounding apparently like it's getting more and more to kind of train these frontier models but the smaller models that kind of follow that that are a bit more localized a

bit more efficient are actually much more cost effective so I hope that we are going to see a future in which you don't need to use the most advanced the biggest model constantly and you do get these kind of more energyefficient compute efficient models that you can run locally or you know more efficiently over time and I think that's something that I've started to notice as I don't always need the Opus 4.5. I can run smaller, more efficient models that are a lot cheaper, both in terms of cost and also in terms of energy.

So, that might be something that helps with some of the energy reliance that we've talked about. It's not always going to keep growing constantly. We can move to smaller models potentially in >> Yeah, it's a really really good point. I think yeah, it's easy to lose sight of the fact that, you know, just how how much more capable these models are. Um, you know that you have got the classic example of Will Smith eating the spaghetti gets better and better every year. Google's nano banana this year was was you know miles miles better than any previous image generator and in a year and year or two's time the cost of using that is probably going to be you know even lower and and again you are seeing when these models come out especially the smaller models like the flash models and things for Gemini they're they're being advertised on how they undercut price and compute cost of of previous models as well um so yeah that that's actually a nice nice positive sign I think

and yeah maybe we'll reach a stage where the the models that are good enough, right, for 99% of use cases, those kind of retail consumer use cases that we we might use on a day-to-day basis, maybe those are, you know, much much cheaper and have less of an impact on on the economy and on on the environment potentially. Um but yeah, so maybe just to to kind of round off the story on the bubble. Um obviously there's lots of there's naturally there's lots of comparisons going on with the the.com bubble and other previous historic bubbles especially related to tech.

Um and one of the one of the big stories surrounding this has been the behavior bur right. So uh if you haven't heard of him he's he's one of the one of the guys from the big short essentially. So, one of the one of the actual Wall Street traders who uh saw the the do the 2008 financial crash coming uh managed to bet against it and you know made made a made a killing. So, he's often seen as one of [snorts] the harbingers of of of predictions and you know he's made a lot of predictions over the years and just to add fuel to the flame obviously he's now uh you know taken certain positions in the market where it looks like he's betting against the big AO players, right?

So, he's like shorting some of these these major stocks now. What do you make of that? >> Well, what's the phrase? I think you did it in the notes. It says he is early until he's suddenly right. So, it might be that he's like predicted something that could happen in like many years time or something like that. If you've watched the big short, you see that he predicted it like several weeks if not months before it actually happened. Yeah, it's tricky. I think that he's obviously famous for what happened with the mortgage collapse, but it's they're very different, right?

I think that like that was a level of complexity that you couldn't explain in a three-hour podcast. Like there even having watched the big show, I still don't understand exactly what was going on. I think that there's a slight difference here in that value is being created by these systems like value is being created by these models. It's just is it less value than was predicted at this stage. Are they kind of capturing the value um early enough or not? And I think I'm of the opinion of someone that uses multiple models every single day that this value will come to light.

Like even if it's three to five years versus one to two years, I think that it is inevitable. I think that you see people that are completely reliant on these LLMs. They are replacing like entire workflows in businesses. They're replacing entire workflows in day-to-day life. Like people are using chat GPT instead of Google search. And obviously Google search was an absolute money maker. And most people are using things like GPT to find out their content. It's just how do you translate those clicks into money? Like some go down the productivity route like your co-pilot, your enterprise tools, all this kind of stuff when you're targeting businesses specifically.

Some are going down the ads and search route like Google Meta and GPT will probably go down that route more and more and then others go down like the kind of the overall infrastructure lock in like your Azur, your AWS. It's just how do they start to turn this attention into real money? And the only difference being things like Meta that kind of just had attention, just had ads, the burn rate of money was is nowhere near as high as these AI firms. So you kind of have the difficulty of making money versus the difficulty of spending a lot of money.

But I think it's inevitable. I think like there's a complete dependence on these systems now and people will not move away from that dependence. It's just how does that translate into value and it might just take a bit longer than was previously expected. Yeah. Yeah. I think that that's that's a good way of looking at it. And you know, when you compare against previous bubbles, you know, from my understanding, take the.com bubble for example, it's very easy to reach for that as a comparison. And you know, is this just history repeating itself, but by all accounts, there are lots of differences in how the market is is structured for the AI cycle and the previous, you know, do bubble cycle.

So obviously lots of the companies are as we said making a lot of revenue. They may not be profitable yet but there is real revenue there. There's clearly demand for what these companies are building and they are building genuinely valuable things right. So that is that is a difference you have [snorts] uh with a lot of the com bubble where you know you just have people putting the internet or www into their into their you know company names and getting funding off that. Um, and I think it's not only the new entrance in, you know, it's it's arguably a new category, but it's not just completely new entrance like OpenAI into the market.

It's obviously these these huge existing players, right? You know, the rest of the Magnificent 7. I mentioned Microsoft, Oracle. A lot of these companies are so well established. They've been they've been the biggest companies in the world for many many years. They are also in many cases quite cashrich. So, they do have deep pockets to to invest in this infrastructure. they're not startups, you know, that are like growing at an a kind of unnecessarily inflated rate. >> Um, and I do I I kind of just on a a very subjective view, I get the sense that people are relatively wise to the whole uh, you know, putting AI in the title way of getting money, right?

It's like we saw it a bit more with crypto I think a few years ago and blockchain but I think people are wise to the fact that you know there's not a huge amount of difference between a lot of these models and people are veryware aware of the best companies the best performing companies the best products in the space like it's not quite so finger in the air when it comes to how the market seems to be investing in these companies. So yeah, I'm I'm less worried that there's a bubble here related to, you know, pure misvaluation and and kind of completely missing the mark.

It's more I think the bigger problem is just the fact that it's on such a big scale and it is the >> you know, I arguably the fastest biggest growing industry right now and how big the bets are going into it, how long-term the bets are. I guess just it's purely the scale for me that is is is the worrying aspect of this alongside with all the other you know the trigger point with with TSMC and the uh and the kind of and the the the the energy kind of resource uh questions as well.

I think yeah I think the the evidence for I think a big thing is I do think there are not the magnificent 7 per se I think that one's a much more complicated one but when you see these startups that are just AI enabled insert whatever term and they're you know >> yeah exactly just GPT rappers and then they have incredible incredibly high valuations without any usage they're just the whole game plan is just to have AI plus some workflow and then like you said there's these memes that are popping up everywhere where it's like, okay, well, open AI just killed my entire business model and then a thousand startups get killed just from one upgrade from GPT.

I think that stuff makes a lot of sense that a lot of investors are losing a lot of money because people are getting overly excited. They're thinking that, oh, maybe the Magnificent 7, all the kind of all the price now is cashed in, all the price is locked in. I need to find a smaller, more disruptive model where I can actually make money. And imagine that's where a lot of people will lose money. I think a big thing you see is so many companies are investing in this space and enterprise level and not seeing the ROI.

Like a big thing that I've seen more and more is that it's because a lot of these people are using Copilot. And no offense to Copilot, but C-pilot sucks. Like when you're kind of a big enterprise, you want to go with something that's safe, secure, built into your workflows, but you're not using the top-end models. Like I've used C-Pilot. I don't like C-Pilot. It's probably where GPT was at like two years ago. And I think we're going to see a maturity of enterprise use cases. And I think, you know, it's still fairly early.

Like the enterprises are only the big enterprise anyway were only jumping on AI usage in the past year. I think this is so disruptive and so novel and so new. It's going to take a few more years for enterprises to really feel comfortable deploying this stuff and getting the ROI and potentially replacing entire workflows with this stuff. So yeah, I personally think it's coming. I think like not financial advice, there will be a small correction, but I think like it's long-term inevitable. The short-term correction will probably just coincide with like wider geopolitical kind of recession fear factors and all this kind of stuff, but I think long-term on this space like AI is kind of inevitable in my opinion.

>> Yes, doing your Thanos impression there, AI is inevitable. [laughter] Um yeah, I think that it's a really good point maybe to end on is the just the fact that obviously there are economic cycles, there are recessions, there are booms and busts and uh people have been predicting that there will be you know a huge crash in the global financial markets anyway for for a little while now. Um so you know these things happen and I think maybe this one's just being branded the the AI bubble because it is the hot the hot ticket right now.

the biggest tech story and the biggest company in the world Nvidia is is is you know driving the growth of this and you know along with the other magnificent seven. Um so yeah I think it's not a completely Grinch story it's it's it's it's an interesting one and I think maybe just to end the purpose of you know today's episode was just to dig into what are the arguments around you know is there an AI bubble and some of the more nuance stories around the tech behind that and and the uh markets that are propping up the industry right now.

So yeah, lots for us all to learn and lots to, you know, stay safe out there with. But yeah, this is quite a fun one. Alec, I enjoyed this. >> Yeah, I enjoyed it, too. I think like the promise of AI is huge and I think there's so much it can do. But there are a lot of bigger dependencies outside of just the AI companies that we've talked about. Some of the manufacturing dependencies, the chip dependencies, the wider geopolitical dependencies, the energy dependencies. There's a lot of dependencies that I think these big AI companies will require to keep growing the way in which they're growing.

And with dependencies comes risk. So yeah, stay safe out there. Do your own research. Don't just rely on me and Jack. And yeah, thanks for listening wherever you may be. And join us next time to untangle a little more of Web 3. Thanks for tuning in to another episode of Untangling Web 3. >> Don't forget to like and subscribe and to follow us on social media. It really does help us out. The views we express here are our own and do not reflect the views of our employers.

And with that, we'll see you next time.

Topics: AI & Robotics · Blockchain & Crypto

Related episodes

All episodes