82. Is the AI Bubble About to Burst? The Warning Signs You Need to Know

0:00Welcome to The Difference Engine, the show for tech founders, investors and innovators.

0:11okay this time we're talking about the problems of succession no not the teleprogramme succession in tech should you go for the insider or the outsider approach and everybody's talking about the ai bubble so here's three reasons why we think it ain't going to burst anytime soon

0:31so the next tech crash is it going to be soon will it be the biggest isn't it about time we maybe called out some of the doomsterism which which is appearing at the moment don't you think i don't know i mean this big tech crash everyone's talking about looks like it could be imminent actually you know we need to put this in some sort of perspective i mean human human societies it seemed to have accepted that the march of it is is somehow inevitable and you know that's been continuing since the first world war with a few blips you know we've we've seen we've seen it move into our lives from you know office and factory automation to gaming and mobility and so on and and you know we've seen relentless category creation going on in this um but you know are we now actually reaching something different a sort of tipping point where you know Real numbers and societal acceptance or not of AI and the data centers that drives it are actually accelerating an inevitable classic black tulip mania crash driven by what I would see as an over-exuberant supply of capital into what is actually a very resource-restricted space.

1:46And is someone going to cry, Emperor's New Clothes, and some form of rational reality dawn, and as this sort of irrational hope evaporates. Oh, yeah, I think so. Well, I think the guy who's going to call Emperor's New Closes out there, good old Ed Zitron, Ed and I, spoken a while back. And of course, back in the days when Ed and I were speaking to each other, he was a PR man. In fact, if I recall, it was EZPR. But Ed Zitron has now reinvented himself as a commentator, a researcher, no less. And he's talking about a lot of founder cycle type stuff, like end of an era. I thought we'd look at three sort of, on the surface level, unconnected, but connected, I believe, moves here.

2:32The fact that data center bills are being stalled in the US, and you're the eco guy, so you'll know why that's all to do with power outages, etc. etc um it's also talked about um the chinese who are now threatening to to withdraw their ai models like take their toys home just when we all got used to them a bit harsh and then um meanwhile over here in good old eu where they say we legislate don't innovate we've got the eu issuing massive fines to google and apple and apple's saying right if you're going to do that we'll withdraw our tech from europe so this feels a bit like top of the bubble um is it or can we hold on Something I heard the other day from some very, very well-informed consultants was that the financial picture for OpenAI and Anthropic is now so weak, actually, they should not even be allowed to IPO.

3:25Yeah, I hear that a lot. Ed's also argued, and this is the, you know, God bless him, he's got a turn of phrase here. Speaking of generative AI, his quote is, there's not really a business there. Wow. Wow. That's pretty short. That's going to come as a bit of a shock for some people. Yeah, but it said some pretty sharp things. I mean, for instance, that companies encourage waste because they charge for usage rather than outcomes, right, in this day and age. Yeah, tokenomics, baby. Yeah, they haven't changed the basic model. They've applied a completely different technology to a very different model.

3:58It's sort of like broadband versus dial-up. That went really well, didn't it? So it's also been pretty tough about AI agents and Dash co-workers. Well, we love those guys. They're brilliant people. But didn't he say they want everybody else to do the innovation for them and spend as much time as they can on tokens? That doesn't sound to me like a very productive idea. What he's saying there is that it serves the interests of LLM folks to hype up AI agents and co-workers, and we'll all have hundreds of them. why because they burn tokens and what do they sell they sell tokens so he's pretty pretty neg on the whole thing this is bubbly stuff i would say yeah but he's you know he's he says something i thought was stunning but also the bleeding of his because we've seen it over and over again is is that you know big tech and venture capital is is investing in our ai because it's run out of other growth stories and frankly through its own actions it's reduced the value of many of its investments in in SaaS companies.

5:05They've basically run out of hyper growth ideas so this is the new hyper growth idea and something else he's warned about which is very real world to my mind is that infrastructure build out could really turn into a financial problem if um if capex and debt keep rising yeah so there is a danger that data center debt stops being issues um and you know when that happens it's bye-bye industry right yeah it's like musical chairs right when the music stops you got to sit down that data center build slowing or be delayed is probably the most important issue in all of this right so an investment boom has to assume things and in this case it's assuming that you know tens or hundreds of gigawatts of new power and data center capacity will come online in in the next few years nobody knows but if you think about it there are some genuine bottlenecks in that right so you've got the delays from utility interconnection you've got shortage of transformers you've got gas turbine supply constraints and frankly you shouldn't be using gas turbines to power these things.

6:19You've got power transmission build out, which often can't even service the basic houses in California that it was supposed to be serving. You've got the whole issue of planning permission. No matter how fast you drive that through, we're seeing more and more local and political opposition to data center build and indeed AI take up. There's just a massive environmental acceptability issue about this. What we have seen are the likes of Google and Microsoft completely breaking their commitments to carbon reduction at a stroke with these investments. So if those constraints persist, then one half of the investment equation ceases to work.

7:07You know, AI revenue could arrive more slowly than investors expect, right? um however there's a weird economic quirk in all of this which is not the jevons paradox again i'm not sure if this is jevons but it might be tell me if you think this is jevons so delayed projects don't necessarily mean mean cancelled demand so companies like hyperscalers have enormous capital budgets and just increasingly build wherever power is available so out they go and put and you know generate their own power the problem as i said before that could be gas turbines which is not a good idea well i mean that could be jevin's paradox we just don't know yet the jury's out which is why so much fun speculating but what we do know real and present is power is a strategic constraint already oh god yeah i mean we've seen some amazing headlines on that i mean if anybody has not not been listening hasn't the economist and reuters and fortune been going on about that yeah yeah so the economy says will the will the data center backlash derail the ai boom sort of what you're just saying uh talking about how the american tech giants and they're focusing in on America here, but of course, we've got similar issues here in Europe.

8:15The Germans switched off their nuclear. That was clever, because you could always rely on Mr. Putin to help you out. We've got a terrible national grid here in the UK. And having just come back from France, they are quite ensconced in nuclear, but that has its own issues. But the American tech giants in particular, according to The Economist, are going to face lots of local opposition, are facing lots of local opposition to data centers. And the newspaper said that the backlash has made it difficult to find places to build them. Ouch, that's going to slow things down. You know, this is a problem when the grid operators themselves, in whose interest it is to look like they're in control of the thing, are already saying that the demand for electricity to power AI and manufacturing actually is rising much faster than historical trends.

9:02So what they mean by that is they've got no idea how they're going to meet the demand based on what they know from the past so i think that this creates really three possibilities so ai grows slows simply because electricity isn't available um massive investment could flow into generation transmission and storage problem with that is it's not nearly as sexy as ai and and will the capital follow or, and this is probably what will happen, is both happen simultaneously in different regions at different speeds. So it's a big mishmash. However, if I was going to be positive about this, the idea of power shortages coming up with this key enabling technology might actually increase spending on infrastructure or infrastructure companies.

9:53So spending on infrastructure investment in infrastructure companies across the world, which of course has a quite a nice um spin-off which just increases national security but it will slow returns for ai infrastructure users so you know bit of a dance with the devil that's possibly the first time uh you've given financial guidance or advice and and share tips no it is it is not financial advice it is not share tips it absolutely not it's an opinion let's put it this way there may be a return in um infrastructure companies possibly what about the Chinese? Well, you can't invest in the Chinese.

10:28It's very hard, very hard. But the Chinese AI companies are now saying, you can't use our stuff. This is definitely more geopolitical than fundamental, but nonetheless, it is a real threat. The Chinese are saying, we might withdraw our LLM models, which, let's face it, are a tenth or less of the cost of the US models, and a lot of folks are relying on them and if you are somewhat sophisticated as a user you can choose the model which you um deploy your generative ai agent on so um you know economics being economics the chinese uh open generally open source um models below that are booming and now the chinese are saying we might switch them off yeah right but you know if they do here's the positive i've I've been trying to think about it positive for this.

11:21And it may actually, and I'm not sure if this is the positive, but it may have the effect of strengthening US model providers, you know, providing they don't bend to the whims of Dementia Donny, because that is another massive spanner that could be thrown into the AI works. This stuff is disruptive, but I don't think it's massively to worry about. Well, I don't know. I mean, like, talking about the geopolitical side of this, chinese takeaway forgive the pun is that um anthropic is now making eyes at the u.s government and saying we will give you not you not offer you to buy give you five percent then you've got a lot of alignment between the owners of that five percent capital i.e the u.s government and the people they're competing with in the market many of whom are chinese so sorry what's what's government ownership of the commanding heights of the economy have they gone communist or something what are they thinking so ironic sliding into socialism what will president trump think about that i don't know so what do you think what do you think about this uh chinese takeaway then if they do take it away it'll reduce competition for sure um what else do you see it could it could strengthen it could strengthen innovation in in the rest of the world i think this is just a balancing act.

12:41I'm actually not really too worried about it in terms of its impact on the industry. It's just another geopolitical complexity there. This sort of backs up Ed's point, doesn't it? And you've just said the same thing. You said that you're not too worried about it. So if AI is not as important as we are led to believe, and there isn't really a model in generative AI as Ed purports then maybe maybe it's not something to worry about and maybe we should look close to hand with the eu of course ai is important but the the ultimate impact of ai won't be from large language models it'll be from its selective application right through the economy um and you know we we see the whole ai through the application of llms at the moment and that is not where the real value will actually happen.

13:32And as we've said, it's the actual application of AI where the Europeans may have a Trump, pun intended, card. So the EU fining again, it's all fine in the EU. What is it with everybody getting antsy and throwing their toys out the ground? Apple has changed its app store policy because it doesn't want to be fined even more by the EU. What the bloody hell? Well, that's clearly the EU policy actually having an impact. Let's be clear about this. Apple wants to look tough at home. So throwing its toys out of the pram is frankly a current American way of dealing with things, it appears. It'd be easy to say that Europe is becoming a more difficult regulatory environment for big tech.

14:22My argument would be it's all it is is a counterforce to their pegaminy and you know in a proper capitalist market economy that actually has to be a good thing otherwise we're faced with monopoly and taking what we're given I mean I think that's a bit of a broad brush some of the um you know the GDPR obviously we like that um but some of the other moves that are being legislated may lead us to everybody having a mandated ID. I think that is the issue, the free speech issue that the US big tech's hanging its hat on, you know, to get the moral high ground here. And if they do that, then we'll have Europe where everybody's tracked phenomenally and the US arguably, although I don't know the reality of this, a little less so.

15:13So it's becoming an existential sort of fight, not just about the, you know, 500 million euros here, 700 million euros. It's funny how the same fine in the Wall Street Journal is 500. In the BBC, they've upped it to 700. And it's strange what adversely makes strange bedfellows. You've got both Apple and Meta attacking the EU now. It's really spicing up. And this is the sort of stuff I think that sort of exemplifies a bubble. So are these three pointers of a bubble? We've got power issues. We've got the Chinese declining to serve the rest of the world their amazing technology. We've got the EU legislating like crazy.

15:55Are these enough to agree with Ed and say, AI ain't really a business, generative AI isn't really an industry? Well, allow me to try and segment this and think about what would be a bearish case in that or a bullish case. So my argument for the bearish case would be, from an economic point of view, AI has become highly concentrated. And that means some of the largest companies are priced in a very particular way. It is expected that they will sustain 20, 40 or more percent revenue growth. And on their numbers, those are very big numbers. So there is enormous sale and monetization at the moment.

16:45So a very small group of people are making an enormous amount of money. There is a continuous hyperscaler capex because these companies do have the money to invest. And there is plentiful pricing, financing, and that goes on. And let's assume there's enough electricity to power this. but if even one of those five things weakens i think the multiples could compress sharply and and when that starts to happen it's downward you know and again if we go back in you know a history lesson major technology booms often hit physical or demand bottlenecks before the financial ones. There's a lag. It's sort of emperor's new clothes.

17:39It's obvious something's going on, but the financial markets don't want to admit it and carry on and carry on and carry on until they're forced to correct. That's super interesting. Let me try and make the opposite case to that. Go bully. Go bully. So I'm sorry, Ed, with greatest respect. Let's look at the folks that are generating this cash flow. They are themselves cash cows. Microsoft may have got a lot of people at Xbox, but it's making a ton of money from its old, and I mean like 40-year-old product suite that's clouded. Amazon, the Everything Store. Now, the Everything Cloud. Alphabet, Google, still the world's largest advertising company, and Meta does horrible things with your data, but makes a fine business doing it.

18:26So these guys are largely funding their ai stuff and i say largely voter comments from operating cash flow they've got they've got cash machines in the middle and that's a big difference the big difference is from the dotcom boom where the same thing was happening but it was happening on other people's savings so you know this this is a classic maturing of the market there is you know an internal cash machine going on it's still not enough to feel what's going on but it's different but we're missing a third possibility We're in a bubble. We're not in a bubble. We're in a bubble whether we like it or not, and things just continue as normal.

19:03Could that happen? Some examples would say that it can for the near future. So the late 1990s internet boom went on for way longer than it was obvious that the real world numbers just weren't there. Massive discontinuities between the availability of applications and people's ability to access those applications. Is it Wile E. Coyote that goes over the edge of the cliff and his legs are still going until he looks down? I also think you could argue that the US housing bubble before about 2008, it was obvious that the whole thing was being built on mud and shifting sand. And also the Japanese equities market in the late 1980s also went on for way longer than the fundamentals should have supported.

19:54um so you know the warning signs appeared years before the ultimate uh peak all right so let's pull this together what would in fact convince two old skeptics like us that we are nearing the peak that ed's right that it's all over this is bubble and and the pops coming and cash your savings cash your investments cash your pensions and get out what are you what would you look for for that to happen oh right okay you could put me on the spot now so first thing i think is have a look at the hyperscalers. Are they actually cutting AI capital expenditure, capex, rather than just delaying projects? I'm sure we'll see a bit of delaying projects going on.

20:38My view would be that Microsoft, and you just referred to this, recent downsizing of its gaming business would tell us that this is not yet happening. Well, to counter that though, there's a lot of people, Meta's just started selling excess AI capacity. If there's such demand why are they selling it yeah well that sort of links to what i think would be my second indicator would be that cloud ai in utilization is actually falling um yeah so is supply overtaking demand again clap that is that is classic bubbly popply um what about nvidia i mean they're sort of the bellwether in this if they started reporting several quarters of slowing data center revenue growth.

21:22That's a real sign. Because if this starts happening, then confidence evaporates ahead of it. And I think one of the big thing, and I think we're seeing some examples of this already, is that everybody, we're talking about enterprise customers here, have visibly piled into AI, because they think that's what their investors want to see. And sometimes already it isn't delivering sufficient returns to justify the continuing spending. In layman's language, it ain't working properly. So one of the recent stories I love is Ford in the States recalling workers because AI wasn't delivering as expected on quality.

22:10Now, I know a Ford car on quality are not two things you would probably put together as an idea. But apparently they care a lot about quality and they've taken out all these massively experienced guys from their production lines looking after quality, replaced them with AI. Guess what? The AI wasn't as concerned about quality as the people. And there's another one, I think, which is very financial and it would be the credit markets tightening, which would mean that it'd be more difficult to finance large infrastructure. Can't see any of that yet. There you have it, right? That is the best we can do in terms of are we having a bubble.

22:48You've got hyperscalers cutting AI, the cloud AI utilization dropping, NVIDIA, bless them, I don't want to see Jensen recording a bad quarter, I don't want to see that. And then enterprise customers not buying AI, there are examples of that, and the credit market's tightening. Yeah. So what's our key takeaway, you think, from our discussion on bubbles? The first thing is to understand that there's a key distinction between an AI technology bubble and an AI equity valuation bubble. So it's not about AI not being introduced to the economy and doing stuff. It is about what the markets think it's actually worth.

23:27And we know at this stage in the cycle, if you look at things in retrospect, you will think, wow, why did we overvalue this at the time? because the technology itself, I think, will continue to see rapid adoption for years to come because it's actually based upon generations of innovation in IT hardware, software, and usage. The infrastructure is there to absorb it. The reason to apply it is there in terms of efficiency, if for nothing else. But the problem is, in all of this, the share prices of some of the companies that are massively valued at the moment can experience multiple compressions of valuation, i.e.

24:10significant drops over time, if it's quite clear that the expectations of the financial markets get ahead of actual realized profits from application.

24:29building a technology company is brutal we know that it's years of early mornings late nights working weekends long hours lost sleep and neglected loved ones you keep going because of a relentless belief in what you're creating at hambleton partners our senior advisors have walk that path, experiencing at first hand the challenges that technology leaders face every day. We know selling your tech business isn't just a transaction, it's one of the most important life decisions you'll ever make. That's why we take a different approach to technology M &A. We combine deep transaction expertise with hard-won industry savvy, helping founders navigate every stage of the transaction process with empathy, insight and the confidence born of deep experience.

25:17When the time is right to explore what's next, work with advisors who understand every step of the journey. Hambleton Partners.

25:30The problem that any great leader faces, and anybody who's been in a lead position has to recognise this in themselves, is that the strategies that made you exceptional, certainly at the beginning of your tenure become a form of institutional dogma. Of course they do because you're the leader of the organization. The problem is the world is going to change around you and faster and faster, as we know. And what happens is you become less willing or even less available, able to, to adapt, you know, become risk averse. And ultimately by the time you leave, because most people leave later than they should have done many of the difficult decisions that you should have made have been postponed because you're still surfing the same wave right and what happens successor comes in and what they inherit is is not the great conditions that produce the success or the external conditions and the conditions that you made as a leader but what can be regarded as just the accumulated consequences of years of deferred change, right?

26:40So what happens is the successor comes in and is then judged on the problems that they didn't create. And at the same time, nostalgia grows amongst the viewers of this for the leader whose later decisions contributed to the very problems that they're blaming the successor for. It's the organizational problem of what they often call in political circles, the incumbency problem. Now, from our point of view, this happening is ruinous for continuous category domination. Successful leaders naturally gain confidence in the formula that brought them success. You just don't notice that everything is changing.

27:24But this isn't just the fault of the individual. It's the nature of organizations that those individuals become surrounded by people who are reluctant to challenge them. Now that can be a career strategy from those individuals, or it can be a deliberate choice amongst the leader that they insulate themselves from what they see of as any threat. And if they do that, they can optimise all their strategies around protecting their past achievements, rather than taking the risk of preparing for what is inevitably going to be future disruption because they'll be long gone by the time the consequences that become apparent and they'll just defer painful restructuring because it would tarnish they think an otherwise successful legacy oh nothing went wrong nothing was difficult when x was in charge wasn't what weren't they great all they're doing is deferring um and this is often fed badly by a particular worldview which was good at the beginning of their tenure but that will will undoubtedly not fit a changing reality so you know the result is the organization actually appears to be healthy until external conditions shift um and and then when they do years and years and years of accumulated issues which have been brushed under the carpet emerge like a tsunami and, you know, threatened to just roll straight over any successor, right?

28:59So you're there, you've accepted this big job you've probably been working your whole life for, you know, you're succeeding the great god of any one particular company. You've got a big dilemma on your hands, right? So if you make radical changes, which are the things which you probably need to do, you get the blame for disrupting the company. if you preserve the continuity all you're doing is inheriting inheriting and exacerbating some some declining performance in that company either way you know comparisons with your predecessor are going to be unfavorable you're basically screwed let's um rattle through a couple of tech examples.

29:43So let's go with the guys whose chips are in pretty much everything, and I don't mean NVIDIA, talking about old school Intel. Obvious to anybody who's been around the industry for a while is that Intel is one and still is one of the giants of the industry, and they dominated semiconductors for decades. And naturally, and sort of like the auto industry, they prospered, in inverted commas, and the leaders who quite understandably focused on protecting its integrated manufacturing model, even though it was quite clear that other companies were coming up, you know, Arnding, one of the greatest example of Fabulous Simi, who were being very, very successful.

30:29And what happened over the time that this model was being pursued is that manufacturing leadership slipped because of the of you know leadership capex changes in fab and so on their competitors embraced outsource fabrication and ai accelerated demand for specialized chips so nvidia almost came out of nowhere from inhabiting a tiny little niche for graphics into becoming the major bellwether of the industry. Let's talk about their little journey then. So the big white hope, so to speak, was Pat Gelsinger. But he, of course, has now been replaced by Lip Bhutan. And he took over in mid-March, I think, 2025.

31:20Initially, so far so good, right? The press sentiment's cautiously positive about the appointment, mixed on the near-time upside. so your boiled frog analogy um seems to apply here because pat didn't make it second time around the new the new sheriff's in town um bank of america think it's good the appointment is positive lit boo has a solid track record this is a clever decision yeah but in point by by appointing somebody who is an absolute outsider and you know american companies in tech in particular do have this habit of appointing insiders until the company is in such a bad shape that an outsider has to come in.