The AI Bubble Debate: What’s Really Going On

0:28Is AI in a bubble? But a lot of the indicators I'm seeing are pointing in that direction. But the financial cycle and the technological cycle are not the same thing. And they move at different speeds because they're driven by different forces. And this distinction matters because AI is a general purpose technology like the internet, like electricity. And understanding this nuance is important because it tells us a lot about how this technology is going to behave. These technologies do reshape societies and economies, but slowly, steadily, and then all at once. Historically, these technologies have taken over a decade to make their full impact.

1:09But even more importantly, the technology we have right now is already disruptive enough to change society. And this time, it directly and immediately disrupts the individual, not just the enterprise. It changes the skills we have to bring to the table because the systems we're working with and working alongside are learning from us in real time. My concern is that people are going to over-index on the bubble headlines and mistake a financial correction for a technological one and assume that means they can tune out. So yes, a bubble may burst, and I'm actually starting to expect that. But that's the financial cycle.

1:50The technological cycle is going to keep moving, and we need to be prepared for that. I'm Sinead Bovell, and this is I've Got Questions. Before we talk about whether AI is in a bubble, we need to get clear on what a bubble actually is, particularly when it comes to technology. There is no single agreed upon definition from economists, but a bubble happens when the financial story around a technology starts to move much faster than the real world progress of that technology. So the expectations, the investments, the valuations, they all start to get out ahead of what the technology can deliver in the present.

2:28And then the smallest change in belief or in expectations can cause a massive market correction. So is artificial intelligence in a bubble? And the indicators that I've been looking at are definitely pointing in that direction. So first you have a lot of the money is running far out ahead of reality. So the investment gap. In 2025 alone, for instance, U.S. companies are on track to commit about$400 billion to AI infrastructure, while the revenue from AI products is around$60 billion. And less than 10 % of companies have meaningfully adopted the technology. Now, that data point alone, a lot of people are pointing to it, 400 billion versus 60 billion, to me isn't sufficient to say that we're in a bubble.

3:10Because we do have to expect the infrastructure for a technology that reshapes society and lays the foundations for a new economy to be expensive in the beginning and for revenue to take time to catch up. So that data point alone, my alarm bells wouldn't go off. It's the part two of the investment in infrastructure that I'm really starting to pay attention to. Some of the biggest players are planning trillions in future infrastructure investment, which would mean they would need to be generating hundreds of billions of dollars in revenue in the next couple years. If you look at a company like OpenAI, right now it's making about$20 billion in revenue.

3:46So by 2030, it's generating hundreds of billions. Not impossible, but I think it's really tricky to justify that level of investment and that level of revenue in such a short time span. And it isn't just them making these big infrastructure bets. A lot of the tech companies are. So for me, this is where I'm starting to see a big investment gap. The classic bubble pattern number one, building far more than the world is ready to use, not just today, but the planned spending curve getting even steeper than the current one. And then the way that this is all getting financed is starting to become very fragile.

4:20And we're starting to see the strange kind of circular investing happening where the infrastructure companies are now investing in the AI companies. And then the AI companies are in turn investing in the infrastructure companies. And this tight loop can inflate valuations because everyone's success depends on everybody else's survival. And then when confidence starts to wobble in one part of the loop, it ripples through the whole stack. And on top of that, the boom is no longer being funded by traditional corporate profits. The rest of the 1.5 trillion-ish is going to start to come from private credit, asset-backed loans, these opaque debt structures that can disappear very quickly if sentiment shifts.

4:58So this fragile money pattern is something I'm also tuning into. And then finally, it's this expectation gap, right? Bubbles start to form when people's expectations about what the technology is going to deliver rise faster than the technology can actually deliver right now. And when we spoke to Ajay Agarwal, one of the world's leading AI economists, he puts it perfectly. A bubble doesn't burst because the technology fails. It bursts when expectations about the future earnings of the companies tied to that technology start to change. Right now, the Mag7, so Apple, Amazon, Alphabet, Meta, Tesla, Microsoft, and NVIDIA, the leading AI companies, make up about 30 % or more than 30 % of the entire U.S.

5:37stock market. And their valuations are mostly tied to the promise of AI, the storyline of AI. And that's that psychological gap that we can start to see before bubbles. And so if a bubble is going to pop, where might we see that happening? Ajay told us, likely around the data centers. And here's why. The data centers are the biggest, longest-term bets in the AI ecosystem. They require massive upfront spending, the multi-year power contracts, multi-year hardware contracts, and hardware that depreciates really fast. So if expectations around AI slow down, people get spooked, even temporarily, these enormous investments suddenly don't make any sense in the near term.

6:19And then this creates stranded assets. We could also see a bubble burst because of a breakthrough in artificial intelligence. So there's a lot of talk about how this current architecture, the current way we are building AI systems, large language models, might be reaching their limits. meaning we can't get much more advancement from this version of the technology. If there is a breakthrough in how AI is getting built and designed, and there likely will be, right? The story of AI is one of continuous change and evolution and improvement in how this technology gets built and designed. But if there is a breakthrough, it could reallocate capital away from the companies that are currently heavily invested in and towards the company or towards the new infrastructure or architecture of where AI could be going in the future.

7:08And we saw this with DeepSeek. So remember when the Chinese AI company that launched DeepSeek came out on the scene in January, and it demonstrated that an AI model could be built much cheaper and much more efficiently. This wiped out trillions of dollars off of U.S. stock markets and off of the Mag7 in particular. It wasn't a sustained wipeout. But if there is another breakthrough in the architecture and it comes from somewhere else and it's not going to be a breakthrough that's going to be easily disseminated throughout the rest of the market, we again could see investors get spooked. And this is an area I'm really paying attention to because there are academic labs and small startups all around the world working on new architectures that could replace or build on top of large language models.

7:50And I also think a lot of the tech companies know the current infrastructure, the current large language model architecture isn't going to be sufficient to take us to the promise of AI. And they are working on what could come next, but nobody is talking or sharing papers anymore because the industry is getting so competitive. So these indicators are pointing towards a potential bubble bursting, and I am starting to expect that that's what's going to happen. And I chatted about this with Ajay, right? We can expect some ebbs and flows, some ups and downs, but the long-term trajectory of the technology is still going to move forward.

8:25So the question becomes, why are AI companies throwing all of this money at AI to this extent, if it's getting quite obvious that we're going to see some form of some crashes likely in the near future. And my guess, and this is just a guess, is that they know historically every major technology, especially these general purpose ones, go through these overinvestment phases. They know markets overreact at some point. And they know that even if the financial cycle dips, the technology cycle will keep moving. So from their calculus, and this is me just making a guess, it's better to slightly overinvest and survive a correction than to underinvest and miss the internet 1995 to 2010 moment of the AI age.

9:09In other words, they want to be the future Amazons and the future Googles of AI, even if it means living through a financial crash on the way there. And we're starting to hear more talk from them of the market being overheated and a potential dip coming along the way. And here's what's important to understand about foundational general purpose technologies like AI. So first, even when the technology is real, and even if it's a technology that transforms everything, the financial story usually always runs out ahead of reality, especially in the early years. But part two, why does it take these technologies so long to actually generate a return and to transform the world as we keep hearing about the promise.

9:52And the reason is it takes time for the ecosystem around the technology to evolve, for the co-invention to happen that's going to make these technologies actually make sense in the new economy. And it takes time for businesses and institutions to redesign their business models around the new foundational technology. When electricity first came on the scene, companies did not want to rip up their factory floor to invest in it. They only wanted to swap where the steam engine was for electricity. That didn't really make any sense. It led to almost hardly any noticeable productivity gained. It wasn't until over 20 years later, after its invention, that a new crop of companies started to say, huh, maybe we can actually apply power throughout the whole plant, not just in the one area of the wall where the steam engine was.

10:43And they redesigned factories from the ground up. And that's when the assembly line was born. And that's when we saw an entirely new way to build and manufacture. And then we saw electricity finally take off. And the same thing happened with the internet. It took time for the ecosystem that made the internet make sense, the co-invention, to form. So in 1998, there was a bunch of companies trying to build on the internet. A lot of them were overvalued, didn't have business models that made sense. The technology wasn't fully ready. But also a company like Amazon was there. Once the expectations and the beliefs started to change about where the technology really was and its true potential and how long it would probably take before it did anything, the market crashed.

11:27We saw the dot-com boom. It wiped out 50 % of the companies. Amazon lost 90 % of its value. But over time, the co-invention, social media, smartphones, the advertising model, supply chains getting redesigned around e-commerce. Once all that co-invention started to happen, the intranet started to make a lot more sense. So where I think we are in this moment, it is 1998 and people are starting to step on the intranet. Companies know that the technology is important. A lot of them that are building on it are probably not going to make it. But eventually the technology is going to go on to reshape societies, economies and business models around it.

12:07And the companies that don't see it, the companies that write off this moment as nothing but financial hype won't make it. And the ones that understand the trajectory of general purpose technologies are currently taking the time, because it takes a lot of time, to redesign their business models for the future metrics that will matter in an AI-first world and the future workflows that will matter in an AI-first world. So what does all of this mean for us? And why does a financial crash still matter for all of us? Because the companies making the massive bets on AI and the Mag7 alone make up almost 40 % of the entire U.S.

12:48stock market. That means anyone with an index fund or a retirement account or a pension fund is indirectly invested in this AI story, whether they realize it or not. So if expectations shift around AI, and I'm saying we should probably expect that to happen, and these companies' valuations reset, even temporarily, that could show up in the value of people's savings or their pensions or their retirement plans. Maybe you're young and you have decades before you need that money, so that's not a huge problem. You're playing the long-term game with investing. But if you're retiring in five years, the timing matters.

13:29And to be clear, this is definitely not financial advice. But moments like this are good reminders that we have to look under the hood of our own financial life. And a lot of people may not realize how much of their retirement savings or their index funds are tied to the big tech companies. And then the other side of this, even if all AI companies stopped innovating tomorrow, that we've hit peak AI, markets wouldn't like that. But even if we did, the technology we have today is disruptive enough that we still need to learn how to work with it. If there are two people applying for a role and one person understands how to work with AI and the other doesn't, it's pretty easy to see who's more likely to get that job.

14:11And the companies that understand the trajectory of AI are, again, hiring for that future AI economy. AI can do a lot of analytics. it can do a lot of research, it can do a lot of summarizing, it's doing a lot of coding. So even if all progress stops, it's changing who gets hired and it's changing what we need to bring to the table. So we aren't learning AI to bet on the hype. We are preparing for the reality that is already here. So is AI in a bubble? Quite possibly, at least parts of it. And could we see a correction? Absolutely. And I am starting to expect that. But does this tell us about the long-term promise of the technology?

14:52Not really. Every major technology, every general purpose technology, had a bubble that burst before the real transformation kicked in. The financial cycle and the technological cycle do not move at the same speed. And this is precisely the reason we need to be understanding this technology more and investing and learning about it more, not less. like the future of work this is a very important topic and we're going to continue to cover it we're going to continue to share where we think we are in this moment where things are going and most importantly what this means for us so thanks so much for tuning in to this episode and I look forward to seeing you at the next one