Welcome back to The Institutional Beat podcast, where we cover interesting and relevant topics for institutional investors. My name is Julie Ducharme. I'm a portfolio manager on the PH&N Institutional team, and I'll be your host for this first podcast of 2026. Now, today, we're diving into one of the most hotly debated topics in the markets right now. You all know it: artificial intelligence.

And the question is, is the current AI boom a revolutionary paradigm shift that's going to transform our economy? And I think we'll all agree on some parts of that. Or are we witnessing another speculative bubble reminiscent of the dotcom era? And perhaps as our guest suggests, it might be both.

And joining us today is my colleague Guido Giammattei, portfolio manager on the RBC Emerging Markets Equity team based in London. And he's recent author of a paper titled “Artificial Intelligence Paradigm Shift or Bubble. Maybe It's Both”. Guido, welcome to the show.

Thank you. Julie, great to be here.

It's great to have you, Guido. Now, your paper tackles a fundamental question about AI that I think many investors are wrestling with, and certainly I'm curious about. You draw some fascinating parallels to the dotcom boom of the late 90s. Now, can you give us your take on today's AI investment landscape as we sit here discussing this in January 2026?

Absolutely. There are moments in markets when you know you're looking at something truly extraordinary, because the only relevant historical comparison takes you back over 200 years ago. When we look at the current scale of investments in AI today. The scale is so unprecedented that that when measured, for example, as a percentage of U.S. GDP, it has already surpassed what we saw during the dotcom era. In fact, the only historical parallel that I could find requires going all the way back to 1820, when the U.S. and the UK were building railroads.

Well, so we're talking about investment levels that we haven't seen for nearly two centuries. Which is kind of mind boggling when you think about it. But then if I stop and look at where we're sitting today, how little the common company or individual is using AI on a regular basis, it really feels like we're in the very early innings. So in my head, this has much further to run. Would you agree with that? What are your thoughts?

Well, the excitement, at least on the technology, certainly appears justified on quite a few fronts. I can think at least four strong supporting factors. For a start, there is no sign whatsoever that capital expenditures in AI infrastructure are slowing down. If anything, I think that they are accelerating. The leading companies – take your, Microsoft, Google, Amazon, Meta – the U.S. hyperscalers, but also the China-based ones like Alibaba, Tencent, ByteDance, they maintain very strong balance sheets. I sense that for these companies, the perceived cost of losing to competition is far greater than the risk of low or no returns at all on these investments.

Secondly, demand continues to grow exponentially. Take the token, which I think everybody has heard of by now. In AI, it represents the basic unit of text or image that AI models use to read and respond to queries. We use words to communicate to others. AI models use tokens. Now we can use the growth in tokens, in my view, as a measure of AI demand. And because I could not find global data on these, I asked ChatGPT. So I asked it: what is the rate of growth it has seen its own use, measuring tokens? And it responded to me: about 300% in 2025. And then it also broke it down by how – more and more users, more frequency of use, longer conversations, and also follow ups. The rise in token usage aligns with what economists call the Jevons paradox. As the cost per token falls, AI becomes cheaper to use, which actually stimulates even greater demand.

Thirdly, and this is the kicker, enterprise AI adoption has barely started.

Right.

Corporations are only spending a few percentage points of the global AI CapEx in 2025 on AI tools. In my view, we're really just scratching the surface of what enterprise demand could look like. My sense is that this number will definitely grow a lot. I mean, I don't want to say exponentially, but possibly so over the next decade. And lastly, technological progress on both the hardware and software fronts also continues at a remarkable pace. The computational requirements for new AI models continue to increase at much more than twice the rate of Moore's Law. In simple words, what this means is that both models and hardware continue to get better, and justify further investments.

Yeah. And it's you know it's easy to get excited about all this growth of potential – it’s really, really impressive. And then I go back to your paper and its title, and you're alluding to: yes, there is this paradigm shift. You've made some good points here, but there's also the bubble characteristics. And I'd like to dig into that part of it. What's giving you pause?

Yeah. There are three key areas of concern that I think as investors we all need to carefully consider. The first one is rising leverage across the whole AI ecosystem. Second, the uncertainty and the challenges related to the monetization of AI investments. And thirdly, the growing infrastructure and energy constraints that could become real bottlenecks.

Yeah. So I'd love to unpack these one by one. So, can we start maybe with the leverage concerns?

Yeah, sure. AI investments are consuming a growing share of cloud service providers’ free cash flows. So I mentioned that they have – both the U.S. and the Chinese-based – very strong balance sheets and very strong free cash flow generation. But now we're talking of investments that are taking up to 80, 90% of their free cash flow generation.

But what is particularly concerning is the flood of new entrants over the last 12 to 18 months, who are relying exclusively on leverage to fund that operation. These new entrants have no profits, no free cash flow generation, and just relying on leverage. In my view, one area that needs careful consideration over the next few years is private credit exposure to AI.

About two-thirds of venture capital funds in 2025 has been allocated to AI infrastructure, and about one-third of U.S. private credit exposure is now tied to AI. Most of these funds have been lent, as I was mentioning, to new entrants in AI that have weak balance sheets and no profits. It would not be too much of a stretch to think that many of these new entrants will not make it.

Outside of private credit, there are also many other instances of let's call them, creative financing. GPUs are increasingly being used as a collateral to raise additional capital, and we're seeing more leaseback arrangements or even sales on consignment with suppliers. Last but not least, there’s the well-documented circle of financing happening through M&A between suppliers and customers, which is also a form of, in a way, financing your customers.

Yeah. So it sounds a little bit reminiscent of some of that financing innovation, if I can call it that, that we saw in the late 90s – a little bit concerning. And for our listeners out there, if you have a private credit portfolio and you see some AI themes, maybe look under the hood and look at the balance sheet health of some of those holdings, because you may be sitting on a pile of nothing when it comes to fruition.

But we won't delve too far into that rabbit hole. We'll leave that for maybe another topic. In a future podcast. But I'd like to come back to the monetization one, Guido. And this is a head scratcher for me because you think that with all this demand, companies should start seeing returns. But it hasn't happened yet.

So what are your thoughts on when that's going to be? And is it soon enough for these companies to stay afloat? I'm not asking about the mega ones. We know they certainly have the revenue streams for that. But for everybody else who's been rushing into this and maybe doesn't have a healthy balance sheet, I wish your thoughts on that.

Yeah. So what is important to understand here is the long lag between when investments are made and when the monetization happens. So investments in GPUs and data centers typically begin generating revenues two or three years after the initial investments. For profits, it can take five or six years before you actually are in the black.

We are probably – I mean, I would argue we are probably like in year three. So the path to profits, is still, you know, relatively distant. Because of this long lag, there are risks to monetization, and I think even more so to profitability. There are quite a few risks, actually. But if you ask me, in my view, the most important one is competition.

So I mentioned that there has been a flood of new entrants. We’re not talking only about the usual suspects, those stronger hyperscalers, that we mentioned earlier. The success of a technology – and perhaps this is the analogy with the dotcom era – the success of a technology does not guarantee profitability for all the participants.

And what we've seen in the last 12 to 18 months is, as I was mentioning, a growing number of debt-funded entrants that could undermine capital returns for the entire industry. In a similar fashion of what happened during the dotcom bubble. Internet adoption back in 2000 – 25 years ago – internet adoption and data traffic boomed in the years from 2000 to 2010.

But the price of data collapsed under massive oversupply and brutal competition. There are also other risks I can possibly mention: slower, incremental advance in AI models, misjudgment in depreciation estimates. But to me, competition is probably the biggest risk to future return on capital.

Right, so this notion that tomorrow's winners might not even be around today strikes me as a reality that we're facing right now. And it's hard to say who is going to benefit from all this AI investment.

But, I'd like to unpack another area that is we're hearing a lot about, and that is around infrastructure gridlock that's emerging, right? And I want to steer us back to this challenge that you also address in your paper – which you can find the link to in the description to this podcast. You talk about the race to build data centres and their impact on the energy grid, and that's a real physical constraint on growth. And to the point where we're hearing about it in mainstream media on a regular basis. Do you have some stats that you can share about AI's need for energy and what we're looking at?

Yeah, absolutely. The rapid growth – as you mentioned, the rapid growth in the data centre construction – is straining the U.S. energy grid. Private construction spending on data centres has nearly tripled from US$14 billion annually in 2022. So only in 2022, to over US$40 billion. And this was by end of last year, according to a recent the Schneider Electric survey. Schneider does a lot of the electric parts that are basically used in these data centres. Ninety-two percent of respondents to the survey see grid constraints as the key challenge to growth, with longer wait times for getting their electricity power supply sorted cited as a one of the main concerns and as a factor that could slow expansion.

Now, if data centers CapEx continue to grow at this pace, there is an estimated potential shortfall of about 45GW of power in the U.S. between now, or between the end of 2025 and end of 2028 – in three years. Now, I know that when I first read that “45 gigwatt,” did not tell me much. I actually struggled to put that number into context.

And those of us who watched “Back to the Future” remember the lightning bolt with 1.1 gigawatts. But what does 45 gigawatts mean for the U.S. economy? How much how much does that represent?

I Googled around a little bit, and that 45 gigawatts of electricity is an immense amount. It's basically equal to powering a whole country. And I'm Italian, so, that's what I did first. I went to try to measure it with the country, and I went to look at Italy. It happened to be exactly the same amount of power that you need to supply a whole country like Italy. So it's a huge deficit.

So in the next three years, the U.S. has to add enough power to power a country like Italy onto its existing grid, just to keep up with demand.

And that excludes, you know, the normal growth in demand. That is only to supply the growth in data centres.

Yeah. Okay, so that's over and above normal growth. That's a lot of inflation on the energy grid.

Yeah. If you can, take a long-term contract now. Or build your own off-the-grid solar plant in the garden.

Haha, I think we might need a few more than just a few. Okay. So now you're starting to put some of these caveats on the growth, the exuberance. You know, everybody's very excited about the potential that AI has for our economy. As investors, where we've benefitted a lot from getting behind some of these investments.

But in your paper – and this was what surprised me and why I wanted to talk to you about it – is that as an equity investor, you're actually dozing your enthusiasm. You're not quite as exuberant as you have been in the last five to seven years when we talked to you about AI, chips, and the tech sector. So I want to switch now to maybe your tempered enthusiasm and what you would share with our institutional clients about how to take advantage of the next AI chapter in the portfolio, considering we've come off a great run.

I think the key lesson from the dotcom era is instructive here. The investment thesis wasn't wrong. The internet has indeed transformed the world and created multi-trillion-dollar companies. But the excitement around transformative technology can result in too much capital chasing too few opportunities. And it's in a way, what we're seeing now. The market is very, very narrow. Even within the technology sector, only a few companies are really performing. And the valuation of these companies – or in some cases, perhaps not as much in emerging markets yet, but we're starting to see some stretch valuations. The valuation of these companies have been pushed perhaps beyond the reasonable expectations. From a portfolio construction perspective, we're finding still, you know, good opportunities in emerging markets. I think what is also important to consider is the relative valuation argument rather than the absolute valuation argument. So from an absolute standpoint, valuations in EM (Emerging Markets) are not extreme, but they're definitely at the top of the range that we've seen in these stocks for the last 20-25 years. So the valuation is not nearly as compelling – alluding to what you were referring earlier, like the case was even just two or three years ago, even a year and a half ago. However, when you compare them to their developed markets peers, they trade at meaningful discount on a price-to-book basis – price-to-book value basis. They trade at 65-70% discount. This substantial discount, combined with the fact EM tech is less widely owned than U.S tech, could provide some resilience in the event of a market correction.

So just to make sure it's clear, are you suggesting that investors maybe should be rotating some of that developed market or U.S.-dominant AI exposure towards exposures in emerging markets?

I think that this is actually what has been happening over the last six months. So when, there is always the chart that we look at that compares the price-to-book value of developed markets tech and emerging markets tech. And that discount has been consistently narrowing over the last six months. And I feel that this is a combination of exactly what you were referring to.

People are thinking: if there is indeed a bubble and it’s going to burst, I have less downside in tech because valuations are not as extreme. But if this continues, EM tech valuation will catch up because earnings growth is very similar – if not actually better. And the profitability is also very similar. So the reason that, you know, perhaps other than your perception about risk and cost of equity in EM versus developed markets, there isn't perhaps as much of an argument for such a large discount rate.

Right? So from a valuation perspective, we could say that there's still some room to run, or a little more room to run, if you're an emerging markets player versus developed markets. So let's think about wrapping up now, Guido. If you had to leave our institutional investors, the listeners today, with one key takeaway about navigating the AI investment landscape, what would it be? Looking forward, other than moving into emerging markets.

Markets, and the technology industry even more, are fast-paced environments, and things are constantly changing. Even when you have, like we have, a very long-term investment horizon, accidents can happen. But it appears that, at least based on the guidance that we are seeing coming from most of the companies in the tech industry that we follow, 2026 could be another strong year for AI stocks.

However, in my view, tech is and will continue to be a cyclical sector. And the up and downs can be very material and very painful. So as I often say, for me it's important to have a long-term view, decide what weight you want to give to this team or to this industry, and be invested above all in companies that you believe in, and that you are prepared to buy more when weakness happens –because eventually it will happen.

And by the same token, you should also be prepared to reduce in time of exceptional strength. So, I have a long-term weight in mind. And when that weight changes – it deviates from your ideal weight because of exceptional strength – you reduce. And when it's falling below because of weakness because we’re in one of the downs, then you need to be prepared to back it up.

Right. And so there is still growth ahead. The waters may be choppy at times. You want to keep your eye on the long term, stay invested. And if you can't handle the chop, rely on professional to do that for you – who can navigate into the opportunities when they present themselves and trim when it's hard to do so.

So. Okay, there's some words that we've heard before in the past. I think we're going to keep living by those going forward. Thank you for your advice and your recommendations, Guido. It was a pleasure to have you on the podcast today.

No, my pleasure, Julie. Thank you very much for having me.

Great. And that wraps up today's episode of The Institutional Beat. For our listeners, you can find a link to Guido's full paper in the episode description and on our website. And if you enjoyed this episode, don't forget to subscribe to the podcast so you can catch future episodes where we continue to explore the topics that matter most to institutional investors.

Thanks for listening.

This content is provided for general information only and does not constitute financial, tax, legal or accounting advice and should not be relied upon in that regard, neither an institutional nor any of its affiliates accepts any liability for loss or damage arising from the use of the information contained in this podcast. Securities mentioned are for information purposes only and do not constitute investment advice, a recommendation or an offer solicitation.