Transcript
Speaker: Hello from HSBC Hong Kong and welcome to Under the Banyan Tree. I'm your host Harold Winderlinde, head of Asian Equity Strategy. And if you're a regular listener to our sister podcast, The Macro Brief, you'll be familiar with my guest this week.
Speaker: Mark McDonald is our head of AI and data science here at HSBC. He recently moved to Hong Kong and what better time to get him under the banyan tree. We're looking at the bull and bear cases for the story that continues to dominate financial markets here in Asia and the rest of the world.
Speaker: Plus, what do we make of the call for more regulation from AI industry leaders themselves? Plenty to talk about, so let's get to it. From HSBC Global Investment Research, you're listening to Under the Banyan Tree.
Speaker: Mark, welcome to the podcast. Thanks for having me on. So you are our in-house AI and data scientist, or the head of the team, right? So I presume you are very positive on all the things that are developing in the artificial intelligence. Yes, absolutely. I think AI is the most exciting thing to happen throughout my whole career.
Speaker: Do I take it that perhaps you're a bit more bearish on AI than I am? Well, there's a couple of things I'm a little bit worried about, but maybe we can talk a bit about what makes you so positive. You see, to be honest, just before we recorded this, I had to suddenly make a presentation for the in in the afternoon. They asked, can you put a presentation together on a topic that I had to read up about? So early this morning, came in a bit early, I used AI to read up about it, and then I asked it to make a presentation. That it would normally take me two days. So there are clearly things that are productivity enhancing, and that's what you're looking at, I presume, right?
Speaker: Yes. And I think with any new technology, it often takes people a while to really understand how to reorganize business processes and reorganize their life in order to best make use of this. And so I think even if the AI models and the AI tools that we have today, even if they never get any better, like this is as good as it gets. Yeah.
Speaker: um I think we've got five to ten years of productivity improvements as companies work out. Already with what we have at the moment. Yes. It's very unevenly distributed how people are using this and the degree to which the implementations are being material um or not for the companies that are doing them. And you can see it in our industry. Like when we go and talk to clients.
Speaker: At one end of the spectrum, you've got the systematic community who their life has been completely changed by AI. Systematic community is sort of the Quants guys. Yes. So they run like models and say, buy me any stock that is down 90% from its peak and trades on these and these multiples. And the program simply buys it for them in the market, right? Exactly. It's um it's very little human intervention and it's yeah it's rules-based. And so for those guys, their their investment process has always been like a large part of it has been focused on process and procedure.
Speaker: And, you know, the sort of critical bits where humans had to do a lot of work in that world were in coming up with new signals that you thought might be additive to your portfolio. and then doing the work to test whether they actually helped and doing the back tests and doing the simulations. That first part still needs humans. You still need humans. That's where your market intuition can add value to the process. But that second step of screening through new signals and seeing if they're additive to the pot earlier, AI can do that so much more quickly. And so when I speak to systematic clients, their throughput has gone up massively and they've often reorganized their entire process so that you have human decision making at the beginning and the end but the middle part is mostly automated and is doing it a much higher volume. There's a however coming up now. There is indeed however and that's really people like you and I. The more discretionary the investment process and cell-side research is very discretionary. So we don't run computer programs we just think and write and these sort of things right? Yeah. Discuss. Exactly and so that part of the investment community when I speak to them.
Speaker: There are lots of people who have examples like you had, where there is a task that used to take a bit of time, and now takes significantly less time. But whenever I speak to this community, that nothing has fundamentally changed about the process. They haven't reorganized their life around this new tooling. It hasn't upended everything. And the most common question I get from this community is, what's everybody else doing? Because I think people are paranoid that they're missing a trick and all their competitors have suddenly found a way of like completely transforming their life. exactly yeah And so I think this this is a you know a good example
Speaker: of how you know the same technology can have different sort of implications for a business. But the adoption rate of this yeah is different for different types of companies, you could you could say. That is true. I mean, the um the AI governance has been... something that large financial institutions have been very focused on. It's ah it's a critical part of doing this well because you know if you if you don't have trust as a financial institution, then you you don't really have anything. And so I think people would rather not risk that and so move carefully.
Speaker: yep So why are you more negative than me? There are two things I'm a little bit worried about. What we've spoken of so far is the sort of changes in what we can call maybe the demand side. So how are people going to use this and things are gonna be faster. i actually, I went a holiday this summer and a large part of my trip was organized by ai So I'm not so worried about this, but there's also a supply side. Supply side means we need to build data centers, we need the energy, and this is having a big impact on markets. um
Speaker: We see the demand for funds is driving up bond yields globally. We see certain markets, Korea, for example, Taiwan, all the attentions on those markets and all the investments are going in there at the moment. so It is distorting markets, and I think it might also mean that capital is moving away from other sectors that might just need money now to recover, for example. So that's that's one thing. There's distortion going on.
Speaker: I think also on a bigger level. In the 1960s, my own country, I wasn't there yet, I wasn't born, not that old, ah discovered gas. the Netherlands. And when they discovered gas, they suddenly became more more wealthier. And you can think about some of the Asian countries now experiencing this as well, right? ah Korea seeing a massive improvement in, say, national wealth. Taiwan to a certain extent as well.
Speaker: the yeah The Dutch suddenly found a lot of gas, became more wealthier, they started to invest, people wanted to be there, the Dutch guild were appreciated. And two things happened. First of all, the exporters that were not in the gas business, they really started to struggle because the currency got richer. So actually, you really put a lot of pressure on another part of the economy. Secondly, because that wealth meant that people started to spend money, they build out a whole welfare system that was probably a little bit too much. And now we have to scale it down.
Speaker: So you get these sort of distortions in in nations as well, not just in American nations as well. And you can see this maybe in Korea. And I think the Korean policymakers are very much aware of this because they are already talking about trying to deal with this. But this is called in in economics the Dutch disease.
Speaker: There's a lot of money being made by the DREM industry in in Korea. These people pay bonuses. They go then then spend the money locally. There's already evidence that this this is taking place. And that might mean that within Korea that certain people say, I want to become ah a private chef for a guy who works for Hynix because he's paying me more money than... My business, what I'm doing now, which is ah kimchi business, and I export it all around the world, but I make more money now by just making fresh kimchi for for locals that are paying me now more to do so, right? So you you are, again, putting a lot of pressure on an export business that in itself was very unique and and very good. So it can create distortions. And so in that sense, maybe we're not too different. It's just that I look at it from another angle as well. Yeah, two different viewpoints on the same.
Speaker: Exactly, yes. um You looked once also at how companies, I want to go back to this adoption sort of thing. You rated companies once on a scale of one to five or so. And you can you explain a little bit what that was? Yes, of course. Because I i think your comments about you know the the impact of of AI on on markets...
Speaker: There are certain areas of markets where the um the AI trade is most clearly focused and particularly the AI infrastructure trade. And so for those areas, I think it's it's hard for any investor to really believe that they truly have a different and superior investment process for analyzing those companies because everybody is analyzing those the same data, everybody's analyzing the companies through the same AI lens. um One less exciting area of the AI trade, but one which has also been successful, is the AI implementation trade. um Here the idea being that you get sort of shifts between winners and losers so within a sector that's not a classic AI sector. If you have companies that are using AI well and are implementing it into their operations, these companies are likely to see lower costs or higher revenues and just generally improved margins. That's a bit what you said earlier on, right? The systematic company that use competitive programs adopting it very fast and then discretionary companies, as we called it, being a bit slower and thinking, hey, what are the other ones doing, right? Yes, exactly. And so you'll see this within a sector that's not classically exposed to AI. You should expect to see the shift between winners and and losers. And initially we were using, it's bit self-referential, using AI to read through discussions about AI to see are they really AI discussions.
Speaker: and But we were looking for companies talking about implementing AI on earnings calls and those companies were were outperforming. But of course, that's now everybody. Which company is not saying that they're using AI for something? And so really the importance now is to try and find companies that are using it in a way that is material.
Speaker: And so we now use AI to score these discussions on a level of materiality, where a sort of one out of five would be like pure buzzwords. Yeah, we're going to use AI. It's going to be great. Yeah. Something very, would get a score of one. That get score of And what would get a score of five? A score of five would be when a company says that they're doing something, they talk about it in enough detail that it's clear that they are actually doing something. we are saving $100 million because we've implemented this system.
Speaker: Something like that. Something concrete, right? Yes, it's it's it's concrete and quantified. So it's where they they say what they're doing, they say how they're doing it, and then they also give numerical KPIs. They say, we've saved this many hundred millions of dollars through this AI implementation. So something specific like that.
Speaker: And we're increasingly seeing companies falling into that bucket. um And those companies are obviously very exciting. I think what's probably interesting as well is those that are in the middle. So like a three out of five, you might have a company that they're clearly doing something. They might say, yeah, we've got 87 different AI projects. for doing this, we're doing that. But they don't give details of any of them.
Speaker: And so clearly, at the moment, nothing is moving the needle. But when you see that in you know the most recent reporting season, the next reporting season, we would hope start seeing... need to see some details quantified. So they go from a three towards a four towards a five. yeah But if these companies are saving money on it, do they actually can keep these savings? Who is actually really benefiting? Is that good for the companies or...? well I think this is where you still need a discretionary analysts to look on a case-by-case basis for each company. For you know for instance, you know our lives as research analysts, we we have basically three chunks to our lives. So there's the part where you're coming up with your investment thesis. yep There's the part where you write it up into a report that gets published. There's the part that the bank really wants us to do, which is going and talking to clients. yeah We were all hoping that with more AI tools we'd get more of that and less of the first two chunks. Now, the writing the research report, that is basically binary. It's either done well enough or not. And anything you can do to make that more efficient is a real productivity game for us. It's like me making my presentation now in two hours. Precisely. So if something used to take you three days, it now takes day and a half. that is a day and a half extra that you get to speak to clients about your report.
Speaker: If you used to spend two weeks coming up with your investment thesis, and the same amount of research can now be done in two days using some of these AI search tools and AI research tools, you don't just do two days of research because it's a competitive market and all our competitors have AI and all our clients have AI. And so if you were an analyst who were to do the same amount of research as before, but do it in this more half-hearted manner and just do it quickly, then nobody would want to read your research. And so the competition is actually eating the productivity gain there. It's the consumer that gets the benefit. I think for a lot of companies, if they don't have pricing power, it will be the benefit. Well, this is a major question actually, right? Because who is going to benefit from this and what is the monetization we very often talk about? Are the companies really making money out of their investments? And there's simply a risk is that it creates such an intense environment, a lot of competition that ultimately the consumer just gets free apps or very cheap apps and the companies don't really make that much money out of it despite the fact that they've invested billions of dollars in it. Yes. Right.
Speaker: And i think it you definitely get productivity gains amongst companies who are competing. And so you will get shifts again from winners to losers. But it won't solve the productivity problems that maybe our economics colleagues are worried about. You're not going to get an aggregate productivity boost. For instance, if... A while ago we had Alibaba release some research showing that some of their Gen AI experiments were leading to like measurable and economically significant productivity benefits. But these were often coming as a result of higher conversion rates.
Speaker: Not everybody can have the higher conversion rates. Presumably in the pre-Gen AI days, the people who didn't convert on Alibaba bought that product from somebody else. yeah You can't just get higher conversion rates across the whole market.
Speaker: And so I think, yes, Companies have an incentive to do this because they want to be the winners within their sector, but it's not like you get an aggregate economy level productivity boost from a lot of these AI implementations. now Fantastic. Well, I mean, this sets the stage a little bit for two more questions that I want to pose to you. One is over the weekend, we've had some tech companies coming out and said, hey, we need to slow this down. So I'd like to get your view on that. And I also like to discuss about the difference between the adoption in China versus ah the U.S. So let's do that after the break.
Speaker: Over the weekend, we had a couple of the leading tech companies come out and say, we need to slow this down. And some people have suggested that they got models, they know more than we do. So maybe they are very worried about this, or maybe more alternative business motives might be behind their statements. what or What is your view on this?
Speaker: Yes. ah So I think both can be true. For example, Anthropic for a long time have been positioning themselves as a leader in the AI safety space. They've been mourning about AI safety for a long time. They definitely have credibility in this area. um But there clearly is a conflict of interest here where if they...
Speaker: are able to stimulate increased regulations and make it much harder for open source models to compete and open weight models to compete. Just very quickly, yes open and closed models. So they are in the closed models, right? They are running closed models where it's their ecosystem and they control everything and they can see all the questions that you ask the model and it's running on their servers. Exactly. An open model is but basically a package that you can download and you can implement and change it, whatever you want, according to your particular needs, right? Precisely. And those have been recently much more common to come out of the Chinese AI labs than the US AI labs. And so you know we're starting to see, you know if you look at data for recent data from OpenRouter, then you know a large proportion of the tokens, the usage of AI models, um even in the West, is now going towards these open source models from Chinese AI labs. And that is clearly a business concern to companies that are running closed ecosystems. And so even if they are genuinely worried about ai safety, which is believable that they are, this isn't a sudden road to Damascus event. This is a you know this is a something they've been warning about for a while. They also have a business interest that is aligned with with stimulating concern from policymakers about AI safety. Well, this is important because just bringing it back into the Asian sort of context, the build out of the Americans is driving the demand, of course, for product out of Asia.
Speaker: So do you think is that something that I need to keep a very close eye on if they are going to slow down those investments or will they continue those investments and actually there's an incentive for them to speed it up? what How do I need to think a bit about this? So I think in in general with the AI monetization race, um I suspect there'll probably be sort of two types of two types of winners. There will be a winner or maybe two winners of the enterprise AI systems race. And so probably somebody will become like... the Microsoft Office of AI. And that would just be the default way in which we all use AI and it's all installed on our systems. It's just just like Microsoft Word. Everybody uses it. It's very difficult to go to another one. You can't write in it. You're not used to it. But also, communicating with other systems is is there going to be different. So everybody wants to use the same system, right? It's a bit like Adobe has in in PDF files. Exactly. And then once you've installed something like Microsoft Office on on your system as an enterprise, um and on i switch imagine how complicated it would be to move to an alternative Office system, even if it was dramatically cheaper or even free. yeah um the The cost of all the change of like moving removing it from all your systems and implementing the new system It's enormous. It wildly impractical and probably will end up in a situation where there's a similar piece of software that it does that for AI and everybody uses it. yeah But we don't know who that is yet. We don't know who that is, and that is actually very bullish for the AI infrastructure trade because whilst that prize is available, it's worth competing like astronomic amounts of money today to become that winner because that is going to be a very sticky, very high-margin business. At the other end of the spectrum, anything where you have companies who are writing code that periodically the code calls an AI API, that sounds fantastically commoditized because switching from an anthropic model to an open AI model or open source And that's everybody doing this right? I mean, people using four or five sometimes different different models to design the holiday or do interior design, whatever it is. And that actually is almost free.
Speaker: Yes, and so that business, I feel, will be very highly commoditized, at very low margin. That probably plays towards the strengths of these Chinese AI labs. They've been much more focused on efficiency.
Speaker: So it's very interesting. So we're still in the midst of this sort revolution. rollout of this of this new technology. All sorts of questions are being asked and sometimes the questions might have different alternative motives behind it but the big prize is who will become ultimately the AI sort of company that everybody will use and interact with And and and for us, from an Asian point of view, that probably means, yes, you're right, that the demand for to try to build out data centers and and run these models will continue to be fairly robust.
Speaker: But still, a water caution is that within markets we see, therefore, all kinds of things happening. Profit growth with these sort of companies, resources being reallocated to one sector, maybe taken away from another sector, bond yields moving higher, that's good for some companies and not good for other companies. so This is leading to distortions in economies and markets that we need to be a little bit thoughtful about this as well.
Speaker: Absolutely. Fantastic. It was great talking to you, Mark. Likewise. Well, that's all we've got time for on this episode for Under the Banyan Tree. We're available wherever you get your podcast, including YouTube.
Speaker: Please do listen, like, and subscribe. Under the Banyan Tree is an HSBC Global Investment Research podcast. I've been your host, Harold van der Linde, and our producer is Graeme McKay. Thanks and take care till next week.
