Summary

00:00 – Introduction: Meet Billy Leung from Global X
03:00 – The AI Opportunity: Why AI is becoming a major investment theme
07:00 – Hype vs Reality: Is AI investment justified or overheating?
12:00 – AI Adoption: How businesses are already using AI to drive productivity
18:00 – Building the Backbone: The infrastructure powering the AI boom
25:00 – Beyond Big Tech: The overlooked beneficiaries of AI
31:00 – Portfolio Construction: Diversifying AI exposure beyond mega-cap tech
39:00 – GXAI ETF: How Global X provides access to the AI value chain
45:00 – Looking Ahead: What investors should watch next in AI
40:00 – Conclusion: Breaking the "old world" habits of low interest rates and low volatility

Paul O'Connor:
Welcome all to another installment of the Netwealth Portfolio Construction Podcast series. I'm Paul O'Connor and my role is as head of strategy and development, investment choice for net wealth investments. On today's podcast, we have Billy Leung from Global X. Billy is a senior investment strategist and based in Sydney. Great to have you on the podcast today, Billy, on what will hopefully be an informative and educational discussion on artificial intelligence.

Billy Leung:
Thanks, Paul. Thanks for having me.

Paul O'Connor:
Global X ETFs is a subsidiary of Mirae Asset Financial Group and has a global ETF platform with over US 100 billion in assets under management. ETF Securities was acquired by Mirae in 2022 and rebranded to become Global X. Mirae, established in 1997, is a privately owned company based in Saul, Korea, with over US 628 billion in assets under management. Mirae is one of the largest standalone financial groups in Asia providing a broad range of services to clients worldwide, including asset management, wealth management, investment banking, and life insurance. Billy joined Global X in 2024 and is responsible for investment research and ETF analysis in the technology sector. Billy has over a decade of experience in financial services focusing on equities and technology previously working as an equity analyst at Optiva in Sydney and was the director of equity research for China internet at Hao Tong International in Hong Kong.

Billy has been a top ranked equity analyst for regional software and internet by Asiamoney. Billy holds a Bachelor of Commerce from the University of Melbourne and is a qualified CPA Australia. There are currently 29 Global X ETFs available on the Netwealth super investment menu and all Global X ETFs listed on Australian securities exchanges are available on the Netwealth IDPS investment menu.

Artificial intelligence has grown from the mid 20th century theoretical foundations into a transformative technology advancing through periods, I guess, of high optimism and AI winters. Early symbolic AI gave way to machine learning, which gained momentum in the 2010s mainly due to increased computing power and growing datasets. The development of deep learning and more recently transformer architectures propelled AI from simple pattern recognition to generative AI models capable of creating high quality images and texts. Today, AI is widely integrated into global industries for tasks such as medical diagnosis and automation with recent advancements focusing on agentic AI that plans and takes actions independently.
So AI is now transforming many companies and industries at a rate not seen since the industrial revolution. So we'll have a growing influence on markets and securities pricing. So I'm really looking forward to discussing this with Billy and understanding his views.

So maybe to commence, Billy, AI has been one of the most dominant drivers of markets over the past 18 months, but there's still debate around the structural impact on companies and markets and whether this is hype to a degree or something more structural. From your perspective, what evidence suggests this has moved into a real infrastructure and the investment cycle?

Billy Leung:
Great first question, Paul. And I think this is something that we get a lot whenever we discuss artificial intelligence and all its adjacent investment cases. But I think taking one step back, I think one of the data that I like to share is that when people say there's so much being invested in AI, there's so much overinvestment in AI is at hype. I think one of the first things that I usually share is that as a percentage of how much we are investing in AI right now annually, whether it be $800 billion or $1 trillion, as a percentage of total global GDP, this is roughly about 0.7% of what the global GDP is. And when to put this into context, when you compare this with, I guess, previous innovation breakthroughs, for example, the 3G network or the internet network, we were actually spending much more as a percentage of annual GDP at that time, probably over 1% or closer to 1.2%.

So I don't think we are overspending. I think all the hype and the concerns about overspending is not valified, but more importantly, if you look at the potential addressable market pool of artificial intelligence, which is estimated to be closer to $4 trillion over the next 10 years or in 2025, that's about 7% of our global GDP or the global economic value.
And again, this addressable market is actually much larger than what we've seen historically in the internet or in the 3G smartphone network or even during the railway breakthrough into 1800. So I would really refer to these numbers, putting things into context to really understand that we are not really hype and AI is really seeing a lot of different things benefiting from this investment cycle.

Paul O'Connor:
And I guess we're still trying to get our head around how AI will grow, how we will use the various applications. So I do understand your point there that we are in the earlier stages still of the CapEx investment into AI, given how transformative that it can be right across society. If we think about AI moving into real-world deployment, where are we actually seeing the greatest demand show up today across the economy?

Billy Leung:
Yeah. I think one thing that we've seen surprisingly, I guess rapidly is obviously in the enterprise productivity or the enterprise, I guess, adoption of the platforms. And what I mean by that is I'm sure that you yourself or many other firms that you know are already adopting AI, whether in a sense of using Copilot for their, I guess managing their data, managing their outlook, managing their inboxes, whether they're using ChatGPT or Gemini or even Claude to manage their infrastructure in terms of how the company HR or even accounting works and even to more specific, I guess, more relative business in terms of the financials. We've been seeing investment houses using AI, especially Claude or Gemini to actually improve their trading systems. So I would say that we are actually seeing a lot of adoption of AI to improve the efficiency or the productivity of a lot of these enterprises right now.

And to look even further is we're already seeing somewhat of an impact. And I guess a lot of the mismatch or misconception is that a lot of people don't see AI appearing on, for example, the profit and loss lines. They don't see it on the revenue line and they believe that this is not materializing. But on really good example is that if you look at a lot of the internet companies in the US that have applied artificial intelligence in the past six or seven years, they've actually seen their margins improve over the two, three years. And even when you look at the latest earnings of results in the US market for the quarter one results, we actually saw revenue increasing by an average of 9%, but we've actually seen earnings of US company rise to over 20% for the first time in a few years.
So we're seeing margin expansion, earnings acceleration and this is simply because we're seeing productivity improvement from artificial intelligence. So I would say that we are seeing a lot of adoption, especially into enterprises and especially in the recurring workloads.

Paul O'Connor:
That's very much common sense to me, the comments you made there, Billy, given that the impact AI can have on productivity. And I know Australia has an economy needs to really focus on uplifting productivity. So let's hope that AI is one of the key drivers to our own economy. A lot of the focus has been on end applications and models, but less so on what enables them. Where do you see the biggest bottlenecks that could shape how quickly AI scales further from here?

Billy Leung:
Yeah, this is an interesting question that myself and my team have been doing a lot of work on recently. The honest answer is actually we are at an inflection point and I hate using this inflection point word, but the thing is what I see is that we are going to see an explosion of usage and AI in the coming 12 months. And the reason for that is, A, we're actually seeing technology rise rapidly, but more importantly, the cost of running AI modeling or inference or training has come down significantly.
So just to give you a context, in the past four years, the cost of running AI models have come down by about a hundred times, a hundred times. And how I look at this is that this is going to cause an explosion of usage, whether it be enterprises or industrial. And what that means is that in the past few years, market has generally been complacent in terms of this explosion.
And so to answer your question, I actually see a lot of the bottlenecks across the whole AI value chain. So this is anything from semiconductors to power to grid capacity. This is all going to be involved because of the pace that we've missed. And this happens all the time, Paul. If you look at internet, if you look at smartphone, everyone missed how big the 3G to 4G explosion in terms of network usage would mean as well. So I would say that we are going to see a lot of, I guess, demand or supply issues across the value chain, but if you had to really ask me where the biggest bottleneck is, we're definitely more focused on the power and the grid infrastructure that is supporting all this learning and all this AI cloud computing.

Paul O'Connor:
We've seen significant capital commitments from the large tech platforms into AI as highlighted by the big four being Microsoft, Alphabet, Amazon, and Meta projected to exceed 725 billion in 2026, which is up 77% on last year. How should investors think about where that capital ultimately flows and who the key beneficiaries will be?

Billy Leung:
How I look at that, Paul, is I would say that if you look at it from a very higher level perspective of this AI investment case, it probably started probably close to seven years ago or six years ago where in 2019, people realized the importance of semiconductor chips. And I always thought that the first cohort of companies that really benefited from AI were the semiconductor names such as NVIDIA, the Chinese TSMC, the Netherlands we had ASML, and also the Korean companies like Samsung. If you look at this cohort of companies, their market cap increased by about 150% from 2019 and for two and a half years. So that was the first beneficiary of the AI trade, so to speak. The second phase will be what I call the hyperscalers, which is essentially the last two to three years where we saw companies such as Amazon, Alphabet, Meta, Microsoft, using all these explosion and improvement in chips to actually advance on modeling, advance on cloud computing, advance on, I guess, processing all this data.

And that's when we saw this cohort of hyperscalers or infrastructure enablers increase their market cap by 150%. Now, to answer your question is I think the next or where there's most opportunities there or where all the spending is really benefiting right now is really this cohort of what I call the AI infrastructure. So these are the picks and shovels that goes behind building out the grid networks, building out the data centers. And this is where I see the next phase, so to speak, of AI and this is where all these beneficiaries are going to be really seeing a increase in value in the next five or 10 years.

Paul O'Connor:
I guess we do hear the stories of the Microsofts and Alphabet, et cetera, all trying to, I guess, shore up their own power generation and supply. So I guess that's exactly to your point about the next cohort of AI infrastructure, the opportunities in the picks and shovels is really the build out of the ability to deliver it at such a scale and then incorporate it into their own businesses. So yeah, I guess that's another area for the listeners to consider and think about in terms of their portfolio exposures.
Investor retention has largely been concentrated in the US mega cap tech. Where do you see the more underappreciated or less obvious parts of the AI value chain today and do you believe there are any bottlenecks? And I guess this flows on from the previous discussion we had about the next opportunity set being the picks and shovels.

Billy Leung:
Yeah, I think so. I think if you dig deeper into what we discussed in the previous question was that if this is the cohort, then a lot of the, I guess, investors and clients would come to me and they say, "Billy, why don't we just buy land? Why don't we just buy all these data centers which we'll see capital appreciation and probably paying a good yield as well?"
And the interesting aspect is I actually looked into a whole cost of building out a data center and in terms of the cost of just buying the land and even just the labor that goes into it, these only account for about 20% of a total data center construction cost. And then when you look deeper, about 70 to 80% of building out a data center actually involves your IT infrastructure equipment, all these cables that connect all these service together.
There's also a large part which is cooling equipment. Obviously large data centers require a lot of cooling in order to maintain their sustainability and ongoing processing. And there's also a lot of what we call, I guess, the energy and materials that go into it. So what is really powering it, especially are we going to use more renewable energy, whether it be I guess solar even to an extent, uranium and even the wiring requires a lot of copper as well.
So in terms of how I see the opportunities there or what's really being, I guess, underlooked to answer your question is that yes, we are seeing a lot of spending by the big tech names that we always see on a daily basis. We see a lot of data center names as well, but we don't really look at the companies that are, for example, building out the cables, building out the optical cables, building out the cooling equipment systems, or even just the copper miners or even the uranium miners.
I think these are probably the ones which have been doing well, but I would say that underappreciated from a AI structural trend perspective.

Paul O'Connor:
And I guess as you're answering the question there, I'm just thinking about how broad the whole infrastructure value chain is across AI and even some of the old traditional companies like miners that could benefit significantly out of this investment in the infrastructure space. So yeah, it's quite interesting how broad that value chain is.

Paul O'Connor:
There's been some volatility in semiconductors recently and I guess partly driven by changing expectations around demand and efficiency. How should investors interpret these moves in the broader context of the AI cycle?

Billy Leung:
Yeah, it's true. And if you look at, I guess one of the key leading benchmarks for semiconductor in the US is the Philadelphia Semiconductor Index or what we call the SOX Index and that's just been rising rapidly in the past few weeks and hitting all time highs and we've seen a lot of comments and also a lot of incoming questions of whether or not this has been overbought. I'll be in I guess what we call a bubble situation. But I think that goes back to one of the first questions that we were talking about, Paul, about how I believe, or I've seen data that suggests that we are going to see an explosion of usage of a lot of the AI development. And what we call in terms of AI development is obviously the unit of token usage and we are actually seeing a rapid rise in token usage, which means there are more and more, I guess, people using AI, whether it be you or me using ChatGPT or Claude.

And like I said, the cost is coming down. And what that means is that when there is higher demand for all this processing, there is going to be a real demand for the brains that power this and this is going to be exactly the semiconductor part. I see a huge demand for semiconductor and a huge bottleneck to an extent for semiconductor in the next five years, mainly because we are going to enter this rapid usage of AI, especially agentic AI. But more interestingly is if you look at the whole semiconductor industry, there's actually two things that we should be concerned with.
One, there's actually a development in technology. There's actually improvement in what we call, for example, high-bandwidth memory. So what this high-bandwidth memory allows is that these kind of memory chips allows more and more or more efficient processing. So within this very old traditional semiconductor industry, there is actually innovation that is happening on its own.

So I think that is one thing that's also increasing the value of this value chain or this part of the value chain. Number two is, I guess, more straightforward is that we need time to build out capacity, especially for semiconductors. And a very good example is when the Taiwanese company TSMC was invited to build a plant in Texas. It is still building right now. It takes time to actually construct a semiconductor plan. It also takes time to actually ramp up the capacity because this is a complete closed off facility that no dust can actually go into. So the lead time that goes into increasing semiconductor capacity actually is quite long, probably about three to five years. So what we have now is an explosion of usage, explosion of demand, increasing or improving demand, which improves the pricing power of semiconductor companies. And at that, we've also got a lead time in terms of when the actual capacity will come in.
So I guess what we're seeing is a very perfect storm, or not a perfect storm, but an ideal environment where semiconductor companies will benefit from stronger pricing power and I guess more demand with limited supply. I do think that there is a lot of justification in terms of how semiconductors are being priced right now.

Paul O'Connor:
So I guess you're still holding onto NVIDIA in your own portfolio.

Billy Leung:
With all disclaimers on that, but I would say that semiconductor is one where our firm feels very strongly convicted on and we do see a lot of value coming up from this side.

Paul O'Connor:
More broadly, we're seeing that shift that you've been talking about towards physical infrastructure, energy, industrial capacity. So how do you see that trend connecting with the AI investment theme and does this broaden the opportunity set even further?

Billy Leung:
Yes, I think so. I think a lot of the things which I'm seeing now would not have been made possible by AI. And I think this is something that we need to learn from the past. I think one thing that we looked at, especially for the smartphone or even the internet age is that it does impact a lot of existing industries. It does impact a lot of adjacent industries. I mean, to your point about spreading to traditional industries, I still remember that Ford right now, 100-year-old auto company, has recently been, I guess, classified as an AI company now because they've been building a lot of these energy storage batteries. So if you look at recently the share price of Ford, it's gone up quite significantly and the reason is because they have these batteries which will be used for data centers. So I find that incredible because this is 100-year-old company where now it's benefiting from AI as well.

So I think this is one case, but just to be more specific, I think in terms of how AI has been impacting or allowed now new industries is I think humanoid robotics is a very good example, Paul. I don't think humanoid robotics would have been possible without the advancement in AI. Humanoid robotics or just general advanced robotics requires sensory because something that is learning from new tasks, not just repetitive task. It requires vision as well. All this have been made possible with AI. And something that's also interesting that we've been looking at is that a lot of the space travel or space development in the industry right now would not have been possible without the development of AI.
So I think we are seeing a lot of this AI shifting to more the physical nature where whether it be humanoid robotics, automation, I think you mentioned, healthcare as well and this is all actually coming together now because we've reached this point where AI is cheap enough for application and use cases and monetization and it's also seeing a lot of use cases now because people know that it could be actually helping and improving productivity.

Paul O'Connor:
If this is increasingly becoming a multi-year build out cycle, which parts of the market are most directly leveraged to that dynamic?

Billy Leung:
That's a good question because we obviously do have a very long-term view on AI and we do think that when we look at it, we split artificial intelligence into three buckets, so to speak. We like to say that there is a compute or what we call the hardware. So these are obviously your semiconductors. We also see that as a middle stream where we call it the infrastructure or enablers. So these are the companies that build out the infrastructures that allow or use the hardware and allow better learning and better adoption cases. And the final part is the adopters. So this is as we were talking about where we are seeing a lot of adoption of the better technology, better infrastructure and using it for real world application. So we do think that all these three buckets are going to se a lot of development in the next 10, 20, 30 years.
With all this build out right now and with all the spending and I guess something that I've been repeating about the lower costs and potential higher usage, I think one thing we have to accept that is we've moved to a part where we're going to see a lot of the adopters or a lot of companies adopting AI really starting to see commercialization or really starting to see benefits from it.
And one example I would like to give is we're actually seeing this from a lot of the Chinese companies. And what I mean by that is we are seeing, for example, e-commerce companies in China, such as Alibaba, they're actually adopting AI, improving their AI, and then now implementing AI on their e-commerce platform. And how they do that is they're offering the service to merchants where users can go on their platform and they can look at certain accessories, certain bags and with AI they could visualize themselves wearing it, they could visualize themselves wearing it outside on the street or even in their room and that entices people to buy it.

And this is all made possible with artificial intelligence. And obviously they're also using it to better target users to get the most out of users or even to provide better service for users and not just in e-commerce, but also in development of online games, also development online advertising as well. So we're seeing, I guess, I would say a faster adoption and also commercialization, especially in these Chinese companies. So I would say that post this build out or during this build out, I think a lot of interesting aspects is that people will be or companies will be leveraging all this improvement to actually make more money. And I think one of the things we have to look at is eye companies such as online gaming, online advertising, e-commerce companies really applying this in the right way to actually see, I guess, better margins or better profitability, so to speak.

Paul O'Connor:
The AI value chain seems to be materially broadening as you've been articulating across hardware and cloud computing data and foundation models, AI development and deployment to business applications. Bringing it back to, I guess, investing in portfolio construction, from that perspective, how should investors think about gaining exposure to AI in a diversified portfolio and the related themes without ending up overly concentrated in a single part of the market.

Billy Leung:
I think that's an excellent question. That's something we deal with on a daily basis because the markets are moving very fast, whether it be trying to invest in hardware, trying to invest in software, trying to invest in data center infrastructure, it's very hard to capture, which is why I think we have the benefit of operating and issuing ETFs where a lot of these ETFs, when we look at indices that these ETFs track, we try to really look for a way to solve a solution for investors, especially in Australia as you would imagine a lot of the Australian investors don't have that natural exposure to technology and especially when the technology is moving so fast. So when we create or when we issue ETFs, we really think about how is it possible to encompass the thematic in a way where I guess it's true to label first of all, but it's also very timely as well.
And one example that I used to like to give is we have a product called the Global X Artificial Intelligence ETF, the ticker's GXAI. The thing that it does is that it is a global portfolio. It tracks a global index of 85 companies which are screened for whether these companies are using AI or developing AI or the hardware that goes into supporting AI. And it's exactly what we just talked about, Paul, where it does select companies from a string of whether or not it's semiconductors, it's the infrastructure enablers or even the adopters. And this one has been popular because it exactly solves a lot of solution, a lot of problems where investors can't catch up with all the things that's moving in AI right now or technology right now and also allows them to have an exposure to across the world, whether it be US stocks or even European Japanese or even Asian stocks or Chinese stocks.

Paul O'Connor:
So it's aiming really to give, I guess, investors fairly diversified exposure across the whole value chain of AI there. So who's the index provider?

Billy Leung:
The index provider is called Indxx.

Paul O'Connor:
Yeah, I have heard of them.

Billy Leung:
Yes. So the index is the Indxx Global Artificial Intelligence and Robotics ETF. Yes. And that's, like you said, it tries to encompass the entire value chain.

Paul O'Connor:
We could go on, I guess, all day on this discussion, Billy, but we better bring it to an end. But looking ahead over the next 6 to 12 months, which I guess for mine seems to be a lifetime in AI advancement, what are the key signals or indicators you're watching that will determine whether this theme continues to involve or starts to fade? And I guess particularly in various underlying parts of the value chain.

Billy Leung:
Good question. So I think in terms of things that to look for in the nearer term is obviously the infrastructure spend. So as long we are seeing the Mag 7 or the big four continue spending, we are confident that this should be something that's still developing and we're still seeing a lot of the infrastructure adjacent players really benefiting from it.
Another data point that's really interesting is it's actually from private companies, Paul. It's actually from the more well-known or getting well-known Anthropic. So if you look at Anthropic, they're reporting, I guess, less officially their annual recurring revenue and this gives us a really, really good outlook of how much people are using all these large language modelings. And just to give you an example, Anthropic's, I guess, recurring revenue is close to $20 billion right now on an annual basis. So we're seeing very strong demand for these large language modelings, but I think that the last thing I'd really like to share is that whenever people come to me and go, "Oh, Billy, is AI a bubble? How do we identify it?"
I would always say that a bubble, whether it be AI or any innovation really comes from two things. It comes from a lack of demand and a lack of capital. And that's exactly what I was saying. Whenever we see a lack of demand for AI, when we see a lack of capital for AI, that's when we should be start to concern and we are seeing neither of them. I think a much more closer to home, I guess, observation is that if you are ever spending less on your AI, whether you're spending less on Claude's, less on Gemini, less on ChatGPT, that's when you should be concerned. And I don't think anyone that I've met would be saying that they've been spending less on these large language modelings. And I think that's a very strong indication that the pricing power is not with us, but actually with the providers, which means that this is still a long way to go.

Paul O'Connor:
Yeah, I guess I get the feeling that we're still very much in the early days of it there, Billy. Thank you very much for joining us this morning on the Netwealth Portfolio Construction Podcast series. It's been a really educative discussion for myself. I guess I've learned a lot more about the CapEx going into, I guess, the whole AI value chain and how that's projected to just continue to increase and potentially just a little bit CapEx starved at the moment. Certainly been educative around the value chain and just you've got to really think laterally because you've even mentioned old world companies like Ford and the way that they're adopting and being able to monetize opportunities in AI there. And I guess just the real transformative nature that it is having on society today and the increased investment, I know certainly network we've been adopting AI very strongly and very fast.
And it's amazing the way when I use it myself in my day-to-day work that it just from what was occurring three months ago, it seems to be a hundred times more powerful today. So it's certainly such a transformative tool that will, I think, benefit society to a great degree. So again, Billy, thank you very much for joining us today.

Billy Leung:
And thank you for having me.

Paul O'Connor:
And to the listeners, thanks again for joining us on the podcast today. I hope you enjoyed the discussion that I had with Billy Leung from Global X ETFs. And I look forward to you joining us on the next installment of the podcast series.