Words on the Next Frontier of AI

In this episode of Words on Wealth, Evans and Partners CIO Tim Rocks is joined by Harry Morrow and Raymond Tong, Portfolio Managers at Loftus Peak, to unpack the current state of AI and where it’s headed. Fresh off five weeks in Silicon Valley, Harry shares what companies are really saying about demand, and why the on-the-ground confidence doesn’t match the headline pessimism. The conversation covers semiconductor supply constraints, the explosive revenue growth of frontier AI labs, and why this infrastructure buildout looks nothing like the tech bubble of the past. The conversation also looks ahead to the next frontiers of AI, from agentic workflows and physical robotics to breakthroughs in health and life sciences. Tune in to find out more.

This episode is also available on Apple Podcast. 

Tim Rocks (00:00)

Hello, welcome to Words on Wealth. I’m Tim Rocks, Chief Investment E Officer at Evans and Partners. today I have with me two representatives of Loftus Peak that should be well known to us, Ray Tong and Harry Morrow, ⁓ who are PMs on the Loftus Peak Fund and have been great sources of insight to us on the tech and in particular the AI side over the past couple of years. So ⁓ it’s great to have them back.

 

⁓ let’s perhaps start with Harry who has spent the last five weeks touring Silicon Valley and and other sort of hot spots in the US. ⁓ and it’s certainly an interesting time to have done that because the sector is being perhaps sort of reassessed at the moment. But like over to you, Harry, you know, what what what was the sentiment on the ground? What were your kind of key takeaways from that trip?

 

Harry Morrow (00:55)

Awesome. Thanks. Thanks, Tim. so yeah, I I spent five weeks over in the US, ⁓ kind of going between West Coast, East Coast. ⁓ spent, you know, I I think I probably met with about a hundred companies ⁓ across semi software, media and tech, three broker conferences and then there was one industry conference that we went to as well. ⁓ so maybe I’ll just go in that order in terms of the semi side of things and then the software side of things. ⁓ obviously semiconductors, I think they’ve they’ve

 

⁓ come got more attention recently than they have in the past. And I think, you know, we’ve been investing in this area for, you know, a decade plus, just because we think it is the heart, it’s at the heart of all ⁓ disruption across all industries. And I think, you know, all a lot of the names have run quite hard. It’s been a a very ⁓ performant area of the market. And I think if you read the AFR consistently, you might think that there’s doom and gloom ahead and it’s the end of the AI trade. But having been on the ground for five weeks and and talking to a lot of the companies there.

 

you you just don’t get that pessimism ⁓ from any of them. I think the the surprising thing for us was the amount of visibility that a lot of these guys have. ⁓ they’re talking about, you know, numbers out to twenty twenty eight, twenty twenty nine. ⁓ and I I think one of the one CFO that I met with of a multi hundred billion dollar company was saying that we’re seeing demand signals for twenty twenty eight and twenty twenty nine that are crazy. Right. And so I think th there’s possibly a bit of a red flag in there. You don’t want to take management

 

know, ⁓ you want to add a take with a grain of salt. ⁓ but it’s important, I think, to note that that a lot of companies have the same visibility. They’re talking about the the same type of demand signals and, you know, Ray might touch on it a little bit later, but some of the ⁓ other data points that have emerged recently, whether it’s OpenAI or Anthropic or SpaceX, they all lead us to believe that the deb demand environment is is not weakening and if anything, it’s probably getting ⁓ a little bit stronger. ⁓

 

So I

 

Tim Rocks (02:55)

I thought the most encouraging news is that evidence of take up and monetization by anthropic is is just pretty incredible that it’s its its revenue growth rate seems to be growing at sort of ten X per year. And that’s that’s kind of the the the the the proof point, isn’t it?

 

Harry Morrow (03:11)

⁓ totally. And I think ⁓ you know, Dario made an offhand comment ⁓ maybe a year ago. He was like, Look, we ten X revenue the first year we ten X it again the previous year and I think they were sitting on maybe ten billion dollars of ARR towards the end of last year and everyone was like, Can’t possibly do that again and there’s obviously limits to that with the law of large numbers, but ⁓ it looks like they’re gonna ten X it again this year and and that was ⁓ it seems as a result of the agentic side of things, which I think Ray has has spoken about before, just in terms of

 

you know, flawed being able to go off and do things in the background, come back a few hours later with, you know, close to a a a completed project or mostly on the coding side of things. But I think the advancements that have been made in the past, you know, ⁓ year, months have been incredible and that’s resulting in much, much more revenue growth, ⁓ and return on the the kind of AI infrastructure that’s been invested in.

 

Tim Rocks (04:05)

Yeah, yeah. And then aside from just that that big, you know, big picture sort of growth story, what were the other topics of conversation at those conferences that, you know, perhaps surprised you or interested you?

 

Harry Morrow (04:18)

I think one of the things that we were seeing in markets ⁓ was probably confirmed for us just in terms of markets are moving very quickly and I think investors are looking for the next bottleneck. ⁓ kind of and it’s mostly in the semi supply chain. ⁓ and I I think that could be, you know, somewhat behind us for the most part. Like, you know, memory was a big bottleneck for a bit and then ⁓ energy was also a bottleneck probably before that, right? And so you had a lot of names run as a result of that. Optical names, ⁓

 

without getting too much in the weeds. ⁓ you know, the guys that help out with the networking side of things, ⁓ they all all had a run. And I so I think that just kind of confirmed for us that, you know, markets were moving quickly and a lot of investors over there were looking for the next bottleneck. And, you know, as far as memory goes, as, you know, boring as memory is, Samsung, SK Heinex and Micron, ⁓ I think one thing that has changed for those guys is

 

the long term agreements that they’re signing up because historically memory’s been an incredibly cyclical business. It’s mostly consumer end markets where it’s like PCs, smartphones, which kind of move with the economic cycle and you have massive boom bust kind of moves within the memory market. And what’s happening now is that because all of the hyperscalers are massive customers for the memory players, you know, almost half of the supply ⁓ could be locked up in long term agreements, you know, with Micron.

 

SK Heinrichs and Samsung through the hyperscalers. And you would have seen it recently, right? With Apple having to raise ⁓ you know, a lot of the product pricing. And I think smartphones, when they get released in a few months, are probably gonna have higher prices again because of what’s happening in the memory market. And ⁓ it doesn’t look like that pain is gonna be alleviated anytime soon. The supply-demand imbalance I think is is remains tight. And if anything, I think it could be getting a bit worse next year.

 

Tim Rocks (06:10)

Yeah, yeah. Well, why don’t we move on and talk about ⁓ that CapEx, I think specifically in the in the data centers. ⁓ that’s where say the bubble word is often kind of used. ⁓ perhaps ⁓ maybe this might be initially to you, Ray. What w ⁓ what what are your thoughts there? You know, are the use cases coming through

 

Do we think there’s going to be sufficient return on investment to to to justify the amounts being spent there?

 

Raymond Tong (06:44)

Yeah, I I think I’ll probably split this answer into sort of three sort of areas that we look at. So in terms of just the CAPEX itself and the funding of it. So I think the market now is forecasting close to a trillion dollars of CAPEX next year. And that continues to increase. And so far anyway, most of it has been being funded out of the hyperscaler free cash flow. So the likes of Amazon, Microsoft, Google.

 

meta. And though that free cash flow is starting to compress, but to date, most of it is that and there hasn’t been that much debt. But we are starting to see probably more debt into the system. We’ve seen, you know, some equity raises happen in the last month or so. So SpaceX, obviously, raising 85 billion. ⁓ Google did the equity raise of 80 to 90 billion as well.

 

So we are seeing some more equity funding, but also going forward, I suspect that some of this capex will need to get funded out of a bit more debt. But at this stage, that still remains reasonably low. ⁓ In terms of the fundamentals and the demand, think, as Harry said, I mean, the consistent data points that we’re seeing is that demand remains off the charts and way ahead of supply. And, you know, I think whether that’s the hyperscalers talking about it.

 

⁓ And that’s why they’re investing all that money or even likes of the Frontier Labs, OpenAI and Anthropic consistently beating their own revenue forecasts. And I think Tim, as you mentioned, and Harry mentioned before, Anthropic’s growth rate this year has been just off the charts. I they started, ⁓ ended last year at close to $10 billion.

 

I think in May or June, they’re at $47 billion was the last time they revealed those numbers. And many of the market potentially suggesting that they could sort of exit this year close to $100 billion of annual revenue run rate. And what’s probably just as important from our perspective is that ⁓ the last month or so, ⁓ it seems like there’s data come out to suggest that they are actually becoming profitable.

 

even just for a month and we’ll see how sustained that is. all that sort of bring that back to the point is that most of the sort of GPUs that are getting installed out there, they’re getting monetized and used straight away. So they’re not underutilized. It’s not like back during the tech bubble where you had dark fiber, which remained unused for a long time and a lot of it remains unused.

 

or the infrastructure that’s getting rolled out at the moment is getting used and monetized. And I think Harry sort of mentioned it as well. mean, one of the things that gives us a bit of comfort at this point in time anyway, that could be holding a bubble back is that there are supply constraints. ⁓ So ⁓ not only on the memory side, ⁓ but also now ⁓ on the CPU side of things with the agentic AI, you need more CPUs. That’s a bit of a bottleneck at the moment, but…

 

probably more high level is that there is just not enough supply of chips at the moment. And part of that is that TSMC is the ⁓ majority ⁓ sort of manufacturer of advanced leading AI chips ⁓ and they are expanding their fabs or their capacity, but it’s probably not at the right that the industry ⁓ would like it to be. And that in itself is probably meaning that this ⁓

 

bubble is not forming, if anything, it’s gonna be high for longer for a period of time.

 

Tim Rocks (10:32)

Yeah, exactly. It actually saves the companies from themselves. I’m sure they love to be doing multiples of the investment they’re doing, but they simply can’t because they keep coming across these these kind of constraints.

 

Raymond Tong (10:44)

Yeah, no, I think that’s absolutely right. And I think, you know, I’m sure Nvidia could sell a lot more chips that they sell and get a lot more revenues this year and next year. But, you know, the fact is that there’s supply constraints on the sort of the GPU side of things, but also on the memory side of things. I think Harry, I think Hinex, Micron and Samsung are saying that things are sold out to 2028.

 

and supply probably really isn’t expanding properly until probably late 2027.

 

Tim Rocks (11:21)

Yeah. Yeah. All right. So I’m I might go back to you now, Harry. So I I’d I’d love to know where we are in the whole AI rollout. So if you think that AI started as just sort of the reasoning, co kind of a glorified search engine. And then we’re we’re going through a gentic now. ⁓ and then there’s gonna be whatever there is kinda after that. ⁓

 

Love to hear your thoughts about where we are in that cycle and and what we know, what those next phases might be.

 

Harry Morrow (11:53)

Yeah, no, I think it’s a good question. ⁓ look, I I I think we’re still pretty early on in the deployment. Like I th I think the technology is there. ⁓ and you know, we started with Chat GPT and LLMs to begin with, and then, you know, we had the models move towards more of a reasoning ⁓ type of output, which resulted in better outputs. And then, you know, recently we’ve obviously had the emergence of Agentic, which is kind of multi turn.

 

Work like entire workflows that are now being done by this stuff and

 

Tim Rocks (12:23)

Basically

 

th the the models actually doing things and you know making their own decisions effectively. Is is that’s what a genetic is, isn’t it?

 

Harry Morrow (12:30)

Exactly. And I I I think ⁓ exactly it’s like a an agent doing something on your behalf, right? And I think it is it’s complex. ⁓ I think OpenAI and An Anthropic are probably ahead in that respect with Cloed Core cord cloud, sorry, clawed code ⁓ and codecs from open AI. ⁓ and they’re also able to like tap into, you know, regular applications a lot of the time as well, right? Like

 

I don’t know if you’ve used it recently, but Claude Code can spin up a an Excel spreadsheet for you if you’re trying to do some sort of data analysis. Right. So it’s it’s that type of of stuff that’s that’s being done. But I think we’re still very early in terms of the overall penetration of the opportunity, right? Like I think ⁓ you know, both of our firms are probably using it at a re relatively base layer. I think we have Claude or ⁓ ChatGPT, ⁓ but in terms of actually having ⁓ recurring integrated workflows, we’re probably

 

not that far along. So I j I just think that the with the the technology is getting there. ⁓ but in terms of deploying that across a variety of different different industries, there’s still a fairly long way to go. And ⁓ on the agentic side of things, you know, one of the data points that we saw the other day, you know, Mark Zuckerberg, CEO of of Meta came out and said that they were not making as much progress as they would have liked on the agentic side of things. And maybe that’s on the consumer market specifically as opposed to enterprise where

 

Anthropic and OpenAI do somewhat dominate, but it it does seem like it’s a hard problem. ⁓ they’re throwing a lot of resources at it, but we’re still in that journey ⁓ on the agentic side of things.

 

Tim Rocks (14:05)

Yeah. And then what’s after that? Is it robotics or or is it perhaps more specific use cases and the benefit going over sort of to to other industries like people talk about h health research and the like? Well like what w wh where are we like if we look forward five years and move away from the AI companies or data centre companies themselves, what what should we be looking at?

 

Harry Morrow (14:34)

Yeah. Well I might I might let Ray touch ⁓ on the the the physical AI side of things. ⁓ but I mean I I I think, you know, on the on the healthcare side of things, we actually have ⁓ an expert on our team, Rob Forage, who’s you know, he was at Harvard and Oxford. he did bioinformatics and he was talking about the fact he reads a lot of kind of ⁓ journals and ⁓ that come out on a daily basis leveraging this technology and

 

What he was seeing was that like previously when he was kind of I think he sequenced first person to sequence the genome of a sheep, ⁓ you know, back then and even now, it would take an enormous amount of time to do that sequencing, you know, like you would do it one at a time. Whereas now, ⁓ you know, I think there’s a thousand, if not million ty million X lift on the amount of times that you could do that, right? So I just I think from a velocity perspective, and this goes for health and life sciences as well as pretty much every other industry, I just think the

 

velocity of the speed at which people can move is so much faster than it’s ever been, right? And so I think that’s probably one of the things that’s gonna hit the health and life sciences. And and you won’t necessarily see that today, but to your point, maybe five years from now, ⁓ the number of drugs that are moving through the pipeline might be significantly larger, or the ones that are identified are more likely to be successful, right?

 

Tim Rocks (15:56)

Yeah.

 

Raymond Tong (15:57)

I think probably just to add to that as well, think both Dario and Modi and Sam Ottman from Anthropic and OpenAI, mean, obviously they’re seeing what is down the pipeline in terms of the models. Both of them sort of have consistently said that what they’re most excited about in terms of the breakthroughs coming forward is in science and in medicine. So as these models get smarter,

 

⁓ I suspect that we’re going to see more of this sort of breakthroughs in the scientific and healthcare side come forward. And what we’re noticing as well is that a lot more healthcare companies are starting to partner with the likes of like an Nvidia or an Anthropic to ⁓ utilize these models and AI in their R &D a lot more than what ⁓ we saw a couple of years ago.

 

In terms of the physical AI side of things, that is going to be one of the next legs that sort of Jensen from Nvidia talks about. Probably two parts of that at this stage that we’re seeing some real evidence of. So firstly, I think is in things like automata driving, it is happening. In the US, you’ve got Waymo, which is Google’s robot taxi service.

 

As a clear leader in the US, they’re in more and more cities. They’re expanding internationally to the UK and into Japan. And then you’ve got Tesla, who with their full self-driving software trying to ramp up their service. But you’re also seeing the likes of other companies coming in and really trialling that service and developing it. So, know, companies like Zoogs.

 

which is owned by Amazon Wave, which is a startup. We’re just seeing this over the next two to three years, probably more and more of this happening. And then in the robotics side for sure, I mean, think more more companies are leveraging robotics. I Amazon is probably, I think they announced last year or earlier this year that they’ve hit 1 million robots in their warehouses across their business. But no, clearly in terms of humanoid robots,

 

Tesla is doing that. ⁓ We’re probably gonna see a bit of progress in that later this year. But the ones that which are probably we’re seeing the most progress at the moment is in China, ⁓ where you’ve got plenty of companies ⁓ investing a lot of capital ⁓ to develop ⁓ humanoid robots. ⁓ a couple like Unitree, which you’ve probably seen a lot of videos on, ⁓ seems to be, and a company called Ubitech seem to be ahead. ⁓

 

but it’s one where it’s a ⁓ priority of the Chinese government in their five-year plan in robotics and humanoids. And so there’s a lot of capital being invested, a lot of competition being fostered, and probably definitely a lot of innovation that is gonna be driven ⁓ today and in the next coming years.

 

Tim Rocks (19:09)

Yeah, yeah, but s the China thing’s interesting. Whether it’s a ⁓ whether it’s a threat or not, it feels like China’s approach is a bit different. It’s very focused on the robotics on the industrial side. So it i i is it complementary to what’s going on in the US or is it is it a threat?

 

Raymond Tong (19:30)

I mean, I think it’s definitely a race between the two countries for AI at the moment. ⁓ And I suppose one of the things with China is that it feels like from the top down, there’s a really big push to adopt AI across not only the consumer side, across businesses and enterprises. You you just think about, I mean, there are surveys done recently.

 

in terms of the public, what they think about AI and what kind of adoption. And China is a lot more embracing of AI across the country. And also I think it is part of their five-year plan. And when you just think about it, just on the robotic side, China’s population is forecast to decline by a fair bit in the next 20, 30, 40 years.

 

population is aging. So they need to fill those factories up or increase the productivity somehow. And robotics and humanoids is a clear priority that could do that.

 

Harry Morrow (20:40)

Yeah. But I it’s actually a I think it’s an interesting point, right? ‘Cause I think the willingness to adopt, ⁓ and the need to adopt is very different between the two countries. And so our CIO, ⁓ Alex Polak and one of our analysts, were over in China towards the end of last year. And I mean, you know, they went to a a cattle battery fac factory, ⁓ which, you know, a big chunk of which was automated, but even just in everyday life, right? I think at the hotel, ⁓ there was a robot that would take ⁓ you know, in room dining.

 

up to the door ⁓ as opposed to having a person do that, right? And I think if you saw that in Australia or in the US, you’d think that that was probably a little bit weird. But I think, you know, in in China they’re much more accepting of robots in instances like that and in restaurants too, for example.

 

Tim Rocks (21:25)

Yeah, yeah. All right. Lo it might be worth finishing by talking about how you’re putting this together in terms of your portfolio. just in general terms, and perhaps not at the stock level, but you know, w where are you in terms of the semiconductors versus the the cloud companies? How how are you focusing your investments now?

 

Harry Morrow (21:51)

Yeah. so I think on the portfolio side of things, we might have mentioned this before, but you know, we’ve got the tools, which is most of the semiconductors, ⁓ the enablers, which is the hyperscalers, and then the beneficiaries, and that’s kind of the the life cycle that we expect it to flow through. And so, you know, we’re still ⁓ quite bullish on the tools side of things in terms of, you know, the the logic guys, the the you know, a little bit of memory, ⁓ as well as a few other things on the tools side of things.

 

I think on the hyperscalers, ⁓ you know, we still hold the companies, ⁓ but it does look like CapEx is probably going to continue going up and and you know, how the market feels about that I think is still up for debate. ⁓ they need to accelerate revenue growth, I think, to be able to justify the CapEx bill, but that’s gonna be good for the semis that the portfolio does hold. ⁓ and then on the beneficiaries, I think still think we’re pretty early. ⁓

 

And I mean, I know you said no no specific not specific companies, but companies like Meta, I think do have obvious ⁓ you know, applications of this technology. And they released a model the other day which is for image creation. ⁓ there’s a video one coming out shortly. ⁓ and that should turbocharge their business model. but as far as, you know, pure ⁓ I guess ⁓ generative AI L L businesses, we’re still ⁓ moving down that path outside of ⁓ anthropic and and open AI. But

 

⁓ outside of the the AI complex, I would say that we have increased the exposure to the life sciences. So, you know, Lily’s a large holding and I think we’re incrementally positive on that. It’s, you know, top five position in the fund. ⁓ and and we do think that the GLPs is a a market that people keep ⁓ you know, underestimating. And then there’s a couple of other names that have entered the portfolio recently on that side of things. So it’s we we would

 

view it as a a a a well diversified portfolio. Yes, there’s AI in there, but there’s also a bunch of other interesting ⁓ things too in life sciences, media. ⁓ EVs is also super important, which Ray might touch on briefly.

 

Raymond Tong (23:58)

Yeah, I mean, I think that the whole ⁓ energy security side and the electrification ⁓ sort of thematic is ⁓ really interesting at the moment. And it sort of does actually cross over with AI in some sense, but a company like ⁓ CATL, which is the largest

 

that should make it globally. mean, it’s seeing real strong tailwinds, not only from the increased EV penetration. And we’ve seen a real acceleration in last few months, ⁓ not only in Australia, but ⁓ globally. But also the next leg ⁓ of growth is these energy storage systems, which are getting deployed globally. ⁓ And part of that is just,

 

Tim Rocks (24:43)

Yeah, but you need more grid, right?

 

Raymond Tong (24:45)

Yeah, need more. No, no, exactly. And also, in many countries at the moment, renewables plus batteries, actually, the cost of that has come down to be cheaper than your fossil fuel systems. ⁓ And then you’ve got ⁓ these AI data centers, which all this capex is getting spent to getting rolled out. A lot of those are now getting paired with sort of these battery storage systems as

 

⁓ Number one, maybe just a second source of energy. ⁓ But number two also, just given the energy demands from AI can be quite variable and can spike. These battery systems get to, I suppose, even those out. And so they’re becoming increasingly important around the world as these data centers get rolled out. So I think in terms of the battery demand,

 

looking at sort of five plus years, know, there are market forecasts that suggest that it will be growing at 30 % plus overall, with these energy storage systems growing ⁓ quite a bit above that.

 

Harry Morrow (25:54)

Wh which one one interesting point just from the trip quickly was that like I think there’s a lot of rhetoric that comes out of the administration in the US that’s probably against renewables and batteries. ⁓ you know, for somewhat valid reasons sometimes, but I think the the what’s actually happening on the ground is a little bit different where there is an enormous amount of solar that’s going in in Nevada specifically, ⁓ which is where I w drove past massive solar farm that I think was gonna, you know, five X over the course of a few years.

 

And, you know, that’s all batteries, that’s all solar. And and to Ray’s point, there’s a lot of data centers that are are also going to be ⁓ you know, powered by solar and batteries. So I think there is, you know, despite some of the headlines that you hear coming out of the US, I think, you know, the business case and the economics always end up winning and solar is is pretty cheap when you compare it to a lot of the other options out there.

 

Tim Rocks (26:45)

Yeah, well it’s what China itself is doing in solar plus battery in its deserts are pretty amazing as as well. All right, great. Look, that’s been really interesting. I really like that, ⁓ that ⁓ you know the tools enablers sort of beneficiaries kind of split that you’ve done there. But it feels like there’s just so much more ⁓ to go in the tools and enablers first before we even think about the the the the beneficiaries. But gosh, there’s so much sort of going on and

 

Harry Morrow (26:52)

Exactly.

 

Tim Rocks (27:13)

Yeah, really appreciate your your insight. So so thank you very much, gentlemen.

 

Harry Morrow (27:17)

No, awesome. Thank very much.

 

Raymond Tong (27:18)

no problem at all.