Global Perspectives: Investing in the accelerating trend of generative AI

Lara Castleton: Hello and thank you for joining this episode of Global Perspectives, a Janus Henderson podcast created to share insights from our investment professionals and implications they have for investors. I’m your host for the day, Lara Castleton.

The tech sector took a sharp hit earlier this year with AI skepticism over China’s DeepSeek release and the potential impact tariffs would have on the global economy. However, as of this recording on July 23^rd, broad tech indices have not only recouped losses but are showing gains on the year.

As my team engages in mid-year portfolio consultations, clients frequently ask us how to think about their allocation to technology. It’s a sector that shows no signs of slowing down. However, it also comes with valuations that make some investors cautious. To talk to the future of the tech sector, we must address the current state and outlook of AI. And to do that, I am thrilled to be joined by no other than Denny Fish, Portfolio Manager on our Global Technology and Innovation team. Denny, thank you for being here all the way from Silicon Valley.

Denny Fish: Great, glad to be here.

Castleton: So, let’s just level set again: What happened with the AI volatility and tech volatility earlier this year, and maybe your take on that DeepSeek moment we saw.

Fish: Yeah. You know, coming into the year, there were two things at work. One is there was a degree of skepticism in the investment community as to whether the CapEx [capital expenditure] that had been deployed broadly to AI was going to start translating into revenues and actually economics and seeing some returns on that. So that was number one; there was some skittishness associated with that.

And then number two was, that coincided with what you referred to as the DeepSeek moment. And it’s fascinating; it feels like it was an eternity ago, you know, that it happened. But why investors pretty much freaked out was because DeepSeek, at first blush, even though it turned out to be a complete head fake, appeared to require far less compute to be able to prefer to perform the same tasks that previously had required a lot of CapEx. The reality was it still required the similar levels of CapEx; it was just the last iteration didn’t require as much. So there there was a real misunderstanding of kind of the CapEx behind it.

But secondarily, they also did some novel things that were then replicated or were already in the works by all the big model providers. It’s just this leapfrogging effect that we’ve seen. And so we got through that. And if anything, go back and look at where the expectations were for CapEx: They’ve only gone up, and they’ve actually gone up a lot across the industry. And that’s because inference is really important. Reasoning requires a ton of CapEx in addition to training. And so, we’ve just seen this uninhibited growth in in CapEx associated with AI that.

Castleton: Makes sense. But we’ve still seen these hyperscalers then come back in spades. How does that large investment then …. we know CapEx is required, how does that pay off for these hyperscalers in particular?

Fish: Yeah, absolutely. And I think maybe the best moment that we had around that
was when Microsoft reported its Q1 earnings in April. And there was a lot of consternation about when we were going to see an uplift in revenues. And they effectively came out and said, look, we’ve been telling you we’re deploying all this CapEx. As we deploy it, we’re going to see revenue contribution. We’re still supply constrained, but we were able to actually provide more supply, and it resulted in a commensurate amount of revenues and therefore they exceeded expectations. And suddenly the narrative shifted real quick to like, wow, the returns are actually coming through for companies. And then you heard all the use cases for Meta, you’ve seen Google with AI summaries, and you just go down the list of these use cases that are developing.

And I think I saw in Q1, I think the number was roughly 50% of companies in the S&P 500 referenced artificial intelligence in their transcript. So, the trend is just accelerating. The reason it’s accelerated is because I’ve never seen this in my career before, where every single CEO of the most important tech companies in the world are all chasing the same shiny object. And it’s because it’s existential both to the upside and the downside to these companies. It is existential for governments around the world. So, you see every government thinking about how they’re going to protect their economic interests, their social interests, and their defensive military interests as well. And it’s because AI is going to have such profound effects globally.

Castleton: So, AGI, that’s artificial general intelligence. Obviously, I’m sure you’re watching which company is going to get there first. Is it just the first one’s the winner, or is there more nuance to that?

Fish: It’s a little more nuanced than that, but the reality is, whoever gets there first, it’s a big prize. And so, yeah, we follow the evolution of every model, what they’re doing, how they’re leapfrogging each other. And so, if we’re just talking about the big model providers, the hyperscalers, they’re all pursuing AGI at the same time. They’re pursuing niches within their models to enhance their businesses. Whether it’s Meta with Llama, whether it’s, you know, Alphabet with Gemini, whether it’s the Microsoft Open AI partnership, whether it’s Elon [Musk], you know, Grok … everyone’s got some unique use cases for what they’re trying to accomplish. So you kind of get the best of both worlds. You get the call option of AGI and you get these very specific models that are enhancing these businesses.

Castleton: So I guess near term, just what risk do you see with the AI spend, the CapEx, and is it just earnings as we go forward, or what other risks are on the near-term horizon that you’re watching?

Fish: There’s always the risk that we could go through some period of digestion, OK. And so that could potentially impact the CapEx profile for semiconductor, semiconductor equipment, foundry, data center builds. But the reality is, if you start looking through, these companies aren’t making six-month decisions. These data centers are 10- to 20-year decisions. They’re things that you have, you know, two, three, four years of lead time to get the permitting, to get the power, all the things that you need to do this to support the deployment of these clusters. I mean, it’s just a year ago we thought 100,000 GPUs was a big cluster, and now it’s clear that they’re going to be a million GPU clusters out there, and it’s going higher from there to 2 million.

And so, coming back, we could see a period of digestion, you know, we could also see just disruptions to the global economy that would actually lead to maybe softer than expected revenues for the companies, for non-AI business; maybe AI revenue contribution ramps a little slower for consumer internet or software. But, you know, the trend is firmly intact. And if we look out over the next three to five years, the entire world’s in the pursuit of what’s referred to as AGI.

Castleton: And you just used the word, I mean, “existential,” right? So this is existential for so many companies. So maybe if you put it that way, every company going after one thing, valuations might actually be reasonable. Despite, again, what I mentioned earlier, investors just feeling, you know, this is really pricey. It’s run up quite a bit actually. With the potential growth going forward, you might say that it’s actually realistic.

Fish: Yeah, and I’ve had this debate for, you know, 20 years. I went through even like the dot-com bubble, and valuations got very extreme then, right? And that wasn’t even as profound as a shift as we have right now with AI. If you look at what sector is going to continue to increase as a percentage of global GDP for the next 10 or 20 years, it’s technology, OK? So you have that going for you. The other thing that I think gets less attention right now when we talk about valuations is these tech companies are the early adopters of AI. Their employee bases are actually, they have the profile that can lead to the most productivity enhancement within their employees. So, I think even though valuations may look optically a little rich, the productivity enhances we’re going to see are going to be meaningful.

And so, I think margins could be up a lot as you start looking at R&D, software development, customer support, marketing … there’s just a lot of areas that these companies are going to get a ton of efficiency, and then that’s going to broaden out to the rest of the economy as well. But the tech companies are going to see first, they’re going to see the revenue contribution first, they’re going to see the productivity enhancements first, and then it’s going to broaden out.

Castleton: And you mentioned that hit to margin. So, I do think we need to address then the tariff uncertainty as you mentioned, I mean at the beginning of the year, it still is ongoing. What’s going to happen with tariffs? Can you just address how you’re modeling that in into the technology sector? It’s obviously not one holistic sector that’s all treated the same, but how are you thinking about tariff volatility going forward?

Fish: Yeah, and it’s complicated because tech has many different subsectors. But if we look at software for example, there’s really no risk there, right? You look at the consumer internet companies, there’s a little bit of risk, you know, in terms of digital taxes and other things. We seem to be getting through that, but there’s not like a direct tariff impact. So those are relatively safe.

And then you actually look at areas like AI servers, you would think, OK, we’re at risk there because, you know, so many of the chips get manufactured in Taiwan. But the reality is most of the AI servers actually get built and assembled in Mexico. And so those are not subject to tariffs as part of the USMCA agreement. So, we’ve been able to skirt that, which has been good.

There is something called Section 232 right now that isn’t resolved. That’s how the tariffs are actually going to be deployed to electronics probably have more of an impact on smartphones, traditional PCs, traditional servers, things like that. But I think for where the growth is, once again, hopefully we’re going to be all right. And then there are other things that haven’t gotten as much attention. But, you know, we were able to evolve the diffusion rule so that now we’re actually able to support companies like Nvidia or AMD in terms of
selling GPUs to other countries around the world without limits, assuming they don’t find their way to countries they’re not supposed to. And so, so far, so good for tech in terms of the tariff landscape.

Castleton: OK, maybe double-click on that a little bit too, because it has seemed like obviously, AI, everybody’s after this, and the U.S. government in particular wants to have dominance within AI. So, the plan is to get get access to our GPUs as much as we can across the world to the good actors. Maybe just address how that’s going and what your view and outlook is on that a little bit more.

Fish: Yeah, it’s evolved positively, you know, with the diffusion rule, which is helpful. And what’s important is we’re never going to be able to completely keep this technology out of the hands of adversaries or whoever. It just … it is what it is. But what we do need to do is we need to support our companies here in the U.S. because those cash flows then get reinvested. We continue to keep our technology lead, and it’s a virtuous cycle. And one of the reasons that the U.S. has been the leader in technology for many, many years, is because of that constant evolution and reinvestment of the free cash flows that are actually generated from these companies. So once again, I think we’re in a pretty good spot.

Castleton: So, outside of hyperscalers, the large language models, or the LLMs, AI covers a wide span of opportunities. How do you even come up with a framework for thinking of AI and investing there?

Fish: You know, shortly after ChatGPT was actually released, we got the team together, and we put a framework together where we did what I call a whiteboard exercise, and we just drew a line down the middle of the whiteboard. And you were either on the wrong side of time as a result of AI or you’re on the right side of time, or [you’re] straddling the middle and you’re not quite sure which direction you’re going. And that was our job, to figure it out. So we start with that. It’s simple, but it’s very complicated in actually trying to get to the answer.

And then the other thing that we do is we really kind of think forward, because AI is impacting the entire supply chain for technology, whether it’s a digital or physical supply chain, and figure out who has dominant competitive positions where the company is actually better because of AI, and it would have been a great business if we didn’t even have AI. And so you kind of think through, you know, businesses like foundries, for example, where all of these chips actually have to get manufactured. Those are great businesses, and there are only so many of them. If you think about the companies that are needed to actually design the chips, it’s called electronic design and automation. Those are, you know, great businesses that benefit from AI, and AI is infused into their products to capture more value from their customers. You think about the data center infrastructure that we’re going to need built, and it is continuing to be built and will for a very long time. That’s a really interesting area, the amount of power that’s actually necessary as well.

And so there are a lot of different areas. And then we think through things like the software ecosystem, you know, they’re agentic for software companies now that are being funded in the venture community. They’re going to be some big winners; they’re going to disrupt their incumbents in the public markets that are actually going to evolve their products and services to embed AI, and then they’re going to be winners as well. There are going to be a lot of losers, be a lot of winners. Same thing’s going to happen in consumer internet.

So it’s really putting this lens and this framework onto every sector and really trying to get to a conclusion [about] who’s a winner, who’s a loser. And sometimes it’s not obvious and it just takes some time. And you’ve got to follow the data and, by the way, the data is going to change and and the facts are going to change. You’ve got to change your investment opinion when it does.

Castleton: A lot of expertise needed in figuring out the winners of AI going forward. And as you mentioned, I’m glad you brought up power, because does seem like sky’s the limit in terms of the of AI to generate innovation productivity gains. But there’s a clear issue, and that is the amount of power generation that we have. So how do you think about that framework? I know maybe some big hyperscalers or tech companies partnering with utilities to solve this creatively. What’s our path forward to get enough power and energy demand for all of this AI?

Fish: It’s really complicated. There is a supply demand mismatch. If we look at the size of a mega data center now, I mean that used to be, I don’t know, 100 megawatts. Now it’s two gigawatts, potentially three gigawatts, right. I mean, these are massive numbers. If you look at, you know, the Stargate announcement and stuff like that, it just requires a tremendous amount of power. It takes a long time to get it permitted; it’s very difficult to get it transmitted. And so, it’s really working in partnership with independent power producers, alternative energy, thinking about nuclear … just a lot of things that we’re going to need, and we’re going to need to do it globally as well. This is a problem in the U.S. We definitely need power so that we can maintain our competitive advantage. They need power in the Middle East, they need power in Europe, they need power everywhere. And so it’s, as I said, complicated, but a really, really interesting area of the economy right now. And if you would have told me 10 years ago that I would be sitting here talking about utilities and natural gas, I would have called you crazy, right? But we are. We are.

Castleton: You’re optimistic there will be creative solutions, yes.

Fish: We’ll get there. We always figure it out. It’s just this one’s going to take time because you just can’t spin it up in six months. It’s a multi-year process.

Castleton: Great. So, I guess as we look forward to the end of this year, even in next year, maybe two or three things that you’re going to really be watching that might help give you a guess as to the outlook of technology and the performance going forward.

Fish: Yeah, I think as I mentioned earlier, I think that clearly AI is appreciated. I think there are areas within tech and out of tech that you can invest in that benefit from this theme, like whether it’s data center infrastructure, power, you know, things like that. But I think the thing, and I mentioned it earlier, that I still think is really underappreciated is the margin leverage that we’re going to see. And I’m expecting that to start coming through, and I’ll give you an example. Microsoft has revenues that are accelerating in their most important businesses. They’re actually reducing headcount. I mean, that’s just, that is simple math: Revenue up, headcount down, margins up. And I think we’re going to see a lot of lot of that across the tech sector.

Castleton: So, exciting stuff to come. A lot of the listeners here are mostly in that financial services world. And so as we think of AI, you know, taking our jobs, I don’t know that that’s the case here, but how do you see in our particular sector AI having the most amount of impact?

Fish: Yes, I can talk about how it’s really having an impact for us and it’s really in the research process. I mean, the ability to use things like a ChatGPT-type tool, whether it’s ChatGPT or not, to do deep research, to prepare for meetings, to aggregate information from expert transcripts, earnings releases, management calls, filings. We can just do it a lot faster, a lot more efficiently, and we can connect the dots in a different way than we have historically. So, it’s really enhancing our research process, which is great.

I think just broadly speaking, across financial services, we’re going to have unique products that are automated and tailored for individuals. Over time, it’s going to make advisors more efficient. Our organization’s just going to get more efficient in terms of content creation and how we approach media and things like that. And so, it’s an efficiency gain. We’re going to see it across all sectors of financial services. And I think it also then ultimately gives a lift because it’s more inclusive for people as well.

Castleton: Nice. Well, who knows, maybe in a year these AI avatars will be replacing us for this conversation.

Denny, that was awesome. Thank you so much. A lot of exciting things to still come, obviously in technology and then AI broadly, very exciting stuff.

Thank you all for listening. We hope you enjoyed the conversation. For more insights from Janus Henderson, you can download other episodes of Global Perspectives wherever you get your podcast or check out our website at janushenderson.com. I’m Lara Castleton. Thanks. See you next time.