Polar Capital Technology: the AI stocks I’m excited about in cognitive era

Thank you.

Well, it's been around for a long time. It's very carefully managed. It's the flagship fund of the company that we look after at Polar Capital. And I've been managing it now for what feels like a very long time, but 20-plus years. And what distinguishes it, I guess in the end, beauty is in the eye of the beholder, but what we love about it and how we've positioned it is a very broad church product. It's a technology fund that's really focused on, I suppose, all of the themes that happen within the technology sector, and it has done that for a very long time. So there are always very fascinating sub-things happening at any given moment in the tech sector. Let's say robotics, or earlier it might have been the cloud or software as a service, and we've tried to capture those sort of smaller themes within a sort of broad church technology product.

That's certainly been the history of the trust since I've been managing it. Having said that, what's unique about it today is actually quite different, in that we made a very bold or big decision at least in May 2023 to pivot very hard into the AI thing. We were extremely excited about the AI story. We were waiting for that moment where it would, I suppose, really permeate hard into or be adopted or really happen. We launched a fund actually outside of the trust to focus on the AI story earlier. So ChatGPT arrives, and we decided at that point that this was one of those rare moments — a general purpose technology that was about to take off. We actually collapsed the way that we'd been thinking about the trust into those sort of six to eight core themes and actually focused very heavily on the AI story, which, so far, so good. We've really seen very rapid adoption of the AI story. So right now, it's an actively managed trust. It has 90 stocks or thereabouts, so it's well diversified, but it's very focused on the AI theme and trying very hard to avoid the other side of the trade: the disrupted companies we believe could fall into the shadows, if you like, as AI continues to evolve.

The numbers are a movable feast. I would say the vast majority of the trust is directly exposed to continuing progress of the AI story. And when I say progress, I mean models continuing to improve, capital spending — capex — continuing to be revised higher. You know, we're at a trillion dollars or thereabouts now, from zero four years ago, and continued adoption of the technology in various different forms. And so the vast majority — it could be two-thirds, it could be 80%, depending on how the portfolio is positioned. On the other side of that, I think it's worth just pointing out that yes, the disruption story has again also continued to evolve. Because if you go back just three years, the transformer model — the architectural underpinning of large language models — was producing a sort of probabilistic and error-laden output. And so at that point, it was very easy to dismiss it as, you know, interesting, but ultimately not going to disrupt anybody.

All technologies begin as complements and they end as substitutes — or at least the important ones. And we believe and continue to believe that that's still true. And as the technology has got better, as the output's got better, as the models have got bigger, more capable, and when you see the error rates of these models collapsing, so the scope of disruption has increased. And I think it's a really important point to make, if I may, which is that this idea that you can coexist for a while means that you can be, as an investor, lulled into thinking that a company is not on the wrong side of something because, at a lower level of capability, it's been fine. But as the AI models have continued to move, and now we are at points where most of the frontier models are already well beyond human capability. Which means that, in our opinion, what started with sort of weaker application software companies that have effectively been sold just by using a large language model has crept into other parts of the market, like information services businesses that we've been actively avoiding now for a while. Companies in the software space that have become also sort of embroiled in a debate at least about what their long-term value prop is. Because in the end a large language model is already looking capable of doing a lot of what software does today. And then longer term, I think you have to start thinking about — and this is where we believe the current debate is now moving — is what the businesses look like in a world where the incremental interaction with a business isn't by a human, but by a machine, by an agent.

And in that world, what place does advertising have? What does advertising mean in that world? Who are you advertising to? And for example, if you're aggregating traffic because you're, I don't know, you're in the hotel business and you've been sitting between consumers that have been sort of overwhelmed with choice, and hotel companies that are looking for customers or an audience — those businesses have been fabulous in a world where humans have ultimately been booking hotels. But what if the incremental hotel booking is a machine that doesn't respond to advertising and isn't attention-scarce? So, sorry for a long-winded answer, but I think it's really important to understand that this is a movable feast, that as model capability continues to improve, so we expect the shadow cast or formed by AI to continue to broaden. I think you have to be absolutely alive to that risk within a technology portfolio but also in a broader equity investment portfolio.

Yeah, we're actively avoiding the software space, and I mean that's not to say that there aren't pockets of interest. Of course, ever more capable AI models in the wrong hands can be ever more capable sources of cyber attacks, for example. So we have exposure in the portfolio to the cyber theme, as you would imagine. Longer term there are some question marks even there. You know, if the large language models are so capable, why couldn't they at some point be productized to perhaps compete with some of the cyber companies? But right now the attack surface and the capabilities in the wrong hands mean that companies will invariably, I think, be spending more on cyber going forward. Same thing applies to companies like Snowflake, for example. Again, it's not an individual recommendation of a stock; it's one of our 19 names. But that's a company that seems to be benefiting very well from just the idea that, you know, companies are needing to get their data estates in order, and with that data plus large language models bring incredible outcomes, and so Snowflake benefits.

So there will be pockets in the software space that we're excited by, but I just think as a rule this idea that productivity software is ultimately a tool for humans to be more productive, but AI is becoming a substitute for human cognition. And I think that, again, that's a big statement to make but that is how we see it. That's why we believe this is one of those kind of key moments in life or in history where non-human scaling takes place. We saw it in agriculture, we've seen it in the industrial revolution, and I think we're seeing it now in the sort of cognitive era.

Look, there are lots of risks to the AI story, and I think that, if I may, it's worth just highlighting that AI news cycles are always characterized by volatility. Just recently, we've heard about companies at the frontier talking about wanting to slow down potentially the pace of development. And again, before that, we've had plenty of other sorts of headwinds to the AI story in the form of, I suppose, skepticism around revenues or what returns people are going to get on this capex. So I think that if you're involved in the story, or if you're as excited as I am in it, you should still be guarded against the prospect of volatility. There's going to be buffeting. In fact, the summer has been a pretty tricky time for a number of the sectors, and the semiconductors are really amongst some of the heaviest hit. The memory stocks have been really a focal point for investors because, you know, memory being kind of a commodity product that in the smartphone or in the cloud world there was plenty of it, and so essentially the industry was a price taker. We've moved into a world now where memory, particularly high bandwidth memory, is a critical component in the training or building of compute clusters that train AI models. And going forward, as we do more stuff with AI — so inference workloads — the memory that your AI agent has of prior information that you shared with it may be stored in something like NAND flash. So the products that were really commodity components in a previous world seem to have taken on a completely different level of importance today. And as a result of that, the prices have moved meaningfully higher for those products. The industry is a price giver, not a price taker, and the stocks have responded.

Since the summer, a number of those companies, particularly ones in South Korea where there's been lots of retail involvement in the stocks, have been under quite sustained pressure. So how are we thinking about it now? I mean, I think the market, investors generally, haven't seen anything like this before. I think we were chatting before we started recording just about the idea of historical parallels helping to perhaps illuminate where we are, because it is an unusual moment. I really do believe that. And I think the market — you know, we look at the forward P/E that people are putting on stocks like Nvidia, let alone companies in the memory area that I've referenced. I mean they're very modest forward P/Es, and I think that speaks to the idea that the market is very skeptical that the growth that we've seen so far can persist. There's a real question mark about the duration of this trade because it's very unusual for an industry to go from spending nothing to a trillion dollars. We believe that the cycle will persist, and therefore we've been holding and adding to actually some of our semiconductor holdings.