Investing in AI's next chapter

Jeremy: Everyone these days seems to have an opinion on artificial intelligence. Some believe we're in another technology bubble. Others think we're only at the beginning of one of the biggest investment cycles in decades.

So, to separate the hype from reality, two of Investec's global equity analysts have just returned from Boston and San Francisco, where they attended some of the industry's biggest conferences: J.P. Morgan's Global Technology, Media and Communications Conference, and Bank of America's C-Suite Technology, Media and Telecom Conference.

They spent several days speaking to executives from many of the world's largest technology companies about where AI is going. So, what did they learn?

Hello, I'm Jeremy Maggs, and you're listening to No Ordinary Wednesday, an Investec Focus Radio podcast that tracks the macro moves shaping our world today.

Joining me from London are Investec global equity analysts David Smith and Zane Bezuidenhout. Gentlemen, welcome back to the program.

Jeremy: All right, David, when you and Zane joined me on the podcast, if memory serves, last October, you argued that AI still had a long runway as an investment theme. So having spent some time now with some of the world's biggest technology companies, as I referenced at the start of this interview, I'm wondering if anything has changed. Is demand for AI still as strong as you expected? 

David: Jeremy, to answer your second question first, it absolutely is as strong, if not stronger, than we had expected in October. The opportunity is bigger, and it's coming way quicker than we expected. Google, for example, has just reported 82% growth in its cloud division last night. This was largely driven by the demand for AI products.

Anthropic, which has been in the news a lot recently, is looking at a $70 to $75 billion recurring revenue run rate at the moment. I'll give you context, when we spoke in October last year, it was about $5 billion. Four years ago, they didn't make a single cent in revenue. The pace of acceleration has been quite extraordinary. 

Jeremy: So Zane, to you now. If demand isn't necessarily the issue as far as this is concerned, then I'm wondering what is. We hear a lot about shortages of advanced semiconductors, for instance, but it would sound as though building the infrastructure to support AI has become a lot more technical, a more complex challenge. What's your reading of the environment?

Zane: Absolutely. From the conference in San Francisco, I think there was a common thread through all the executives that presented, was that given that, as David alluded, that demand’s got out of the starting blocks at a rapid pace, the supply chain, maybe once bitten and twice shy from past boom-bust cycles, was a lot more skeptical and cautious and has now been convinced that AI demand is real.

But there is a great degree of catch-up that is needed from the supply chain, whether that's across the advanced chips, memory chips, the networking side which is connecting the chips, and then, you know, demand keeps on moving. And the big impulse late last year, early this year, has been agentic AI, which we'll get into.

And this was all never mind the constraints around power and energy that is needed to fuel this demand growth. So definitely the scale of demand has outstripped supply, and supply needs to scale rapidly, and the complexity of it is ever-increasing. 

Jeremy: So David, let's pick up on that. So much new terminology these days. Agentic AI is obviously the new buzzword. For those of us who are still getting our heads around the term, maybe just tell us what it is and why the industry is so excited about its potential.

David: The best way to think of it is as an online employee. In your personal capacity, the best way to frame it is probably think about someone like Jeff Bezos or Elon Musk, who's a multi-billionaire.

They will employ 10 to 20 people to run their lives behind the scenes, from their finances, to their health, to their diaries, organizing household chores, food.

In 5 to 10 years, if we are right in our view in how agentic AI play happens, we think everyone will have access to the same opportunities to have someone run those chores for you, just this will be done with AI rather than people and will cost a lot less.

In a work capacity, it's an employee that you can ask to do a task for you that never sleeps. You will need to manage them, of course. You'll need to direct them and check them and make sure the output's what you wanted.

But each person is probably going to have the opportunity to have a veritable army of digital employees working for them, and you will be the manager of those employees.

Jeremy: So when do you think this capability is really going to start to kick off? We really are on sort of the slope period right now. A long way to climb?

David: Yeah, you’re spot on. So, we met with the CFO of OpenAI fairly recently, and her view was they think it's going to happen soon. We're talking six to 12 months for use cases that are genuinely value-adding and potentially can act on your behalf.

My expectation is that definitely in two to three years, we will have a meaningful step-up in the value that agents can provide, and it's going to be fairly broad-based in terms of adoption. So somewhere between six months and three years would be my best guess. Hopefully sooner rather than later. 

Jeremy: Zane, a year or so ago, I think it was all about building bigger and more powerful models, and that's fairly understandable. But people are now talking about “intelligence per dollar”. What does that tell us about where the industry is heading? Intelligence per dollar. Rolls off the tongue nicely.

Zane: Yes, this industry's full of a lot of the unique acronyms and phrases, but yeah, it's basically a way of saying, "Are you getting bang for your buck in terms of AI compute?"

And I think for us it's a very rational and healthy signal that people are now focusing on this, what's also called “token optimization”, as opposed to previously we've been in a period of what was called “token maxing”. And I think token maxing is just where people were encouraged to go and spend as much as they can in terms of their IT budget on input tokens. That's queries or prompts, and then what you receive back from these large language models being output tokens.

And I think it's a signal that we are shifting from experimentation, playing with this. I think it shows that enterprises have identified use cases, and now it's around optimisation in terms of making sure that these tokens and the cost associated with it is directed in the right direction.

I mean, it pretty much means you don't need a powerful foundational model for every query. It's that ability to flex between low-cost models or free models for maybe trivial queries or prompts, and then you use the powerful models when, you know, super intelligence is ultimately required. So for me it's a very healthy signal that we are making that transition from experimentation to implementation ultimately.

Jeremy: David, let me circle back to that bang for your buck then. For the past two years or so, we've heard about the enormous sums being invested in artificial intelligence. So to push on a little further from what Zane was saying, do you think companies are now starting to see measurable returns, or is this still very much an investment story?

David: We are seeing genuine returns. Obviously, it's been particularly amazing for the infrastructure layer, which have been direct beneficiaries of all the spend. But the cloud providers, those who are spending a lot of the money, are seeing an acceleration of demand from their clients.

They're not able to keep up with demand, that is consistent, and that wouldn't be happening if their clients weren't seeing value for money in using the cloud service providers.

So, we obviously have particular use cases where it's very easy to see value, right? So you’ve see massive upside to people who want to code. The effectiveness of online advertisers, like Google or Meta, has exploded in terms of what they can extract from on a revenue basis. Customer service - business lines have changed fundamentally. The chances of you speaking to a person is quite small on your first call into anywhere.

And we are starting to see operational businesses, things like you need to go check this bit because it's on security or compliance or legal. Those bits of the businesses are being impacted, and quite rapidly. 

Jeremy: So, Zane, if companies then are beginning to see real returns, as David has alluded to, what do you think is going to separate then the long-term winners in AI from the rest? And maybe more importantly, how should investors be thinking about that?

Zane: Yeah, so a framework that has helped us, you know, in terms of our analysis and thinking for the long term is we ideally are looking for companies that can be the bridge between the infrastructure layer, that's the compute, and basically the end use, which is workflows or even personal use.

So, some examples with that could be in enterprise software, e-commerce, payment providers. So we're looking for these companies that can serve as a bridge. But being a bridge isn't enough on its own, and a way that we look at it is, I think in a fast-changing environment, I think customers, whether they're enterprise or, you know, small-medium businesses, are looking for almost one-stop shops in the provision of these services.

So we would call those platform companies, where they have a portfolio of products which they can essentially wrap around their customer. And ideally, I guess a holy grail would be companies that have a degree of control over their ecosystem or a degree of network effect. So this is the interplay between developers on their platform as well as the end customers.

It almost becomes a network of systems providers. And an example of this is, you know, which we favour, is a company like Microsoft as an example of a company with strong ecosystem control. 

Jeremy: On Investec Focus Radio, you're listening to No Ordinary Wednesday. Today, Investing in AI's Next Chapter. My guests are Investec global equity analysts David Smith and Zane Bezuidenhout. Gentlemen, hold all those thoughts. We'll be back to the conversation in just a moment.

Many of the technology companies mentioned today feature in the Investec Global Leaders Portfolio. The portfolio invests in 30-50 high conviction, global quality growth companies with enduring competitive advantages, taking a long-term approach to capital growth. To find out more, visit investec.com.

The Global Leaders portfolio invests exclusively in equities and carries a high degree of risk. The value of investments can go down as well as up, and investors may get back less than they invested. Past performance is not a reliable indicator of future results. 

And welcome back. This is No Ordinary Wednesday. Now, David, one topic that's impossible to ignore is China. So is the AI race still America's to lose, or are Chinese technology companies, do you think, beginning to close the gap?

David: We think the USA is likely to stay ahead on leading models, but China has some big advantages, and that's generally around access to energy.

They don't suffer from the term of NIMBY, which is “Not In My Backyard”, which is playing out in a lot of the developed markets, in the US and Europe in particular. They have a lot of government support, and they don't have the same level of guardrails or regulation that's starting to show its head in other parts of the AI world.

Jeremy: So Zane, back to you. And last week, OpenAI disclosing that one of its autonomous AI agents independently hacked into another company's systems during testing. So, incidents like this would obviously increase the likelihood of tighter regulation. What do you think that might mean then for the pace of AI innovation?

Zane: A very scary incident, and ultimately already demonstrates a pace at which these large language models are developing. And directionally, which I think is probably consensus, is that regulation is naturally going to increase. That being said, we’re of the view that given the speed and change of innovation, that typically regulation follows innovation.

But we do view it as a necessary step to safeguard use and provide the necessary guardrails that will ultimately help long-term adoption of artificial intelligence. 

Jeremy: David, let's get back to the investment case. When most people think about AI investing, they immediately think of Nvidia, of Microsoft, or Alphabet.

After everything that you've seen on this trip to the United States, do you think the opportunity is beginning to broaden beyond the Magnificent Seven?

David: It's been interesting, 'cause the best opportunity this year has by far been in the bottlenecks in the AI data centre build-out. Think of things such as memory players or fibre optics. They have absolutely rallied. They've had a stormer of a year, where the big large names that you've mentioned have actually had a pedestrian year would be quite generous. They've actually had a pretty bad year so far.

So with that, we think that where we stand today, and because of valuation, which we want, the best risk-adjusted return, and that doesn't mean you're going to have the best share price, but the best when you adjust for risk, we think sits with the large players who have what Zane referred to as having control over an ecosystem and having a platform or an ecosystem effect. So, we actually really like those large names. We think that investors will be very well rewarded to own them over a long period of time.

But, I have to say that on a slightly longer-term time horizon, we think that robotics or physical AI is likely to be the major opportunity in a few years' time. It's just probably not there yet. 

Jeremy: Zane, a little earlier you were talking about, uh, regulation. So- Beyond cybersecurity risks and tighter regulatory environment, maybe a view on what you see the biggest threats to the AI investment story are right now.

Zane: Gosh, there are quite a few curve balls that have potential, and I think that's where you see investor behaviour be quite skittish in markets, which is natural given, you know, this is a new and fast-developing technology.

From the demand side, I'd say a key risk is the rate of improvement from the large language models, whether that's out of the US and China, whether there is a diminishing return to the improvements in these models. That could be a potential risk.

On the supply side, and you know, Dave alluded to the rollout of data centres where there's obviously the unpopularity of having these data centres built in people's, you know, call it backyards, as it were.

And a very realistic constraint in the short term is around power supply, energy supply, which is linked to the unpopular view with a lot of consumers potentially experiencing high electricity bills off the increased electricity demand.

Another area, you know, and kind of delves into our space of capital markets is around access to capital and the behaviour of investors in the market. It's something we always keep an eye on, whether we see irrational behaviour rearing its head within the public market space. 

Jeremy: And David, finally to you, there's no doubt that both you and Zane have come back from this visit with a head full of information. But if there's one message that maybe investors should take away from this fact-finding trip, uh, what would it be?

David: That we are very, very early in the AI cycle. There's a massive runway for the entire ecosystem to benefit, and this is probably the most transformative technology we are going to be exposed to in our lives. So we don't think it's one or two years, we think this could be decades' worth of growth.

Jeremy: And that's where we are going to leave it. To both of you, thank you very much indeed, and we look forward to welcoming you back soon, whether in person or via your agentic counterparts.

Now, before we go, just a quick favour. Thousands of people listen to Investec Focus Radio every month, but many haven't hit the follow button. So, if you enjoy these conversations, please follow Investec Focus Radio on Spotify, Apple Podcasts, or YouTube podcasts. It's the easiest way to make sure that you don't miss an episode. Until next time, goodbye.

Disclaimer: The views expressed are those of the contributors at the time of publication and do not necessarily represent the views of the firm and should not be taken as advice or recommendations. Investec Limited and subsidiaries, authorised financial service providers, registered credit providers, and long-term insurer.