AI revolution: decades of innovation behind the 'overnight' success

As with any investment, your capital is at risk. Past performance is not a guide to future returns.

Stewart Hogg: Hello, my name is Stewart Hogg, and I am an investment specialist in the Long Term Global Growth Strategy. Now, AI is a topic that everyone is talking about right now, especially with the invention of ChatGPT just a few years ago. But AI is something that we've actually been thinking about and investing in in long-term global growth going back a couple of decades.

Back in the early 2000s, for example, one of our first investments in Amazon, and they were using machine learning algorithms to predict book recommendations for their customers. Now, Jeff Bezos definitely knew the potential of this technology early, so much so he was asking his Amazon executives every year to answer the question, how do you intend to utilize machine learning in the business?

And at the end of the day, AI is just a tool that helps businesses make better predictions. And there's such huge potential because better predictions translate to better business outcomes. whether that's trying to predict a product recommendation to prompt a purchase on Amazon, whether that's predicting what show a user will binge on Netflix to help them stay more engaged, or in Tesla's case, how to predict how a human drives to achieve autonomy.

Now, artificial intelligence is nothing new. It goes back a couple of decades in the Long Term Global Growth portfolio, but actually it goes all the way back to the 50s in terms of how it was studied as an academic discipline.

In 1956 in Dartmouth College, you had computer scientists like Marvin Minsky and the engineer Claude Shannon get together at this research project to start to study AI as an academic discipline. Now, since then, there's definitely been periods of exuberance. There have been periods of despair, so-called AI winters, as they are now known, until now, in the current era that we're in. And I think there's general feeling that this could be the next technological revolution.

Now, the interesting thing about technological revolutions is that they have been occurring for centuries. Now, this is a piece of work by Professor Johan Schott and also Carlota Perez, where they wrote about technological revolutions over the last, say, 250 years. And you can see in the bottom left here, you've got the Industrial Revolution that then saw the rise of steam and railways. Steel and electricity made way to oil and mass production before developing into the information age we live in right now.

Now the thing consistent with each of these technological revolutions was that they lasted for many decades. Now the most recent information age that we're in right now harks back all the way 50-plus years ago to the 70s.

Now, the interesting thing about each one of these technological revolutions is that there was a new general-purpose technology that came along where a key input was falling in price pretty precipitously, and there was a suitable infrastructure to help distribute it. So for example, you know, steam and railways, it was all about coal reducing in price, and the railways were the infrastructure that allowed that to be distributed. More recently in the information age, We've definitely benefited from the falling cost of compute combined with the infrastructures to distribute it such as the internet.

Now, the important point with each of these revolutions is that what came before set the conditions for what came next. And if you think about it, that makes a lot of sense. Without electricity, there would be no computers. Without computers, there would be no internet. Without the internet, there would be no widespread use of AI.

And the thing that has really driven forward the information age has been Moore's Law. And that is depicted in the chart on the left-hand side here. Now, this was this observation from Gordon Moore at Intel that every couple of years, we'll be able to cram more transistors on a chip, making them more powerful whilst they continue to fall in cost.

And you can see that this relationship has held for the last half a century or so. And this has driven computing platform shifts that we've tracked over time. So people may remember the mainframe era of computing. That then made way for personal computers start appearing on our desks. And that has then made way for smartphones and phones appearing. And now we're at an era where really chips are pervading all sorts of different industries.

Now, the interesting thing is that the amount of compute used in machine learning systems at the same time was actually doubling at a pretty similar rate, as you can see on the right-hand side of this page. And that is defined as the first era on the right-hand chart.

But the really interesting thing is that something interesting happened around 2012. And what you saw was that this progress seemed to accelerate. Now, instead of a two-year doubling, the amount of compute doubled every three to four months. And the big change or discovery back then was actually an image classification competition around 2012.

And the interesting thing is that instead of using traditional central processing units, CPUs, they actually used GPUs to train these neural networks. And what that basically did was accelerate computation time and the result drastically improved. And what this did was kick off this new trajectory in training compute performance, which we'll come back to.

Now, many of you may know who this is in the image. This is chess grandmaster Garry Kasparov being challenged to a match in the 90s by IBM's Deep Blue computer. Now, the interesting thing was, even though this was seen as an impressive feat of computing intelligence back then, actually it wasn't really. What it really highlighted is that sheer brute force computing power could outsmart a grandmaster if you had enough of that computing power.

Now, interestingly, in 2016, something similar happened. Again, a computer outcompeted a world champion again. In this case, it was in the game of Go. Now, the difference with the game of Go is it's infinitely more complex than chess. After two moves in chess, you have 400 possible moves. After two moves in Go, you have 130,000 next possible moves. In fact, there's this amazing stat that there are more configurations in the board in Go than all the atoms in the universe. Now, the interesting thing is the result was made possible because of that accelerating progress in the last 10 years that I mentioned in the last slide. And it was this progress that interests us as long-term investors. And we observed this progress, and we also observed that the chips that were making this possible were produced by one company, and they had 90 per cent market share. And that was NVIDIA, and that's why we first got interested in the business back in 2016.

Now, the interesting thing back then, [is] that the investment case was very much predicated on the virtual reality and augmented reality opportunities. The automotive part of their business was still nascent, but potentially large. And there was a small, very nascent part of their business called the deep learning opportunity. It was really hard to grasp what was possible, but we definitely had an optimistic case that could be significant in size. Another part of the investment case that is incredibly important to us was the culture of the business.

Now, the CEO, Jensen Wang, who's been running the business since the early 90s, he is very well known at investing very astutely in what he calls ‘zero-billion-dollar markets’. Now, it's his fun way of saying that, yes, the market might be zero billion dollars right now, but it could be absolutely monumental in size. And that is what he's been very astute at doing over the last 20 or so years.

And then so we had the starting market cap of about $18bn. And one of our analysts at the time was actually pushed to go back and research further because we didn't think her upside case was being optimistic enough. In fact, in our research piece, she actually said, “I'm coming up with an upside case here, but I think it's bordering on science fiction.”

However, one of the interesting things she said in that research report was that deep learning opportunity and the applications of that technology were seemingly endless. So what she did was come up with an ultra-optimistic, actually 25x upside case, super sunny blue sky, and that was a $500bn company by 2026. The rest is history.

Everyone knows what the share price is right now and the market cap. Not even eight years later, the company is worth $3.5tn. And what that told us, and we've had experiences of this over the period of running long term global growth, is that we tend to underestimate how great, great growth companies can ultimately be.

You can even see there throughout 2021 and 2022, it had a 50 per cent drawdown throughout that time. So our investment process is very much centred on finding the right company with huge potential and being very patient. So the interesting thing is, but where does NVIDIA go from here?

So one of the things we have within Baillie Gifford is something called Sidekick, and that allows us to interrogate certain data stores. And we have a research library going back the past 30 to 40 years, that institutional knowledge all in one place. So I thought, you know, let's ask Sidekick what our investors are thinking in terms of the long-term case.

So here I am, tell me what the most optimistic upside case for NVIDIA is from here. And interestingly, Sidekick came back with, well, AI systems could achieve genius-level capabilities across every domain. Rather than simply replacing humans, AI could become an intellectual force multiplier, allowing people to operate at previously impossible levels of sophistication and creativity. And then finally, NVIDIA's role in this transformation could be foundational.

And so the question really is, you know, in the next five to 10 years, could NVIDIA be a $15tn company? Could it be even higher? And I think it's really important that we continue to think positively and optimistically about where companies can get to.

Now the way we invest in Long Term Global Growth and at Baillie Gifford more generally is we're trying to stretch our time horizon. We're trying to envisage what the world looks like in 10-plus years. And we try to do that because there are not many other investors that are thinking over that time horizon. And we also know that progress doesn't happen overnight. Company progress takes time. And I think what maybe a lot of investors do and succumb to is you tend to overestimate what happens in the short term, but tends to underestimate what happens in the long term. And I think NVIDIA is a really good example of that. That progress did take time.

And so this is what we think the world looks like in 2035. And these are the themes that are most reflected in Long Term Global Growth at the moment. Now, AI is definitely pervading every industry around the world. We're seeing impacts of that already. But I think the biggest impacts are still going to be felt over that next five- to 10-year period. And what I'm going to show you today is just a few examples where we're seeing that already. But the interesting thing is it may not be the most obvious of areas and industries.

Now, one area where we can continue to see AI being used effectively is in e-commerce. Now, to many, this is pretty well understood. It's understood to us as well. We've been investing in Amazon over the last 20 years. We’ve had a number of other e-commerce businesses in the portfolio. It's this idea that clicks and orders, as in online commerce, is increasingly displacing the bricks and mortar way of commerce. And what we're seeing is penetration rates still incredibly low in certain parts of the world. So we still think there's a huge growth opportunity ahead.

In fact, back in, I think, 2007, we were chatting with Jeff Bezos at the time. And one of the things he talked about was the raw materials in his business becoming significantly cheaper and more powerful over time. What he was talking about there was computing power and that power of Moore's law, making his business more efficient over time, where compared to say a Walmart, where if you wanted to grow, you needed to put significant capital investments in there. And that's really been a key tailwind for those types of businesses.

But another area where we're seeing AI transforming industries is in the world of advertising. Now, just think back to the days of maybe the 60s and 70s, if you were in Madison Avenue in New York. If anyone has watched Mad Men on television, that's a really good summary of what it used to be like. This was the era of broadcast media.

Now, back then, that is when advertising spoke to the masses rather than necessarily with the individuals. And that makes a lot of sense. Brands would place ads on billboards. It could be magazines. It could be TV commercials. But it was very much a one-way relationship.

Now, the advertising giants back then relied on their wit, their intuition. They created these campaigns, and they would hope they would resonate with consumers. And what they did was pay for pretty expensive advertising space where they thought our eyeballs would be.

Now, it wasn't until the 2010s and 20s where we started to see this seismic transformation happen. And that was reshaped by digital technology, huge amounts of data, and the power of connectivity.

Now I'm using Netflix as an example here, but actually they're relatively new to the world of advertising. If anyone is a subscriber and is using the advertising tier, the real behemoths of this type of era are companies like Meta and Alphabet. And they've been using advertising very astutely to drive their businesses for the past couple of decades.

Now in this era, this is an era of personalization. Now, it's far more important who sees the ad compared to what the ad actually says. And what we're seeing is AI being used incredibly effectively in this area.

A company that we invest in is The Trade Desk. And it's interesting that their software analyses 900 million ad impressions per minute, and it matches the right advertiser with the opportunities that are going to generate the highest return for their given budget.

Now, The Trade Desk use these predictive engines to automate ad decisions. There's a huge amount of data helping them make hyper-personalized ads to each individual. And we think the opportunity is vast. And really, it's the opportunity is everything outside the so-called walled gardens of meta and alphabet.

At the moment, it's mostly in connected TV, but they now have tie-ups with Netflix and Spotify. So if you're starting to hear advertisements in any of those platforms, it's likely going to be down to The Trade Desk technology.

Now, another area where we're seeing interesting developments is in the world of healthcare. Now, one of our contentions is, you know, what if healthcare becomes codable over time? Now, we're used to programming in zeros and ones, you know, in computing, we've been doing that for decades. But the interesting thing now is you can code with the building blocks of life.

Now, those building blocks is our DNA. And one area where we're seeing interesting developments is actually in the surgical setting. And the reality is that most surgeries are still mostly conducted by humans. And this is a pretty traditional picture of a surgery here. And interestingly, more than 95 per cent of surgeries are either open surgeries or minimally invasive laparoscopic keyhole types of surgeries.

Now, the interesting thing and the issue, unfortunately, with these types of surgeries is that this can tend to lead to longer recoveries, greater blood loss, scarring, tissue trauma, etc. And that is in comparison to robotic surgeries.

Now, a company that we've been investing in since 2009 is Intuitive Surgical. Back then, I think there was something like half a million surgeries. Now they're up to over 15 million from a cumulative perspective. And if you lived in the US and you needed to get a prostatectomy, about 90 per cent of them are performed using their robots.

Now, you can see how a surgeon would operate the device. And this is the one that you can see here is actually the new device that has been benefiting from a 10,000x uplift in computing power since the last version. Now, what does that mean? Well, one of the benefits is haptic feedback. Now in the past if you were a surgeon you may be somewhat wary of using robots for certain procedures because say you were operating on a tumour and you needed to know whether it was a hard or a soft tumour, because if you cut into it and it exploded, you do not want cancerous cells going into the operating cavity.

But the interesting thing about this new system was that there is now haptic feedback. It allows the surgeon to feel the force they're actually applying, similar to using their fingers. And these advances are going to help with bruising. They're going to better identify different types of tumours. And the real advances are going to come from the vast data sets that these procedures accumulate. There are something over nearly 50 million procedures now. They've got 25 years of kinematic and video data. And what they're using is artificial intelligence to make that predictive ability even better.

Now, in time, in the next five years, training is going to be improved due to these technological advancements, precision and efficiency is going to improve, and who knows, we could even have completely automated surgeries over time.

Now, another area is, obviously, AI is pervading many industries, but in the world of automation, now, this is an area where we've started to have an increased exposure in Long Term Global Growth.

This image really shows you what a traditional warehouse looks like at the moment. Something that we would know relatively well. And really, it is a complex valley of human labour, as you can see on this page. The reality is that workers have to walk miles. They're doing manual picking up and packing and tracking inventory. And the reality is that that leads to errors. It leads to inefficient routing. And unfortunately, that is hard to scale.

Now what is coming in now and starting to displace traditional warehouses is automated warehouse. And I think a really good way to think about it, it really is like a symphony of robots within these warehouses now. And when you look at some of this technology, it does look like Tetris meeting Formula One.

Now, a company that we invest in is Symbotic. Now, Symbotic provide kits to allow warehouses to become more automated. Each one of their platforms costs around $50m, so they're definitely not cheap. You get 400 autonomous robots that can zoom around at 25 miles an hour. and it throws off huge amounts of data, something like 10 terabytes per day.

Now the interesting thing is that companies like NVIDIA have chips that allow them to orchestrate this effectively. And so ultimately this differentiated technology is driven by machine learning algorithms being able to understand where each of these robots are at any one time. And that is powered by NVIDIA graphics processing units and NVIDIA chips. So there's another example of where NVIDIA could benefit over time. And the benefits are very clear.

Existing pallets or moving boxes around the warehouses when it's done by a human costed about 55 cents per pallet. Now the cost reduces to something like five cents. And so it's understandable that Walmart in the US are going to be retrofitting all 42 of their distribution centres with this technology. And you can see there's a huge potential for other businesses to automate too.

Now, the final area I want to talk about is maybe one of the most obvious areas, it's in the world of chips. And those are obviously the picks and shovels of the AI revolution. And really it is standard to facilitate the AI revolution, chips are vital to that. And it's very clear that there's a huge opportunity set there. Over 1 trillion chips are produced every single year, and that equates to about 140 per person. And so as a result, this opportunity set is obviously incredibly appealing for competitors. And what you're starting to see is some disruption and hopefully competition coming in from the public markets, but also in the private markets.

Now, we've been investing in private businesses since the early 2010s. We have about $10bn of our clients' capital invested in about 150 private companies. And one of them that we're really interested in at the moment and been investing in is a business called Tenstorrent.

Now they are very much a chip manufacturer similar to what NVIDIA have been doing over the last 10 or so years and are really trying to disrupt that value chain of ASML, TSMC [and] NVIDIA. Now, what Tenstorrent are trying to do is bring a differentiated strategy in terms of integrating AI and CPU platforms. Not many chip companies have tried to do that. And in terms of their software stack, they believe in completely open sourcing that, rather than say NVIDIA, which is locked down, you have to have and use their specific CUDA software platform.

The interesting thing for us is we're always trying to follow industry luminaries and visionaries, and we think the CEO Jim Keller of Tenstorrent, is definitely someone to back. He's got a huge amount of experience, and it's interesting that he was at Apple when Apple were first trying to produce their first proprietary chips that powered their iPhones. He was at AMD, instrumental in the production of their Zen chips, which ultimately revitalized AMD as a business. And he was also at Tesla, instrumental in producing the first autopilot chip as well.

So we think the next generation of challengers, especially in the chip industry, is going to come from other public companies, but also private markets. So we think it's really important that we're there.

So maybe in conclusion, I think everyone is excited about ChatGPT and artificial intelligence, but that is just one application of artificial intelligence. Progress is happening, but we always know that these rapid changes, or so-called rapid changes, still happen incredibly slowly. So it is important to have a long-term mindset because we know that progress takes time.

We know that AI is going to pervade every industry, and you've seen a couple of examples there. The Trade Desk in advertising, Intuitive in the healthcare space, Symbiotic in industrial automation or warehouse automation, intense Tenstorrent in the chip industry. And I think it's very clear that AI is not a tide that's going to raise all boats automatically.

You will find companies use AI in more interesting and innovative ways. And so we still think that a few big winners are going to prosper in the stock markets as a result of artificial intelligence. Now, that's all for me. Thank you for listening.

Risk factors

This communication was produced and approved in January 2025 and has not been updated subsequently. It represents views held at the time and may not reflect current thinking.

The views expressed should not be considered as advice or a recommendation to buy, sell or hold a particular investment. They reflect opinion and should not be taken as statements of fact nor should any reliance be placed on them when making investment decisions.

This communication contains information on investments which does not constitute independent research. Accordingly, it is not subject to the protections afforded to independent research, but is classified as advertising under Art 68 of the Financial Services Act (‘FinSA’) and Baillie Gifford and its staff may have dealt in the investments concerned.

All information is sourced from Baillie Gifford & Co and is current unless otherwise stated.

The images used in this communication are for illustrative purposes only.

Financial Intermediaries

This communication is suitable for use of financial intermediaries. Financial intermediaries are solely responsible for any further distribution and Baillie Gifford takes no responsibility for the reliance on this document by any other person who did not receive this document directly from Baillie Gifford.