To say that AI will start doing what it wants for its own purposes is like saying a calculator will start making its own calculations.
We are sometimes asked whether we haven’t made a mistake by not classifying Artificial Intelligence (AI) as a megatrend. As you are probably aware, underneath the seven megatrends we have identified, we have also detected a number of associated investment themes and have, in this case, chosen to classify AI as an investment theme rather than a megatrend. In Exhibit 1 below, we have summarised all seven megatrends, their associated investment themes, and how they interact. As you can see, in many cases, a rather complex picture emerges.
In the case of AI, the investment theme is named The Advent of AI, and we have identified four megatrends that all affect it. In no particular order, they are:
- The Rising Gap between Rich and Poor;
- Rise of the East;
- Changing Demographic; and
- The Era of Disruption.
Before I go any further, allow me to spend a minute or two on the underlying logic why we have decided to tie this theme up with no less than four megatrends. As far as the rising gap is concerned, one could in fact argue that it works the other way around; that AI is one of the underlying reasons why the gap in wealth is rising, and there is indeed more than a flicker of truth in that. However, there is also plenty of evidence to suggest that the super-rich, many of whom are tech billionaires, are using AI to further their wealth. In other words, it sort of works both ways.
Rise of the East is an interesting one. Asia is ahead of the rest of the world in terms of implementing AI. This is partially a function of the rising living standards in Asia (hence The Rise of the East); however, it is also a function of the miserable demographics in many countries in the region (hence Changing Demographics). Korea, Japan and China all face millions of retirees within the next ten years and are preparing themselves for a material decline in workforce numbers.
Disruption is the easy one and probably doesn’t require many comments – only one from me: Unlike other technology inventions, AI will threaten the livelihoods of educated people. Therefore, it will affect society dramatically differently when compared to other recent IT inventions, where education has offered protection against unemployment.
AI is a very powerful theme and is likely to become even more powerful in the years to come. Having said that, whether we classify it as a megatrend or an investment theme won’t matter much. Investors, who can identify the winners, stand to make a lot of money, whether it is classified as one or the other. This paper will, as much as possible, try to quantify the impact of the AI revolution, and it will point you to some of the most likely winners.
One last point before I start. AI is a gigantic topic, and I could write the first 50 research papers on it. This paper will serve as an introduction to the subject and will therefore skate over various aspects. Feel free to drop me a note, if you think I have ignored one or two important attributes that you would like to hear my views on.
What is AI?
Opinions on AI are deeply divided. The bulls argue that it will have a phenomenal impact on economic growth, whereas the biggest sceptics claim it will ultimately lead to the demise of the human race. Humans will become what dogs are to humans today, the critics argue.
I find it hard to quantify that risk; however, when Elon Musk insists that the use of AI must be regulated, it is tempting to conclude that, perhaps, there is a flicker of truth in those doomsday projections (unless he does it because he is falling behind the curve). That said, I do not intend to paint an Armageddon-like picture in this research paper, as I am convinced our political leadership will do what is necessary for the human race not to become redundant.
In preparation for this paper, I spent a great deal of time looking into the definition of AI, as it is not always the same. Allow me to start with my own definition. I think of AI as software, and I agree with those who argue that AI can only be used to do things we have programmed the software to do. Furthermore, I disagree with those who argue that robots and AI are largely the same. They are not. A robot is a machine whereas AI is software. And a robot is only under the control of AI, if AI software has been programmed into it.
One of the most concise, and also one of the earliest, definitions comes from John McCarthy of Stanford University. In his 2007 paper, “What is Artificial Intelligence?" (which you can find here), McCarthy suggests that “it is the science and engineering of making intelligent machines, especially intelligent computer programmes”.
McCarthy goes on to define “intelligent”, and it becomes clear that he doesn’t think it will ever become possible to put the human brain into a computer. McCarthy’s paper is a fascinating read that I would strongly recommend you spend some time on. It is only 15 pages and can be comfortably read in about 20 minutes.
McKinsey & Company goes a step further and defines AI as “a machine’s ability to perform the cognitive functions we associate with human minds” and provides a few definitions on terms frequently used in the context of AI:
Machine Learning is a form of AI based on algorithms that are trained on data. These algorithms can detect patterns and learn how to make predictions and recommendations.
Deep Leaning is a type of machine learning that can process a wider range of data resources (e.g. images). It requires even less human intervention and can often produce more accurate results than traditional machine learning.
Generative AI is an AI model that generates content in response to a prompt. Generative AI tools like ChatGPT and DALL-E have the potential to change how a range of jobs are performed.
Why AI will change the world
The use of AI is gaining momentum at a mind-boggling pace. In 2021, almost 25% of all new patent applications were AI-related (source: builtin.com), and there are no signs of this trend slowing down. Take for example ChatGPT. It is increasingly being used in business, as credible writings can be provided in seconds and subsequently adjusted to make the text fit for purpose.
There is hardly an industry that won’t be affected by AI, although the impact will vary a great deal from industry to industry (Exhibit 2). It is a misconception to think that the IT industry will be the only industry affected. Take for example ChatGPT. Any industry that produces written material on a regular basis stands to benefit from it. The technology may be used in a sales & marketing context or in operations. R&D can also take advantage of it, and so can lawyers. In the financial industry, the tool could prove particularly useful for risk managers.
According to McKinsey, approx. 75% of the added value will fall across four business areas – sales & marketing, operations, R&D and software engineering, and the impact will be significant across all sectors in the economy. That said, as you can see in Exhibit 2, the impact will vary significantly from industry to industry. In the industries most affected, hundreds of millions of dollars of value could be added every year.
Over the next few years, the biggest opportunity is probably generative AI. According to McKinsey & Company, $3-4Tn of value (roughly the size of the UK economy) could be added to the global economy annually, once AI has been rolled out. And that number could double if the impact of embedding generative software into other software is included – software that is currently being used for other tasks.
It is impossible to predict how this will all end, and that is probably why so many find it frightening. I am in two camps on that issue. On one hand, I certainly agree that it must be regulated. On the other hand, the workforce is retiring at a troubling pace, leaving us with too many elderly to serve and not enough workers to serve them. And, before you argue that this is only a problem in the healthcare industry, think about the growing shortage of workers everywhere. AI can relatively easily address that problem.
In terms of the impact on the workforce, again, it will vary from industry to industry. According to McKinsey, as much as 70% of all work can be automated in certain industries, and I note that this estimate has recently been raised from 50%. As the ramifications of AI are better and better understood, researchers realise how dramatic the impact will be. This is obviously bad for job security but good for corporate profitability.
It goes without saying that none of this will happen overnight. McKinsey is of the opinion that approx. 50% of all jobs will have been lost to AI by 2060 – approx. ten years faster than the company’s previous estimate – but that estimate is surrounded by plenty of uncertainty. Factors such as legislation and access to capital (rolling out AI is expensive) will have a major say on the timing and pace of the rollout.
One final point worth mentioning before I move on. The impact could differ dramatically from country to country. Whereas some countries face a worrying decline in workforce numbers, others don’t, and the latter could be faced with much bigger problems. In fact, AI could prove the saving grace for countries such as Germany, Italy, Japan and Korea, where the demographic outlook is particularly punitive as far as economic growth is concerned. Other countries, such as the UK and the US, face less-worrying demographics but potentially far more troublesome AI implications.
To continue reading...
A few years ago, Darrell M. West and John R. Allen produced an intriguing paper on AI for the Brookings Institute (see here). As Messrs West and Allen pointed out, AI depends on data that is accessible for exploration, and the output is only as good as the quality of that data. That is precisely what worries me. Access to sensitive data is widespread and the management of it rather casual at times. In 2015, Tim Cook, CEO of Apple, derided competitors Google and Meta for greed-driven data mining. As he said:
“They’re gobbling up everything they can learn about you and try to monetize it. We think that’s wrong […] Advancing AI by collecting huge personal profiles is laziness, not efficiency. For AI to be truly smart, it must respect human values, including privacy. If we get this wrong, the dangers are profound.”
The output produced by most AI models may look convincing, but that is by design. Biases of different sorts can actually make the output unsuitable if not entirely unusable. A simple example comes from Harvard University. A recent study of theirs reveals that black people are 16% less likely than white people to find accommodation through Airbnb, because the AI used is racially biased.
During my preparatory work for this paper, I came across a long list of ethical issues one should worry about. I won’t bother you with them all, but some of the more serious ones would include:
1. little respect for privacy when handling the data;
2. the risk of algorithm-inflicted harm;
3. discriminatory or other biases in the underlying algorithm;
4. legal liabilities raised as the consequence of the use of AI; and
5. the risk of ‘singularity’ – the industry’s term when superintelligent machines take over and permanently alter human existence through enslavement or eradication.
While #1 is of concern, I also find it rather comical how the issue is (not) dealt with in some countries. Take for example the Americans, who are hardcore opponents of personal ID systems, arguing it is an invasion of their heralded right to privacy. Meanwhile, they are happy for the ‘sharks’ in the IT industry to collect virtually endless amounts of data on them, and that data can be abused just as much as the data the government holds on them.
#5 is the ultimate nightmare scenario. Having said that, as Diego Klabjan, Professor at Northwestern University and Director of the school’s Center for Deep Learning, says:
“Currently, computers can handle a little more than 10,000 words, [i.e.] a few million neurons. But human brains have billions of neurons that are connected in a very intriguing and complex way, and the current state-of-the-art [technology] is just straightforward connections following very easy patterns. So going from a few million neurons to billions of neurons with current hardware and software technologies — I don’t see that happening.”
I am leaning with Mr. Klabjan on that point, although I am also fully behind Elon Musk when he asks for the industry to be regulated. As Tim Cook says: “If we get this wrong, the dangers are profound.”
Only a few days ago, the Guardian ran a fascinating article on AI and how it can be abused to manipulate the outcome of parliamentary elections. Disinformation reimagined: how AI could erode democracy in the 2024 US election is the title of the articles, and I warmly recommend it. As pointed out by the author, AI lowers the bar for disinformation, and the 2024 election in the US could turn into a stern test in terms of what we should and shouldn’t accept in a civilised society. As the article concludes: “There’s not really going to be sufficient control of dissemination. There’s no shortage of robo callers, robo emailers or texters, and mass email platforms. There’s nothing limiting the use of those.”
When looking to invest in the growth of AI, you can take one of two avenues (or indeed both of them). You either invest in those companies that are early beneficiaries of the AI rollout, most of which are technology companies, or you invest in companies (in any industry) that are prepared to make the investment in AI and therefore stand to benefit over time.
However, there is a problem with both of those approaches, though. As you can see in Exhibit 3 below, investors have not exactly been asleep at the steering wheel more recently. The Magnificent Seven, most of which are early winners of the AI revolution, have performed phenomenally well this year, and I find it hard to throw my money after a handful of companies that already sell at ridiculous earnings multiples.
Alternatively, you can invest in companies in other industries likely to benefit over time when the technology is implemented in their company. According to research provided by Goldman Sachs (GS), the IT industry is by no means the only industry likely to benefit from this trend. As you can see in Exhibit 4, the analysts at GS expect the overall impact to be meaningful across the economy.
The problem with this approach is that it is still too early to say which companies that are likely to embrace the new technology promptly and which companies that are not. Admittedly, Exhibit 2 provided some guidance on that question. A portfolio of retailers and banks will probably benefit a great deal more from AI than a portfolio consisting of insurance companies and agricultural companies. However, no assurances can be granted that all retailers and banks will roll out the new technology promptly. It still depends on the action taken by each and every company.
The analysts at GS have assumed widespread AI adoption (where appropriate) over a 10-year period. By using the firm’s dividend discount model and its productivity forecasts, they conclude that US trend GDP growth will be 1.1 percentage point higher per annum over that 10-year period when measured in real terms, and that the fair value of the S&P 500 (holding all else equal) over that same period will be 9% higher. The two other estimates in the chart above – +5% (smaller boost) and +14% (larger boost) – represent GS’ worst and best case scenarios as far as the productivity impact is concerned. In a global context, the analysts at GS expect AI to contribute 7% to real GDP globally over that same 10-year period.
So, what should one invest in? GS divides its list of recommended AI companies into:
(1) global hyperscalers (e.g. Microsoft and Alphabet);
(2) global enablers (e.g. Nvidia and Marvell Technology);
(3) AI empowered companies (e.g. Adobe and Pearson); and
(4) China platforms (e.g. Alibaba and Tencent).
I would start by eliminating (1) and (4) – the global hyperscalers because they have already performed extraordinarily well this year (see Exhibit 3 again), and China platform companies because I am concerned about (i) excessive debt levels in China and (ii) the massive construction of apartment buildings in urban China despite (a) the urbanisation wave slowing down and (b) the Chinese facing an extraordinarily negative demographic outlook.
We get regular updates on China from our contacts out there, and one of them is particularly informative. I ought to tell you that he doesn’t share my concerns, although it is probably fair to say that he has his own concerns. He told us that he finds it hard to get excited about big platform companies like Alibaba and Tencent and considers them beta plays that trade on sentiment
Ignoring (1) and (4) above leaves me with global enablers and AI empowered companies. On the list that is about to follow, I have taken a number of issues into consideration, such as liquidity, current valuation, anticipated growth of revenues and earnings as well as recent performance. You may have spotted that I have picked far more enablers (technology companies) than AI empowered companies. Indeed, that is no coincidence. In the early days, the enablers will make it all happen, following which the empowered companies will step in and take advantage.
One final point before I begin: during my research, I have found that Asian AI enablers are trading at much lower earnings multiples than their US peers. Therefore, if one could easily access those names, they would be my top picks. However, Taiwan is not an easy place to do business. With that in mind, in no particular order, my favourites would include:
Marvell Technology (MRVL on Nasdaq)
MRVL (Exhibit 5) produces the so-called optical PAM4 digital signal processor which enables high-bandwidth connectivity on AI platforms. The company is not profitable, but revenues should grow fast over the next couple of years, which should make the company profitable. Meanwhile, the stock is down some 25% since December 2021.
Credo Technology Group (CRDO on Nasdaq)
Credo’s (Exhibit 6) existing AEC business (Active Electric Cables), which is used to transmit data from 5G to servers in data centres, is growing fast. Now, as the company is entering the AI business, the growth rate should further accelerate as AI servers require up to five times the number of cables vis-à-vis standard servers.
SK Hynix (000660.KS on Korea Exchange)
SK Hynix (Exhibit 7) is a leader in the high-bandwidth memory (HBM) space, which forms a critical part of the AI server market – a market that is expected to grow fast over the next few years, Also, bear in mind that Asia is at the leading edge of the AI revolution, and the company is Korean. Although the stock hasn’t performed poorly more recently, it is still off almost 20% from its February 2021 peak.
Unimicron Technology (3037.TW on Taiwan Stock Exchange)
Investing in Unimicron Technology (Exhibit 8) is, indirectly, an investment in Nvidia – one of the Magnificent Seven – and I find that quite appealing, particularly because the earnings multiple on Unimicron is only a fraction of that of Nvidia’s. The company accounts for almost 30% of Nvidia’s supply chain of ABF substrates, a critical material when producing the powerful chips used in AI models, so robust growth in Nvidia’s order book translates into attractive business for Unimicron Technology.
Nan Ya Printed Circuit Board Corporation (8046.TW on Taiwan Stock Exchange)
NYPBC (Exhibit 9) is another producer of ABF substrates and trades, like Unimicron, at a very reasonable earnings multiple. The company is currently upgrading its production facilities to better serve the high-end ABF substrates market and expects the revenue contribution from high-end chips used in AI models to double over the next few quarters. The stock can be acquired at less than half the price it traded at in late 2021.
Adobe (ADBE on Nasdaq)
Adobe (Exhibit 10) is another beneficiary of Nvidia’s strong growth. Having well-established relationships with two global hyperscalers (Nvidia and Microsoft), the company benefits from the growth derived from (in particular) generative AI without having to invest silly money. Furthermore, the stock can be acquired some 23% cheaper than in late 2021, when it peaked at $670.
Intuit (INTU on Nasdaq)
INTU (Exhibit 11) specialises in matching its customers with tax experts and is already using AI to streamline its operations. Being an early mover, INTU possesses a significant competitive advantage vis-à-vis its rivals – an advantage the company should benefit from for years. The stock is not exactly cheap, but the early-mover advantage justifies that, I believe.
The research team at GS list plenty of other interesting names in their research paper, Thematic Stock Exposures – A special AI edition, which I have used as a source of inspiration for the names I have just shared with you. Having said that, I believe the names just mentioned add up to a good day- one portfolio. Over time, we will gradually switch away from enablers and invest more in AI empowered companies but, for now, there can be no doubt that enablers (and hyperscalers) stand to benefit the most.
One final question is are we making a mistake by deliberately ignoring the Magnificent Seven? Only time can tell, but the market has developed a bit of a habit to go hot and cold on various investment themes and, right now, megacaps can do nothing wrong. Sooner or later, that will change, and we will be ready to take advantage when the megacaps return to the doghouse.
Niels C. Jensen
21 July 2023