Podcast

Welcome to Beyond the Benchmark, the EFG podcast with Moz Afzal.

Moz:

This conversation is part of EFG's Future Leaders Initiative, our entrepreneurial innovation hub focused on sharing knowledge and identifying innovation across AI, geopolitics, leadership, and the digital economy.

Moz:

This is where we work with people like Jordi Visser to bring actionable insights, and in this case, artificial intelligence and how it plays into the global economy.

Moz:

Welcome, Jordi.

Jordi:

It's great to be here, Moz. Nice to spend time with you.

Moz:

Yeah. Well, I'm certainly looking forward to this and certainly over the last few weeks, so catching up on your latest thinking has been a real joy. So let's go straight to it in terms of your background, your history. What I wanted to do was just maybe spend a bit of time in terms of what does AI macro nexus actually mean from your perspective?

Jordi:

Well, it's a good question. I haven't been asked it before, but when Dennis DeBuscher and I started sitting down, Dennis and I have known each other for over 20 years, I guess, at this point. And he wanted to find a way for us to work together for me to do research for him. And he knew that my viewpoint was that the world was going to go through a dramatic shift in terms of the next five years being all about artificial intelligence and the impact that that would have on macro. And since macro and most people who would be on this podcast with you, like myself, who are macro people, we're also historians. We think about business cycles. We think about things that have gone on forever. I can sit here and talk about the early 1900s all the way through, what the economy did, what labour did.

But my viewpoint was that AI would be the nexus to destroy or at least to disrupt businesses and macro in every possible way. And I think this year is when we're finally starting to see that become, I'd say, more evident. But the other part was the speed of the change was going to be dramatic and it would feel a little dystopian. And to be fair, and let you ask the next question, this has been going on for a while now, and I want to make sure people realise when you hear artificial intelligence, machine learning and artificial intelligence has been disrupting the economy and the global marketplace for at least the last 18, 19 years. And I know from our industry, if you just think about WorldQuant when it really began having an impact on a firm like Millennium, that was before the great financial crisis.

So you start having a realisation that artificial intelligence is accelerating now at a fast pace. And so the name was meant to represent both the disruption, but also as the speed increases, I think it's going to be very, very difficult for humans to adjust to that environment.

Moz:

So maybe talk to us about the speed, because I suspect the speed thing is actually the biggest challenge because certainly adoption, we're talking a little bit earlier about OpenClaw, the adoption numbers are just completely unprecedented.

Jordi:

Yeah. The speed of progress, and I think the listeners need to definitely put into concept something I'm going to say. There's the speed of the technology, and then there's the friction by human beings to accept it or believe in it. And all of last year, AI was called a bubble. It wasn't doing anything. It was hallucinating. I mean, I heard this repeatedly throughout the year, and I think even today, there's a lot of people that think it's hype. They don't believe it does anything. And as someone who's on LLMs all day long, who pays the highest price for all five of the major ones in the US, uses them all during the day for different tasks, thinks of them as different employees, and now has moved into the world of agents with open claw, multiple machines. I can say with certainty, I think the big thing that people have to start to realise is this is a reality that is here for the next decade and will only disrupt things more in a compounding fashion.

And the course from the end of November of last year when Opus 4.5 was released until today when we sit here, I have been completely shocked and I think so has Silicon Valley about the progress and what has happened. And the Agentic world was kind of pushed majorly forward. And it's hard for me to describe to people how important this is, but let's just say we're going from eight billion possible consumers and eight billion possible people to work to trillions and trillions of consumers in terms of AI agents and workers. And I think that is going to lead and has led to the disruptions and the scarcity issues that are showing up around the globe.

Moz:

So maybe just rewind a little bit, I guess when for most people at least, the ChatGPT moment was where the world's sudden eyes suddenly just opened and the realisation that AI world was upon us. Although absolutely what you said, machine learning and some of those techniques and neural networks. I mean, I remember doing my first neural network back in the '90s on FX. And by the way, it didn't work at that time, partly because I didn't have enough data or compute power to actually do it properly. But what's interesting is that we had that ChatGPT moment and it was all about the large language model. Talk us through how we've gone from large language model, which is still there, but now to the next stage, which is much more around Agentic AI and now agents.

Jordi:

Yeah. So I think the easiest way for me to just describe is first to go through the chatbot, let's say growth. When you first got ChatGPT, IQ on it was around a hundred, slightly below. By the end of the following year in 2023, let's assume that we were somewhere between 100 and 110. So you're past average US human intelligence, and that means the answers are getting better, but you're still getting mistakes. By the end of 2024, you're up around 120. Now you're starting to get into a much, much higher. And if people don't realise, starting at 120, the answers you're getting for, again, the best model. Because if you're not using the best model, you have to decrease the IQ points that you're getting, which means you're getting a lot, you're not seeing how smart this is. Well, when we finished last year, we were up around 130.

The current models are higher than that. The reason the intelligence had gone higher was not because of more compute. We've actually run to the point where the data centres and all of the things that will eventually allow these things to get even smarter, that really didn't cause it. It came from reasoning. It came from reinforcement learning. It basically came from, let's just say, tricks of making the model get smarter by slowing down its thinking, by eventually bringing agents in where you had more than one person. So the way I want to kind of take it into the agentic world is if you were sitting in a conference room and you had a meeting and you had two employees in there with you and you guys were discussing a project that you were going to do for the firm and you set these two people off and you gave them 30 minutes to come back with an answer, okay?

It's not going to be a great answer. Well, now if you had a hundred people in that conference room, you gave those hundred people the time to come back and you gave them a month to come back with a project answer, you're going to have a far better answer. You're going to have a far better project a lot. You're going to have time to check it to go through it. That is the difference between a chatbot, having a one-on-one conversation as opposed to an agent one where they're taking more time, they're correcting each other's worst work, they're being specifically specialists at a given task, and now all of a sudden they're running twenty four seven, which is why the time is increasing. So if people haven't thought about what it means to have an AI agent, that is the visual that I've said is if you want a better answer and you could bring in a hundred people, the top hundred people in the firm to go get it, you're going to get a better answer, especially if you give them a longer time period.

That's what's happening with agents.

Moz:

Yeah, it's certainly moving very fast, but let's go to the enabling of that, and which is obviously part of your thesis is around the shortages that are now a bit rapidly being created. So we've moved to this sort of agentic world. We have now something that is very interesting in that it's not just about the GPU, it's also about the CPU, it's about the memory, it's about connecting servers in a fast way and even the pipes within each server. So maybe talk us through how that has transformed and what Ruben and whatever comes next to Ruben looks like in the future.

Jordi:

Yeah. I think this is another point. I have to use an analogy for people to understand how we're going from just GPUs. And think of GPUs as the training model for your children going to college and really what they're doing is memorising things. They don't have a job, they're not taking down personal experiences. They're actually using things that are taught to them to gain memory, to use when they get more into the experiential side. When you get to the work, all of a sudden now it's experiential. You're not sitting in a class, you're not memorising thing. You are memorising movements, you're memorising the way people are talking to each other in a whole bunch of things that way, but you have to remember them all and you have to put them into context. It's not just a regurgitation. So I think of school as more like Google search.

You put something in, you get a response, a teacher asks you a question, you fill it out on what you think they want to hear. That's great. That's a chatbot situation, but now we're entering a different part. And for that, you need to have memory. You need to have the prior conversations. Right now, the only way you can get that with an LLM, even though it's improving, is within the same context window. If you're using ChatGPT, it'll remember some of the things that ... It remembers that I work for 22V. It remembers stuff like that, but it doesn't remember every single conversation or every paper I've written, but eventually it will know every single thing that I've done. And if I wanted to, I could store them in a folder. I could use ClaudeCode and I could go reference it. You're starting to get into the point where the ability to do all that stuff without any latency, with supreme speed, you need more than just the in- out response.

You actually need east-west traffic where it's getting stored. It's getting pulled back very quickly. And for that to happen at the speed necessary for it to be usable for AI agents and to go through, you have different hardware needs than you did for the prior three years. So we're entering a different point. I've just started to write about it. I wrote papers about this last year, but with a lot of the papers that I wrote on AI, if I write about humanoids right now, you're not going to see one walking down the streets of Manhattan or London. In five years, you will, 10 years you certainly will. Autonomous vehicles, you're unlikely to see them at this point, where next year you're going to be very likely to see them. So the hardware thing became more of an issue this year once the Agentic side came up, and this surprised even Silicon Valley.

We never know when these breakthroughs are going to happen. OpenClaw was done, and I think this is one of the most important stories. OpenClaw was not created by someone at one of the big companies. ChatGPT came from the transformer papers, which came from Google. This was actually a breakthrough by someone just sitting at home. And this is one of the points that I want to make to people. Public companies are not going to be the ones creating the great technologies going forward. They will be doing a lot of the things that are important, but right now, because they are restricted in many cases from using OpenCloth, from being ... They have to worry about the data they have. They're being very secure. People sitting at home are going to come up with massive inventions. And I think that's where we are on this. And that just brings up that OpenClaw needs a lot more memory.

It needs to hold these models and be able to run things. And I'm doing mine on open source Chinese models at this point. There's just a lot more needs in terms of the makeup. So the hardware situation where we had been underinvested for a long time because we just didn't have a non-cyclical area of needing memory, of needing CPUs. And so all these companies, their multiples were depressed. They were waiting for some kind of an upgrade cycle for phones and computers and cars. Well, now what you're having is the realisation that if you're going to use an Agentic AI agents, you need more hardware. And so this is going to trigger something that to some people is going to look a lot like Nvidia. They're going to underestimate how fast this will go. And the first kind of idea of this was really with Corning and Optical Fibre, and you've seen it spread, and no one thought that a company from the 1850s could see this kind of a parabolic rise, but you're going to see this around the globe, just like you've seen with memory.

There's other components, and I just started writing about those this week.

Moz:

So it's moving very fast, but that speed has also meant shortages being created. And I think you probably hit the nail on the head there, is because we've underinvested over the years very consistently because we see these cyclical companies and not, as you said, sort of perpetual companies, if you like. So let's talk about software, and obviously you have very strong views about software. Take us through your thinking, and then maybe we touch upon the credit issues that we're currently seeing.

Jordi:

Don't get caught in the trap of saying this is cheap relative to history. The uncertainty looking forward is what you need to do. There are two parts that I will say to you, if you could get rid of this uncertainty, which is what the market is saying, then fine, you're okay with this. The first one is, is there really growth for these companies? First, they have to get around the seat-based model and they have to be to transition into an accomplished based model. They have to hope that somehow or another their agents that are bolted on are going to be able to do this better than someone then like Palantir that sits above things and is orchestrating the Agentic side. So that's the first question that also involves something that I just want people to believe. Whether it's Eric Schmidt, whether it's Mark Andreessen, the one thing I am certain of that these guys have said is the rise of millions of startup businesses, millions, not thousands, millions of startup businesses.

I'm a startup business. I'm doing software. I'm providing that software to people. They don't need to go out and pay certain software companies for what I do. I see it in every single data provider. Their software costs are going down because the ability to create. So I'm more worried that the startups of today never get to the size that they would ever be customers of Salesforce and Adobe. So you're going to have a growing ecosystem of companies that actually never go public. And I think that is a risk. The second thing, and probably equally as important for these companies in terms of the future, gets into this rise of competition and how quickly it comes up. If you believe that you actually know what the world is going to look like in three years with your own customers, I believe enterprises have a huge risk because they have friction in adopting AI, and I think their plans are going far slower.

So if you're a salesforce or an Adobe, one of the things you want is a company that actually knows what their plan is going to be for AI. I don't think most companies know because the progress is going too fast. Consultants don't know what the answer is. And I've worked at a big firm. I just know the committees that are put together take longer time to come to an answer. And I think that hurts these companies from getting more revenue in the door when they need to build out a massive amount of hardware in servers, in memory, and all the things we talked about, because it won't all be cloud-based, and most of these providers are cloud-based. It's going to have to be something that is more inside their environment. I just think when you add all those questions up, that's the reason that there's been multiple compression because it's very difficult to value these companies three years from now at this point.

Moz:

You're listening to a conversation from EFG Future Leaders, our entrepreneurial innovation hub, where we work with experts like Jordi to share knowledge and identify innovation for our clients and colleagues.

Moz:

We've seen all the cycles previously and there is displacement that lasts, I don't know, five years, 10 years. Do you think this is longer?

Jordi:

So I'm going to say nobody knows for sure based on one thing. If you could tell me with certainty that we'll be putting out hundreds of thousands of humanoids a month in five years, I will tell you there's no way that it won't change forever. Displacement is the keyword, Mos, and the data speaks for itself. So what does the data say, and I'll just use the United States. The data says in the US that we've created zero jobs over the last 12 months. If you take out healthcare, we've had negative jobs over the last 12 months. I really want people to think about that, that in the United States of America where it wasn't before COVID, we were talking about a normal pace of growth being 150 to 200,000 jobs a month. Now we're talking about zero over the course of a year where nominal GDP was relatively high.

So I think we've already proven a point on that metric. The second thing is if you go through all the survey based work, whether it's consumer confidence, whether it's the data points that show up on the job postings, everything has been in decline. Wages have been in decline. What I feel is happening, considering the fact that the jobless claims are not rising, is if someone gets fired, we still have a tremendous amount of job shortages in the United States because of demographics and now because of immigration. The problem is knowledge workers, people that have a college education, that is the direct target on your back for AI, and that will continue to be disruptive. You are seeing thousands of layoffs at companies that are massive. Meta just announced that they're going to lay off 20% of their workforce. You've had other companies that have done bigger numbers, IBM's done them, Amazon.

All of these companies are massive companies. You've never seen this before where as the economy grows, as earnings grows, companies are firing people. So there's no question in my mind that economists have this all wrong. This is happening. The difference is if someone loses their job in New York City, about 70% of people live paycheck to paycheck in the United States, meaning they don't have a cushion in their savings account. If you don't have a cushion in your savings account and you lose a job, you have to go get another job. You can go get another job, but if someone graduated from a good school in the US, do they really want to work at a supermarket and be an Uber driver? If they want to, they can. And they can make probably enough of a wage to keep them surviving, keep their kids in school, keep the daycare going.

But the reality is for people, I think the displacement has happened. I think it shows up in the sentiment, the way people are voting in every single thing. It's not like this is some ... This is not the dotcom bubble. I remember the dotcom bubble. I remember everyone in this country being like the stock market was going to go up forever. This is actually a very, very dystopian feel to it that as the economy's getting better, as the stock market's going up 15% a year, you still have the labour situation.

Moz:

So kind of moving on, in this sort of environment that we've been talking through, and let's move on five or 10 years, what do you think the sort of investor playbook needs to look like going forward?

Jordi:

Well, first, let's put the current situation in place that's getting all the news headlines. And I'll even start last year with the tariffs. So Donald Trump gets elected in the United States. The electorate wants him to do certain things. He goes with this tariff approach because he's trying to help manufacturing in the country. He's trying to redistribute money in some way. Doesn't care about the wealthy people. He cares about his voting block. What causes him to get voted into office rather than his decision making, in my opinion, is what's been going on now since really the great financial crisis. And I just want to take this back and make sure people realise, because at the beginning I talked about machine learning. The world has been disrupted by exponential innovation really since that point. I don't think people associate the great financial crisis with the birth of the iPhone, but it happened exactly the same time.

When you're fighting over materials and there's a shortage of them, like we've seen with memory, like we've seen with optical fibre, you're just going to have more of these episodic things. And the more episodic you get, the bigger the issue. Now, the other thing that has happened, and again, I think leverage has increased again, as is always the case in the Fiat system. The leverage increased from the great financial crisis. It's not in the banks, but I think it was definitely in long duration assets. I think it was in quant strategies. I think it was in anything that has been more of an optimization. We probably have a bubble in optimising risk based on historical correlations. And I think you're seeing kind of that. I have a turbulence model that I send out. That turbulence model is meant to show the disruption in cross-asset volatility.

So I think when you put it all together, I think people should be trading more. I think software is going to be a very challenging thing that is more long, short, and I think you have to pick your names that'd be long and pick your names to be short. And then on the flip side, I think for the commodities, you have to get used to these things where you could have a 25% year in a commodity name, but it might involve two, 20% corrections as opposed to a software name where you pretty much could have a steady state, high, sharp ratio. So that's the way behavioural, I think people have to adjust. And I think some of it's going to take some time to make sure that the managers either adapt or they die.

Moz:

Well, it certainly has profound impact for portfolio construction, right? Because as you said, we're an optimised world, so we're looking at, I guess, past data to try and sort of predict the future and also future volatility. So this will certainly disrupt things. And I guess you could see that in performances over the last sort of 12 to 18 months where it's been quite difficult for the traditional money manager to make money or outperform, should I suppose the right word, because in general, they have made money. But it's look, using a lens that you probably, or many people are just not used to.

Jordi:

Yeah. One thing I want to make sure people hear, because I don't say this enough, but I think it's an important point. So one thing about the 1970s were commodities caused recessions. I do not think there will be a recession. I don't care. I mean, I don't know where oil has to go to, but the one thing that people need to keep the back of their mind is the CapEx numbers are going to continue to increase. The only way that wouldn't happen is if the central bank and the treasury, mainly of the United States, because this is where most of the CapEx is happening, if they allowed private credit to unwind and become a contagion across other parts, they're not going to do that. AI is a government military supremacy issue. It is a race for technology. So I don't believe at any level, since this is being funded by the government, but also by the most cash rich equity massive companies in the globe, the numbers we're talking about are not that big.

The benefits and profit margins will be there enough to at least justify it. So I don't want people to think that this is some recession-based story. I actually don't believe that. Commodities, they're price inelastic to this situation. It does not matter how high they go, meaning the physical commodities. If memory were to triple from here, it'd be an issue, but memory is the thing that'll slow some of this progress down, and you're going to see that with iPhones and stuff like that, because you won't be able to sell those as well if it's $300 more. So you will see demand destruction to some degree, but our economy has a lot of transfer payments from the government. It also has a lot of healthcare expenditures. These are different numbers than what was before GFC. And I think people have to adjust that. We actually have a less cyclical economy, and this is a mistake.

I just think it's less certainty, and that's the way that people should adapt their portfolios. I do think there's a migration for the next year and a half that involves wider credit spreads, because I think those are completely mispriced for the uncertainty that we've begun to see in certain industries. I think volatility is going to be higher across assets.

Moz :

We kind of briefly touched on humanoids. Do you think that maybe a snap question on this one, US wins or does China win on this one?

Jordi:

I think there is no winner in AI. I think this is one of the ironies is I don't know what winning means. I really don't. I don't know how you define it. I think their models are just behind ours and they've done it without our compute. They have more energy than we have, which is an issue. And I think that is a rising cost. I think Chinese government is investing so much money on the memory side that their ability to get the models out there, but they have a deflationary situation because they're charging only 10% of what US companies are charging for compute. So maybe they win in terms of having more models in people's hands, but they don't win in terms of the capitalist side.

Moz:

Yeah, no, absolutely. It's a good point. I always think that people don't realise that Apple makes ... I don't know, last time I saw it was like 90% of all profits made out of mobile telephones is because it's all closed source versus open source and Android phones where competition kind of rules. And I guess we know there's more Android phones than there are iOS phones. So I think the analogy I think is kind of well, or certainly at least for me at least, is well understood with respect to the making money versus winning by having a greater adoption. So I think that's quite good. And then humanoids, where do you think that go? Obviously Elon Musk is now sort of taking ... Well, he's taken one of his production lines to make Optimus, but where does China fit in? Where does US fit in? Obviously China's very, certainly in terms of manufacturing and robotic and automation and manufacturing seems to be substantially ahead.

Where do you think this lands?

Jordi:

So again, I think it's going to end up being the exact same thing. I think China is going to win in volume for sure, but I think that's a critical part to this question is if the brain has been dealt with with XAI, with SpaceX and all of the things that he's built, think about the vertical integration that he's created to be able to do rocket ships, to be able to do Starlink, to be able to do the car, collect all the data, and to have XAI, to create a competitor, a real competitor, one of the five LLMs, and to do it in less than three years. I mean, it's amazing what he's been able to do. And so I hate to sit there and talk, but I know a lot of people have a negative viewpoint of him. But I think when you think about the issues that Anthropic and OpenAI are having, Ilan Musk said this on a podcast recently, he said, "Software boys trying to build hardware can get very complicated." I think vertical integration in the supply chains for the scaling of humanoids will be a deciding factor.

And he has one of the world's largest battery factories in the world. He's already thought about all of these issues. So I think Tesla is going to win because of the brain, but I think China's going to win in volume for sure.

Moz:

Yeah. Yeah. No, very, very, very interesting. So Jordi, we're going to wrap it up there. It was as good and as fascinating as I was expecting. So thank you very much for that. A lot to think about. It's probably your podcast I will replay, I'm sure, and listen in and make furious notes. So Jordi Visser, thank you very much for coming on Beyond the Benchmark.

Jordi:

It was a pleasure, Moz. Look forward to it again.

Moz:

This episode was part of EFG's Future Leaders Programme, where we work with regular and guest contributors to give our clients and colleagues the insights they need to navigate a fast changing global environment. Look out for more future leaders episodes coming soon.