Episode Transcript
The Finance Ghost: Welcome to episode 243 of Magic Markets. We come to you in a week where we've just done our premium report on Oracle, which has got to be one of the most hyped-up AI stories of the lot right now. We come to you the day after I had a particularly outrageous experience with AI from start to finish, and I guess that inspired us, Moe, to talk about just AI eating the world, really, and how much hype is actually in this thing or not.
And to our listeners, before you audibly groan and think, oh, no, another AI podcast, I think Moe and I probably bring some unusual views to this thing and I think it's important to actually just listen to as many of the different views as you can on the world of AI, some of which is from people in the industry - and then you've always got to be very skeptical of people talking their book and trying to hype this thing up, the hype train that they are on - we are obviously not in the industry, so maybe that's good and bad.
With that out the way - Moe, welcome to the show. It's fun to do this with you, as always.
Mohammed Nalla: Yeah, Ghost, I want to have some fun with the show. AI is eating the world. The reason I want to have fun, and firstly, maybe just as a caveat that's out there, we're not Luddites. I believe AI does have a lot of good to bring to the world. There's also a lot of bad. And we always take a very balanced view on the world. What we want to try and showcase in this particular show, I mean, obviously the hilarious examples you gave me over the course of the last two days, your own use case and experience with AI, we want to bring some of that into the mix. But what I also want to bring into the mix, maybe later in the discussion, is just where some of the other pressure points are emerging because we're seeing all the hype in AI stocks, but there are pressure points emerging in the debt markets. There are pressure points emerging in the energy markets in the US particularly. And so we want to touch on some of those points.
Let's maybe first start off with the hilarious part of the story, which, Ghost, I want you to share with our listeners. Maybe not the full detail, because that would probably take up most of the podcast, but take us through your hilarious use case on AI over the course of the last couple of days.
The Finance Ghost: Yeah, so Moe you're right, we're not Luddites. We are actually trying this stuff out to the extent that we can, obviously, I am trying to make Copilot a part of my day. I will say that I am Googling a lot less than I used to, except when I'm desperate about Copilot, but more on that to come.
So it has actually been quite useful. What I've personally enjoyed is you can ask the natural language questions. It's been quite a useful part of the research process. But the thing where it just went completely wrong is we host our podcasts on a platform called Castos, and I had recorded a Ghost Stories podcast the other day and I needed to go and get a transcript for it. The system that hosts our podcast can do these automatic AI transcripts. They're not very good - they always have issues. The transcript that you read as one of our listeners has already been quite heavily edited to get it right, because we've had some absolute howlers that have come back. Believe me, you only need to see the word “incest” instead of “invest” once in your life to realize that you need to carefully edit an AI transcript. But anyway, I uploaded said podcast, which was free of incest in any form, and I got back a transcript in Dutch. Now, what makes this kind of funny is my guest has quite a strong Afrikaans accent, so I can only imagine that he said “ja” in a very strong accent and then the transcript AI tool, whatever you want to call it, went, oh, look, Dutch. And then what it had to do to arrive at that conclusion is it has to listen to an English podcast, transcribe, and then auto translate it. And there's no way to change that setting. It's not possible.
So then I thought, okay, this is great. You try that a few times. Now you have a Dutch transcript, which is really not very useful. So I go into the chatbot, I'm like, okay, hi, help, why is my transcript in a different language to the podcast I uploaded? To which, of course, you get an AI-generated response obviously, because it's now impossible to talk to a human being at any international software-as-a-service type platform. It's just not possible. And so you get this answer back that says, well, the reason why your transcript is not in English is because your podcast is not in English. You can imagine at this point in time how spectacularly irritating this is.
I realised I'm not going to get anywhere. So then I go onto Copilot - hi, Copilot, can you do a transcript from an audio file? Yes, of course I can, upload the MP3, I've got this. Wonderful! Try to upload the MP3 - doesn't work. Try to make the MP3 smaller - doesn't work. Ask it again: why can't you read the MP3? Oh, just try a wav file. Okay, go make a wav file. Of course, it's gigantic - too big to upload. Try compress it. Too big. Just doesn't work.
Eventually I'm like, I think this thing's lying to me. Go onto Google. Hi Google - can Copilot read MP3s? And you find Reddit threads and all of that where people are saying, no, Copilot is lying to you. It cannot, in fact, listen to an MP3.
So now I am deep in AI hell at this point, late at night, under pressure, trying to just find what should be a pretty simplistic solution in a world where we have the Oracle CEO making big noise about all the incredible things they'll be able to do soon. I'm just trying to get a transcript and eventually, hilariously, I figured out that the plan C at this point was to go into Word and use the dictate function. And before you panic and think that I played the podcast to the microphone into Word, which would really be very sad, it does actually let you upload an MP3 and then it does a transcript.
So in summary, Microsoft Copilot, which is a Microsoft thing that is designed to work with Office, can't read an MP3. But Word, which is a Microsoft productivity suite program, can read an MP3 and give you a transcript, albeit of worse quality than Castos, but at least in the right language.
Anyway, I feel actually quite a bit better now. I'm glad I could share that. It wasn't fun though at the time. Laugh, Moe. Enjoy it, enjoy it. I'm gonna make you do transcripts!
Mohammed Nalla: I'm so glad my microphone was actually muted throughout this entire time because I'm just laughing so hard. Even having heard the story before. I mean, it's literally a comedy of errors. But again, I think this is so important because a lot of people out there are buying into the narrative that AI is actually solving all of our problems and it's going to make things a lot more efficient. And in fact, we have a real-world use case, we are very big on the whole do-it-yourself investment thing, we do a lot of our own bottoms-up, ground-up research in Magic Markets Premium. I want to highlight the importance of this is again in playing around with the technology to see how useful it can actually be.
You had an hilarious example on a stock we covered quite a while ago now where I had put into the report that the company didn't have a credit rating, because they don't have a credit rating, just to put that out there. And I think you had gone in and you had punched something into either Google or into one of the AI bots or whatever it might be: Google, what is the credit rating of Company X? And came out with a very authoritative sounding output. It was like, yes, Company X has a credit rating of this as issued by this ratings agency with a little bit of a narrative.
And again, just be so careful because if Copilot can lie to you about whether it can read an MP3 or not, it can lie to you about its analysis and just sound very intelligent while doing that. It's called hallucination sometimes. And so I would really caution anyone out there who listens to this podcast, who maybe subscribes even to Magic Markets Premium, go out there, validate your information yourself. AI is not going to do this for you. It is actually remarkably, let's call it unreliable. I've had many experiences of this where I've done side-by-side analysis to see what the AI gives me versus what my own analysis gives me.
I've seen a whole raft of errors that AI can actually generate. And even after you try and educate and you say you've made an error, this is where you've made an error. It's very apologetic, it says, oh, I'm so sorry, yes, thanks for picking that up. And then goes and generates again very authoritative sounding but incorrect information. So I would say it's very good as a narrative tool. It's good on the linguistics side of things, sure, no problem. But when it comes to hard analysis, just pay attention to what you're getting there.
I'm going to stop there with kind of moaning and groaning about AI because I want to pivot from that into another talking point. And this is talking about some of the pressure points that are coming through in other parts of the value chain, other parts of the ecosystem, because we know that AI is sucking up a tremendous amount of investment. A lot of the capex in the US is going into AI. You've got data centres coming up and this is creating a number of flashpoints that I think we need to pay attention to. One is that you've got municipalities, for example, that are going out there and they're providing massive tax incentives to the companies to actually build the data centres in their respective municipalities and in fact, I read that at least 10 states so far have lost more than $100 million per year in tax revenue to data centres.
How they do this is they have to provide the incentives they want, the jobs they want, the big flashy sticker to say hey, we've got AI data centres in our municipality. But that comes at a cost. And that cost is being borne, albeit indirectly by some of these states, by some of those municipalities.
Another flashpoint is pay attention to what some of the big players are doing in the market. A little while ago we had a report of Microsoft that backed out of a $1 billion data centre investment, I think it was March this year. And then we had a report from UBS that was analysing the company that Microsoft was transacting with that then had lease obligations on the other side of around $175 billion across their portfolio. And they were concerned that that company may have over committed because you're now saying we've got Microsoft coming on stream, there's a lot of hyperscalers coming on stream, we're going to put the supply on stream. And if one or two of those deals go sour, you've got a market that is quite stretched. There's a lot of debt going into this as well. There's a lot of structured finance that has now become a very popular way for players to actually finance their data centre projects. So just pay attention to what's potentially lurking in the background of this AI mega theme. Certainly when it starts coming to things like mortgage-backed securities that are maybe backed by leases for these data centres from large tech companies that have proven that they can maybe pull those leases and try and build these things themselves.
The Finance Ghost: Yeah, the whole value chain is super interesting in this space. And some of the hype behaviour just cannot be ignored either. I mean, literally hot off the press this week was this whole NVIDIA investment in OpenAI. But then OpenAI is actually buying chips from NVIDIA. So in other words, is NVIDIA just using its money, essentially, its investment budget to go and then lock in supply which then comes back as revenue? I mean it's pretty weird. You don't really see that very often. And when you see it at this sort of scale, then you've got to wonder - vertical integration is fine, but it just feels like the timing is a little bit weird Moe? A little bit hypey, don't you think?
Mohammed Nalla: I saw this meme which for me was just perfect. It's like one of these power cord extension things. And it had the one plug that you normally plug into a wall socket just plugged back into the extension. Because that's NVIDIA...
The Finance Ghost: That's great!
Mohammed Nalla: And that for me sums it up, right? At the end of the day, we've got to pay attention to what's happening. Where are we in the hype cycle here? Because, yes, I do believe AI is real. Do I believe that the use case, the real use case in businesses has been proven? I think we're somewhere along that road. I think some of it's been proven, but I think there's a lot of optimism that is out there and I'm concerned about that.
I'm concerned about other parts of the value chain. I haven't even touched on the energy story in the US which I'll touch on maybe a little bit later on, or maybe it's the subject of another show. But this, for me just - it proliferates. And this is exactly - I'm not calling it a global financial crisis or an AI crisis, let me put that out - but I think it's really reminiscent of where we saw systemic risks slowly seep through the system in the global financial crisis. Initially it seemed like a great idea bundling all of those mortgages together because you get this nice blended credit profile. But eventually those chickens come home to roost.
And my question is, I don't know how far along the curve we are in terms of those chickens coming home to roost in the AI space, but I would watch out for some of those points I'd indicated. Watch the credit markets. Watch how difficult or easy it is for these companies to actually pay back some of the loans they've taken to build out the data centres. Ghost, are we plugging ourselves into our own extension cords? Potentially, yes.
The Finance Ghost: Yes. It's a very funny way of putting it. The meme that I saw, that I enjoyed was that Spider-Man one, where there's three Spider-mans (or Spider-men perhaps?) on the screen and they're all pointing at each other and it's like, look, AI capex and revenue - and it's just this absolute circle of money going around and around. And look, the reality is that Microsoft is my biggest individual position. It has been for a while. And part of that is AI.
That’s what makes us difficult for investors, because we just don't know. We absolutely don't know. So if you go and have a look at Oracle like we did in Premium this week, they talk a lot about how so much of the spend up until now has been on training the models. Not so much the inference by the models, the ability for them to actually be used to do stuff. The “do stuff” is still coming, but therein lies the problem is no one's quite sure what the “do stuff” is going to entail yet. There are definitely undoubtedly big pieces of what's going on out there in AI that is exciting and is going to change things and make things more efficient. I'm sure of that. Stuff like machine learning - even for me personally, for all of the nonsense we talked about earlier with AI, there are elements of the research process that are being made easier.
I think where AI is dicey at the moment is when you need a very precise answer. So, for example, while we were researching the Oracle show, I tested Copilot and I said hi Copilot, can you do a table of Oracle's remaining performance obligations - that's basically their unearned revenue the way they disclose it - by quarter, for the past two years? So the first thing it does, is it does it for the wrong period. OK. I said, by quarter for the past two years. Oh, okay, now it does it, fine. And then it can't really read it for a few quarters and then it gives up, and then the latest one was just completely wrong, completely wrong for the last quarter and I even told it look on SEC filings.
So for me, AI is really good at something like if you want to test your thinking around a particular story or if you want to ask it something where it needs to go and just bring you a stream of thought. So for example, what are some of the recent acquisitions by company xyz? It does a very good job (usually) of going and reading the news, summarising it, telling you what they paid for it, cross-checking sources because there's multiple sources. It's not trying to get a super technical answer, it's going and reading the internet and bringing the answers to you.
But here's the thing that also doesn't really make sense, is where does this leave the media industry? Because what came out in the past week in South Africa was a quite frightening report on readership of some of the leading media platforms in South Africa. And there were some incredible year-on-year drops there. Daily Maverick - I'm going to get the number wrong, but I think it was more than 60% down year-on-year. There were only two publications that were in the green on that table. I think one was IOL and the other one was Daily Investor, slightly green, and the rest were just red, red, red, red, red in terms of website traffic. How much of that is because AI is going in our reading those websites and giving people the answer? Now that's all very clever right now, but for the AI to work, someone needed to go and write the original report. But if the economics of these platforms is collapsing, where's the original content going to come from? AI can't do that. It needs to read something somewhere else. So I don't know what that looks like in years to come, but it's going to be wild. That's all I do know.
Mohammed Nalla: Yeah, that's actually another good analog for that circular reference, right? Like plugging it back into the same extension cord. Eventually there's just no new power coming into the system!
Ghost, I actually saw a similar report here in Canada talking about exactly that point about traditional media and how website hits are way down, again because of AI. I don't think it's a uniquely South African problem and it wouldn't be because AI is a global phenomenon. That media report with the impact on traditional, even digital media is one that I actually saw replicated up here in North America.
Now I just want to touch on two last very interesting points that I wanted to raise. I mentioned the energy constraints, right? What a lot of people don't know is, yes, okay, there are energy constraints. And I'll throw out some statistics I've read recently where local governments, again, Phoenix, Arizona is really looking to grow their data centre power capacity over 500% in the coming years. And the amount of power that's going to require is enough to support 4.3 million households. For context, that's around - that's more than 1% of the total US population. That's just Phoenix, Arizona's data centre capacity, the power that will require in the coming years.
You've got other states like Virginia looking at 50 new data centres in the works at present, and they've now gone and contracted to the state's largest utility. So again, look at where you are in the value chain. Dominion. I don't think that's a stock we've covered. We've covered other energy producers. They need to build 40 gigawatts of additional capacity to meet that demand alone. Just 50 new data centres.
So again, this is going to look absurd in the context of a South African listener, for example, that looks at Eskom's total production capacity. That's just what's needed for the new investment in a single state in the United States.
Now it goes beyond energy, we've spoken a lot about energy, but it goes beyond energy because a lot of people don't know that data centres use water to cool their servers and they roughly consume – a 100 megawatt data centre will consume roughly 2 million litres of water per day and that's the equivalent of 6,500 households. This is now putting strain on water supply for nearby residential communities. This is starting to become a pressure point, so just pay attention to some of those.
Last point I want to touch on Ghost because this I found super interesting. I wanted to see what is the historical context here in terms of these mega trends that are changing the world. And I read this very interesting article that looked at Britain back when the railways were being rolled out, so the mid-1800s. And there was an interesting stat that I saw here which is that from 1844 to 1847, investment in railways as a percentage of British GDP rose from 5% to 13%. Okay, now I don't have to tell you that was a bubble that eventually burst. But for context, if we look at it versus the spend on AI over the last several years, let's call it four years here, American firms have invested roughly 3% to 4% of American GDP over the past four years. So it's still a long way off those double digits that we saw in Britain with the railway boom that actually happened. But eventually when that railway boom ended, not only did investment collapse, but it killed a lot of stuff further down the value chain. It killed some of the smaller rail operators.
Expect to see this. I don't know how far down this air road we are, but expect to see that as that ratchets up. Maybe we still have some way to go. But I'd be very concerned specifically because as we've indicated, the use case hasn't really been proven in terms of a tangible real-world output, but at the same time it's consuming tangible real-world resources: energy and water. Ghost, every time you say thank you to your Copilot for messing something up, apparently that utilizes a significant amount of energy as well. So is that sustainable in my book? I don't think so.
The Finance Ghost: Yeah, there are a whole lot of considerations here and I guess it begs the question as we finish this up: why am I so long Microsoft? And why are so many people invested in this space? And I think the reality is no one really knows where this goes. But I like to think that if anyone's going to figure it out, it's probably the likes of a Microsoft, for example. They've been at the top of the tech game for a very long time. I've actively avoided buying any of the more obscure, very hyped-up names where there's just no indication of what the valuation could really be. So what's happened with Oracle recently, for example, I haven't held Palantir. I mean, obviously there are times where I wish I had because a lot of this stuff has gone crazy. But I feel like if we are going to reach a point at which this bubble pops where everyone goes, hang on, we've actually way over invested in AI. Yes, we need some of this stuff, but we're in way too deep now. And it's big impairments and it's pullback and it's everything else - the big guns like a Microsoft or a Meta will be fine. It might hurt, it might sting, but through the cycle they will be okay. But the companies that have really hyped it up like crazy, thrown all their cash flow at it, etc. - they might not be so okay. Those are the share prices that can then halve or drop to a third of what they are currently.
I genuinely don't believe that that can happen to the likes of Microsoft. Maybe some will say that's a naive view or it's an overly bullish view. I mean, fair enough, that's what makes a market is everyone has different views. But that's why I'd rather own some of the leaders at a time like this than the very hyped-up small players or smaller players certainly.
Mohammed Nalla: Yeah, Ghost, I think that's a good place to wrap. I mean, I still have exposure to the sector, to the mega trend. I've chosen where I get that exposure. We covered ASML quite recently in Magic Markets Premium. Go and have a look at that, I've got an ASML position. I've historically had a Taiwan Semiconductor position and so there are different parts of the value chain you may choose to play in, but at the end of the day when the tide goes out, that's going to be painful across the ecosystem and like the railways, eventually some of the stronger players will remain and there will still be something left. But just be very sensitive to how much you're paying for the exposures that you're getting out there. That's something we always try and land on in Magic Markets Premium is that yes, there's a story, yes, there's a narrative, but at the end of the day, make sure you're paying a price, a multiple for a stock that you are comfortable with at the end of the day, or bearing in mind all of the risks.
Now, unfortunately, that's where we got to leave it this week. We're not Luddites. Let us know what you thought of the show. Hit us up on social media. It's @MagicMarketsPod, @FinanceGhost and @MohammedNalla or can find us on LinkedIn. Pop us a note on there. We hope you've enjoyed this until next week, same time, same place. Thanks and cheers.
The Finance Ghost: And as a passing goodbye, when you read this transcript, I hope you appreciate the pain that goes into getting it right. Just kidding. Thank you for listening and we'll see you next week.
