Latest / Merryn Talks Money / Can the AI Boom Survive a Margin Squeeze?

Transcript

Can the AI Boom Survive a Margin Squeeze? · Merryn Talks Money

• 0:02
  Speaker 1: Bloomberg Audio Studios Podcasts Radio News.
• 0:08
  Speaker 2: Last night I was walking around West Village back from
• 0:11
  Speaker 2: dinner and I heard just overheard with no context someone
• 0:14
  Speaker 2: said oh it's the big guys building data centers something
• 0:17
  Speaker 2: along that line in other words this is very much
• 0:19
  Speaker 2: in the zeitgeist you know so I think People say
• 0:22
  Speaker 2: that buy the news, sell the rumor. And, you know,
• 0:25
  Speaker 2: or when your barber starts pitching you a stock in,
• 0:28
  Speaker 2: you know, in something, it's time to get out.
• 0:40
  Speaker 1: Welcome to Merrin Talks Money, the podcast in which people
• 0:42
  Speaker 1: who know the markets explain the markets. I am Merrin
• 0:45
  Speaker 1: Somerset Webb, and this week I'm speaking with Steve Hu.
• 0:48
  Speaker 1: Steve is head of research at Silicon Data, a firm
• 0:51
  Speaker 1: founded in 2024, which aims to revolutionize compute markets by
• 0:56
  Speaker 1: providing price transparency and GPU performance data to traders, financial institutions,
• 1:01
  Speaker 1: and builders. Before he took on that role, he was
• 1:03
  Speaker 1: a quantitative researcher on the Bloomberg Index Research Team responsible
• 1:07
  Speaker 1: for systemic equity and multi-asset strategy research. And of course,
• 1:10
  Speaker 1: he has been on this podcast before. And that is
• 1:13
  Speaker 1: why I wanted to speak to him. He's previously recommended
• 1:16
  Speaker 1: reading material to us that has gone viral later. And also,
• 1:20
  Speaker 1: he really is an expert on everything digital. So I'm
• 1:24
  Speaker 1: hoping he'll be able to guide us through what is
• 1:26
  Speaker 1: going on at the moment. Now, during our conversation, we
• 1:30
  Speaker 1: get into everything from the price of GPU rental to
• 1:33
  Speaker 1: whether restaurants are going to be overwhelmed. Are your agents
• 1:36
  Speaker 1: trying to get the best tables? But before all that,
• 1:38
  Speaker 1: we start off where we left off last time we
• 1:41
  Speaker 1: had Steve on the show with situational awareness. Steve, welcome
• 1:45
  Speaker 1: to Merrin Talks Money.
• 1:46
  Speaker 2: Yes, that's right. Thank you for having me back, Merrin.
• 1:49
  Speaker 1: It's a pleasure. Anyway, back then. The last thing that
• 1:52
  Speaker 1: Steve said to us that we should all read is
• 1:56
  Speaker 1: something called situational awareness, which is kind of a pamphlet,
• 1:59
  Speaker 1: kind of a, I would go so far as to
• 2:01
  Speaker 1: call it a book. I think at the time you said,
• 2:03
  Speaker 1: you know, we should read this pamphlet. But in fact,
• 2:05
  Speaker 1: it's a good 200 pages long. And that gave a
• 2:10
  Speaker 1: sort of sense of what the author thought was going
• 2:14
  Speaker 1: to happen in the AI marketplace. It was written in
• 2:18
  Speaker 1: 2024 or 2023 originally.
• 2:20
  Speaker 2: I think that's right. Yeah, around then.
• 2:23
  Speaker 1: Anyway, you recommended it. I had a skim through it
• 2:25
  Speaker 1: and thought, wow, okay, that's kind of interesting. And then
• 2:28
  Speaker 1: about a year off was the whole thing went completely viral, right?
• 2:31
  Speaker 3: And by the.
• 2:32
  Speaker 1: Beginning of this year, everybody was raising situational awareness and
• 2:35
  Speaker 1: getting themselves quite upset about the whole thing. And I
• 2:38
  Speaker 1: wanted to start with this because it is fairly extraordinary
• 2:43
  Speaker 1: that the AI landscape is unfolding almost exactly the as
• 2:49
  Speaker 1: this pamphlet slash book suggested, in that we're now at
• 2:52
  Speaker 1: the bit where the extraordinary power of what is being
• 2:56
  Speaker 1: developed is being recognized. And the conversation has moved on from,
• 3:00
  Speaker 1: oh gosh, we can use this to help organize our
• 3:03
  Speaker 1: email to, okay, well, this is a massive military asset
• 3:06
  Speaker 1: and needs to be contained slash regulated by government. Well,
• 3:09
  Speaker 1: we're at that bit, right?
• 3:10
  Speaker 2: Yeah, indeed. The implications, I think, are enormous. And this
• 3:16
  Speaker 2: is regardless of how this whole investment cycle transpires, going
• 3:21
  Speaker 2: to be a transformative technology that's going to shape and
• 3:25
  Speaker 2: change society and our world beyond recognition. Yeah.
• 3:30
  Speaker 1: So we're mostly interested in the investment cycle, of course,
• 3:32
  Speaker 1: because that's what the podcast is for. But obviously, as
• 3:35
  Speaker 1: a background, the way that it's changing the world is
• 3:38
  Speaker 1: kind of important. But before we get into all that,
• 3:40
  Speaker 1: tell me about your new role, Steve, because you used
• 3:42
  Speaker 1: to be a colleague of ours here and then off
• 3:43
  Speaker 1: you went. Is that all of that?
• 3:46
  Speaker 2: That's right.
• 3:46
  Speaker 1: Yeah.
• 3:47
  Speaker 2: So I left Bloomberg in May to join a startup
• 3:52
  Speaker 2: that also happens to be founded by a former Bloomberg colleague,
• 3:56
  Speaker 2: Carmen Lee. And the company is called Silicon Data. It
• 4:00
  Speaker 2: is a data company that tries to bring data transparency
• 4:04
  Speaker 2: to the AI CapEx industry. And we are launching benchmark
• 4:08
  Speaker 2: indices on GPU rental rates and GPU token prices. And
• 4:13
  Speaker 2: we have also announced plans of launching futures contracts and
• 4:17
  Speaker 2: options on the CME pending CFTC approval.
• 4:20
  Speaker 1: Explain to us about that a little more. What are
• 4:22
  Speaker 1: these indices you're launching? On GPU pricing and GPU.
• 4:27
  Speaker 2: So GPU rental prices. So essentially, how much does it
• 4:30
  Speaker 2: cost to rent a H100? So all these, you know, GPUs,
• 4:35
  Speaker 2: Nvidia GPUs that are running in data centers that produce,
• 4:38
  Speaker 2: you know, AI tokens from which we run these models,
• 4:41
  Speaker 2: they are all named after famous, former physicists, right? So
• 4:44
  Speaker 2: Hopper 100, Blackwell, you know, 200, you know, and Ampere
• 4:48
  Speaker 2: 100 was a chip that's launched in 2020. So we
• 4:52
  Speaker 2: have data that tracks how much it costs to rent
• 4:56
  Speaker 2: an hour of these GPUs around the globe from a designer,
• 4:59
  Speaker 2: whether it's sitting in Western Texas or Finland or Southern
• 5:03
  Speaker 2: Europe somewhere or Asia or Middle East. And we aggregate
• 5:07
  Speaker 2: these numbers into a single, you know, apples to apples
• 5:10
  Speaker 2: comparable benchmark number. The same way you can think of
• 5:13
  Speaker 2: it is almost like creating a single-bedroom rental apartment index
• 5:17
  Speaker 2: for New York City, right? So you aggregate all the
• 5:19
  Speaker 2: rental agreements from across the city, whether it's in Soho, Harlem, Brooklyn,
• 5:23
  Speaker 2: and so on and so forth. You control for all
• 5:25
  Speaker 2: of them. Variables like square footage, you know, term of
• 5:29
  Speaker 2: the length of the agreement and so on. And you
• 5:31
  Speaker 2: make all these rental prices, apples to apples comparable, and
• 5:34
  Speaker 2: you create a singular index so that people can track it.
• 5:37
  Speaker 2: And this is an important variable if you are a
• 5:40
  Speaker 2: cloud provider, if you are one of the data centers
• 5:42
  Speaker 2: that's providing the compute, this is a source of revenue
• 5:45
  Speaker 2: for you. And if you are one of the AI labs,
• 5:48
  Speaker 2: whether it's OpenAI, Anthropic, one of the Neo labs, or
• 5:51
  Speaker 2: one of the inference platforms that post-train these open models
• 5:54
  Speaker 2: that allow people to, say, serve Chinese models, this is
• 5:58
  Speaker 2: a major cost for you, right? And we, Carmen, had
• 6:03
  Speaker 2: this vision that eventually we're going to enter a very
• 6:05
  Speaker 2: capital-intensive part of the cycle. There's going to be a
• 6:09
  Speaker 2: lot of people with a lot of exposure to AI,
• 6:11
  Speaker 2: and there's going to be need for hedging. There's going
• 6:13
  Speaker 2: to be need for... hedging on the revenue side, on
• 6:16
  Speaker 2: the cost side, and there's going to be traditional financial
• 6:18
  Speaker 2: instruments that help solve this problem. That means futures, that
• 6:21
  Speaker 2: means options, which is the reason why we have created
• 6:25
  Speaker 2: these trusted benchmarks against which futures contracts on compute as
• 6:30
  Speaker 2: a new commodity will settle. And we have to believe
• 6:33
  Speaker 2: that this will be one of the biggest, if not
• 6:35
  Speaker 2: the biggest commodity rivaling Okay, interesting.
• 6:40
  Speaker 1: So we've got basically what you're looking at here is
• 6:43
  Speaker 1: the price of renting compute from a data center. You've
• 6:47
  Speaker 1: got suppliers, their revenue, and then you've got everybody who
• 6:51
  Speaker 1: is using AI in anything is one way or another,
• 6:54
  Speaker 1: either one step or two steps removed from the price
• 6:57
  Speaker 1: of renting compute. the chips inside the data centers, right?
• 7:00
  Speaker 1: So that's really interesting information telling us what might or
• 7:04
  Speaker 1: might not be going on in the AI boom or
• 7:06
  Speaker 1: the AI bubble. What is that information telling you right now?
• 7:12
  Speaker 2: So right now we are in this very interesting era
• 7:15
  Speaker 2: where we have this, you know, very, I think, well
• 7:18
  Speaker 2: cited chart on our Twitter account that just poses price
• 7:23
  Speaker 2: of AI tokens or how much you can charge for
• 7:25
  Speaker 2: the models. with against how much it costs to rent
• 7:28
  Speaker 2: this different type of chips. And you see this interesting
• 7:31
  Speaker 2: pattern where since about the end of May, the token
• 7:35
  Speaker 2: index or the expenditure rate, the effective price paid for
• 7:38
  Speaker 2: the models has been coming down pretty significantly, almost half
• 7:43
  Speaker 2: of where it was back in the middle of the year.
• 7:46
  Speaker 2: In the meantime, rental rates for these chips have been
• 7:50
  Speaker 2: going up rapidly, continuously. So The analogy I like to
• 7:55
  Speaker 2: give is almost a bit like if you are inventing
• 7:57
  Speaker 2: the car for the first time, there's a luxury car,
• 8:00
  Speaker 2: say Mercedes, then there's a cheaper car, say Toyota, Corolla.
• 8:04
  Speaker 2: In the beginning, everyone was buying their Mercedes. These are
• 8:07
  Speaker 2: these frontier models, the top-of-the-line models from Cloud or OpenAI.
• 8:14
  Speaker 2: But eventually, people realized that some of these non-frontier models,
• 8:17
  Speaker 2: the cheaper models that have less capability but still evolving
• 8:21
  Speaker 2: over time, are saturating most of the everyday use, and
• 8:25
  Speaker 2: they're using a lot more of that. So everyone's buying
• 8:27
  Speaker 2: a ton and ton of Corollas, right? A lot of
• 8:30
  Speaker 2: Toyotas all over. So the thing is, you can have
• 8:33
  Speaker 2: this proliferation of car usage, and every car burns gasoline
• 8:38
  Speaker 2: or benzene, I guess, wherever, I guess, even in the UK,
• 8:41
  Speaker 2: that would drive up consumption of the underlying input. So
• 8:45
  Speaker 2: the same analogy applies in AI, in that the proliferation
• 8:49
  Speaker 2: of AI and everybody using it these cheaper models to
• 8:53
  Speaker 2: power their agents, right? And to read their emails, to
• 8:57
  Speaker 2: respond to things, that all consume compute, right? Even if
• 9:02
  Speaker 2: it's not necessarily using the highest margin and the most
• 9:06
  Speaker 2: profitable models, the most powerful models coming from the frontier labs.
• 9:09
  Speaker 1: Okay, so you've got the cost going up, but the
• 9:13
  Speaker 1: price coming down.
• 9:15
  Speaker 2: If you put it that way, yes, there is a
• 9:16
  Speaker 2: certain sense that's the case.
• 9:18
  Speaker 1: Okay, that seems problematic for some part of the market.
• 9:23
  Speaker 2: Indeed, for some part of the market, it could be problematic, yes.
• 9:26
  Speaker 2: In fact, interestingly, through sheer coincidence, the time when the
• 9:32
  Speaker 2: token index peaked was also around the time when, according
• 9:34
  Speaker 2: to some measurements, The anthropic revenue growth also slowed and
• 9:39
  Speaker 2: came to grow, you know, sort of either flatlined or,
• 9:42
  Speaker 2: you know, sort of much slower than it was growing
• 9:45
  Speaker 2: in the beginning of the year. So something tentatively related,
• 9:49
  Speaker 2: but not conclusive by any means.
• 9:52
  Speaker 1: Yeah, I mean, this does seem to be one of
• 9:53
  Speaker 1: the bigger problems for the anthropics and the open AIs,
• 9:56
  Speaker 1: which is that cheaper, simpler models work generally. as well
• 10:00
  Speaker 1: for the things that most of us do. Your early
• 10:03
  Speaker 1: adopters are going to be very different to the people
• 10:06
  Speaker 1: who are going to use your product long term. So
• 10:09
  Speaker 1: the majority of us are always going to use, I'm guessing,
• 10:11
  Speaker 1: I'm speaking for myself here, always going to use cheaper,
• 10:14
  Speaker 1: simpler models. So what does that mean? for the anthropics
• 10:20
  Speaker 1: and the open airs long-term?
• 10:23
  Speaker 2: Well, long-term is a long time away, but.
• 10:26
  Speaker 1: All right, medium-term, short-term.
• 10:29
  Speaker 2: I suspect this is going to create some questions surrounding
• 10:34
  Speaker 2: their upcoming IPOs and the, I think, high, shall we say, valuations,
• 10:41
  Speaker 2: relatively high valuations. The last time I think I heard
• 10:45
  Speaker 2: numbers anywhere between 1.5 $ 0. 5 to $ 2 trillion for Anthropic
• 10:50
  Speaker 2: and OpenAI, I think, a tad lower than that. I mean,
• 10:54
  Speaker 2: these are still non-public startups, young companies, but growing extremely rapidly,
• 11:03
  Speaker 2: scratching the soul. Anthropic has gone in almost no time
• 11:09
  Speaker 2: at all from having very little revenue to $ 60 billion
• 11:12
  Speaker 2: of annualized revenue run rate. and projected to go much
• 11:18
  Speaker 2: higher based on the extrapolated growth previously. So if there
• 11:22
  Speaker 2: are signs that the growth rate is indeed slowing, and
• 11:26
  Speaker 2: there could be structural reasons for why, for competitive reasons,
• 11:30
  Speaker 2: that it may not grow as rapidly as people have
• 11:32
  Speaker 2: previously thought, and that will put into question some of
• 11:35
  Speaker 2: the valuations and rosier hopes for their long-term value.
• 11:41
  Speaker 1: But it's also the case that at an IPO, the
• 11:44
  Speaker 1: buyers at IPO don't want to pay a price that
• 11:47
  Speaker 1: reflects every single piece of potential growth already priced in, right?
• 11:51
  Speaker 1: The buyer at IPO wants to buy at a price
• 11:54
  Speaker 1: that gives them something. And the problem possibly with these IPOs,
• 11:58
  Speaker 1: particularly with Anthropic and OpenAI, is that We look at
• 12:02
  Speaker 1: them now and I think increasingly people don't think of
• 12:04
  Speaker 1: them as an opportunity for new investors to get a
• 12:07
  Speaker 1: slice of the growth. They look at them and think
• 12:08
  Speaker 1: of them as fabulous exit opportunities for people who already
• 12:12
  Speaker 1: own the shares. And that's not what the market is
• 12:15
  Speaker 1: supposed to do or it's not what people expect the
• 12:18
  Speaker 1: market to do. You get an IPO, you buy something
• 12:20
  Speaker 1: that has growth left in it. And even if these
• 12:23
  Speaker 1: companies grow, if you've paid in advance for all the growth,
• 12:28
  Speaker 1: what's in it for you? And so you get the
• 12:31
  Speaker 1: sense that the market is beginning to shift the way
• 12:34
  Speaker 1: they're thinking about these IPOs.
• 12:36
  Speaker 2: Oh, you ask very good questions. Yeah, I think those
• 12:38
  Speaker 2: are questions that people should be asking.
• 12:49
  Speaker 1: Steve, before we go any further, it occurs to me
• 12:51
  Speaker 1: that we have been talking about tokens, tokens this, tokens that,
• 12:54
  Speaker 1: tokens this. And, you know, there is a tendency for
• 12:57
  Speaker 1: all of us at this point in the cycle, we've
• 12:59
  Speaker 1: been talking about things for a while now, not to
• 13:01
  Speaker 1: define our terms. So would you mind just for our
• 13:04
  Speaker 1: less experienced listeners explaining what we mean by a token?
• 13:09
  Speaker 2: We've all used ChatGPT, right? And how you use it
• 13:12
  Speaker 2: is you enter a bunch of texts, say, you know,
• 13:14
  Speaker 2: sort of, uh, help me, uh, plan a little trip to,
• 13:17
  Speaker 2: you know, uh, North England. This, you know, coming weekend
• 13:21
  Speaker 2: with my family is going to give you back an
• 13:24
  Speaker 2: itinerary telling you, go here, go here, go here. And
• 13:27
  Speaker 2: if you are here, you check out this restaurant and that.
• 13:29
  Speaker 2: So a token is basically a length of text that
• 13:34
  Speaker 2: you're entering. So if you're entering a long prompt, a
• 13:37
  Speaker 2: long set of conditions about what you want, there's a
• 13:41
  Speaker 2: large number of tokens. And if the output is a
• 13:43
  Speaker 2: large output, long output, that is a large number of
• 13:46
  Speaker 2: tokens too. Now, it doesn't necessarily have to be the
• 13:50
  Speaker 2: case that your input or output are just text, right? Increasingly,
• 13:54
  Speaker 2: we're seeing people inputting a screenshot into a prompt to say, hey,
• 14:00
  Speaker 2: explain how this works or what is in the picture
• 14:03
  Speaker 2: or whatever. Nevertheless, the computer is basically, the AI is
• 14:06
  Speaker 2: translating the text, you know, the picture into essentially text
• 14:11
  Speaker 2: or tokens, right? And that's the only way the computer
• 14:14
  Speaker 2: is able to think. Okay.
• 14:16
  Speaker 1: So the way that an LLM, a large language model,
• 14:18
  Speaker 1: what we all think of when we talk about AI
• 14:20
  Speaker 1: these days, although obviously it's not, there's more than that
• 14:22
  Speaker 1: to it these days. But nonetheless, an LLM is what
• 14:25
  Speaker 1: we think of when we think AI. So you give
• 14:27
  Speaker 1: it a prompt, ask it a question, and then you
• 14:29
  Speaker 1: keep developing that. You ask more and more questions, you
• 14:31
  Speaker 1: put in more and more prompts. The reason it becomes
• 14:34
  Speaker 1: so expensive or historically expensive. has been expensive, is that
• 14:37
  Speaker 1: the tokens are rerun each time, right? So it's not
• 14:40
  Speaker 1: like you're starting from where you left off. The model
• 14:43
  Speaker 1: has to rerun every token. And that's what historically built
• 14:46
  Speaker 1: the price up.
• 14:47
  Speaker 2: Well, a core challenge with AI, with LLM, has been
• 14:53
  Speaker 2: managing memory and long context, you know, exactly as you say, right?
• 14:58
  Speaker 2: When you are in a conversation, you know, with an
• 15:00
  Speaker 2: AI person, just like we know in a conversation with
• 15:03
  Speaker 2: a human being, you don't want to start over. You
• 15:05
  Speaker 2: want people to remember the context of where you got to,
• 15:08
  Speaker 2: where you are, so that the next sentence makes sense
• 15:11
  Speaker 2: within the context of what you were talking about. But
• 15:13
  Speaker 2: of course, as the conversation gets longer and longer, the
• 15:17
  Speaker 2: amount of information that has been talked about, the amount
• 15:20
  Speaker 2: of memory, things that needs to be memorized, it just
• 15:23
  Speaker 2: becomes bigger and bigger, right? And at some point, you
• 15:26
  Speaker 2: run out of room to store that memory, but also
• 15:29
  Speaker 2: memorizing it. So there's all these attempts of essentially compressing
• 15:34
  Speaker 2: the memory, right?
• 15:36
  Speaker 3: Yeah.
• 15:36
  Speaker 2: You know, it's a little bit not so different from
• 15:40
  Speaker 2: when we human beings talk to people, we don't remember
• 15:43
  Speaker 2: every single detail in a conversation that's been mentioned. We
• 15:45
  Speaker 2: make mental notes of the key points that have been
• 15:48
  Speaker 2: touched upon, right? And we make recall. And what a
• 15:51
  Speaker 2: computer is able to do even better, increasing better, you know,
• 15:55
  Speaker 2: algorithms is actually able to sort of make that recall better.
• 15:57
  Speaker 2: even more uh refinely being able to actually go back
• 16:00
  Speaker 2: oh actually let's dial back to that moment when we
• 16:02
  Speaker 2: talk about something in the greater detail stuff that you
• 16:05
  Speaker 2: may not remember very well and that makes ai super
• 16:07
  Speaker 2: powerful and in terms of the practical implications part of
• 16:12
  Speaker 2: the reason why we have seen such an explosive growth
• 16:15
  Speaker 2: in the stock prices of memory and you know storage
• 16:20
  Speaker 2: stocks of such as sandisk or micron in the us
• 16:24
  Speaker 2: or sk hynix and samsung in in uh In Korea,
• 16:28
  Speaker 2: it's exactly that, right? We suddenly have an exercise where
• 16:33
  Speaker 2: the demand for memory and demand for storage of data
• 16:37
  Speaker 2: has just gone exponential, right? Every single conversation everybody's having
• 16:43
  Speaker 2: is generating tons and tons more of stuff that needs
• 16:47
  Speaker 2: to be remembered and stored down. So that's the reason
• 16:52
  Speaker 2: why Korea is suddenly seeing such intense attention from across
• 16:58
  Speaker 2: the globe.
• 17:00
  Speaker 1: That's not particularly comfortable, by the way, but thank you
• 17:03
  Speaker 1: for that explanation. Very useful. Can we go back to
• 17:06
  Speaker 1: data centers? We were talking about the cost of GPU
• 17:12
  Speaker 1: and rental going up. Is there still, despite this massive
• 17:16
  Speaker 1: CapEx boom, is there a shortage of data centers? I mean,
• 17:20
  Speaker 1: we're seeing, as you know, protests against data centers, people protesting,
• 17:24
  Speaker 1: People don't really like to have them built near them.
• 17:26
  Speaker 1: We've seen some cancellations most recently in the UK, a
• 17:28
  Speaker 1: big cancellation because they can't get connected to an energy supply, etc.
• 17:32
  Speaker 1: And that's a big problem, globally connecting your data center
• 17:36
  Speaker 1: to an energy supply. So is there still a shortage
• 17:40
  Speaker 1: despite all this capex? And is that what those GPU
• 17:43
  Speaker 1: rental prices are reflecting?
• 17:45
  Speaker 2: So shortage is, you know, a complicated concept, right? So
• 17:50
  Speaker 2: As it stands today, everybody is able to get on
• 17:53
  Speaker 2: an AI and get some output from it. But, you know,
• 17:57
  Speaker 2: shortage is mostly reflected in terms of the anticipated growth
• 18:02
  Speaker 2: in demand of a lot more people using a lot
• 18:05
  Speaker 2: more AI and the ability to supply into it. And
• 18:09
  Speaker 2: to that extent, on the reasonable estimates, you know, of
• 18:13
  Speaker 2: everybody just using, I think, AI more meaningfully, as you
• 18:16
  Speaker 2: said earlier, The average person is not using AI in
• 18:19
  Speaker 2: a very, very high power way, right? Everyone is using
• 18:22
  Speaker 2: it mostly as like a glorified search. But there is
• 18:25
  Speaker 2: actually a lot more ways you can actually use it
• 18:28
  Speaker 2: in an agentic way where it's continuously running. And that
• 18:31
  Speaker 2: is actually a lot more demand for compute. And we
• 18:35
  Speaker 2: multiply that across the global population, especially within developed economies.
• 18:39
  Speaker 2: And that would imply a level of demand that is
• 18:42
  Speaker 2: not currently matched by the growth in supply. So to
• 18:45
  Speaker 2: that extent, yes, there is a shortage. In data centers,
• 18:49
  Speaker 2: not necessarily.
• 18:50
  Speaker 1: But it's a forecast shortage as opposed to a current shortage.
• 18:52
  Speaker 2: Of course, of course. It's a forecast shortage. Nevertheless, I
• 18:56
  Speaker 2: think it's a reasonable forecast given the recent trend. I mean,
• 19:00
  Speaker 2: I do think people are going to be using more
• 19:02
  Speaker 2: AI and it's going to proliferate, especially with more, I think,
• 19:07
  Speaker 2: products that are built with better product market fit, with
• 19:10
  Speaker 2: good understanding, anticipating consumer needs, like the latest product from Meta, Muse,
• 19:15
  Speaker 2: and so on. There's going to be more of those
• 19:17
  Speaker 2: and combine that with the fact that there is a
• 19:20
  Speaker 2: shortage in supply on data centers because of the shortage
• 19:23
  Speaker 2: of just sheer physical constraint of power, right? The lack thereof, right?
• 19:28
  Speaker 2: There's not enough sites for us to build large data centers.
• 19:31
  Speaker 2: We're having to build smaller ones in different corners of
• 19:34
  Speaker 2: the world and contending with energy shortage. I think local
• 19:39
  Speaker 2: political pushback from local residents is, which I think is
• 19:43
  Speaker 2: maybe a little bit overblown in terms of the actual
• 19:45
  Speaker 2: net impact.
• 19:46
  Speaker 1: Is there a case to be made for demand not developing?
• 19:50
  Speaker 1: in the way you expect? I mean, I was surprised
• 19:53
  Speaker 1: by a stat, I don't even know if it was true,
• 19:55
  Speaker 1: but it rang a bit true, that more people pay
• 19:57
  Speaker 1: for weight loss drugs in the US than pay to
• 20:01
  Speaker 1: use AI in any way. Fascinating at this point in
• 20:05
  Speaker 1: the cycle. And the other thing to say on that
• 20:08
  Speaker 1: is that you know, look at Muse and the other
• 20:10
  Speaker 1: agents that people are starting using fairly regularly. Is it
• 20:14
  Speaker 1: sustainable for everyone to use an AI agent in this way?
• 20:18
  Speaker 1: And that we're already seeing, for example, local government getting
• 20:22
  Speaker 1: clogged up by people having their agents complain about their
• 20:25
  Speaker 1: parking tickets and potholes, planning getting clogged up with masses
• 20:31
  Speaker 1: of objections generated by agents and even restaurants that are
• 20:35
  Speaker 1: hard to get into being overwhelmed by... By phone calls
• 20:39
  Speaker 1: and emails from agents. And there comes a point when rather,
• 20:43
  Speaker 1: and this is at the very beginning of this cycle,
• 20:46
  Speaker 1: there comes a point when instead of reducing friction in
• 20:50
  Speaker 1: our lives, if everyone's using agents to do all these things,
• 20:54
  Speaker 1: the world just collapses. I mean, how does this actually
• 20:59
  Speaker 1: work if already when only a tiny percentage of the
• 21:03
  Speaker 1: population is using an agent to sort their email box,
• 21:06
  Speaker 1: to do their shopping, to do their restaurant renovations, to
• 21:09
  Speaker 1: complain to the council, to do their taxes? Tiny, tiny
• 21:12
  Speaker 1: already is overwhelming some of our institutions and our businesses.
• 21:17
  Speaker 1: How does that play out?
• 21:20
  Speaker 2: So you bring up two main points that are both
• 21:23
  Speaker 2: issues are very good. So let me address each of them.
• 21:25
  Speaker 2: The first one you, you, you raise is regarding demand
• 21:29
  Speaker 2: growth and how reliable is it? Um, in the, it
• 21:34
  Speaker 2: is the case when I say demand, we are looking
• 21:36
  Speaker 2: at the current condition of how much, um, you know, uh,
• 21:39
  Speaker 2: LGBT and tropic is selling, you know, they are sort
• 21:43
  Speaker 2: of monthly plans at, uh, you still a fixed price
• 21:45
  Speaker 2: or you can eat, you know, kind of a plan. They,
• 21:48
  Speaker 2: they throttle usage and they give you an amount of
• 21:50
  Speaker 2: credit at which point you run out. Muse, by the way,
• 21:53
  Speaker 2: is free. So, um, At current prices, demand is growing
• 21:58
  Speaker 2: in a certain way. It is unknown at this point
• 22:02
  Speaker 2: how much people are going to be willing to pay
• 22:05
  Speaker 2: for AI per usage, especially if it's going to be mediated.
• 22:09
  Speaker 2: It sounds as though that Mela and Zuckerberg have been
• 22:14
  Speaker 2: talking about the fact that they want to make this
• 22:16
  Speaker 2: widely available to people either for free or at very
• 22:20
  Speaker 2: low cost, and they will find ways of monetizing it
• 22:23
  Speaker 2: via other means such as e-commerce and collecting a fee
• 22:26
  Speaker 2: that way, and so on. I don't think the exact
• 22:29
  Speaker 2: channel of the economics of how AI companies will be
• 22:34
  Speaker 2: paid while demand grows has been fully articulated. I think
• 22:39
  Speaker 2: there are competing narratives. So you are right to question
• 22:42
  Speaker 2: or ask the question that what is going to be
• 22:46
  Speaker 2: the growth of demand for AI if costs start going up, right?
• 22:51
  Speaker 2: I don't mean to say that nobody's paying for AI.
• 22:55
  Speaker 2: That's clearly not the case. I'm just saying that how
• 22:58
  Speaker 2: AI will be monetized, I think there are competing narratives
• 23:03
  Speaker 2: and also depends a little bit on whether or not
• 23:05
  Speaker 2: you're looking at consumer use case or enterprise, right? Enterprises
• 23:08
  Speaker 2: are clearly paying very large bills. And so enterprises clearly
• 23:14
  Speaker 2: are paying these models by the token prices and that are,
• 23:17
  Speaker 2: you know, publicly known. And then there is cashing, there
• 23:20
  Speaker 2: are ways in which you get some discounts. But generally speaking, yeah,
• 23:24
  Speaker 2: there is an anticipation that this is going to be
• 23:27
  Speaker 2: a metered usage for a lot of it. You know,
• 23:30
  Speaker 2: the question remains, as to whether or not once this proliferates,
• 23:36
  Speaker 2: you know, the usage will keep up, right? And that
• 23:40
  Speaker 2: is part of the discussion when we are talking about
• 23:43
  Speaker 2: the economics insofar as who actually captures the rent. If
• 23:47
  Speaker 2: you invent this extraordinary thing and, you know, open AI,
• 23:52
  Speaker 2: you know, anthropic, and then do you actually sort of
• 23:56
  Speaker 2: manage to recuperate, right, those initial discoveries, the training runs, uh,
• 24:02
  Speaker 2: that came up with all these models, uh, if, uh,
• 24:05
  Speaker 2: subsequently there are models that are based on these models, uh,
• 24:09
  Speaker 2: distilled versions, either, you know, themselves distilled cheaper versions of
• 24:13
  Speaker 2: those models or, you know, others, uh, um, such that
• 24:17
  Speaker 2: they are sold at a much cheaper price and people
• 24:20
  Speaker 2: optimize on cost and quality, then, you know, so that's
• 24:24
  Speaker 2: the open question.
• 24:25
  Speaker 1: Okay, second bit. Second bit.
• 24:26
  Speaker 3: Yes.
• 24:27
  Speaker 1: Obviously, this is, you know, the use case for AI
• 24:29
  Speaker 1: is now, or for LLMs at least, is very clear.
• 24:32
  Speaker 1: We know what they can do and we know what
• 24:33
  Speaker 1: they can't do. And we know that we need new
• 24:36
  Speaker 1: types of models to do the things they can't do.
• 24:38
  Speaker 1: For example, you know, precision, audit, compliance, regulatory stuff, et cetera.
• 24:42
  Speaker 1: We know that. So the use case is certain. But
• 24:45
  Speaker 1: when you are going back to my question about retail agents,
• 24:49
  Speaker 1: agents working for ordinary people, and the way this can
• 24:53
  Speaker 1: very quickly overwhelm the systems that we already have in place.
• 24:57
  Speaker 2: Indeed, right. If you have proliferation, everybody just activates agents and, hey,
• 25:01
  Speaker 2: go pick up, you know, go find me a reservation,
• 25:03
  Speaker 2: go find me a ticket, go do whatever. On the
• 25:06
  Speaker 2: receiving end, you know, you're going to find yourself overwhelmed
• 25:10
  Speaker 2: with all the agentic requests And we're going to find
• 25:14
  Speaker 2: ourselves in a situation where everyone can generate a ton
• 25:19
  Speaker 2: of AI output, AI slop, but we don't have the
• 25:21
  Speaker 2: ability to process it all, right? And in the enterprise context,
• 25:26
  Speaker 2: I think it's a similar thing going on where you
• 25:28
  Speaker 2: can have enormous efficiency gains where someone's generating tons and
• 25:32
  Speaker 2: tons of new PPTs and presentations or code and whatnot.
• 25:37
  Speaker 2: But then you have other parts of the business that
• 25:39
  Speaker 2: actually process that output to actually take it into a
• 25:41
  Speaker 2: finished product to turn that into a business outcome and revenue.
• 25:45
  Speaker 2: And I think that analogy applies and that is a
• 25:49
  Speaker 2: very big open question as to what happens You know,
• 25:52
  Speaker 2: I don't have good answers for you. I suspect some
• 25:55
  Speaker 2: of the questions and pain points you're pointing out are
• 25:57
  Speaker 2: going to actually surface in the coming months.
• 26:01
  Speaker 1: Yeah, I mean, I have this vague worry that we
• 26:03
  Speaker 1: end up with everyone has an agent trying to make
• 26:06
  Speaker 1: them a reservation at the new great restaurant in London.
• 26:08
  Speaker 1: The new great restaurant in London then has to, you know,
• 26:11
  Speaker 1: create a big pile of agents to deal with the
• 26:13
  Speaker 1: agents bothering it for reservations.
• 26:15
  Speaker 2: Either that or they shut down agentic inbound, you know, altogether.
• 26:22
  Speaker 2: Or you've got to basically force a new type of
• 26:24
  Speaker 2: restaurants that is AI native that can actually handle these things.
• 26:28
  Speaker 2: So you have to be a new type of business.
• 26:30
  Speaker 2: sort of, you know, they say shortage or whatever is
• 26:33
  Speaker 2: always the mother of innovation. I mean, there's got to be, yeah,
• 26:37
  Speaker 2: there's got to be things that we've got to give, right?
• 26:39
  Speaker 1: Exactly.
• 26:48
  Speaker 3: Anyway, so we've talked about a lot of things. We've
• 26:50
  Speaker 3: talked about a lot of the uncertainties. We've talked about
• 26:52
  Speaker 3: who's using, who's demanding, et cetera.
• 26:55
  Speaker 2: Yeah.
• 26:55
  Speaker 1: Bubble or boom.
• 26:57
  Speaker 2: I think the question is not so much that whether
• 26:59
  Speaker 2: it's a bubble or not. I mean, every single investment cycle,
• 27:02
  Speaker 2: every Quebec cycle has always gone through the boom and bust.
• 27:06
  Speaker 2: The question is a matter of timeline. The market has
• 27:09
  Speaker 2: already topped in the local fashion in the middle of
• 27:12
  Speaker 2: the year, right? So there is definitely a sense in
• 27:16
  Speaker 2: which people, a lot of the questions you were just
• 27:17
  Speaker 2: raising with regard to ROI, with regard to how does
• 27:21
  Speaker 2: this investment produce a return and how do these frontier
• 27:26
  Speaker 2: labs actually justify their valuation, which has been the source
• 27:31
  Speaker 2: for a lot of their CapEx spending commitments because they've
• 27:35
  Speaker 2: been selling shares, right? They've been raising money from VCs
• 27:38
  Speaker 2: and other investors in order to raise capital, which in
• 27:41
  Speaker 2: turn to acquire, compute, and build data centers or commit
• 27:45
  Speaker 2: to renting data centers. So that whole narrative, that whole cycle,
• 27:50
  Speaker 2: cyclical narrative, could be put into question if more and
• 27:55
  Speaker 2: more people are going to start questioning, you know, some
• 27:57
  Speaker 2: of the questions that you're just raising and good answers
• 28:00
  Speaker 2: are not forthcoming. So I do think we are in
• 28:04
  Speaker 2: an environment of the market where these questions are growing
• 28:09
  Speaker 2: louder and we're not early in this trade. I think
• 28:12
  Speaker 2: last night I was walking around West Village back from
• 28:15
  Speaker 2: dinner and I heard, just overheard with no context, someone said, oh,
• 28:19
  Speaker 2: it's the big guys building data centers or something along
• 28:21
  Speaker 2: that line. In other words, this is very much in
• 28:24
  Speaker 2: the zeitgeist. So I think people say that buy the news,
• 28:27
  Speaker 2: sell the rumor, or when your barber starts pitching you
• 28:31
  Speaker 2: a stock in something, it's time to get out. I mean,
• 28:35
  Speaker 2: I think there's some truth to it in that we
• 28:38
  Speaker 2: are probably in a pretty feverish moment of the market
• 28:45
  Speaker 2: that some of the smarter money has already either taking
• 28:49
  Speaker 2: chips off the table or having their chips taken off
• 28:52
  Speaker 2: the table for them. Also, I do think this is
• 28:54
  Speaker 2: an important point to make, is that nothing ever happens
• 28:57
  Speaker 2: in a straight line. Even if this was the end
• 29:00
  Speaker 2: of the bubble, we can have extremely violent counter rallies
• 29:04
  Speaker 2: as uncertainties regarding whether or not this really was the
• 29:08
  Speaker 2: end can actually cause that to happen, number one. Number
• 29:12
  Speaker 2: two is that I think There's a lot of people
• 29:17
  Speaker 2: in the bearish circles. Some of them actually sort of
• 29:22
  Speaker 2: have this position that this whole thing is vaporware and
• 29:25
  Speaker 2: it's all useless. And it's just been a con job
• 29:28
  Speaker 2: that none of this is ever going to be useful
• 29:30
  Speaker 2: to anybody. Nobody actually wants it. I think that is
• 29:33
  Speaker 2: fundamentally untrue.
• 29:34
  Speaker 1: That's a man who's never going to get a restaurant
• 29:36
  Speaker 1: reservation again.
• 29:37
  Speaker 2: Yeah. he's short in the market, in which case he
• 29:42
  Speaker 2: might very well just, yeah. But I do think this
• 29:46
  Speaker 2: is a fundamentally transformative technology. I do think it's going
• 29:50
  Speaker 2: to reshape society. And I eventually think the ball is
• 29:53
  Speaker 2: that people who are bullish on the technology are going
• 29:55
  Speaker 2: to be proven right. And there is a chance in
• 29:58
  Speaker 2: which things work out right in that you find products
• 30:02
  Speaker 2: with good product and product market fit you have an
• 30:04
  Speaker 2: acceleration of adoption and people actually do want to pay
• 30:08
  Speaker 2: for things that give them utility. I want to be
• 30:10
  Speaker 2: a little bit contrarian just because these days, it seems
• 30:14
  Speaker 2: to have become very fashionable. I have not met a
• 30:17
  Speaker 2: single person who doesn't think this is just a repeat
• 30:19
  Speaker 2: of the dot-com bubble that we're staring at within the
• 30:21
  Speaker 2: next 12 months, some sort of spectacular crash and blow
• 30:27
  Speaker 2: up of the US tech bubble and stock market. And
• 30:31
  Speaker 2: we're going to repeat the 2010 I think there is
• 30:36
  Speaker 2: a chance in which the mismatch, I think this is
• 30:40
  Speaker 2: fundamentally a problem of potential mismatch of timing, not necessarily
• 30:44
  Speaker 2: there won't be demand for it, but rather the demand
• 30:48
  Speaker 2: could show up slower than some of the most bullish
• 30:51
  Speaker 2: people in SF seem to think. They seem to think
• 30:54
  Speaker 2: that because it works for coding, all the programmers adopt AIT,
• 30:59
  Speaker 2: rapidly overnight. This is going to happen everywhere. Ignoring a
• 31:03
  Speaker 2: lot of the practical bureaucratic hurdles to adoption that could
• 31:07
  Speaker 2: slow things down such that the supply could actually show
• 31:11
  Speaker 2: up or show up in 2027, 2028 and then you
• 31:13
  Speaker 2: find yourself not having as much demand because enterprise adoption
• 31:18
  Speaker 2: wasn't as rapid and maybe consumer adoption is not as
• 31:21
  Speaker 2: rapid because you didn't find the right product market fit
• 31:24
  Speaker 2: for the products that actually really took off the same
• 31:27
  Speaker 2: way iPhone took hold with mobile. So I think that
• 31:31
  Speaker 2: remains to be seen. This risk of an air pocket
• 31:34
  Speaker 2: is very much real, and we are clearly living through
• 31:37
  Speaker 2: it in different manifestations. But I think there's also a
• 31:40
  Speaker 2: risk of of a right tail risk by things that
• 31:43
  Speaker 2: really do go right, in which case we can actually
• 31:45
  Speaker 2: see a pretty violent melt up as well.
• 31:48
  Speaker 1: Okay, Steve, so you've given me two options here. One
• 31:50
  Speaker 1: is that we're closer to the end than the beginning,
• 31:52
  Speaker 1: and the other is that we might have a massive
• 31:54
  Speaker 1: melt up. Which one do you think is most likely?
• 31:57
  Speaker 1: Melt up or closer to the end than the beginning?
• 32:00
  Speaker 2: The funny thing is, those two things don't have to
• 32:02
  Speaker 2: necessarily be.
• 32:04
  Speaker 1: They don't have to be mutually exclusive. They could both happen.
• 32:08
  Speaker 2: Indeed. You know, what happens if the market melts up
• 32:11
  Speaker 2: and then crashes? You know, you're not necessarily going to
• 32:14
  Speaker 2: be nimble enough to time yourself perfectly to, you know,
• 32:18
  Speaker 2: catch one and then boy the other.
• 32:19
  Speaker 3: So.
• 32:21
  Speaker 2: I'm sorry, that's a little bit of unsatisfactory answer if
• 32:23
  Speaker 2: you are looking for me to give you like a
• 32:25
  Speaker 2: tummy signal.
• 32:26
  Speaker 3: Yes, I am.
• 32:26
  Speaker 1: Okay, so what do we do? What do we do?
• 32:29
  Speaker 1: We're an ordinary retail investor. We're like everybody. We are overweight,
• 32:34
  Speaker 1: probably by mistake, overweight US tech, by mistake, overweight AI.
• 32:39
  Speaker 1: What do we do now? Is the answer just to rebalance,
• 32:42
  Speaker 1: pull back? But then, of course, is everything AI now?
• 32:45
  Speaker 1: Is there a way out? Is there a way that
• 32:47
  Speaker 1: we should be investing right now to protect ourselves a little?
• 32:53
  Speaker 2: Well, first of all, I am not here to give
• 32:56
  Speaker 2: anybody any investment advice. No, no, this is not advice.
• 32:59
  Speaker 2: In general, if we are just chatting casually, I think
• 33:05
  Speaker 2: It is fair to say there are plenty of things
• 33:07
  Speaker 2: that are not directly AI-related. You know, energy, you know,
• 33:13
  Speaker 2: like traditional fossil energy stocks and, you know, oil companies.
• 33:17
  Speaker 2: They are not AI, right?
• 33:19
  Speaker 1: Yeah, okay. Thank you. Now, the other thing I wanted
• 33:22
  Speaker 1: to ask you was about this business that is all
• 33:24
  Speaker 1: over the news this week, about Anthropic is saying that
• 33:28
  Speaker 1: they intend to give 5% of their shares to into
• 33:32
  Speaker 1: a foundation that will support people who are affected by AI.
• 33:38
  Speaker 1: So it should be nice to people who've been affected
• 33:42
  Speaker 1: by it. Maybe they lost their job or to look
• 33:43
  Speaker 1: into this kind of thing, et cetera. And then at
• 33:45
  Speaker 1: the same time, we've got Elon Musk saying that he's
• 33:48
  Speaker 1: going to give a SpaceX share to every kid in America.
• 33:53
  Speaker 1: And all this looks to me a bit like, well,
• 33:55
  Speaker 1: we better give it away before they bring in a
• 33:57
  Speaker 1: wealth tax. What does it look like to you?
• 34:01
  Speaker 2: I did not make that connection. I knew that anthropic
• 34:05
  Speaker 2: has always had this longstanding worry, very vocal, you know,
• 34:09
  Speaker 2: about the danger of AI poses on society, which is
• 34:13
  Speaker 2: a bit ironic because, you know, they are the one
• 34:16
  Speaker 2: actually that's spearheading or, you know, at the forefront of
• 34:20
  Speaker 2: inventing this technology. So it's almost as though like, oh,
• 34:24
  Speaker 2: we'll have this product. I look, look, you know, look,
• 34:27
  Speaker 2: the danger is actually, you know, posing. So, Yeah, it
• 34:30
  Speaker 2: is a bit interesting, I think, from a behavioral point
• 34:32
  Speaker 2: of view. But I also kind of appreciate from their perspective,
• 34:37
  Speaker 2: I think, in a much more charitable fashion to understand
• 34:39
  Speaker 2: that there are other people developing this technology. One of
• 34:43
  Speaker 2: the many competing forces, this is going to be developed.
• 34:46
  Speaker 2: So they want to pay their own, make their contribution
• 34:49
  Speaker 2: to help society deal with the consequences of I don't
• 34:53
  Speaker 2: know what 5% of their market cap, you know, put
• 34:56
  Speaker 2: aside towards, you know, how this exactly will pan out.
• 34:59
  Speaker 2: I mean, you know, there's not a lot of details
• 35:02
  Speaker 2: yet as to how the program will work. But I
• 35:04
  Speaker 2: do think the end goal and the question as to
• 35:08
  Speaker 2: how this will, big society that I have no doubt.
• 35:11
  Speaker 2: I think AI is going to be hugely transformative in
• 35:14
  Speaker 2: good and potentially bad ways as well. Not just only
• 35:19
  Speaker 2: in the popular narrative of, you know, disrupting jobs, but
• 35:22
  Speaker 2: also this idea that you suddenly have a technology that
• 35:27
  Speaker 2: enables everybody to do things that previously could not, right?
• 35:32
  Speaker 2: You know, and there could be all sorts of risks
• 35:36
  Speaker 2: unleashed onto the system. that we previously didn't have the
• 35:39
  Speaker 2: capability of dealing with. So, yeah, I mean, within that context,
• 35:44
  Speaker 2: I don't know what the 5% share will do. But
• 35:49
  Speaker 2: it's good that somebody is funding somebody to study these issues.
• 35:53
  Speaker 2: And I hadn't heard about the SpaceX one. So I'll
• 35:56
  Speaker 2: have to learn more about that.
• 35:57
  Speaker 1: The SpaceX one is interesting. I mean, it smacks to
• 35:59
  Speaker 1: me a little bit about, you know, buying indulgences. We
• 36:02
  Speaker 1: may destroy the human race, but at least, you know,
• 36:04
  Speaker 1: we paid up a little up front.
• 36:06
  Speaker 2: Yeah. You know, the one question I've got to ask you,
• 36:09
  Speaker 2: maybe we'll have to do this on a follow-up conversation.
• 36:11
  Speaker 2: So what this all looks like from the perspective of
• 36:13
  Speaker 2: a European person, a British person, you could looking, you know,
• 36:18
  Speaker 2: from across the Atlantic, I think this seems mostly of
• 36:22
  Speaker 2: an American activity, right?
• 36:23
  Speaker 1: Yeah, we're just irritated we're not going to get a
• 36:26
  Speaker 1: SpaceX share.
• 36:27
  Speaker 3: Why can't my kids have SpaceX shares too? What's with
• 36:31
  Speaker 3: all the borders?
• 36:32
  Speaker 1: I don't know. Anyway, listen, the last question I must
• 36:35
  Speaker 1: ask you before we go, because I've had you far
• 36:37
  Speaker 1: too long, is, so last time I said to you,
• 36:40
  Speaker 1: what should we read? You said read situational awareness. And boy,
• 36:43
  Speaker 1: were you ever right. So now I'm going to say
• 36:45
  Speaker 1: to you, Steve, what should we read?
• 36:47
  Speaker 2: I think people should read some poetry. rediscover our humanity oh.
• 36:54
  Speaker 3: God you know that it's an awful time on the
• 36:57
  Speaker 3: other day said something very similar i said what should
• 36:59
  Speaker 3: we renew the philosophy philosophy get ready get ready for
• 37:04
  Speaker 3: your existential crisis and you're clearly in the same zone
• 37:07
  Speaker 3: anyway indeed so thank.
• 37:11
  Speaker 1: You so much for joining us today steve we really.
• 37:13
  Speaker 2: Appreciate it thank you very much for having marine always
• 37:17
  Speaker 2: a good time and i hope the next one won't
• 37:19
  Speaker 2: take us another two years It won't.
• 37:21
  Speaker 1: It won't. Thanks for listening to this week's Merrin Talks Money.
• 37:26
  Speaker 1: If you like our show, rate, review, and subscribe wherever
• 37:29
  Speaker 1: you listen to your podcasts. And keep sending questions or
• 37:32
  Speaker 1: comments to merrinmoney at bloomberg.net. You can also follow me
• 37:36
  Speaker 1: and John on Twitter or X. I'm at MerrinSW and
• 37:38
  Speaker 1: John is John underscore Stepak. This episode was hosted by me,
• 37:42
  Speaker 1: Merrin's Unset Web. It was produced by Sam Asadi and
• 37:44
  Speaker 1: Moses Andam and sound designed by Blake Mapleson.