Can the AI Boom Survive a Margin Squeeze?

0:01This is the Bloomberg Tech Minute brought to you by ChachiPT. I'm Carol Masser. Getting hired at SpaceX requires exceptional ability and may include having Elon Musk review a candidate's resume. So writes Bloomberg Businessweek's Lauren Grush. She reports that the company culture is defined by Musk and job candidates must understand that a commitment to SpaceX means adhering to this culture and exhibiting loyalty to Elon Musk. To get hired, candidates must demonstrate their achievements from past jobs, and hands-on experience matters more to recruiters than advanced degrees, with the company looking for people with raw, innate intelligence.

0:39A spokesperson for SpaceX didn't respond to a request for comment on its recruiting and hiring practices, but according to recruiters, advisors, and those who've interviewed for jobs there, it's the space exploration dream that draws people to the company, and Musk invokes that dream regularly to remind workers why they're there. Prospective hires are told that the job will require stamina, including in job ads that candidates, quote, must be willing to work extended hours and weekends as needed. One major perk for all that hard work? Employees' access to equity, either the right to buy shares or payment in the form of restricted stock units that vest over time.

1:16That's the Bloomberg Tech Minute brought to you by ChachiPT. Get started at ChachiPT.com. If you're a business owner, you need a card that can keep up. With the Delta SkyMiles Reserve Business American Express card, everyday purchases transform into miles you can use towards future travel. So business as usual can lead to a well-deserved getaway. And for a limited time, card members can earn 200 ,000 bonus miles, the highest miles offered this year. Whether you're traveling for business or planning your next adventure, now's a great time to make every purchase go further. Apply today at www.delta.com slash podcast.

1:56Offer ends November 4th. Minimum spending requirements and terms apply. For big business, AI opened up a world of promise, but that world of promise turned into a world of pain. Cost spiraling, data trapped, security and governance risks multiplying, and ROI out of reach. That's why enterprises turned to Boomi to connect data, apps, and AI, helping them operate securely, efficiently, and at scale. Boomi turns a world of AI pain into a world of AI gain. As Boomi says, we got you. Head to boomi.com. That's B-O-O-M-I dot com. Bloomberg Audio Studios. Podcasts. Radio. News. Last night I was walking around West Village back from dinner and I heard, just overheard with no context.

2:46Someone said, oh, it's the big guy's building data centers or something along that line. In other words, this is very much in the zeitgeist. So I think people say that buy the news, sell the rumor, or when your barber starts pitching you a stock in something, it's time to get out.

3:12Welcome to Maren Talks Money, the podcast in which people who know the markets explain the markets. I am Meryn Somerset-Webb, and this week I'm speaking with Steve Hu. Steve is Head of Research at Silicon Data, a firm founded in 2024, which aims to revolutionize compute markets by providing price transparency and GPU performance data to traders, financial institutions, and builders. Before he took on that role, he was a quantitative researcher on the Bloomberg Index research team responsible for systemic equity and multi-asset strategy research. And, of course, he has been on this podcast before.

3:45and that is why I wanted to speak to him. He's previously recommended reading material to us that has gone viral later and also he really is an expert on everything AI. So I'm hoping he'll be able to guide us through what is going on at the moment. Now, during our conversation, we get into everything from the price of GPU rental to whether restaurants are going to be overwhelmed or your agents trying to get the best tables. But before all that, we start off where we left off last time we had Steve on the show with situational awareness. Steve, welcome to Merrin Talks Money. Yes, that's right. Thank you for having me back, Mary.

4:21It's a pleasure. Anyway, back then, the last thing that Steve said to us that we should all read is something called situational awareness, which is kind of a pamphlet, kind of a... I would go so far as to call it a book. I think at the time you said, you know, we should read this pamphlet, but in fact, it's a good 200 pages long. And that gave a sort of sense of what the author thought was going to happen in the AI marketplace. It was written in 2024 or 2023 originally. I think that's right. Yeah, around then. Anyway, you recommended it. I had a skim through it, thought, wow, OK, that's kind of interesting.

5:00And then about a year after the whole thing went completely viral, right? And by the beginning of this year, everybody was reading Situational Awareness and getting themselves quite upset about the whole thing. And I wanted to start with this because it is fairly extraordinary that the AI landscape is unfolding almost exactly as this pamphlet slash book suggested. And that we're now at the bit where the extraordinary power of what is being developed is being recognized. and the conversation has moved on from, oh gosh, we can use this to help organize our email to, okay, well, this is a massive military asset and needs to be contained slash regulated by government.

5:41Well, we're at that bit, right? Yeah, indeed. The implications, I think, are enormous. And this is regardless of how this whole investment cycle transpires, are going to be a transformative technology that's going to shape and change society and our world. beyond recognition. Yeah. So we're mostly interested in the investment cycle, of course, because that's what the podcast is for. But obviously, as a background, the way that it's changing the world is kind of important. But before we get into all that, tell me about your new role, Steve, because you used to be a colleague of ours here and then off you went.

6:16Is that all about? That's right. Yeah. So I left Bloomberg in May to join a startup that also happens to be founded by a former Bloomberg colleague, Carmen Lee. And the company is called Silicon Data. It is a data company that tries to bring data transparency to the AI CapEx industry. And we are launching benchmark indices on GPU rental rates and GPU token prices. And we have also announced plans of launching futures contracts and options on the CME pending CFTC approval. Explain to us about that a little more. What are these indices you're launching on GPU pricing and GPU? So GPU rental prices.

7:01So essentially, how much does it cost to rent H100? So all these GPUs, NVIDIA GPUs that are running in data centers that produce AI tokens from which we run these models, they are all named after famous former physicists, right? So Hopper 100, Blackwell 200, and Ampere 100 was a chip that's launched in 2020. So we have data that tracks how much it costs to rent an hour of these GPUs around the globe from a designer, whether it's sitting in Western Texas or Finland or Southern Europe somewhere or Asia or the Middle East. And we aggregate these numbers into a single, you know, app-to-app-all comparable benchmark number.

7:44The same way you can think of it is almost like creating a single Belgian rental apartment index for New York City, right? So you aggregate all the rental agreements from across the city, whether it's in Soho, Harlem, Brooklyn, and so on and so forth. You control for all the variables like square footage, term of the length of the agreement and so on. And you make all these rental prices up or so comparable and you create a singular index so that people can track it. And this is an important variable if you are a cloud provider, if you are one of the data centers that's providing a compute, this is a source of revenue for you.

8:18And if you're one of the AI labs, whether it's OpenAI and Stropic, one of the Neo labs, or one of the inference platforms that post-train these open models that allow people to, say, serve Chinese models, this is a major cost for you, right? And we, Carmen, had this vision that eventually we're going to enter a very capital-intensive part of the cycle. There's going to be a lot of people with a lot of exposure to AI, and there's going to be need for hedging. There's going to be a need for hedging on the revenue side, on the cost side, and there's going to be traditional financial instruments that help solve this problem.

8:52That means futures, that means options, which is the reason why we've created these trusted benchmarks against which futures contracts on compute as a new commodity will settle. And we have to believe that this will be one of the biggest, if not the biggest commodity rivaling oil futures. Okay, interesting. So we've got basically what you're looking at here is the price of renting compute from a data center. You've got suppliers, that's their revenue, and then you've got everybody who's using AI in anything is one way or another, either one step or two steps removed from the price of renting the chips inside the data centers, right?

9:32So that's really interesting information telling us what might or might not be going on in the AI boom or the AI bubble. What is that information telling you right now? So right now, we are in this very interesting era where we have this, you know, very, I think, wild side of charts on our Twitter account that just poses price of AI tokens or how much you can charge for the models with against how much it costs to run this different type of chips. And you see this interesting pattern where since about the end of May, the token index or the expenditure rate, the effective price paid for the models has been coming down pretty significantly, almost half of where it was back in the middle of the year.

10:18In the meantime, rental rates for these chips have been going up rapidly, continuously. So the analogy I like to give is almost a bit like if you are inventing the car for the first time, there's a luxury car, say Mercedes, then there's a cheaper car, say Toyota, Corolla. In the beginning, everyone was buying their Mercedes. These are these frontier models, the top of the line models from Claude or OpenAI. but eventually people realize that some of these non-frontier models the cheaper models that have less capability but still evolving over time are saturating most of the everyday use and they're using a lot more of that so everyone's buying a ton and ton of like corollas right a lot of toyotas all over so the thing is you can have this proliferation of you know car usage and every car burns gasoline or benzene i guess whatever i guess we're in the uk that would drive up and you the consumption of the underlying input.

11:17So the same analogy applies in AI, in that the proliferation of AI and everybody using these cheaper models to power their agents, right? And to read their emails, to respond to things. That will consume compute, right? Even if it's not necessarily using the highest margin and the most profitable models, the most powerful models coming from the frontier labs. Okay, so you've got the cost going up, but the price coming down. If you put it that way, yes, there is a certain sense that's the case. Okay, that seems problematic for some part of the market. Indeed, for some part of the market, it could be problematic, yes.

11:58In fact, interestingly, through sheer coincidence, the time when the token index peaked was also around the time when, according to some measurements, The anthropic revenue growth also slowed and came to grow, you know, sort of either flatlined or, you know, sort of much slower than it was growing in the beginning of the year. So something tentatively related, but not conclusive by any means. Yeah, I mean, this does seem to be one of the bigger problems for the anthropics and the open AIs, which is that cheaper, simpler models work just as well for the things that most of us do. Your earlier doctors are going to be very different to the people who are going to use your product long term.

12:40So the majority of us are always going to use, I'm guessing, I'm speaking for myself here, always going to use cheaper, simpler models. So what does that mean for the Anthropics and the OpenAI long term? Well, long term is a long time away. All right, medium term, short term. I suspect this is going to create some questions surrounding their upcoming IPOs and the, I think, high, I would say, valuations, relatively high valuations. The last time I think I heard numbers anywhere between, you know,$1.5 to$2 trillion for Anthropik and OpenAI, I think, a tad lower than that. I mean, these are still non-public startups, young companies, but growing extremely rapidly, head-scratchingly so.

13:38And Thropic has gone in almost no time at all from having very little revenue to$60 billion of annualized revenue run rate and projected to go much higher based on the extrapolated growth previously. So if there are signs that the growth rate is indeed slowing and there could be structural reasons for why, for competitive reasons that it may not grow as rapidly as people have previously thought, and that will put into question some of the valuations and rosier hopes for their long-term value. Yeah, but it's also the case that at an IPO, the buyers at IPO don't want to pay a price that reflects every single piece of potential growth already priced in, right?

14:23The buyer at IPO wants to buy at a price that gives them something. And the problem possibly with these IPOs, particularly with Anthropic and OpenAI, is that we look at them now and I think increasingly people don't think of them as an opportunity for new investors to get a slice of the growth. They look at them and think of them as fabulous exit opportunities for people who already own the shares. And that's not what the market is supposed to do or it's not what people expect the market to do. You get an IPO, you buy something that has growth left in it. And even if these companies grow, if you've paid in advance for all the growth, what's in it for you?

15:02And so you get the sense that the market is beginning to shift the way they're thinking about these IPOs. Well, you ask very good questions. Yeah, I think those are questions that people should be asking.

15:19This is the Bloomberg Tech Minute brought to you by Chachi PT. I'm Carol Masser. Getting hired at SpaceX requires exceptional ability and may include having Elon Musk review a candidate's resume. So writes Bloomberg Businessweek's Lauren Grush. She reports that the company culture is defined by Musk and job candidates must understand that a commitment to SpaceX means adhering to this culture and exhibiting loyalty to Elon Musk. To get hired, candidates must demonstrate their achievements from past jobs and hands-on experience matters more to recruiters than advanced degrees. with the company looking for people with raw, innate intelligence.

15:56A spokesperson for SpaceX didn't respond to a request for comment on its recruiting and hiring practices. But according to recruiters, advisors, and those who've interviewed for jobs there, it's the space exploration dream that draws people to the company. And Musk invokes that dream regularly to remind workers why they're there. Prospective hires are told that the job will require stamina, including in job ads that candidates, quote, must be willing to work extended hours and weekends as needed. One major perk for all that hard work? Employees' access to equity, either the right to buy shares or payment in the form of restricted stock units that vest over time.

16:34That's the Bloomberg Tech Minute brought to you by ChachiPT. Get started at ChachiPT.com. Running a small business takes everything you've got, but with Chase for Business, you're not alone. They bring together local support and a broad range of resources to more than 7 million customers. With a deep understanding of your day-to-day needs, they provide products and guidance built to help you thrive. Right now, earn$500 when you open a new Chase Business Complete Checking account for new business checking customers with qualifying activities. Offer expires January 21st, 2027. Chase Business Complete Checking has the flexible tools you need to accept payments, make deposits, and manage your finances with confidence.

17:16Learn more at chase.com slash podcast biz offer. Chase, make more of what's yours. Fees may apply to Chase Business Complete Checking Accounts. The$500 offer is available for new business checking accounts with qualifying activities through 1-21-2027. Eligibility and qualification requirements must be met. Additional restrictions may apply. Please speak with a business banker for more information. JPMorgan Chase Bank, N.A., member FDIC. For big business, AI opened up a world of promise, but that world of promise turned into a world of pain. Cost spiraling, data scattered across disconnected systems, security and governance increasingly complex, AI initiatives struggling to scale, and ROI that remains frustratingly out of reach.

18:00The challenge isn't AI itself, it's having the infrastructure AI needs to succeed, which is right where Boomi says we got you. Tens of thousands of enterprises trust Boomi to connect data, applications, processes, and AI. With Boomi, businesses can eliminate silos, automate workflows, strengthen governance, and create the foundation AI needs to perform at scale. The result? Faster innovation, more efficiency, better business outcomes. Boomi turns a world of AI pain into a world of AI gain. As Boomi says, we got you. Head to boomi.com to learn more. That's B-O-O-M-I dot com.

19:07more.

19:13Steve, before we go any further, it occurs to me that we have been talking about tokens. Tokens this, tokens that, tokens this. And there is a tendency for all of us at this point in the cycle, we've been talking about things for a while now, not to define our terms. So would you mind, just for our less experienced listeners, explaining what we mean by a token? We've all used ChatGPT, right? How you use it is you enter a bunch of texts, say, you know, sort of help me plan a little trip to, you know, North England this, you know, coming weekend. My family is going to give you back an itinerary telling you, go here, go here, go here.

19:51And if you are here, you check out this restaurant and that. So a token is basically a length of text that you're entering. So if you're entering a long prompt, a long set of conditions about what you want. That's a large number of tokens. And if the output is a large output, long output, that is a large number of tokens too. Now, it doesn't necessarily have to be the case that your input or output are just text, right? Increasingly, we're seeing people inputting a screenshot into a prompt to say, hey, explain how this works or what is in the picture and whatever. Nevertheless, the computer is basically, because the AI is translating the text, you know, the picture into essentially text or tokens, right?

20:37And that's the only way, you know, the computers in which you think. Okay, so the way that an LLM, a large language model, what we all think of when we talk about AI these days, although obviously it's not, there's more than that to it these days, but nonetheless, an LLM is what we think of when we think AI. So you give it a prompt, ask it a question, and then you keep developing that. You ask more and more questions, you put in more and more prompts. The reason it becomes so expensive or historically has been expensive is that the tokens are rerun each time, right? So it's not like you're starting from where you left off.

21:07The model has to rerun every token. And that's what historically built the price up. Well, a core challenge with AI, with LLM, has been managing memory and long context, you know, exactly as you say, right? When you are in a conversation, you know, with an AI, just like when you're in a conversation with a human being, you don't want to start over. You want people to remember the context of where you got to where you are so that the next sentence makes sense within the context of what you were talking about. But of course, as the conversation gets longer and longer, the amount of information that has been talked about, the amount of memory, things that needs to be memorized, it just becomes bigger and bigger, right?

21:49And at some point, you run out of room to store that memory, but also memorizing. So there's all these attempts of essentially compressing the memory, right yeah you know it's a little bit not indifferent from when we human beings talk to people we don't remember every single detail in a conversation that's been mentioned we make mental notes of the key points that have been touched upon right and we make recall and what a computer is able to do even better in increasing better you know algorithms is actually able to sort of make that recall even more uh refinely being able to actually go back actually let's dial back to that moment when we talk about something in the greater detail, stuff that you may not remember very well.

22:31And that's what makes AI super powerful. And in terms of the practical implications, part of the reason why we have seen such an explosive growth in the stock prices of memory and storage stocks of such as SanDisk or Micron in the US or SK Hynix and Samsung in Korea is exactly that right you know we suddenly have a exercise where the demand for memory and demand for storage of data has just gone exponential right every single conversation everybody's having is generating tons and tons more of stuff that needs to be remembered and stored down so so that's the reason why korea you know is suddenly seeing such a much such you know intense attention from across the globe.

23:24That's not particularly comfortable, by the way, but thank you for that explanation. Very useful. Can we go back to data centers? We were talking about the cost of GPU Yand rental going up. Is there still, despite this massive capex boom, is there a shortage of data centers? I mean, we're seeing, as you know, protests against data centers, people don't really like to have them built near them. We've seen some cancellations most recently in the UK, a big cancellation because they can't get connected to an energy supply, etc. And that's a big problem, globally connecting your data center to an energy supply.

24:01So is there still a shortage despite all this CapEx? And is that what those GPU rental prices are reflecting? So shortage is, you know, a complicated concept, right? So as it stands today, everybody is able to get on an AI and get some output from it. But shortage is mostly reflected in terms of the anticipated growth in demand of a lot more people using a lot more AI and the ability to supply into it. And to that extent, on the reasonable estimates of everybody just using, I think, AI more meaningfully, as you said earlier. The average person is not using AI in a very, very high-power way, right?

24:45Everyone is using it mostly as a glorified search. But there is actually a lot more ways you can actually use it in an agentic way where it's continuously running. And that is actually a lot more demand for compute. And we multiply that across the global population, especially within developed economies, and that would imply a level of demand that is not currently matched by the growth in supply. So to that extent, yes, there is a shortage in data centers, not necessarily... But it's a forecast shortage as opposed to a current shortage. Of course, of course, it's a forecast shortage. Nevertheless, I think it's a reasonable forecast given the recent trend.

25:24I mean, I do think people are going to be using more AI and it's going to proliferate, especially with more, I think, products that are built with better product market fit, with good understanding, anticipating consumer needs, like the latest product from Meta, Muse, and so on. There's going to be more of those. And combine that with the fact that there is a shortage in supply on data centers because of the shortage of just sheer physical constraint of power, right? The lack thereof, right? There's not enough sites for us to build a lot of large data centers having to build smaller ones in different corners of the world and contending with energy shortage.

26:02I think local political pushback from local residents, which I think is maybe a little bit overblown in terms of the actual net impact. Is there a case to be made for demand not developing in the way you expect? I mean, I was surprised by a stat, I don't even know if it was true, but it rang a bit true, that more people pay for weight loss drugs in the US than pay to use AI in any way. Fascinating at this point in the cycle. And the other thing to say on that is that, you know, look at Muse and the other agents that people are starting using fairly regularly. Is it sustainable for everyone to use an AI agent in this way?

26:43And that we're already seeing, for example, local government getting clogged up by people having their agents complain about their parking tickets and potholes. Planning getting clogged up with masses of objections generated by agents. and even restaurants that are hard to get into being overwhelmed by phone calls and emails from agents. And there comes a point when rather, and this is at the very beginning of this cycle, there comes a point when instead of reducing friction in our lives, if everyone's using agents to do all these things, the world just collapses. I mean, how does this actually work if already when only a tiny percentage of the population is using an agent to sort their email box, to do their shopping, to do their restaurant renovations, to complain to the council, to do their taxes, tiny, tiny already is overwhelming some of our institutions and our businesses.

27:41How does that play out? So you bring up two main points that are both interesting, very good. So let me address each of them. The first one you raise is regarding demand growth and how reliable is it indeed it is the case when i say demand we are looking at the current condition of how much you know hgbt anthropic is selling you know their sort of monthly plans at you still a fixed price all you can eat you know kind of a plan they throttle usage and give you a amount of credit at which point you run out muse by the way is free so um at current prices you know demand is growing a certain way, it is unknown at this point how much people are going to be willing to pay for AI per usage, especially if it's going to be metered.

28:34It sounds as though that Mata and Zuckerberg have been talking about the fact that they want to make this widely available to people either for free or at very low cost, and they will find ways of monetizing it via other means, such as e-commerce and collecting a fee that way, and so on. I don't think the exact channel of the economics of how AI companies will be paid while demand grows has been fully articulated. I think there are competing narratives. So you are right to question or ask the question that what is going to be the growth of demand for AI if costs start going up, right? I don't mean to say that nobody's paying for AI.

29:19That's clearly not the case. I'm just saying that how AI will be monetized, I think there are competing narratives and also depends a little bit on whether or not you're looking at consumer use case or enterprise, right? Enterprises are clearly paying very large bills. And so enterprises clearly are paying these models by the token prices and that are publicly known. And then there is caching, there are ways in which you get some discounts. But generally speaking, yeah, there is an anticipation that this is going to be a metered usage for a lot of it. You know, the question remains as to whether or not once this proliferates, you know, the usage will keep up.

30:03Right. And that is part of the discussion when we are talking about the economics insofar as who actually captures the rent. If you invent this extraordinary thing and, you know, OpenAI, you know, Anthropic, and then do you actually sort of manage to recuperate, right, those initial discoveries, the training runs that came up with all these models? if subsequently there are models that are based on these models, distilled versions, either themselves distilled cheaper versions of those models or others such that they are sold at a much cheaper price and people optimize on cost and quality. So that's the open question.

30:49Okay, second bit. Second bit. Yes. Obviously, the use case for AI is now, or for LLMs at least, is very clear. We know what they can do, we know what they can't do, and we know that we need new types of models to do the things they can't do. For example, you know, precision, audit, compliance, regulatory stuff, etc. We know that. So the use case is certain. But when you are going back to my question about retail agents, agents working for ordinary people, and the way this can very quickly overwhelm the systems that we already have in place. Indeed, right. If you have proliferation, everybody just activates agents and, hey, go pick up, go find me a reservation, go find me a ticket, go do whatever.

31:30On the receiving end, you're going to find yourself overwhelmed with all the agentic requests. And we're going to find ourselves in a situation where we don't have enough. Everyone can generate a ton of AI output, AI slop, but we don't have the ability to process it all. And in the enterprise context, I think it's a similar thing going on where you can have enormous efficiency gains where someone's generating tons and tons of new PPTs and presentations or code and whatnot. But then you have other parts of the business that actually process that output to actually take it into a finished product to turn that into a business outcome and revenue.

32:09And I think that analogy applies and that is a very big open question as to what happens. You know, I don't have good answers for you. I suspect some of the questions and pain points you're pointing out are going to actually surface in the coming months. Yeah, I mean, I have this vague worry that we end up with everyone has an agent trying to make them a reservation at the new great restaurant in London. the new great restaurant in London then has to, you know, create a big pile of agents to deal with the agents bothering it for reservations. Either that or they shut down agentic inbound, you know, altogether.

32:46Or you've got to basically force a new type of restaurants that is AI native that can actually handle these things. So you have to be a new type of business. Sort of, you know, they say shortage or whatever is always the mother of innovation. I mean, there's got to be, yeah, there's got to be things I've got to give, right? Exactly.

33:29that a commitment to SpaceX means adhering to this culture and exhibiting loyalty to Elon Musk. To get hired, candidates must demonstrate their achievements from past jobs, and hands-on experience matters more to recruiters than advanced degrees, with the company looking for people with raw, innate intelligence. A spokesperson for SpaceX didn't respond to a request for comment on its recruiting and hiring practices, but according to recruiters, advisors, and those who've interviewed for jobs there, It's the space exploration dream that draws people to the company, and Musk invokes that dream regularly to remind workers why they're there.

34:06Prospective hires are told that the job will require stamina, including in job ads that candidates, quote, must be willing to work extended hours and weekends as needed. One major perk for all that hard work? Employees' access to equity, either the right to buy shares or payment in the form of restricted stock units that vest over time. That's the Bloomberg Tech Minute brought to you by ChatGPT. Get started at ChatGPT.com.

35:17Thank you.

35:25Additional restrictions may apply. Please speak with a business banker for more information. JPMorgan Chase Bank, N.A., member FDIC. For big business, AI opened up a world of promise, but that world of promise turned into a world of pain. Cost spiraling, data scattered across disconnected systems, security and governance increasingly complex, AI initiatives struggling to scale, and ROI that remains frustratingly out of reach. The challenge isn't AI itself, it's having the infrastructure AI needs to succeed, which is right where Boomi says we got you. Tens of thousands of enterprises trust Boomi to connect data, applications, processes, and AI.

36:05With Boomi, businesses can eliminate silos, automate workflows, strengthen governance, and create the foundation AI needs to perform at scale. The result? Faster innovation, more efficiency, better business outcomes. Boomi turns a world of AI pain into a world of AI gain. As Boomi says, we got you. Head to boomi.com to learn more. That's B-O-O-M-I dot com. Running a business means keeping pace with the latest tech. You put a lot into building your enterprise software stack, and scrapping it to integrate AI is out of the question. That's why IBM technology works with all the apps, data, and models you need.

36:48It's another way IBM helps you build on the technology you rely on. Let's create smarter business. IBM. Visit ibm.com to learn more.

37:05Anyway, so we've talked about a lot of things. We've talked about a lot of the uncertainties. We've talked about who's using, who's demanding, etc. Bubble or boom? I think the question is not so much whether it's a bubble or not. I mean, every single investment cycle, every capital cycle has always gone through the boom and bust. The question is a matter of timeline. The market has already topped in the local fashion in the middle of the year, right? So there is definitely a sense in which people, a lot of the questions you were just raising with regard to ROI, with regards to how does this investment produce a return and how do these frontier labs actually justify their valuation, which has been the source for a lot of their CapEx spending commitments because they've been selling shares.

37:53They've been raising money from VCs and other investors in order to raise capital which in turn to acquire, compute and build data centers or commit to renting data centers. So that whole narrative, that whole cycle, cyclical narrative could be put into question if more and more people are going to start questioning some of the questions that you're just raising and good answers are not forthcoming. So I do think we are in an environment of the market where these questions are growing louder and we're not early in this trade. I think last night I was walking around West Village back from dinner and I heard, just overheard with no context, someone said, oh, it's the big guy, it's building data centers or something along that line.

38:39In other words, this is very much in the zeitgeist. So I think people say that buy the new sell the rumor or when your barber starts pitching you a stock in something, it's time to get out. I mean, I think there's some truth to it in that we are probably in a pretty feverish moment of the market that some of the smarter money has already either taken chips off the table or having their chips taken off the table for them. Also, I do think this is an important point to make, is that nothing ever happens in a straight line. Even if this was the end of the bubble, we can have extremely violent counter-rallies, uncertainties regarding whether or not this really was the end can actually cause that to happen, number one.

39:29Number two is that I think there's a lot of people in the bearish circles, Because some of them actually sort of have this position that this whole thing is vaporware and it's all useless. And it's just been a con job that, you know, none of this is ever going to be useful to anybody. Nobody actually wants it. I think that is fundamentally untrue. That's a man who's never going to get a restaurant reservation again. Unless he's shorting the market, in which case he might very well just, you know. But I do think this is a fundamentally transformative technology. I do think it's going to reshape society.

40:08And I eventually think the balls that people are bullish on technology are going to be proven right. And there is a chance in which things work out right in that you find products with good product and product market fit. You have an acceleration of adoption. And people actually do want to pay for things that give them utility. I want to be a little bit contrarian just because these days it seems to have become very fashionable. I have not met a single person who doesn't think this is just a repeat of the dot-com bubble that we're staring at within the next 12 months, some sort of spectacular crash and blow up of the US tech bubble in stock market, and we're going to repeat the 2000 story.

40:51I think there is a chance in which the mismatch, I think this is fundamentally a potential mismatch of timing. Not necessarily there won't be demand for it, but rather the demand could show up slower than some of the most bullish people in SF seem to think. They seem to think that because it works for coding, all the programmers adopt AI rapidly overnight, that this is going to happen everywhere, ignoring a lot of the practical bureaucratic hurdles to adoption that could slow things down, such that the supply could actually show up or show up in 2027, 2028, and then you find yourself not having as much demand because enterprise adoption wasn't as rapid, and maybe consumer adoption is not as rapid because you didn't find the right product market fit, you know, for the products that actually really took off the same way iPhone took hold with mobile.

41:46So I think that remains to be seen. You know, this risk of an air pocket is very much real and we are clearly living through it, you know, in different manifestations. But I think there's also a risk of a right tail risk by things that really do go right, in which case we can actually see a pretty violent melt up as well. Okay, Steve, so you're giving me two options here. One is that we're closer to the end than the beginning, and the other is that we might have a massive melt-up. Which one do you think is most likely, melt-up or closer to the end than the beginning? The funny thing is those two things don't have to necessarily be...

42:21They don't have to be mutually exclusive. They could both happen. Indeed. What happens if the market melts up and then crashes? You're not necessarily going to be nimble enough to time yourself perfectly to catch one and avoid the other. So I'm sorry, that's a little bit of unsatisfactory answer if you are looking for me to give you like a timing signal. Yes, I am. So, okay, so what do we do? What do we do? We're an ordinary retail investor. We're like everybody. We are overweight, probably by mistake, overweight US tech, by mistake, overweight AI. What do we do now? Is the answer just to rebalance, pull back?

42:59But then of course, is everything AI now? Is there a way out? Is there a way that we should be investing right now to protect ourselves a little? Well, first of all, I am not here to give anybody any investment advice. No, no, this is not advice. In general, if we're just, you know, sort of chatting casually, I think it is fair to say there are plenty of things that are not AI, directly AI related. you know energy you know like traditional force of energy stocks and you know oil companies they are not AI right yeah okay thank you now the other thing I wanted to ask you was about this business that is all over the news this week about Anthropic is saying that they intend to give five percent of their shares into a foundation that will support people who are affected by AI.

43:55So it should be nice to people who've been affected by it. Maybe they lost their job or look into this kind of thing, etc. And then at the same time, we've got Elon Musk saying that he's going to give a SpaceX share to every kid in America. And all this looks to me a bit like, well, we better give it away before they bring in a wealth tax. What does it look like to you? I did not make that connection. I knew that Anthropica has always had this longstanding worry, very vocal, you know, about the danger of AI poses on society, which is a bit ironic because, you know, they are the one actually that's spearheading or, you know, at the forefront of inventing this technology.

44:38So it's almost as though like, oh, we'll have these products. I look, look, you know, look, the danger is actually, you know, posing. So yeah, it is a bit interesting, I think, from a behavior point of view, you know, but I also kind of appreciate from their perspective, I think in a much more charitable fashion to understand that there are other people developing this technology. One of the many competing forces, this is going to be developed so they want to pay their or make their contribution to help society deal with the consequences. I don't know what 5 % of their market cap put aside towards how this exactly will pan out.

45:16I mean, there's not a lot of details yet as to how the program will work. But I do think the end goal and the question as to how this will impact society, that I have no doubt. I think AI is going to be hugely transformative in good and potentially bad ways as well. Not just only in the popular narrative of disrupting jobs, but also this idea that you suddenly have a technology that enables everybody to do things that previously could not. And there could be all sorts of risks unleashed onto the system that we previously didn't have the capability of dealing with. So, yeah, I mean, within that context, I don't know what the 5 % share will do.

46:05But it's good that somebody is funding somebody to study these issues. And I hadn't heard about the SpaceX one, so I'll have to learn more about that. The SpaceX one is interesting. I mean, it smacks to me a little bit about buying indulgences. We may destroy the human race, but at least we paid up a little up front. Yeah. The one question I've ever got to ask, maybe we'll have to do this on a follow-up conversation, what this all looks like from the perspective of a European person, a British person, looking from across the Atlantic. I think this seems mostly of an American activity. Yeah, we're just irritated we're not going to get a SpaceX.

46:44Why can't my kids have SpaceX shares too? What's with all the borders? I don't know. Anyway, listen, the last question I must ask you before we go, because I've had you far too long, is, so last time I said to you, what should we read? You said read Situational Awareness, and boy, were you ever right. So now I'm going to say to you, Steve, what should we read? I think people should read some poetry. Rediscover our humanity. Oh, God. This is awful. Someone the other day said something very similar. I said, what should we read in here? Philosophy, philosophy. Get ready for your existential crisis.

47:22And you're clearly in the same zone. Anyway. Indeed so. Thank you so much for joining us today, Steve. We really appreciate it. Thank you very much for having, Maureen. Always a good time. And I hope the next one won't take us another two years. It won't. It won't. Thanks for listening to this week's Merrin Talks Money. If you like our show, rate, review and subscribe wherever you listen to your podcasts. And keep sending questions or comments to marynmoney at bloomberg.net. You can also follow me and John on Twitter or X. I'm at marynsw and John is John underscore stepek. This episode was hosted by me, Mary Somerset Webb.

48:00It was produced by Sam Asadi and Moses Ander. A sound design by Blake Maples.

48:10The world of business is constantly evolving and Comcast Business keeps you totally in step with secure AI-backed networking in more than 100 countries. They're powering over 90 % of the Fortune 500 and millions of small businesses. That's a lot of muscle. And behind it all, thousands of experts answering your call at 2 a.m. like it's 2 p.m. One partner powering how business gets done for companies around the globe. When you add it all up, no one does business like Comcast Business. Most meeting note takers get the easy words right and the important ones wrong. The client's name, the acronym. Whisper Flow Note Taker is built to get the hard words right.

48:51It takes notes on any meeting platform. Zoom, Google Meet, Teams, even WhatsApp calls. keeps every meeting searchable, and connects to Claude or Chat GPTs. So you can ask, what did we agree to last Tuesday? Free to try on Mac and Windows at whisperflow.ai. That's W-I-S-P-R flow dot A-I. If you're a business owner, you need a card that can keep up. With the Delta SkyMiles Reserve Business American Express card, everyday purchases transform into miles you can use towards future travel. So business as usual can lead to a well-deserved getaway. And for a limited time, card members can earn 200 ,000 bonus miles, the highest miles offered this year.

49:31Whether you're traveling for business or planning your next adventure, now's a great time to make every purchase go further. Apply today at www.delta.com slash podcast. Offer ends November 4th. Minimum spending requirements and terms apply. This is Robert Smith from Business History. If you're listening to this, there's a good chance you're a small business owner. And like every small business owner, you started with a dream to do what you love and watch it grow. What you probably didn't dream about? Keeping up with cyber threats. That's where MasterCard can help with access to tools that help identify cyber threats to better protect your business.

50:05Building your dream business? Priceless. For cybersecurity in a changing world, there's MasterCard. Learn more at MasterCard.com slash small business.