Pod Insights Summation with Auren Hoffman Azeem Azhar of Exponential View - AI, hyperscalers, reshaping US GDP About this episode Azeem Azhar is the founder of Exponential View, a newsletter and research platform on emerging technology read by over 130,000 executives and policymakers globally, and author of the bestselling book The Exponential Age. In this episode of World of DaaS, Azeem and Auren discuss: Diagnosing an AI bubbleData centers driving 33% of US GDP growthWhether energy will constrain AI before capital doesCircular financing in AI and funding quality risksLooking for more tech, data and venture capital int...
Transcript 10,004 words · 2 speakers
Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:02 Auren Hoffman? Hello, fellow data nerds. Welcome to World of DAS. I'm your host, Orin Hoffman, CEO of Incubate and GP of Flex Capital. Discover more episodes, get weekly data as a service news, original content, articles on data and more at worldofdas.com. That's world of das, D-A-A-S, dot com. Hello, fellow data nerds. My guest today is Azeem Mazhar. Azeem is the founder of Exponential View, a newsletter research platform on emerging technology read by over 130,000 people. He's also an investor, an advisor. He's the author of The Exponential Age. Azeem, welcome to World of DAS.
0:41 Speaker 01 Thank you for having me and hello to all the data nerds out there. Excited to be talking to you.
0:46 Auren Hoffman? Yeah, I mean, this is great. We're gonna dive in. I mean, you're a data nerd as well, right?
0:49 Speaker 01 I'm a data nerd.
0:51 Auren Hoffman? Do you think we're in the AI bubble?
0:53 Speaker 01 I love how you get straight in there, Orin, no messing around.
0:57 Auren Hoffman? No messing around here.
0:58 Speaker 01 No messing around. I've studied a lot of technology and industrial transformations, and every major transformation from agriculture onwards runs on a feedback loop between intelligence, energy and capital. That is what drives us forward. And AI is simply the latest turn of that flywheel. And so where we sit today is so much capital going into AI. And that question that you've asked me, are we in a bubble? Well, what is a bubble? A bubble is not a bear market. A bubble is not stocks dropping 20%. That's a bad day and they'll come back. A bubble is what you and I know from our experience of the dot com, when
1:40 Speaker 01 stock prices drop by more than 50% and they stay depressed for years, and the active useful capital going into that technology drops by 50% and doesn't recover for a really long time. That is a bubble. And we're going to have ups and downs. Most technology booms turn into bubbles. In a nutshell, I don't think we are in bubble territory right now. That doesn't mean we won't get to bubble territory, but it doesn't feel for a number of reasons I hope we'll discuss that we are there just yet.
2:13 Auren Hoffman? Some people are worried about these inside dealings where these like round trippings or some of these other things that are starting to happen, maybe in the last few months that weren't happening a year ago. Do those things worry you?
2:26 Speaker 01 We have to find some empathy for the entrepreneurs on the other side. Imagine that all the things that we hear about generative AI and the amount of use that's going on is true. Imagine it's really true, as Walton suggests, that 84% of American executives are using it weekly already and 47% are using it daily. And let's assume that it's true that when Kevin Wile at OpenAI or Dario at Anthropic says, "We have run out of servers and we need more GPUs." Let's assume that's true. And let's assume the orders that are then being put to Nvidia and, to some extent, AMD and Broadcom are all real orders and they're just not there for shits
3:11 Speaker 01 and giggles. They're genuine orders because of custom demand. The question is, what would you do? You and I have both run companies. We're entrepreneurs. We know that at those moments, growth matters and you do the deals you need to do to put the cash into your company to buy the things you need to serve your customers. Now, if you are Satya Nadella at Microsoft, you've got 40 years of profits against which you can find the cash to buy GPUs. If you're Michael Imperator at CoreWeave, which is one of these NeoClouds, or you are coming out of Oracle, which is new into this market, you're going to have to figure out where to get that capital to buy the equipment you need to serve your customers. So if you
3:53 Speaker 01 think that this thing is running red hot and if you think the demand is real, you're going to end up needing to do some vendor financing, just as virtually every car dealer in the United States did in the 1920s through the General Motors Acceptance Corporation, GMAC. They were funded by General Motors to help sell General Motors cars and we as consumers were funded to buy them. So we have seen this before and that's good news if there's demand. If it's the things that you and I remember from the telecoms bubble, not Funding a demand that didn't exist or the very wonderful Cosmo dot com, which allowed you to buy a Ben and Jerry's ice cream for 50 cents in five minutes in Manhattan, funded by venture capitalists.
4:34 Speaker 01 That's not a good thing. It depends on which side of the fence you're on. Right now, I think that this is supporting the revenue and the revenue trajectory and customer orders. But of course, it's a little bit messy and it's a bit dirty. And so we look at that and we say, well, this raises a bit of risk, but it's not an absolute disaster. Pull the ejection cord at this point.
4:57 Auren Hoffman? Certain bubbles that we've had in history have had a huge benefit to society. When we laid all these railroad tracks that maybe we didn't need at the time, like many of these railroads we're still using today. The fiber that we laid in the late nineties, like we're still using that fiber today and benefiting from all that fiber. Like maybe now it's a hundred percent used, but it took a long time to get there. But maybe when you have an Nvidia chip that is only lasts for a couple of years and you're kind of going into debt to buy that chip or something like that, maybe there's still use 10, 20 years in, but my guess is like, if the limiting factor is energy, they're not going to be using that chip anymore
5:39 Auren Hoffman? in five or six years. Like, how should we think about that kind of investment?
5:43 Speaker 01 I think I described it once as these aren't steel railway lines. These are more like bananas because nobody wants to eat a six year old's banana. When you dig into these, I think you get a really mixed picture. So the first thing that matters is can you pay that chip back? One of these GPUs that might cost 40,000 bucks. And it seems to be the case that in general, you get that payback within 18 months or 24 months, which is incredibly quick, right? For a capital item. And so in some sense, it's already paid for. The bit that we're less clear of is how much do these chips have economic life in like years four, five or six? And you hear mixed stories. And I think there's something really-
6:25 Auren Hoffman? Right. If you pay back in two years, but only the last three, like, I don't know, maybe that's not the best. If you only have a year to get the benefit of it.
6:32 Speaker 01 Well, if you only do, but you've already made your money back, right? So the thing that matters is like the return on capital that you can get from renting these chips. And it seems to be quite high. And, you know, in this crazy exponential world that we live in, just before I came on to chat to you, I saw a story where one of the Google VPs, because Google makes its own chips called the TPUs, said that some of their seven-year-old AI TPU chips are still being used today, which is really, really remarkable. Let's take it at face value. Maybe he exaggerated, but assume he didn't. I think we're in this very funny situation where the chips do get paid back. Maybe they last a bit longer. But as you say, who wants to train GPT six on a six-year-old chip?
7:15 Speaker 01 You absolutely won't. So that's going to construct demand for ever more chips.
7:19 Auren Hoffman? I mean, you could if maybe energy was free or something, but if energy is your limiting factor and you only have so much energy, you're gonna push that energy to the best chips.
7:29 Speaker 01 Yeah. I mean, I think this is something that Jensen has done, right? The token is the atomic unit of a AI model. The tokens per watt of Nvidia chips is getting better and better and better. And there are loads of companies out there trying to figure out, you know, I've invested in a couple, how to add an order of magnitude or more improvement on that. So energy is going to be the big constraint. And you hear this story. I mean, the boss of Apollo, which is this huge fund, hundreds of billions of dollars, said, the energy demand will not get solved in our lifetimes. He said that a week or so ago, because we are moving for the first time in 40 years into a world where energy demand is growing.
8:12 Speaker 01 You know, I talked about how my work is about the intersection between energy, intelligence and capital. The US and Europe has not been very intelligent about energy for the last 30 years.
8:22 Auren Hoffman? Yeah, we've been terrible.
8:24 Speaker 01 Terrible. We have not invested in our energy grids and our generation capacity compared to, say, what's happening in Asia or China.
8:33 Auren Hoffman? And we've got everything going for us in the US. I mean, we've got- Land, we've got super smart people, we've got super cheap oil and gas. We have all these things, but we have obviously the demand, we have almost infinite demand for the energy, but for whatever reason, like just getting new things online has been just a, and maybe that's going to change. I mean, what do you think? I assume a lot of this is just like a regulatory problem that could potentially change in the US.
8:59 Speaker 01 It's really brutal, but this demand is really good for the US energy system. It's good for innovation and it's good for cutting that red tape.
9:06 Auren Hoffman? And but you don't have to put on the grid, right? Like you could just make it more local to a small number of plants. Cause making the grid have three acts of capacity seems really, really hard to do in the US.
9:19 Speaker 01 You know, there is something that is going on with the energy system that again, you and I as old internet guys will like, which is the energy system looks like the old AT&T, Marbell telecoms network. And the new energy system is going to look a lot more like the internet. So there will be big nodes producing power, pushing it around, and there will be energy producers who are much smaller. They might be people with their electric vehicles, sharing a few kilowatt hours back to the grid. And you'll have tons of behind the meter. I mean, behind the meter, which is when a company gets its own energy generation for its own use is really commonplace in industry.
10:00 Speaker 01 So Dow Chemical and other big industrials generate quite a big part of their energy requirements from their own plants because they are so demanding. And we're starting to see that, of course, Musk has done that in a large part with his big Colossus data center. So I think that they will solve it that way, but it's not just about permitting. It's also about can you get hold of the technologies, right? So gas turbines, which is one choice, backlogged for years and years and years, and it takes a long time to build up factory capacity. Nuclear, we're restarting in the US several plants, but that will take a while. We don't know how to make them cheaply in the US, right? The Chinese or Koreans know how to make them cheaply.
10:44 Speaker 01 I think the thing that is surprising is that There are two new dimensions that could change a little bit of this. One is that solar and batteries are getting much, much cheaper. And you are starting to see some data centers saying, we'll get some of our power mix from solar and batteries.
11:00 Auren Hoffman? For a data center, I assume there's like fairly constant use that they need. And maybe I'm wrong, but at least for an AI training, my guess is like it's 24/7 use of the data center. The power doesn't go up and down too much. Whereas like a lot of other things, you're going to see a lot of more intermittent use. So you're going to have like different uses for power, depending on like how much you're using these things. Is that right?
11:24 Speaker 01 Yes, but I reckon that with training data centers, when they're on, they're on. But they're not always training at maximum capacity. And you have all the complexities, right? That we know when you're building complex computer systems.
11:36 Auren Hoffman? I guess machines have to go down all the time and there's lots of other things that, but yeah, yeah, okay.
11:40 Speaker 01 And one of the biggest problems the grid used to deal with was that when a training run ends, you know, it's like that scene in Star Wars when they blow up Alderaan and there's a sound of silence and voices screaming. Well, when a training run ends, the GPUs go quiet really quickly. And grids used to struggle with the fact that 20 megawatts back in those days would suddenly disappear.
12:02 Auren Hoffman? Oh my God, that's really tough for the grid to deal with.
12:04 Speaker 01 Right. So I do think that one opportunity will be data centers being better citizens to the grid. And we're starting to see that again in the US where effectively a data center might say, we will agree to give you notice if we're going to turn off. We're going to agree to run at this constant level and not be below that level. And all of that contributes to grid stability, which in a way reduces the cost because it reduces the risk and the uncertainty of running the grid. So I'm actually really optimistic about what will happen over the next five years as more data centers come on stream, because they will end up being from being bad, hungry power citizens to being good citizens on a decentralized grid and kind of contributing to its stability.
12:50 Auren Hoffman? There's so much data center construction in the US. How does that look in Europe? Is it tracking similar to the US?
12:58 Speaker 01 I'm sure there's a lot of paperwork being filed somewhere. There is a little bit of data center construction, but it's not like we're seeing in the US or in the UAE or in Asia. I mean, really, the vast bulk of data centers are being built right now in the US for exactly that reason. And, you know, we've seen the impact it's had on the overall strength of the economy this year.
13:21 Auren Hoffman? And we're talking like potentially a third of the US economy is being driven by this kind of like, let's say broadly data center ecosystem, is that right?
13:31 Speaker 01 Yeah, I think it's about a third, and I think it's really great news because building assets, capital expenditure is a good thing. It's a sign of optimism in the future. Investment precedes growth and we're building real physical things and it's creating a demand for all sorts of new skills, like electricians in really short supply. And of course, the US in general has an infrastructure underhang in terms of roads and bridges. These companies are pretty similar companies and they're getting a real feather to their cap that hopefully will help them in years to come as, you know, the US starts to renovate a lot of its infrastructure. So I think this is a good thing.
14:12 Speaker 01 At times like this, it's not uncommon for a single sector to be the starter motor for growth. In the twenties, it was automobiles and manufacturing in general. Seventy percent of US GDP was assigned to that sector. Of course, during the Second World War, about ninety percent of US GDP growth was in manufacturing. And we had this funny situation in Denmark, of course, a much smaller economy where two years ago, if it wasn't for the sale of one of those GLP one obesity drugs, the economy would have been in recession. And in fact, it grew, you know, one and a bit percent. So it's not ideal, but in a way what it can do is it can kickstart.
14:51 Auren Hoffman? Economies themselves have like a power law essentially is what you're saying.
14:55 Speaker 01 They have a power law you will have. I mean, it's not normally as concentrated as this. I mean, normally the biggest sector is driving thirty five percent of the growth. But if other sectors are uncertain because of politics or, you know, with the risk of inflation or the risk of a recession, then having one sector that can march forward, I think is generally a good thing because it's triggering lots of requirements in electrical hardware, in manufacturing, in scaling up electricians, and of course, in finance as well.
15:24 Auren Hoffman? Most Americans, if they're saving, they've got their savings in the S and P 500. And that is very overweight mag seven. I don't know the exact percentage, but I think it's over 10%. It's just in these seven names.
15:37 Speaker 01 I think it might be more than 25% now. Yeah, losing track of it.
15:40 Auren Hoffman? Okay, yeah, sorry. I should have checked that before this podcast. Like, I knew it was high. I didn't even realize it was that high. That's insane. So if you just think of most Americans, like their savings are basically in these seven companies, like they're extremely overweight. These seven companies, and maybe that's a good thing. Because obviously, in the last few years, it's definitely been good to be overweight those companies. But that does seem to me to be somewhat concerning that so much of America has their savings in just seven names. And that, you know, they don't really know that much about.
16:10 Speaker 01 I mean, have you seen that cartoon that goes around Twitter or X? Which is like this really big stack of bricks, a big, big construction. And there's a tiny little toothpick on the right, holding it up.
16:21 Auren Hoffman? Yeah, I love that meme. Yeah.
16:22 Speaker 01 So that meme is like, you know, whatever it is, this is holding up the whole of the economy. And it's like, well, OpenAI's growth rate is holding up. Everyone's 401ks, everyone's sense of optimism, everyone's sense of whether they can send their kids to college. And I think it is, I mean, it's scary, isn't it? But it looks like that.
16:37 Auren Hoffman? Super scary. And it's all one sector. It's not like, I mean, if these seven names, like one of them was an oil company, and one of them was a defense contractor, and one of them was a bank or whatever. Okay, it makes sense. But like, they're all completely interrelated, these seven companies.
16:51 Speaker 01 Look, we are in somewhat uncharted territory, but there are some precedents. If you look back at the big stock returners before 1980, so before the arrival of Microsofts and so on. They're all really clustered around the technologies of the start of the 20th century. So it's all of the spin offs of John Rockefeller's companies, Standard Oil. So you know, Exxon and so on, and Chevron and so on. It is the car companies, believe it or not, you know, GM, it's General Electric, it's AT&T. So the stock returns get clustered around. The companies that are associated with the technology transformation of the time.
17:33 Speaker 01 And the bulk of public companies, I think it's in the order of 90% of public companies in America, between 1926 and 2022, lost money for their investors. There's an epidemic called Hank Bessem Binder, who's looked at the data. So we think of the stock market as being safe, but it's always been quite heavily concentrated, the returns. That's why you need to have a broad portfolio. So it's a challenging thing to navigate. I think if you're a fund manager, you'd have to be a pretty ballsy fund manager to write your letter at the end of the quarter and say, I have no exposure to Nvidia and Tesla and Meta and Microsoft.
18:10 Auren Hoffman? And maybe that's a good thing. The US is riding the tech wave. Like, if you look at the top 10 companies in France or something, like, they're all companies that could have been created in like the 1700s and some of them were created back then.
18:23 Speaker 01 Sure, some of them were. It's all luxury.
18:25 Auren Hoffman? Yeah, that's almost like a bet on the past or something.
18:28 Speaker 01 It's bet on the future because it's all status goods.
18:32 Auren Hoffman? Right, right. It's a bet that the economy is going to keep growing, at least in certain parts of the world.
18:37 Speaker 01 And our wealthier friends will then want to distinguish themselves by spending even more on a, you know, crocodile skin jacket.
18:43 Auren Hoffman? Right, on a Birkin bag or something.
18:45 Speaker 01 Yeah, exactly.
18:46 Auren Hoffman? All right, that makes sense. These luxury goods are just a levered bet on the economy. If you think the economy is going to go up, those will go up 2X. If you think the economy is going to go down, they'll probably go down 2X or something like that.
18:57 Speaker 01 Yeah, I think it's a bit like that, but it's also about what happens when lots of goods get accessible to everyone. So 40 years ago, only a handful of people would fly and go skiing. And then with the arrival of low-cost airlines and lower cost to ski, many, many more people ski. So it's not as elite to ski. So rich people now need to ski out of helicopters or in the Himalayas or blindfold.
19:21 Auren Hoffman? And you got to wear your nice Montclair things while you ski or something to look good while you ski, right? Yeah.
19:27 Speaker 01 Exactly. You know, the Montclair drive. And this notion that maybe status goods will become even more important as the norm of what the average person can afford rises, I think is. Quite a compelling thesis, but you know, it's not one. I mean, I don't invest in these companies because I don't have a clue.
19:43 Auren Hoffman? In the US, the number one status symbol, I would say, is the number of kids people have. Like, high status people seem to be having three, four, or five kids. And this seems like the easy way to differentiate yourself rather than the Birkin bag or something.
20:00 Speaker 01 I mean, it is more expensive than a Birkin bag, to be fair.
20:03 Auren Hoffman? That's true, much more, yeah. Millions, sometimes millions of dollars per kid or something.
20:08 Speaker 01 And if you've not been able to manage it with the same partner, it gets even more expensive, right? So there's a real status signal.
20:14 Auren Hoffman? Right, good point. Richer people tend to stay together longer as well.
20:18 Speaker 01 Right.
20:18 Auren Hoffman? Interesting. You've kind of said, like, the funding quality is one of the major precipitating factors in historic bubbles. Like, what's going on there?
20:28 Speaker 01 If you look at things like the railway busts of the 19th century, like, they were all about funding quality. What that means is, is the capital that's coming in coming from smart, patient, professional investors who understand what they're investing in? Are the terms really straightforward? Direct equity or a simple bond, or is there lots of financial engineering? And going back to, I think, just before we met, when you were running LiveRamp back in the day, that was off the tail of the global financial crisis and the housing bubble. And that is the best example of what looks like bad funding quality.
21:05 Auren Hoffman? It just like got so, so complicated, no one could understand it.
21:09 Speaker 01 So complicated, no one could understand it.
21:11 Auren Hoffman? It's like a derivative of a derivative of a derivative or something, yeah.
21:14 Speaker 01 Yeah, chopped up thousands of times, sold as really high quality to pension funds and institutional investors right across Europe and the world. People who were never going to go down to California or Arizona or Florida and actually look at who was living in that million-dollar house and try to figure out whether they could pay that mortgage. So that was a real example of really bad funding quality. You had about 1.4 trillion dollars of subprime, credit impaired mortgages supporting about nine or 10 trillion dollars of derivatives that were spread all over the world. Something like that is really, really toxic. So the question is, where is funding quality today with Gen AI?
21:57 Speaker 01 It's definitely getting a little bit worse than it was, say, three months ago because companies like Oracle are having to Put together big debt instruments. I mean, they recently did a $38 billion debt instrument because they don't have the cash in their balance sheet to do that. And the market responded by a spike in something called the CDS price, which is the cost to insure against the default of Oracle's debt. So that's a good indicator. If the credit markets think this is a bit funky, they have to pay more to insure against it. And so you are seeing credit quality and finance quality deteriorate, but it's really far from what we saw with Countrywide Financial, the housing bubble back in 2007, 2008.
22:42 Auren Hoffman? World of DAS isn't just a podcast. World of DAS membership is a private invite-only community for founders and executives of important data and AI businesses. It's a place for off-the-record, high-trust conversations focused on the biggest questions data leaders are facing. The community is about peer learning, curated intros, and access to private events. Everyone in the group is building or leading something meaningful in data. If that sounds like you, apply to join at worldofdas.com. That's world of D A A S dot com.
23:24 Auren Hoffman? A lot of the AI build-outs is being funded by sovereign wealth funds in the Gulf or some other types of funds, maybe GIC. For these sovereign wealth funds, is it like strategic from a company level? Or is it, okay, we have a lot of money and we're looking for a good return. This is a good return. Like, how are they thinking about it?
23:46 Speaker 01 I've just come back from the Middle East and, you know, they think about it as a 30- to 50-year journey where you're going to go and move to a society that is built around computational platforms, computing, AI, and all the power that's required. And so they take it very, very seriously. It's really integrated into the politics. I think if you remember, the Saudi government was building this new city called the Line across the desert, and they have just started to divert resources from NEOM into AI.
24:20 Auren Hoffman? It was very much like a 1980s vision of something, and AI obviously is much more of a today vision of something.
24:28 Speaker 01 I mean, Neon was a bit like that film The Martian, right, where the guy gets stranded on Mars and has to survive for a year because that's what that desert is like. It's really inhospitable and the technologies didn't exist. If the market corrected, if there was a 15 or 20% correction in these, particularly these chip stocks, TSMC or Nvidia or Broadcom, I would wager that the sovereign wealth funds would come in immediately and buy those as bargains. I think they understand that you've got to compound over a long period of time.
24:56 Auren Hoffman? Because there's a floor price that they would just come in and buy them at.
25:00 Speaker 01 Yeah, I mean, there's a floor price I would come in and buy that. Yeah, if Nvidia went down 20%, I'd go and fill my boots as well.
25:05 Auren Hoffman? And you have almost as much money as these sovereign wealth funds do.
25:08 Speaker 01 Almost, almost as much, yeah. A couple of orders of magnitude below.
25:12 Auren Hoffman? If I was just, if I'm like, just want to watch it, whether I'm the head of the Fed in the US or I'm just an observer who wants to understand, what should I be watching to understand leading indicators, either as things are going well or things might be going south?
25:28 Speaker 01 The most important right now is the rate of revenue growth because ultimately, if real customers are buying the product and putting real money into the tank, the tank is getting fuller. Everything else is engineering and there might be risks in the financial engineering, but it has to be revenue growth. And the bit that's really difficult is that the two big companies in this, Anthropic and OpenAI, are of course private companies, so only a few people see their numbers. Occasionally, they get leaked. And then a lot of that money flows through to the big tech companies, particularly Microsoft, Google and Amazon. But even though those are public companies, it's part of their cloud businesses, which are part of the whole. So in the case of Google, Google Cloud is
26:11 Speaker 01 like $50 billion of a $300 billion revenue business and the AI part of Google Cloud might be, you know, five or six billion dollars this year. So it's really hard for us to know what the revenue trajectory looks like. So then I look at other things, you know, of course, I look at the leaks on, you know, the different websites. I look at surveys, what a manager's saying, what a CIO's saying. And ultimately, I assume that
26:35 Speaker 01 Andy Jassy at Amazon and Sundar Pichai at Alphabet and Satya Nadella at Microsoft are rational. And if they're ordering unending amounts of data centers and chips, they're doing that to fulfill future customer demand that they can see that we can't see. So those are the types of things that I'm looking at. The moment that people start to reduce their capital expenditure commitments, we're learning something about where they think demand is going.
27:01 Auren Hoffman? And we have seen a little bit of that with Microsoft. They're kind of reducing their frothiness in a way. Or do you disagree?
27:09 Speaker 01 They do occasionally cancel data centers, but you're right, it's froth. And so part of that froth is that you need to make sure you get your orders in for power connections, for the land. You're trying to do that because it's better to over order and cut the odd one or two than not be able to get in the queue. So I think there's a little bit of that dynamic. They also had a situation recently, which I think is super interesting, especially as you're based near the DC area, where there was a political pushback on a data center. So I think the town was called Caledonia in Wisconsin, and the town council rejected a new Microsoft data center build out because they just didn't want a data center on their farmland. That is also part of this picture.
27:53 Auren Hoffman? Some people don't like the noise. Some people are worried about the power usage. Some people are worried about the traffic with the jobs. It's kind of the NIMBYism that happens on any kind of project like that.
28:04 Speaker 01 It is, but you know, we're heading towards the midterms in a year or so. There will be political capital to be made one way or another through things like this. That's also something that's worth looking at. I don't think it speaks to an early indicator of customer demand softening, but it is part of this frothiness that you're alluding to.
28:23 Auren Hoffman? How would you kind of advise investors to think through all this stuff? I could just like stay tied to the mag seven. Obviously, this has been a very good thing for the last five years if someone decided to do that. But there may be other ways to play this. How would you advise investors to kind of think about it?
28:40 Speaker 01 Well, can we agree that this isn't financial advice that I'm about to give?
28:44 Auren Hoffman? Yes, absolutely. Of course. Yes.
28:46 Speaker 01 You know, booms take a really long time and the oil boom took about 50 years from, you know, up to about 1970, maybe a bit more than 50 years. And it's also true that if you invest when stocks are really expensive in general, your returns are much worse than if you invest when they're cheap. But the trouble is, can you afford to sit this out? Can you afford to wait, right, for stocks to get cheap and, you know, put your money in a money market fund?
29:11 Auren Hoffman? I mean, Nvidia, it's just like, three years ago, it seemed like it was rising fast. Two years ago, one year ago, today, you know, it'd been really nice to have put more money into that three years ago.
29:23 Speaker 01 Absolutely. And, you know, I'm just looking over here at my charts. There's one measure that you use, which is the P E ratio. So it's kind of the price over the earnings. And these stocks are a bit expensive, but the NASDAQ as a whole is at about 34 or 35. And Google Alphabet is at 29, so it's cheap relative to the NASDAQ.
29:40 Auren Hoffman? And historically in line.
29:43 Speaker 01 Yeah. And I think the question is, where do you think this is going to go? And over what period are you investing?
29:48 Auren Hoffman? That P is really about growth. It's really about how fast you think these things are going to grow in the future. And then how much capital they might need to sustain that growth. If the capital is only like a one time thing and it's going to grow super fast, then those P should be way, way, way higher. If the capital requirements are going to be even higher and the growth might be slower, then those P E's might be overblown.
30:09 Speaker 01 And I think that they're certainly not in dot com territory. I mean, in dot com territory, the NASDAQ was at like a 70 P E. Today even Nvidia, I'm just looking is at 36, so it's half that level.
30:19 Auren Hoffman? And not only is it 70 P E, but a lot of those earnings went kaput. So, you know, it's like, so you just saw like the earnings from, I don't know, Cisco and these other things really start to contract in the dot com crash.
30:32 Speaker 01 Right, because a lot of the revenue in the dot com was other dot coms advertising on Yahoo or buying routers from Cisco.
30:38 Auren Hoffman? And I imagine that's the same today. Like, I imagine, like, most of Anthropic's revenue is coming from other AI companies who are sending it to, you know, Anthropic, and they're getting their stuff from.
30:48 Speaker 01 Venture capital.
30:49 Auren Hoffman? Yeah, yeah, venture capital, or, you know, they're getting it from other companies or something like that. And I imagine, like, Cursor's, or a huge portion of Cursor's revenue is going to Anthropic or something.
30:59 Speaker 01 A huge portion is going to Anthropic, but Cursor's got real revenue and it's got real companies in the non-tech industry who are paying for it.
31:06 Auren Hoffman? And everyone of our companies pretty much uses Cursor. So it's a great company.
31:10 Speaker 01 It's a great company. And the thing that Cursor will do increasingly is that they will, right now, they're growing really fast. So what matters is growth. But you get to that stage where you start to say, we need to become more efficient and we don't want to be paying 30% of our revenue to Anthropic. So we're going to build some of our own models. We're going to find more efficient ways to do parts of our product to increase our margin. You know, the story, as all the data nerds know, is plant your flag, expand, and then start to worry about margin. That's the journey of the entrepreneur. And that's what the CEO of Cursor and all these other firms ought to be doing at this stage in the cycle.
31:46 Auren Hoffman? So US energy demand's been, like, roughly flat, let's say, for the last couple of decades. We're expecting it to grow quite a bit more. Now, people think, let's say, 4% a year, some people say 5% a year over the next decade. How does that change just like everything in the US?
32:04 Speaker 01 Oh, I think it's going to be great news for the US because energy is the fundamental currency of the universe. You can't do anything without energy. You need the intelligence to direct it well, but you really need the energy. And this is a great, great thing for the US over the next 25 years. It's going to be ugly and painful and a bit like being woken up with a bucket of cold water, but it is going to fundamentally help the US energy system in lots of ways. You know, we're putting money into it, but we're also putting innovation into it. So you're going to get people figuring out how to cool chips far better, right? Liquid cooling and microfluidics. You're going to find better ways to use batteries and to bring those prices down.
32:50 Speaker 01 It's going to make certain types of geothermal more cost effective. I mean, all of this is going to be really, really good news for the production of the thing that we all need as humans and as societies, which is energy. It'll be messy at the beginning and, you know, we've seen prices go up for consumers. We've seen grids getting strained, but if we look back at this from 20 years, I think we'll be thankful for what AI did to the energy system.
33:14 Auren Hoffman? If you were a policymaker in the US, because, you know, one of the problems if you're an energy producer, yeah, you see the trend. So you want to go build more power plants or whatever it might be. But if we overbuild, then the prices could go down and, you know, you might be underwater. Should the government have like a backstop, like, okay, we will buy excess energy for X price or something so people have a sense that they could build? I can imagine people are both, you know, the bubble could be like overbuilding and then you just have there, you know, you might be laden with debt or you may have all these other problems that could happen.
33:48 Speaker 01 I mean, the problem that you've just described is really one problem, which is, should we worry about CEOs and shareholders and providers of debt if they don't get a return on their capital? And I think the US has had a great track record for a couple of hundred years to say, no, we're not worried because that's creative disruption, someone else can come in.
34:09 Auren Hoffman? We'll just figure it out.
34:10 Speaker 01 Yeah, so I think no tears lost there. I mean, imagine how remarkable it would be if the US energy system was 20 or 30% bigger than it is now and what that would mean for reindustrialization, what that would mean for electrification of the economy. It would be really, really remarkable and it would provide a lot of price stability and maybe some capital got destroyed along the way because people overbuilt and couldn't get those numbers to work. But that's just the story of the market. So I think that an overbuild in energy is generally going to be a good thing.
34:43 Auren Hoffman? I would say my bias has been to be, and probably most people's bias is to be bearish on Europe. After talking to you for the last thirty minutes, I think I'm even more bearish on Europe. What's the optimistic case for Europe?
34:56 Speaker 01 The optimistic case is that if you're a failing student and your GPA is 1.5, getting to a 2.5 or a 3 is much easier than it is for the student to go from 3.6 to 4. I do think that the optimistic case is that we have super bright people in Europe. There are really great universities that we've now seen some fantastic companies, Spotify, a whole list of very good companies that are coming out of Europe.
35:24 Auren Hoffman? Even companies that have been around for a long time, like ASML and stuff like that, right?
35:28 Speaker 01 Yeah, ASML or companies that, Zeiss that makes the lasers and the optics for the ASML machines. There is some real expertise and capability, so you've got that as an asset. You also have a decent market that is well-run with a good legal system. What drags it down is some of these issues around the labor laws, I think, are a big problem, especially in an innovation economy. The level of compliance is really, really high for fast-growing companies. So these are things that, they're easier to address than not having any universities and not having any research and not having any patents. So I think the bullish case would be that we can get to 60 or 70 miles an hour in Europe
36:11 Speaker 01 easier than someone can go from 120 to 140. And that surely is something to look out for. And there is movement. I mean, I think that what happened with Ukraine that triggered this re-industrialization and triggered the Germans spending more on their defense and the poles is again a good stimulus. It's still too soon to have reaped the rewards of those changes, which are only a couple of years old. But, you know, look, that's the bullish side of the picture. People in Europe shouldn't give up, but there's just a lot of dynamism in the US and in China that they have to contend with.
36:46 Auren Hoffman? So what you're saying is Europe's like a rich guy who's like 300 pounds. He's smart, he's got a bunch of blubber, he knows what he's supposed to do, but it's still not that easy to get like, and he doesn't have to be like super in shape. We're not talking about he doesn't have to go from 300 pounds to professional athlete. He just has to get from 300 to 220.
37:07 Speaker 01 And the first twenty or thirty pounds should be pretty easy, just like those first bicycle cars.
37:12 Auren Hoffman? Stop doing the stupid shit. Stop eating pizza at ten P M. Stop eating all that candy. Stop drinking all the soda, et cetera. Maybe walk around a little bit more. Okay, got it. So like Europe knows what to do. It's just, it's still just hard to do it.
37:27 Speaker 01 It's hard to do it. And I think there is a forcing function that is going to emerge, which is the fiscal situation of a lot of European countries where debt levels are really high. The population is getting older and older. Ultimately, something will end up having to break, whether it's done amicably or not. And in a point where you have this diminishing workforce age population, at the same sort of time that robotics becomes cost effective and useful in these situations, it might actually create an opportunity for Europe that has got such a demographic deficit. Whereas, you know, in the U S, you don't have a demographic deficit. Your demographics are pretty healthy. You can turn the spigot
38:09 Speaker 01 on whenever you want. And so that changes the relative cost benefit analysis around, say, robotics. This is not a, you know, it's like a simple black and white picture. It's quite nuanced. And I think that makes it exciting. Actually, I think if politicians grab that opportunity and start to say, hey, we do want to compete in these ways.
38:28 Auren Hoffman? These Chinese labs have produced like very good models at very cheap prices. How do you think China plays into the A I game?
38:39 Speaker 01 I'm curious, do you use any of the models in any of your companies?
38:42 Auren Hoffman? Some of our companies, I think, have used some of those because those models now exist on A W S and some of these other types of places. So some of them, I think, do use those models, but probably still the majority is Anthropic, Open A I, Gemini, et cetera.
38:59 Speaker 01 Yeah, because, you know, we're still at the stage where we just want to build the product, right? We're not at the efficiency stage of the cycle.
39:07 Auren Hoffman? Sometimes you know what to do. And if you're, I mean, if you already know what to do, you're just kind of like trying to figure something out. Maybe those models do work really well.
39:15 Speaker 01 No, that's true. And you know, I've played around with Kimi and DeepSeek and so on. And in fact, that's what I run on my local when I go on a plane. And if the wifi isn't working, I will do my work with one of the local models.
39:26 Auren Hoffman? Oh, you have a local model like on your laptop?
39:29 Speaker 01 It's actually quite an old M2 Pro with 36 gigs of RAM. And so I can run reasonably sized models on the laptop.
39:37 Auren Hoffman? Do you basically just take like an application that has got like a model internally? And it runs on 24 gigs of RAM or something?
39:44 Speaker 01 There's a framework app called Ollama, which will essentially allow you to run the models. You know, you have to pop into the Unix shell of the Mac. You sort of pull the model down, probably from GitHub or from Hugging Face. You get the model running and then you've got an API endpoint. And then I've got a chat client that I can use, or I can just use a command line if I want to to talk to it. But I tend to use chat clients. It's helpful in those moments where you're on a plane. The thing is, you really notice how much less good they are than Claude 4.5 or GPT-5. I mean, there has definitely been progression in that year or so.
40:22 Auren Hoffman? I spend a lot of time talking to by voice and asking. So I'm spending less time listening to podcasts, listening to audiobooks while I'm in the car, and I spend more time talking. But there's plenty of times in the car where the internet is slow just because the cell connection is bad and it becomes extremely frustrating. So if there was some sort of like local model on my phone, like that would be fantastic.
40:47 Speaker 01 Oh, and I think we'll get there. And I'm sure Apple will eventually get there.
40:51 Auren Hoffman? Maybe in 40 years, Apple will get there. I don't have a lot of faith in Apple at this point.
40:56 Speaker 01 Understandably. But that modality that you've described, which is talking to the model is absolutely fantastic. I mean, it is so, so good. I do the same thing as you. I'll do my to do lists that way. So I will call into Claude and I would just throw out what's ever's in my brain and it will organize it. And then I can't yet get it to send. Me an email, but you know, when I get to my desk, I just go to Claude and I pull out what I need. But you've been using it to study things, right? Or learn new things.
41:24 Auren Hoffman? Yeah, exactly. So I find it's been very helpful. Even maybe I'm preparing for a podcast with you and I only need to get smart about energy and I can start asking all these questions about it. Or sometimes I just want to go deep on a very, very specific topic. And, you know, it could be on rare earths or something and trying to understand, what does that mean? Where does it come from and how does this work? And it's really nice because what I used to do is I would call up an expert. Sometimes I'd pay them. Sometimes they would just be like nice enough to spend a half hour, an hour with me. But it's not super efficient because sometimes I don't know everything I want. Sometimes I don't feel like comfortable just like cutting them off during that time.
42:04 Auren Hoffman? Maybe if I'm paying them by hour, I do, but it's still like a little bit like antisocial to be able to do that. Sometimes they're telling me things I already know. Sometimes they're telling me things that like way beyond my level. And I've got to bring them down, whereas I can just like much quicker. Get like a chat GPT to get on my level quickly and I don't feel bad about cutting it off and moving it around.
42:26 Speaker 01 I have such a similar experience and the one thing that I have started to do is ask it to get me to replay what my understanding. Save that for a long drive, but you're in the rear earths, you're thinking about dysprozium or whatever it happens to be. And you'll say, listen, chat GPT, ask me a question that I have to answer that will show that I'm learning and understanding what you're telling me. And they very patiently helps me reinforce those messages. So yeah, like you, I've started to do quite a bit of that. And what I discovered is that it's actually such a nice way of working with the AI.
43:06 Speaker 01 I sort of miss it. So, you know, when I'm in a public space in the living room or sitting around the kitchen table. I don't really want to invade everyone else's space with this, but I equally feel. This is how I want to work with the AI. So what, how do I do that? That I think has been for me, one of the marks you said like, what is a tell that people are going to stick with this technology? And it's not just going to fizzle. That's the kind of tell for me that is a softer tell that says this thing is real.
43:32 Auren Hoffman? Yeah. Well, the problems with the voice is that it works quite well with like one on one, like me and the AI. It works well. If I'm in a room with my family or something and we want to have the AI as like a participant in the conversation and jump in occasionally, it doesn't work at all. Like it can't work with multiple voices. It doesn't know when to jump in, when to not jump in. You know, with a group of people with two or more people, like the way you might use AI, you'll have it on your phone and you'll like occasionally ask a question of it or something. Or you'll pause the conversation, talk to it and then stop it and then go back to your conversation. It would be nice if it was like, it could be a passive listener and then occasionally jump in or you guys are all talking about, you know,
44:17 Auren Hoffman? what percentage of is the mag seven of the S and P 500, but you know, it could just jump in occasionally and be nice to have even on this podcast.
44:26 Speaker 01 That reminds me quite a lot of the Alan Kay Knowledge Navigator video from 1987, which was this really famous video coming out of Apple's Advanced Technology Group, where they had just that kind of ambient smart AI. And it's really worth, I think, going back and watching that video on YouTube and thinking they imagined this in 1987. I mean, this is not C three PO style stuff, R two D two stuff. This is really, really thoughtful. And it touches on so many of the things that you and I have just talked about, like just being able to fit into our daily needs and be a bit predictive about how we want to behave and what we next want to do.
45:06 Speaker 01 I mean, I think we're slowly, slowly getting there, but you've thrown a really tough challenge there, which is having it participate in a conversation with four or five people.
45:14 Auren Hoffman? I feel like it's gonna happen. It's just gonna take some time. Now, in the U S at least, so many of the well-known AI researchers are Chinese born. The rest of them are like have Chinese born parents or something like this is so, so much is dependent on that flow coming in from China. You can imagine there on the one side, there's these national security concerns. What's the allegiance? On the other side, it's like, well, let's get more of those people in here. How should the U S be thinking about that?
45:44 Speaker 01 I mean, it's really, really complicated, isn't it? I mean, it's just, it is a benefit, it's also vulnerability. I mean, you can run an AI lab without global talent because that's what DeepSeek is doing. And that's what Alibaba is doing. You know, the US has always done really well with bringing high caliber people of all sorts and attracting them for one hundred and fifty years. It's a recipe that has been so successful that in a way you'd want to continue it and figure out how you continue it. I think one of the challenges, I know you're over in the DC area, so you're closer to the national security caucus of the US. You know, I'm not so close to those groups, so maybe I'm a little bit soft about people's intentions.
46:27 Speaker 01 I do think that human capital and the best talent is really what drives a lot of these waves of innovation and technology. So you need to figure out how to attract these people, how to keep them here. And it's not about the H one B, right? You know, the H one B is a, as an immigrant visa, was very, very broad. It was very mid. This is about the very, very best people and how do you make them feel welcome?
46:51 Auren Hoffman? The ones and.
46:52 Speaker 01 Exactly. And I think that that has still got to be a policy priority in the US as it must be in the UK and across Europe as well, because this, we've seen historically, the differential talent in the power law makes a big difference.
47:04 Auren Hoffman? In AI specifically, it seems like if you go like pre-AI computers in the US, so much of that innovation was driven by people with Indian last names. Now it just seems, in at least in the last few years, it seems that it shifted and so much of the innovation is driven when people went with Chinese last names. Do you have any theory about why that shift has happened?
47:27 Speaker 01 I don't have a good theory for that. I will hazard something. I will hazard that, you know, India produces, as you say, these kind of phenomenal business people. And if you look at any of the major banks in Wall Street or the City of London, they have a lot of really senior Indian traders, MDs.
47:46 Auren Hoffman? And just look at the CEO of Google, the CEO of Microsoft.
47:50 Speaker 01 But also within finance. And I think that that is the nature of the IIM, Indian Institutes of Management, which are their super elite schools, and the IIT, which is their Institute of Technology. These have produced fantastic engineers who turn into great managers or financiers. I suppose with China, my thesis would be, I'd need to look into it, I'll ask ChatGPT after this call. The Chinese education systems emphasized harder sciences over things like management, right? So you flowed talent into those areas.
48:26 Auren Hoffman? So this is like the old Soviet, there were so many great Soviet scientists that came in like the eighties and nineties and stuff.
48:32 Speaker 01 Exactly. I think it could be something like that. I haven't really given it a lot of thought. I can't even think of what the equivalent of the Indian Institutes of Management would be for China, right? So that, you know, I think about, so what is that? What is that? But I can think about, you know, the great science and technical universities.
48:48 Auren Hoffman? And beside for what we mentioned, how are you personally using AI?
48:52 Speaker 01 I don't even know where to begin. It is everywhere. I'll give you one of my favorites. So I use an app called Shortwave, like shortwave radio. So it sits on your Gmail and it runs through Claude. And so all of your email is indexed and accessible through Claude. So I'm looking at the founders who I invest in and they send me their monthly updates. I can get Shortwave to say, look at the updates from, say, I'd invested in you, Orin, for the last nine months. Tell me how consistent he is with his goalposts. Is he meeting his promises? Is his tone optimistic or defensive? What are the key things we need to look out for?
49:32 Auren Hoffman? And you can't do that in the basic Gemini stuff that's built into Gmail?
49:37 Speaker 01 I haven't been able to do it as well as I was doing it with Shortwave. And what it means is I can give much better attention to my founders much more frequently. So I really, I really, really use that multiple times a week to help them, right? Understand what they're going through and where they need help.
49:54 Auren Hoffman? Yeah, interesting. Okay, two questions we ask all of our guests. What is a conspiracy theory that you believe?
50:00 Speaker 01 I have an absolutely mental idea, which is that Satoshi is actually from the future and he was sent by aliens in the future to give us this great technology so that we can progress forward because otherwise we wouldn't have figured it out.
50:15 Auren Hoffman? Ooh, I like that.
50:16 Speaker 01 I want to eventually, like, get to the stage we can join the Galactica Federation or something.
50:20 Auren Hoffman? Oh, okay. I love that. It actually kind of does make sense in a weird way. I love it. Okay, last question we ask all of our guests. What conventional wisdom or advice do you think is generally bad advice?
50:31 Speaker 01 I think the advice of stay in your lane and really focus is now not great advice because the world is changing so much. And because it's changing so much, you have to have a range, you have to be able to explore, you have to feel confident that you can explore. You absolutely need some tools, like you need, I don't know, great maths, you need to be a great negotiator, or you need to be a great empath. But ultimately, your lane is about to get broken up by an earthquake or covered in rubble. You need to be able to spread more widely. And that advice was advice, stay in your lane, that, you know, I'm sure you were also given when you were at high school and a college, but I think bad advice for today.
51:10 Auren Hoffman? That makes sense to me. Yeah, that does make sense. This has been amazing. Thank you, Azeem Azhar, for joining us on World of DAS. I follow you at Azeem, A-Z-E-E-M, on X. I definitely encourage our listeners to engage with you there. I also subscribe to your newsletter. This has been amazing. I knew it would be amazing, but this has been amazing. I really appreciate your time. Thank you so much.
51:32 Speaker 01 Aur, my pleasure. Really enjoyed it. Thank you.
51:37 Auren Hoffman? Thanks for listening. If you haven't already done so, please subscribe to World of DAS wherever you consume your podcasts, YouTube, Apple, Spotify, and more. And please help us get discovered by leaving a review and check out worldofdas.com. That's worldofdas, D-A-A-S, dot com. And of course, connect with me on Twitter, at Aur, that's A-U-R-E-N. Would absolutely love to hear from you.
Previous The LM Brief: Love Beats Hate Next The LM Brief: Why Many AI Projects Fail Page updated October 31, 2026 · transcript and summary are machine-generated