Latest / Complex Systems with Patrick McKenzie (patio11) / Boom, busts, and long term progress with Byrne Hobart

Transcript

Boom, busts, and long term progress with Byrne Hobart · Complex Systems with Patrick McKenzie (patio11)

• Patrick McKenzie & Byrne Hobart - recorded 11/8/24 Hi everybody. I'm Patrick McKenzie, better known as patio11 on the internet, and I'm here with my friend Byrne Hobart. Byrne Hey, great to be here, great to be back. Patrick Good to have you back again. So the thing you never want to say about an investment manager is, “I invited you on to talk your book,” but I literally invited you on to talk your book because you and your friend Tobias – who has also worked in investment management and crypto to my understanding – are publishing a book through Stripe Press, and I just want to dig into that thesis a little bit. Can you tell us a bit about Boom? Byrne Yeah, absolutely. The book is Boom, Bubbles, and the End of Stagnation. What Tobias and I explore in that book is the idea that, when you look at economic growth over time, it does demand an explanation: for a very long time, there wasn't much of it, and then in more recent history, there's been a whole lot of it – and then in even more recent slices of history, the last generation or two, there's been a bit less of it than we might have expected as of the mid-20th century. So I think if you were writing about economic growth circa the 1950s, 1960s, you could have had this view that, not only does the economy grow over time – which we all expect – but the pace of that growth actually gradually accelerates, and that when you start in with a very primitive economy, things just don't change over periods that individuals can actually recognize and measure. (Whatever productivity changes there were in the Neolithic were probably swamped by whether there was a drought or flood or pandemic or something somewhere in that time series.) Then as we move to agrarian systems, growth picks up a little bit, but it's still just barely, barely visible over the course of a human lifetime; then when you hit the Industrial Revolution, things are materially different – materially in the figurative sense, but also in the physical sense of just, “there is more stuff; there are kinds of stuff that were available for purchase that simply couldn't have existed under previous technological systems.” Then things get really, really crazy towards the mid 20th century. There's this great boom in productivity. I guess it's useful to talk about what we mean by productivity. Productivity and alpha are two statistics that I like a lot because they're both residuals, where you run a regression and there's an error term and you eventually put a label on it and say, “this thing is important”; in the case of alpha, you can look at what an investor owns – what asset classes they have exposure to, what kind of leverage, how long they are in the market – and you could roughly predict their returns. But active managers do not have those exact returns; they do not have the return you'd expect from the S&P500 because if they did, then they would just be passive. Instead, they have either something better or something worse once you adjust for all the risks that they're taking. The “something better or something worse” just gets labeled with alpha. Alpha is just the error term on a linear regression that compares someone's exposure to broad asset price changes and the actual return that they get. Similarly, total factor productivity is the error term that shows up when you try to model the economy as, “we mash together a certain amount of physical equipment, physical wealth, and then a certain number of worker hours, and we get some level of output.” It's a very intuitive model, makes perfect sense, and if you look at that model and look at the growth rate of those two terms over time and compare that to GDP, you find a widening gap. That gap is not just that we have more factories, more equipment, bigger ports, bigger boats, etc. – it is that at a given level of fixed capital investment and in a given level of labor input, we actually get more out of it. There are lots of fun debates [to be] had about the exact nuances of that metric and whether that is the right way to think about it, but it's so far probably the best one we have, and at least the one that everyone has agreed to focus on. There was this period in the mid-20th century where output per hour was rising at a faster pace than it ever had before – it was, like, two-ish percent per year. Starting around 1970, that slowed down. There are a lot of different ways to try to explain that and a lot of really interesting ones. You can look at things like the Gordon model: we discover general-purpose technologies, they have numerous applications, and they're so general-purpose that it takes us a very long time to figure all of them out. (Think of the internal combustion engine or electricity, something like that, and then more recently the transistor.) I do think that that is an important part of it. But what we look at in the book is this notion that part of what actually kicks this stuff off is some kind of period of totally wild, irresponsible optimism about what can actually be done, what can be built. In the book, we have a couple of case studies of this: just because of the slice of chronology that we focused on, they do start with large scale government mega-projects – things like the Manhattan project, Apollo program – and then move into more private sector stuff. The bridge between those two is this period of corporate R&D labs investing really heavily in just building cool new stuff – institutions like AT&T and DuPont and Lockheed and Xerox, where they have an existing business, it's a profitable business, it's some flavor of monopolistic, and they put a lot of resources into building something new and different. Sometimes that's very much directed by immediate need: Lockheed Martin’s product roadmap for their skunkworks was pretty much set by, “What is the latest intelligence on what the Soviets are doing? What can they stop us from doing and how can we counter that?” AT&T was kind of in the opposite direction; they realized they have this nice monopoly, they have an understanding with the federal government that they are allowed to run a monopoly telephone company (because it does actually make sense for that network, it's kind of pointless for your country to have two of them that each cover the same area). So, we have one monopoly, it's profitable, but there's this tacit agreement that it's not too profitable and they're not going to use it to extend too much into other markets. But AT&T did realize they could bring together a bunch of researchers and just kind of let them run wild. The company was big enough, it was just a large enough share of the economy that they felt like, if you have a bunch of smart people who are just generally interested in solving problems, they will eventually find something really worthwhile to do. Patrick I remember over the course of my adult life, so many people complained in engineering and tech spaces, “Why isn't there a Bell Labs anymore? I mean, Google has all the money, billions of dollars, and all the PhDs in the world. Why don't they ever write anything meaningful?” I've heard that complaint a little bit less since somebody wrote a paper about Attention being All You Need. (Okay, to explain the subtweet, that was the genesis of the LLM revolution a few years ago. Ironically, [Google] didn't end up being the ones that seem to have actually shipped products to market based on it, but some foundational engineering research ended up (knock on wood) making an impact and perhaps causing one of these booms that we're talking about.) Byrne Yeah, I think that's a fun example. There's another layer of this which is less of a focus in the book, but which I think is an interesting thing to think about: how has the relationship between the private sector and the government evolved over time? It does feel like if you look at mid-20th century America, there pretty much is such a thing as ‘the establishment,’ and people can kind of tell if they're in it or not in it. A lot of important, powerful people kind of know each other and sort of have a common understanding of what the broad goals should be. That can absolutely lead to sclerosis and to corruption, to continuous underperformance, and to an elite that doesn't want to let anyone else in – but this particular elite was actually not that great at preserving its own control of things, and it was actually pretty good at engineering an economy in the mid 20th century where there was an understanding of what the relationship is between government and business and labor and the consumer. It was all kind of a nice understanding. Some of that was things like, “AT&T is going to spend more on research and development than might be strictly rational.” They will occasionally have these big hits; there was a brief note in one of the source books that I read on Bell Labs that was talking about how someone found a cheaper way to sheath the cables, and that AT&T realized that at their growth rate they would have ended up using some huge fraction of (or all of) the lead produced in North America – so Bell Labs did end up pretty much paying for itself through things like that, like someone finding some clever way to save a small amount of money, which at AT&T’s scale is a lot of money. Then socially it definitely paid for itself by leading to the development of the transistor. There was this nice situation where, if the transistor had existed as just a “cool thing that AT&T has,” it would have definitely made the phone system more efficient and would have eventually led to a lot of the things that we enjoy today – but it probably would have taken a lot longer, whereas if AT&T develops it and realizes that it's probably more trouble than it's worth to make this a big business that they own and that they try to retain control over, they license this technology quite cheaply to many other organizations. What we found with the transistor was that it has just a uniquely – well, probably uniquely until solar panels – favorable cost curve. Moore's law is basically a universal thing for anything people do: if you do an order of magnitude more of it, you get a bit better at it, it's just a question of what that curve is, and then whether there are any organizational or raw material or other physical limits to how good this thing can get. There are cases where the asymptote of that curve is, “to build this product, you have some chemical reaction, it requires some energy input, and therefore the cheapest this will ever be is determined by the cost of energy,” which also means that lowering the cost of energy and making energy more abundant does make everything better. It moves the asymptote for everything. (That is a tangent we will, I'm sure, get to later in this conversation, but one worth keeping in mind.) Patrick If I can take a different tangent for a moment, one thing responding to something you said: often in the moment, the choices of which projects to focus on look quite a bit irrational. The line that bored both of us to tears for many times over the years was that “the best minds of my generation are spending time getting people to click on more ads.” Personally, I think that line was largely false when it was said, both in terms of like, “it's not the case that AppAmaGooFaceSoft actually have all the best minds with them,” but also that, to the extent that getting really, really good at understanding what human behavior is when presented with a web page is something that we were spending the best minds on, when we look at the experience of retail interactions with a financial system or government right now and realize that a huge portion of them are mediated by web pages, plausibly having the best minds of your generation focused on the retail experience of government, the retail experience of the financial industry, the retail experience of causing houses to be built is not a crazy idea. But it might look crazy for the first couple of years until you get one or two steps down the tech tree. Ok, sorry. Tangent over. Byrne I think that's true. I think it is always easy to underrate the social utility of making it slightly easier for people to transact with one another. But our economy is made up of transactions, and if we make those transactions a little bit more efficient, we lower the search cost to finding the exact thing that you want to buy, and also finding the exact thing you want to sell – that is, you know, getting a good job. If the search cost for that is lower, then there are huge benefits that are very hard to trace to, “this person ran this A-B test and moved something three pixels over and it added another basis point to the clickthrough rate, and that was really wonderful at Google scale.” That stuff does have utility. Actually, turning back to the question of how these different big categories of society relate to each other, one of the things that the book is implicitly tracking is this evolution in where really talented and ambitious people go. In the early case studies, the government was the only game in town. If it's 1935, you just finished school, you are really, really lucky if you get a job at a big company, and there's really no shot at getting promoted rapidly through the ranks. But if there's some New Deal program that didn't exist before and now it exists today, they will be hiring – so you will get that entry-level job. They also need talent, and because a lot of the other cohort of really ambitious go-getters have moved into this particular sector, there were actually pretty high standards for how hard you have to work, how much you have to accomplish, how well you measure your outputs. The 1930s are also a time when there are some early declines in IT costs, the cost of just getting a bunch of data, putting it on punch cards and running really simple algorithms on it – it made more sense to start measuring more things, so a lot of our oldest macroeconomic time series, they start around then. Patrick I think the thing that people underrate about economic history is that, particularly when folks are younger, they assume it's the eternal present – that the structures of society in the past must have resembled quite heavily ones in the future. So one – you read more economic history than almost anyone, I've read a book or two in my life – one of the singular things about the early 20th century is that the large administrative state and large corporations evolved with each other, but the state came first. It was descriptively not the case prior to, for example, AT&T in the 1910s-ish era, that there were large national employers that were running a large national search for talent – you couldn't simply get a job at whatever the equivalent of Google was in 1930 because there basically wasn't one. Very little of the economy was employed by large corporations. Then run 40 to 50 years of economic history; now a lot of the economy has migrated off of the farm and into large corporations and other relatively novel economic models, and thus we have the results of this talent search. I also think that the societal expectation about, “how do we identify talent and nurture it, inculcate it, run a nationwide search, etc,” went through any number of changes over the years. We got much less exclusionary, to our credit. We got a little bit more optimized with respect to things like, “Ooh, it turns out that widespread testing will allow us to identify people early.” And there's some pluses and minuses to that one. But it's not monotonically up and to the right either. This is one of my enduring mind-blowing moments: there were a couple of social technologies that we definitely possessed at some points in the middle of the 20th century where, for a variety of reasons, we decided to lay them on the wayside or restrict which areas societies was able to make use of them for a couple of decades, and then we're like, “oh!” and thus rediscovered things that were known to our forefathers. The one that I was used as an example because I spent my entire adult life in Japan is that – this is a joke and an exaggeration but has a very core bit of truth to it – Japan ended up with a worldwide monopoly on applied mathematics for a period of about like 40 years, right after the United States (well, the United States and the UK) had effectively invented the discipline to win a war. Having won the war and dealt with an excellent economic boom as a result of many factors (including all of the other competitors being wrecked by losing the war), we were like, “OK, well, give the labor unions what they want. No need to measure GM by using statistical processes; I mean, come on, we're bored of that stuff already.” Then 40 years later, they were like, “Wait, you really teach people at a plant in central Japan, even the people who sweep the floors, you teach them statistics 101? Why do you do that?” And now one of the largest corporations in the world (and I'm subtweeting because I previously worked in an offshoot of them) is – an internal and external reading of their success is partly, “yeah, we picked up the trillion dollars that was left on the table.” OK, rant over. Byrne No, I think it's a fun one! It is really impressive how Japan just picked up that torch of, “we are going to figure out how to measure how good factories are and we're going to continuously make them better.” Maybe for the second edition, we'll do some more institutional technologies rather than physical technologies, because the institutional technology of always assuming that you're not at optimal efficiency and that there's something you're doing that you have to do differently in order to improve the process – that is a very different attitude from US mass production in the 50s and 60s, where it was, you just design these vehicles, and since no one at GM and Ford and Chrysler is especially worried that market share is going to shift all that much, you just make a slightly bigger vehicle with slightly more features and then you crank out a lot of copies. Because you are cranking out so many copies due to all these fixed costs and the fact that you are pre-ordering this huge amount of inventory that's going to be flowing through your factories, any kind of disruption to that process is very expensive, so of course you tell the unions that they can have the raise, they can have the health benefits, et cetera., and for a long time that works. The US did have just a huge industrial base post-war; some of that had existed beforehand, some of that had been funded by the federal government and then sold to the private sector very, very cheaply. There was this sense of, “We just created one of the world's largest militaries and these people suffered a whole lot; they’re home, we probably want them to be very, very happy with how everything turned out. We don't want them to feel like they fought a really brutal war in order to make sure that GM could pay a higher dividend” – so yeah, of course, those companies were pretty generous to their unions, and there was enough consumer demand that it made perfect sense to focus on selling more cars instead of optimizing your cost structure. Japan had a very different set of constraints. There's this wonderful book, The Reckoning – basically the peak of the US auto industry's influence and then its absolute collapse – and one of the things it opens with is a Japanese industrial planner visiting the US, because he's actually thinking many, many more steps ahead of the Americans. He's aware that the amount of oil that the US produces is not rising anymore, and that the US gets its oil from the Middle East, and that the Middle East is not necessarily going to keep producing as much oil as the US wants (as nice as that would be for the American standard of living circa 1970). Japan had always had to rely on external sources of oil; they had a lot of experience thinking about the implications of that, even a little bit earlier than the period we're talking about, so they had done things like encourage people to learn Arabic in college and have more Arabic speakers who could actually negotiate one-on-one with the Middle Eastern countries. They also had much more efficient vehicles, in part because Japan was just a very, very poor country when a lot of these companies started ramping up, and so you couldn't actually make a huge gas guzzling vehicle – you knew that your customers were all economizing on things like that. But what the Japanese industrial planners were actually worried about was “We know oil prices will go up at some point, but we don't know when; we know that when that happens, suddenly Japanese cars are going to be much, much more competitive with American cars, and we know when that happens that America is probably going to get very protectionist and might just ruin the party for all of us.” So the Japanese industrial planners were basically begging American car companies to please make smaller, more fuel efficient cars, to please become more competitive with Japan, because they realized that in a fair fight that the Japanese companies would absolutely pummel the American ones – but the American companies were thinking many steps behind this and just weren't really aware of how much of their entire business model depended on Saudi Aramco not being Saudi, just being Aramco. So yeah, it was interesting to see, but that was also a reflection of the fact that the Japanese industrial planning organizations were just getting the best people – that's where those people were going, and so of course it was a very, very well-planned economy for a long time. In the U.S. it was going into different places. Some of it was actually going to the automotive industry, but there was this weird dynamic where Ford famously got a lot more serious and a lot more rigorous about how they managed and measured their business, but the actual part where you manufacture cars was kind of hard for Harvard MBAs to wrap their heads around, whereas the part where you manage working capital and receivables and you get your depreciation schedules right and all of that stuff, that was a lot more tractable to them, so a lot of these companies had a much better financial organization than manufacturing organization and sort of over-optimized the piece that they were better at over-optimizing. Patrick Yeah. Oh man, I've got so many tiny asides. One tiny aside: I think that the MBA-ization and McKinsey-ization of the American economy is often given too little credit. There is actually some signal there, and firms are broadly more efficient than they were 100 years ago in a lot of ways. Two, a Japanese industrial planner anecdote from slightly more recent than those conversations: back in the mid-2000s when I was in Japan working at a prefectural technology incubator, there was this artifact that was coming out which many analysts in the United States thought very confidently would never have a shot in Japan – it was this “iPhone” device, and it was clearly never going to have a shot in Japan because Japan is, as everyone knows, a rabidly xenophobic nation and would never go with a non-domestically manufactured phone. (Clearly, the people who were saying that had not been on a Tokyo subway recently and seen all the iPods and the paucity of whatever Sony's generation of the walkman was back then.) Anyhow, so the iPhone comes out, the iPhone has a bill of materials associated with it; who is making each of those devices that make up the artifact that is the iPhone? A thing that was whispered in the halls of power back then was something to the effect of, “I remember the 1980s when there were Japanese cars put on the nightly news and burned in the United States, and that was an uncomfortable time. Right now, there is this wildly successful device which the Americans believe to be manufactured in China. Now, if you put a Japanese camera and a Japanese motherboard and Japanese [list a bunch of components], and the BOM is 40% Japanese, but it's in a Chinese paper box, I think they will be burning Chinese flags on the TV and not Japanese flags – and so we are comfortable with this lack of understanding of how the supply chain works.” Now granted, I think that's complicated story because that supply chain shifts over time; I think a lot of the supply chain for Shenzhen has continued to move closer and closer to Shenzhen over the years, But that was one of those funny little bits in time that for weird reasons I ended up having a slight ability to witness shift in real-time. Byrne Yeah, there is this weird thing of, it feels like any sufficiently complicated supply chain will pass through at least one Japanese company that just has a global monopoly in a component you've never heard of, that is a fairly small portion of the final price, but is absolutely indispensable. One of the interesting ways to think about that is that part of Japan's economic specialization – this was an essay from a macro fund talking about this like 20 years ago – is in things with a very high cost-to-weight ratio because shipping is a constraint; if you are going to specialize in some kind of exported good, it's probably good for it to be the kind of good where you're happy to put it on a plane if it gets to the customer a few days faster, because that’s just a less-disruptable supply chain, and you just have more room to keep on selling those indispensable intermediate goods. It's true in just a lot of different industrial applications, and a lot of electronics products as well. Patrick Without specifically identifying it because salarymen don't do that to their best friends, my best friend works at a company which makes electronic flow meters, which are the devices that attach to pipes and other things that tell you how much is going through the pipe at any given time. Basically everyone listening to this podcast is downstream of Japanese electronic flow meters… or acoustic flow meters, they're very big on the difference – as you must be, if you are a salaryman whose job is getting on a plane with a suitcase full of flow meters to take to, without loss of generality, Saudi Arabia. (BTW, [when] you fly through an American Airport on your way to Saudi Arabia with a suitcase full of things that have pipe-looking components and bomb-looking components, you end up with a bunch of very fun conversations.) So be that as it may, one of the case studies you talk about in your book is that the fiber rollout in the dot-com days – so in the late 90s to early 2000s – and how we are living downstream of that. For members of the audience who might not have been around for that or for those who weren't paying paying quite as much attention to the internet, what happened with the kind of rabid over-investment in the 90s, what was “learned” from that, and then how should we think about it in the present day? Byrne Yeah, yeah. So if you were an early adopter of the consumer-facing internet, the internet that was not just grad students – so someone whose parents got the AOL disk in the mid-90s and decided to try it out– you connected very slowly. It was through your phone line, so you couldn't do phone calls while you were online, which meant that you did get kicked off the internet a lot; it is weird to even use that idiom today. Like, how does one get kicked off the internet? You have to really, really mess up. Patrick I remember we were online or we were not online before everyone was online all the time. Yeah, sorry. Byrne Right. So it was slow. And pretty early on, people who were sophisticated and metrics-driven would notice that the faster someone's internet connection, the more time they spent online; it's just a more gratifying experience to click and see the thing you clicked on, rather than you click and things start loading, it takes a while, etc. There was also this hypothesis that turned out to be quite true, that eventually TV is going to move online; it turned out people were a little bit optimistic about how soon that would happen in the mid-90s, but they did recognize that it would happen. It is just more natural to stream what you want to watch on demand rather than having a pre-existing schedule and tuning into something someone else has chosen. To make that possible, we needed much, much faster connections, and that included both faster connections directly to the home and then just a better network backbone. So people went to work. They started digging and laying fiber-optic cables and all that good stuff, and then it turned out that the fiber-optic investment was just well ahead of its time. That even though internet traffic was compounding pretty fast, it wasn't compounding quite fast enough to justify that. And so you had a huge amount of – I believe there was actually a lot more money put into the telco infrastructure than into dot-com businesses themselves in the late nineties – and a fair amount of that infrastructure, like the companies that built it went bankrupt. They just ran out of cash and the traffic didn't quite get there in time. There is this dynamic in a lot of industries where, if you have a product where there's a fixed cost to build it and then little or no marginal cost to use it, that is a business where you are very very happy if there is a slight shortage – because now there's a bidding war for whatever capacity you have – and then you are very very very unhappy if there is even a slight surplus, because now the the economically rational thing to do and just in first-order terms is, if you have excess capacity and someone is willing to pay you a penny for it, you'd rather have the penny than not so you sell it. So those prices tend to be very responsive to these changes in demand. There was a whole lot of spare fiber-optic capacity. There were, one, not a lot of homes that had direct broadband access – fast-enough access that they could actually stream a video – and two, not a lot of people putting video online because the demand wasn't there. So that created a situation where there was a lot of fiber that had been built with the expectation of “we'll flip the switch as soon as demand comes” – but demand had not come, and the company was bankrupt. Now the company is controlled by people who bought bonds and bondholders – you know, very, very smart people. I think there's a case to be made that they are often more sophisticated and think more probabilistically and reasonably than equity investors. You certainly see that a lot with, if you compare the chart of a company's stock versus their bonds in the year or so before bankruptcy, the bondholders are doing the math, the equity investors might be… doing something. (They may be doing more sophisticated math, because there is this sense in which if the company's insolvent, the stock actually becomes an option where the strike price of that option is the debt the company owes; if you think of a company where its assets are worth a billion dollars and it owes creditors half a billion dollars, the equity is worth roughly that half a billion that's left over – roughly – but if the value of the assets goes down to like $400 million, you can think of the equity as a call option, where if they bounce back, you get most of that upside as an equity holder. What you're paying for is that potential upside.) Anyhow, what bondholders generally do not like to do is extrapolate very, very high compounding growth rates and have some amazing vision of the future, because that amazing vision of the future is, you made 100 cents on the dollar and you got the interest payments you signed up for. Not an incredibly exciting thing. So a lot of these companies were liquidated and there weren't very many buyers – but Google was a buyer. They did figure that traffic was growing, they were comfortable enough with their business in the early 2000s, comfortable enough with the fundraising environment that they would rather lock down this fiber at pennies on the dollar if they could, and so they did. Then we did start to see more videos moving online, which was partly for technical reasons – I actually just read a transcript of a speech by one of YouTube's founders talking about just what were the circumstances when it was started. One of them was that there weren't great video players. Then Flash got an update. (For younger listeners, Flash was the universal way that you would create or consume interactive content online if you were a teenager in the late 1990s. It was kind of a nice little playground for just creating interactive elements or creating little animated videos, and at some point they added support for just regular video.) So YouTube realized, “We can use Flash to create a video player!” There were more cameras than existed before and more people were taking videos; the reason that there aren't a lot of videos online is that there hadn’t been a good way to share them. Once you have a good way to share them, we already had the hardware to create those videos; now we could upload them. More and more people had enough bandwidth to actually watch them online. (The experience of watching a video online [before this,] in the lower bandwidth era, was, you would think about the video and then you would click to download and then go away and do something else, or you would look at your computer every couple minutes and see if it still says buffering.) Patrick If I can give one other story that people might not remember for sensible reasons: before our friends in Europe decided that cookies were either a delicious treat or a menace that they needed to save global society from, they came to the conclusion that a computer should not be able to play a video out of the box because that was anti-consumer, and fined Microsoft $1 billion for making that possible. We could do an entire podcast on the Codec wars, but partly the issue was that bundling “codecs” capable of doing high-quality video was not allowed for a while, partly it was the great RealPlayer vs. Flash, et cetera war for standardization of the video market, much like there had been a war between VHS and Betamax earlier and VHS won, and then essentially Flash-based video won… until it no longer won. (Funny stories from a long time ago.) As a consequence of this boom in investment, the public sees the “dot-com” companies get shellacked; the actual huge stack of money was mostly invested into this infrastructure and supply chain that was supporting the coming transformation of the internet. So some people began to believe in the intervening years, “Yeah, it was all just a fad; Paul Krugman was right, the internet will never be more economically impactful than the fax machine” (Every time I read that article I get annoyed.) Byrne I have a contrarian view on that one, actually Patrick I would love to hear it! My contrarian view is that, “darn it, he underestimates the fax machine.” But I'm a Japanese salaryman and need to defend our honor. What's your contrarian view? Byrne So part of it was just, maybe he exaggerates this a little bit; the article was part of this semi-tongue-in-cheek collection of futurism and things in the Times. The idea was, “We're all going to publish our wildest predictions and we're to have some fun with this.” So I think it was a little bit less serious than people give it credit for. But also, a big part of that piece was his claim that we are way too optimistic about the technology sector, and we're not really thinking about other serious economic constraints like natural resources, and if you look at the 10 years after he published that, I think the NASDAQ roughly broke even, while oil futures went up almost 10-fold over that intervening period. So it was actually the case that we were spending a little more time focused on the world of bits, and not thinking enough about some of these atom-level constraints, that there are physical inputs that a modern economy needs and they exist in particular places, have to be processed in particular ways, and their prices are not fixed. It is very easy to ignore things like oil prices when they're under $20 a barrel and just treat energy as a cheap input that will always be there. But that tends to be one of the things that drives that cycle: if everyone is underwriting the assumption of indefinite cheap oil forever, if everybody invests in things that require oil as a consequence of that, like Americans buy larger larger cars and the developing world develops faster than previously anticipated, then you quickly reach a point where that supply-demand balance shifts. You also at that point have this shortage of human capital. There's a wonderful chart that circulates from time to time that is tracking the number of people who graduate with a petroleum engineering major and the price of a barrel of oil; it's very reliable that the number of graduates peaks a year or two after the price of oil peaks. So on average, if you are specializing in that space, you tend to graduate into a really, really bad hiring market, you might pivot into another career, and there's always going to be a surplus of people who know a lot about poking holes in the ground and getting energy to come out, right when you really wish you had a lot of them. Patrick As indeed, people told folks in the technology industry in say 2000 through 2004 when I was at university for a CS degree, “Sorry you missed your shot, clearly there will never be programmers needed in the United States again, have you considered immigrating to East Asia?” The Wall Street Journal – oh man, I owe that article a lot. Seriously, in addition to that being a vibes-based thing, I was literally advised by people who have the title Advisor at my university, “Have you considered it like something that has a future in it, rather than CS?” I think one of the other things – and not to grind the gears of people who declared the internet over too much – but one of the other things that gives credence to folks that think during a bubble snapback that “clearly no value was created” was that financial bubbles and other investment bubbles are often accompanied by very, very large frauds, due to the dynamics of how frauds must expand over time or be exposed. Typically the bubble popping will be roughly contemporaneous with shockingly large frauds being unveiled. And often those frauds will have like the rough structure where, there was a real business a couple of billion dollars ago, and the business had real returns, but the level of investor enthusiasm chasing returns was sufficient to fund a very large business and the actual real economics of it didn't support creating that business, so we just made up the difference – and that worked for a while until there were no longer investors to bankroll us. And so in case of the fiber rollout, this was MCI Worldcom I think. Byrne Delphia is the other one. Patrick [Right.] Also, I think if I remember right, a bit of the Enron story was bandwidth trading too. (Because there was never any fraud that they did not say, “yes, let's do that!”) Byrne Yeah, they did have this business where theoretically they were trading bandwidth, and I think they did some kind of deal with Blockbuster where Blockbuster did not have a streaming service, but they signed a deal with Enron where if they did, they would buy bandwidth at some rate, and Enron was somehow able to structure this deal so they could actually report profit from – again – selling a service that was not being consumed because the buyer did not actually have the business that would consume that service. Patrick Yeah, my recollection was that they booked all the revenue upfront and then said, “we will recognize the expenses associated with this contract as those expenses actually arrive when we are hypothetically in future years, you know, delivering actual electrons.” But yeah, shenanigans happened. We’ll talk about crypto maybe in a little more detail in a bit, but, if you are a crypto skeptic like me, one of the reasons to be crypto-skeptical is like, “wow, there are some very staggeringly-sized frauds there,” and the mechanism of revealing those frauds is often like, wobbliness in the market causes the numbers to look no longer paint-over-able in the way they were last year. The reason we discovered that FTX was less backed than it pretended to be was heavily contingent on there being a earlier, mostly unrelated fraud which had left them with greatly depleted capital reserves; the people attempting to exit that fraud were (perhaps unknowingly to FTX management) largely exiting through Alameda, and Alameda lost a shedload of money and then they no longer had the assets they believed they had, and ramped up their dependence on both the FTX customers and on FTT as an asset. Then fast-forward to some drama between SBF and CZ, and then CoinDesk gets a leaked balance sheet, and we're all off to the races on one of the biggest financial frauds discovered to date. There are frauds associated with many boom cycles. Question for you: if we've identified this pattern where there's a boom, there's an investment mania, the mania often pulls back, but it turns out the mania sort of created the schelling point where we pull forward the right brains at vastly different interlinked institutions to create the substrate that the next few decades end up being built on top of – are there any cases where the manias are just manias, we don't actually end up with a societal value created by them? Byrne Yeah, definitely. One of the ways that we taxonomize this in the book is that there are some bubbles that are about sort of regression to the mean or like just pure extrapolation of history. There's excitement about getting more of what we already have and it working the same way, just slightly better; then there are bubbles that are more driven by this completely different vision of the future where you just can't even speculate about what the long-term impacts of this stuff will be. If you compare dot-com to the housing bubble, the housing bubble was not spurred by some radical innovation in building better houses. It was spurred by the fact that we had a lot more data on housing price performance and that allowed us to underwrite more mortgages, but also to package those mortgages in a way where if you knew the statistical characteristics of a portfolio of them, you could basically turn that into its own asset: if you have a bunch of low-quality mortgages, you know that any individual mortgage has an unpredictable default rate, but you know roughly what the average default rate is – so what you can do is package a bunch of them together and say, “Most of these are not especially credit-worthy, a lot of the borrowers are subprime borrowers, but just statistically, most of these mortgages will actually get paid off. People don't actually like to default on their houses and they'll usually work pretty hard to make sure that doesn't happen. So why don't we say that the first X percent of losses hits this category of investors, and then the next chunk of losses hits this category and so on?” You can actually take a cluster of low-quality assets and turn it into one really large high-quality tranche where it's very, very safe because the world basically has to end for this to default, and then other tranches that are lower and lower quality until you get to one that is actually riskier than the average loan in the portfolio. That works really, really well if you know that the new mortgages that are going into [those tranches] are being sampled from the same distribution as the data you use to price it. But once the existence of those structured products is a driver of the availability of mortgages, now you're sampling from a new distribution: you're sampling from the distribution of people who would not have borrowed in the previous system [because] no one would have lent them money. (Now that that borrowing is going straight into a CDO, it's just the underwriter pays a little more attention to loan volume, a little bit less attention to loan quality.) So not only were the loans worse than they looked on paper, but they were actually more correlated than they seem to be. If you have an increase in housing construction that is partly fueled by this, you'll have this really crazy dynamic where the housing construction is actually enough of an economic stimulus that it makes people seem more creditworthy. Your marginal borrower is just way more likely to either work in construction or work in mortgage finance or work in something that is adjacent to those fields – and meanwhile, the more widespread availability of capital for buying houses meant that housing prices went up, which meant that people could [get] cash out of these houses. And again, we had lots of pretty good data on how housing prices perform, and so it becomes a much more straightforward matter to say that we're looking at, you know, this couple, they own their house for 20 years and they owe X percent of the value of that house and they want to borrow a little more money, maybe they want to do a renovation, maybe they want to take a vacation, maybe they just want to cash out a little bit and put money into their 401k or into something else, we don't really know. But people were withdrawing substantial amounts of money. And again, that money is flowing into the regular economy. It's causing spending. It is creating jobs. And so for a while, things do actually look pretty good. In fact, things look better than we modeled because there is more money flowing into the housing sector and the economy is doing okay under those circumstances. And then, yeah, that does turn out to be not something you can extrapolate indefinitely. You are sampling from a different distribution from the one that you were sampling from before. The actual performance of a lot of the highly-rated tranches of subprime-backed CDOs, was actually… it was okay. It was pretty much what investors expected. The problem was, because the correlation of defaults was actually higher than people expected within each of these vehicles, it was either the case that your AAA tranche is even safer than you thought because almost nothing defaults or no, your AAA tranches are not actually AAA at all. The thing that turned that from another episode of “some banks made some bad loans and they lost money and that's why banks have to have some equity capital to back up all the loans they're making,” the thing that made it really bad was you had a lot of very levered investors who were borrowing very short term and buying only the safest slice. If you have these slices where they might be worth 100 cents on the dollar, they might be worth 95 cents on the dollar, and you are the lender who's making the continuous decision – do I underwrite a loan where someone puts up $3 in collateral to buy $100 worth of these assets? – if you think there's any remote chance that these assets are actually worth 95 cents on the dollar rather than 100 cents on the dollar, then you say, “actually, we're not renewing that loan and you should find someone else to lend.” And of course, if enough people are doing that, suddenly there's very little money available to borrow. That meant that those assets dropped massively in price, even though the actual credit quality was not that seriously impaired. That's what turned it into a globe-spanning financial crisis where, like, very, very sophisticated people who have worked in the financial sector for a long time were going from ATM to ATM in midtown trying to get enough cash just in case the banks did not work the next Monday morning. Patrick Yeah, the seize-up of the repo market in particular is one of those things which is relatively appreciated by specialists and relatively not appreciated by people like congressional committees, etc. who devoted a lot of effort into looking at the kind of object-level factor of the housing market. So there is a literature on financial manias and I'm glad your book is a contribution to it. One of the things that we see over and over again in the experience of them is that they were reflexive: in the case where cohort A is not exactly cohort B, and one of the ways that cohort B is different is that, [people say] “providing cohort A with services was just such a great business that we have constructed an extremely efficient machine to manufacture much more of the next version of the cohort,” and then because the economy did not fundamentally change over the intervening years, that required us to put through people that had fundamentally different aspects of credit quality into cohort B, and we did it because we sent people out into the universe with the mandate to do it. Then another bit is that as house prices kept going up and up, even if you had… they're often called “predatory” terms in your adjustable rate mortgages, the typical one is like, “Okay, you'll be on a fixed rate for two years and then after two years your rate will mechanically increase to a floating rate, a much higher floating rate than the fixed rate is.” That will mechanically induce a payment shock. What our nation believed for a period of years in the middle was that, “Yes, there will be a payment shock; yes, many borrowers who have their mortgage payment increased by a factor of like 3X would be in a really rough way as a result of that. However, because the discipline of writing mortgages is an extremely lucrative one to be in, there will be someone in two years who will write a new teaser mortgage to refinance into.” And that worked like three times in a row, but it won't work the fourth – and then when it didn't work the fourth time, then, you know, economic gravity reasserted itself. So one of those ways [we get] reflexivity [is] through this mechanical method of, “We built a machine where we think we'll have similar results to the ad hoc bit that came before,” and it ends up not having it at scale; then another is the financial engineering aspect where it turned out we relied on something that was not as much a law of the nature as we expected it to be. Byrne Yeah, I think that reflexivity thing does tie into the idea that bubbles, and you alluded to this bit, that bubbles parallelize this process that could happen serially, but often it just wouldn't happen if everything had to happen one thing after the other; it works if everything happens all at once. You can see that in more recent things. Before, the knock against Y Combinator was they only fund AI companies. A couple of years earlier, the knock was they only fund SaaS companies that sell to other Y Combinator companies. But every time you create a new SaaS provider that solves some problem that lots of other startup companies have, you have increased the pace at which these other companies can deploy. So if you were starting a SaaS company early enough in history, you had to have some kind of internal communication, so you implement something that is 5 % as good as Slack, and then you have to have some kind of way to keep track of your customers. So maybe you roll out something that is a tiny subset of what Salesforce can do, and maybe you want to text your customers, so you build a very, very unreliable and expensive subset of Twilio, et cetera; every time one of those problems becomes a thing that you can just buy, the remaining companies become able to execute faster. So it actually creates this ecosystem where, yes, it's more competitive as a space overall, but it's also much, much faster to build your little monopoly and know that all the little components that have to get built for that monopoly to function have been built by somebody else. You can buy them off the shelf. Then as you scale, you eventually figure out, “which things should we actually do internally?” – and I think for a lot of them, the answer is none of them. Like you probably don't need to build your own special version of GitHub or any versioning system and all the other stuff that it does; you should probably just use the one that already exists. It also creates another kind of reflexivity, which is that more people either have skills that are relevant to the companies that are growing fast, or you don't need people with those skills because everyone who has to know every little detail about how to send lots of text messages all at once or how to do a lot of automated phone calls all at once, they're all there at Twilio and they have solved all the problems that you could imagine – and a bunch that you haven't imagined yet, and similarly in Stripe, etc. Patrick There's also an interesting thing where the people spin out of the companies that have the best internal tooling and then look around at their new job and they're like, “Wait, you haven't all figured this out already? I should immediately re-implement the thing that Google had used.” Some of them make 10% of what Google has built at their new firm, and then some of them are like, “No, really, I should just sell like 10% of what Google has built to every other firm in capitalism for $50,000 a year,” and that ends up being a pretty good way to do business. As a minor example, and not to toot the horn of horn of my ex-employer too much, I used to work at Stripe – which is publishing the book (disclosure); my version of the disclosure is that I’m still an advisor there, but that they don't necessarily endorse the following – way back in the day charging credit cards on the internet was really rough, and so in those like 1990s that we were talking about it, it frequently cost $250,000 for your Oracle license and hardware, and then another $250,000 for the engineering effort to charge your first credit card, and when you tell people that, they literally don't believe you. I was there! I was there, on the tail end of my undergraduate career, in an office when someone was discussing the number for what his company had just spent on this. Then a bunch of years later, you type a few things into a web form and it's zero dollars upfront. So one way that we get more wealthy as a society is that we end up with infrastructure layers that are just radically better than the ones that we had a few years ago; some of that is downstream of the overinvestment thesis, and some of them it’s just a regular investment in technological improvement in the economy. Byrne Yeah, and there's also a change just in the attitude. Over time, there ends up being this convergence between how people should think in the industry and how vendors structure their offering. I think one of the things that was striking to me about what you were mentioning earlier was this idea that you pay a fixed cost: the implication of paying a fixed cost is, “Oracle is not actually all that sure that taking credit card transactions online is ever going to lead to anything – like if you want to do that, that is your right, but Oracle wants you to pay upfront.” Now the Stripe attitude is, “No, of course you should do this, and of course, it should be as simple as possible, and we shouldn't have any speed bumps in front of that.” I'm sure there are upfront costs for Stripe onboarding a new user, and so it's entirely possible that some large fraction of them are like, most of them are just a slight cost to the company and never that cost never gets recouped. But if the Stripe bet is that a handful of these companies will do extremely well, and when they are growing at a breakneck pace, the last thing they want to do is a migration of their payment system – there's enough breaking every day when you're growing 10% a month, you don't need extra headaches. By the time they're at really large scale, there’s been two-way feedback for that entire time, so as long as your product is getting enterprisey at the pace at which your biggest customer is getting bigger, you can kind of always maintain that and get dragged up to providing an enterprise solution, because you provided a startup-friendly solution to enough startups that one of them is now enterprise. Patrick So without talking too much about that one specifically… in financial services in software and similar, there have broadly speaking historically been like two-ish models. So the classic Joel Spolsky essay is Camels and Rubber Duckies, where you are either selling low-cost software or you are selling enterprise software and between them is the Valley of Death. In financial services in enterprise software and similar, there was for a very long time a hard distinction between businesses which are running a star search, where almost all the money will get delivered by a very small number of accounts and it is very difficult to identify those accounts ex ante, and businesses where you are providing maybe not a utility, but a service which will be consumed by a very large number of businesses in a very similar fashion where the growth of those businesses is perhaps not uninteresting but it's unlikely to be heavily varied. Then there have been a few software companies over the years – Stripe is one, AWS is one other example – where you actually have both businesses being run at the same time: some of your customers will have extraordinarily high growth and they will encounter extraordinary needs within a few years of you onboarding them as customer, and your business will be dragged in the direction of being able to support extraordinary needs. AWS is dragged in the direction of, like, we may have non-human intelligences within a very short number of years from now, and it turns out that they need you know, like a state’s worth of power generation – “can you figure this out for me?” And so AWS and Azure and etc. are getting pulled in that direction. But also, the local flower shop needs a website, and that website either runs on AWS directly (perhaps less likely), or it runs on a platform that runs on AWS and has 40,000 florists that all have a website in, I don't know, AWS Northeast One or whatever the data center is. So there is this interesting business challenge of, “What do you do if you are one of the relatively few companies that is both running a star search and also providing these sort of, broad, relatively flat services to large portions of the economy?” But I feel like I want to get back to talking your book. Shall we talk about crypto a little bit? This is one where I think we have productive disagreement. As folks who have followed me around the internet for a while [know], I was an early adopter of being a crypto hater. I was a crypto hater way before that was cool! You are something of a fan, and in the book talk about crypto as being something of a rallying cry for, let's say structural change within the financial industry. Do you want to spool that thesis out a little bit? Byrne Yeah, absolutely. One of the reasons that we talk about crypto in the book is that, in some of the bubbles that we talk about, there is an underlying technological change, and the reason people are optimistic is that they've extrapolated that change. With crypto, Bitcoin did some, a handful of novel, interesting things technologically. Like, the Byzantine generals problem: I think almost every mention of it at this point is in reference to Bitcoin, but it was an actual problem of, “How do you coordinate if you have a network where there could be untrusted nodes?” It is kind of clever to have some sort of collaborative mechanism where everyone has an incentive to cooperate, everyone has an incentive to choose the right set of trusted transactions and build off of that. That was pretty clever! But a lot of what makes crypto interesting from the bubble standpoint is that it is closer to a pure bubble. Bitcoin does not have a P/E ratio. It does sort of have a yield now, depending, but it doesn't have a yield in the sense that a lot of other financial assets do. Ironically, the crypto asset that was most amenable to just standard financial analysis was actually exchange tokens like FTT. FTT was actually a pretty clever structure. (It is always frustrating to see people who were very, very smart in one domain and just dumb enough in another domain that it did not actually work for them, and in fact, it would have been much better for them to have been less clever at the things they were clever at.) But FTT had some really interesting mechanisms. This is a token issued by the FTX exchange; they issued that token in order to get funding for the exchange, and I think later on just to have collateral for financial engineering purposes. But if you bought a bunch of it and kept it on the exchange, they would actually give you a break on transaction fees, and then they would take some fraction of the commissions that the exchange earned and use them to automatically buy a chunk of FTT and destroy it. So it is basically equity with some perks – and there are equities that have perks. AMC has some kind of stockholder benefits program where you get free popcorn or something. I think there's a pink sheets-traded winery that used to pay a dividend in the form of wine. Aren't there companies in Japan that have some kind of shareholder gift thing? Patrick This continues to be quite common in Japan. It used to be common in the United States too: you used to get a coupon from McDonald's for free French fries if you were a shareholder, [but that practice has] largely gone by the wayside here. But yeah, there are hotels in Japan where [they’ll say], “congratulations on being a hotel shareholder, here is a discount on your next one night a year.” Byrne So it's close enough to equity with some fun sprinkled on top. But because of that, you could actually model it as a financial asset. You could say, “here's the cash flow this asset is entitled to, and then the way that cash flows return to shareholders is through a buyback,” and so you could actually do fundamental analysis – it made sense to have a model for not FTX the entire business, but just FTX trading revenues. But then a lot of other crypto assets, the market cap does come from some kind of hype. When you think about it, there's a sense in which that is just the definition of a currency is that it's some asset whose value is entirely a function of what other people will pay for it, and with crypto that's very salient because the people paying for it are crypto speculators. With something like the US dollar, you do actually have an irrational buyer who buys a whole lot of US dollars, the IRS. They will give you this wonderful benefit of, “you don't go to jail,” but you do have to give them this otherwise worthless set of tokens. So they have a sort of buy-and-burn mechanic themselves, and that is what supports the value of a currency – but it's not the only thing that can support it. There's this fun situation in Iraq – I forget if this was actually during when Saddam was in charge or well after, but there was a point at which Iraq had their currency, it went through a bout of hyperinflation, they created a new version of the currency – and then the old version of the currency now had a fixed supply because no one was printing it, so that became the actual store of value. The new one was more like, you transact in that currency day to day, it is more liquid, but your savings, you switch to the old currency for that. So crypto is kind of like that, where if you know that there's a fixed supply and you know that there are people willing to buy it, then you can start to think of it as something sort of in the direction of a store of value, and if you think of it that way… a lot of financial systems, they do start with some kind of store of value that is just seen by the rest of the world as objectively valuable, whether it's gold or silver or something like that – some way to settle global trade – and then internally, there's often some token issue that's backed by that. Typically the original backing is very, very straightforward – it's something like, you went to the goldsmiths of London, you gave them a certain amount of gold, they give you a piece of paper saying, “this goldsmith owes you this amount of gold” and that piece of paper is just much more convenient to travel around with, and so that's your more liquid form of wealth. Patrick So I've got two quick reflections on things you said. One is on exchange tokens specifically: I have this theory (based on the reflexivity conversation) that exchange tokens are one of those things where there is a financial model that you can run on them in the early days, and there is some read of an explanation you can say for why they create value in the world: like, okay, it's an equity issuance, which for crypto reasons we don't call equity – but between friends, it's definitely performing the function of equity, and we issue it because, like other businesses, issuing equity is the way that one funds something before one has a business that has cash flows associated with it. Then at some point, there become margin rules associated with any individual tracker of value in the world, whether that is a stock or a token or similar; it turns out that exchange tokens will tend to flow into the place where they create the most margin capability, which is almost always the native exchange. SBF et al. end up holding most of the FTT because they are giving themselves extremely generous margin characteristics with regards to that FTT – you know, infinity-percent margin, it's a great thing to have while you can afford it! (And then they get wrecked by the infinity-percent margin.) That's a story that we've seen a few times in crypto, both with regards to exchange tokens and with regards to things like, “Bitcoin is cool, but a Bitcoin which supports margin is even cooler.” And then that is part of the genesis of the grayscale trade where the Grayscale Investment Trust is like a wrapper around Bitcoin, but for a while it was a wrap around Bitcoin that you turn into margin capability, which caused a lot of people to one, overpay for that wrapper and again, like construct a giant machine which was siphoning Bitcoin out of the various places Bitcoin was to put it into the trust to create this like fantastically valuable marginable Bitcoin, and then it over-created that relative to the actual economic demand for that product and the premium inverted, and much wailing of gnashing of teeth happened. That is my brief thought on exchange tokens. [My second reflection is,] a thing I said early about crypto in 2010, 2011, 2012, somewhere in there, was that the Bitcoin community is this globally distributed emergent boiler room, which is talking itself up in various places. I think I was correct with that diagnosis in a lot of ways; I think I underestimated the impact of that diagnosis. One thing you go into in your book is that there has been a reckoning in the financial sector for a very long time about the relative position of technologists – whether they can get their asks approved, the physical and technological and organizational substrates for how money movement is conducted. For a very long time, people in positions of power were like, well, that's back office. None of the interesting stuff happens over there. Sure, I pay some geeks a bit of money and I even give them extraordinary career benefits, like not having to wear a suit like all the rest of us, but if you can keep those smelly people in the corner and allow me to get through the business of banking, I'll be pretty happy with that. And one of the things that the distributed boiler room managed to do was to say the word blockchain, blockchain, blockchain a few times – stealing this point from Matt Levine a little bit – and then get into the corridors of power and say, “You know what would be really cool? Having an honest conversation about what money movement should look like, and various ways that we could make that happen faster.” [Now,] I wouldn't give 100 percent of the credit for realignment to that particular mechanism and faction. I think another mechanism that’s underappreciated in the wider world is that there came a point sometime after 2008 where not just retail depositors, but the mass affluent, the town dentists, started making decisions on where they banked based on who had good mobile apps, and people in positions of authority said, “Man, the mobile app is no longer just a thing that diverts people from calling our call center and saves us a bit of money on call time, but it's actually strategic priority. What do we have to do to make mobile apps?” The tech people said, “Well, you'd have to give us money and the ability to ship things, and we know the banking industry doesn't do those.” Sometime after 2008, Chase and Amex and similar said, “Well, what if I do?” They poached – that’s sometimes the word, I hate that word – they went to the usual suspects in Silicon Valley and said, “Hey, we're recruiting mobile developers now, and you tell us what we need to do to win this.” Banks now have relatively functional mobile apps in 2024, which stack up reasonably decently among the experiences shipped by the likes of Google and similar, in part because they were written by people who very literally wrote the experiences of Google and similar. And so this is kind of like a one-two punch, which is dragging the financial industry, I won't say screaming into the embrace of technology – the financial industry was an early embracer of technology – but there were a few decades in the middle where it was like, “Well, we've got our mainframes and they're working pretty well; we don't see the need to go all in on the new thing.” Byrne Yeah – a lot of this stuff, it's more cyclical than it looks. Banks were really early adopters of computers for businesses; banks and airlines were the early ones. And then there were a lot of companies where IBM did payroll and inventory tracking and nothing else. Because of that early adoption, they have the classic early adopter’s curse, where you adopted version 0.0.1 of this, and then whoever looks at the end service that you provided and looks at the currently available technologies says “there's a much easier way to do this.” And then it turns out there's a lot of complexity on the backend. So in a way, one of the benefits of crypto is that because there is less legacy technology, the regulatory situation is different and evolving. I think early on there was this optimism in crypto spaces, that if you ctrl-F through various laws and look for cryptocurrency you don't find anything, and therefore we can do whatever we want… and a lot of people had a lot of time in government-subsidized housing to read up on what money laundering is, what it means and how the state defines it. Yeah, it turns out that the state does not actually reify crypto as this ‘new thing that is not money’; it is just another way to transfer wealth around the world. Patrick I will say as a skeptic who has pointed out that the crypto folks did a lot of lethal crime over the years: We do live in a democracy, the laws in society are subject to a democratic process. There were many, many criticisms made of the AML and KYC regime over the years, but relatively few concerted attempts to change the AML and KYC regime (or the securities regime, which is a different facet of crypto's regulatory woes). But a lot of the intellectual effort/lobbying/et cetera over the last couple of years with respect to like, “maybe we need a de minimis regime for KYC, or maybe currency transaction reports should be scaled back a little bit” hasn’t come out of kind of the classical places defined as pro-consumer financial advocacy, but from the crypto industry. [This is] partly because they have a business model [in] which they would appreciate if it was, you know, lighter-weight on their reporting obligations and similar. And while some parts of that are very self-interested and criming all the crimes, and some parts of it are like ideologically-weighted, they have a point of view with regards to financial freedom that I'm not entirely unsympathetic to, and they have in some cases forced a conversation that has been a very long time coming, because the the saliency of AML and KYC to the American political process has been extremely low for a very long time. The people who are impacted worst by it are almost definitionally at the sort of edges of the graph of the socio-economy of the United States, and so finding people who are in one fashion kind of at the edges of the socioeconomic graph, but [who] also have a lot of money and a lot of upper-class American ability to move levers in Washington when they want to – [that] has created more enthusiasm/intellectual effort going into [changing] where society decides to put the pivot point on that regulatory apparatus, than there was for a bunch of years. So I've said my one positive thing about crypto for the interview. Byrne Yeah. I would say that right now, that is one of the big real-world impacts of crypto: it has made a lot of very nerdy libertarians and a lot of very libertarian nerds extremely rich, and also kind of radicalized their politics, in particular in favor of more financial access. The inevitable result of that is that yeah, more money moves in ways that we would prefer that it not [to] – like, more bad actors are able to interact with the financial system – and then the other piece of it is like more normal, decent people are able to [interact with it as well]. One of the interesting crypto applications right now that I do hear more about is cross-border settlement outside of the US, where there are a lot of countries that do not have a lot of dollars but [in which] people do recognize the dollar as a store of value – but it's expensive for them to access the dollar financial system. It is not expensive for them to access say the tether financial system. So you do have some cases where, while stablecoins are strictly worse than the US dollar for an American – they're like a dollar that doesn't pay any interest or has other complications, or like a dollar with weird user interface – if you don't actually have access to digitized dollars but you do want to move money around in the world and you have perfectly legitimate reasons for wanting to do so, it's often your choice is either, (a) interact with a legacy banking system that is very much designed around the needs and interests of economic and political elites in your country, or (b) pretty much go direct. The other piece is just the composability of different crypto systems. Like, smart contracts: they're very funny to me. They are a really interesting concept. I was reading about them before crypto was really big. You were an early crypto skeptic – I did find that there's a Nick Szabo blog post from 2008 talking about Bitgold where I did leave a comment saying, “This sounds like we all just use all of our computers to mine this stuff all the time, that's very wasteful and environmentally unfriendly” and all that stuff – that was 2008. So I have also been early to some of these crypto criticisms, and I'm kind of over some of them. But the smart contracts, like, having a way that you can standardize a set of financial outcomes such that anyone who's involved in this transaction can easily see what will happen, they can also run a pretty straightforward Monte Carlo simulation of what will happen, so each side knows roughly what they're getting into: it allows you to create these arbitrary contracts where you can just say, if X happens, then transfer Y between these two parties. It turns out that the one of the big obstacles to that is just the “if X happens.” The world does not supply a lot of API endpoints that are booleans for arbitrary questions about what events happen, so you have to have an oracle. There are some oracles that you can use where you can be pretty certain – for example, that the weather data is going to keep getting updated, and so if you want a smart contract for, you know, buying insurance against not having much snow for your ski resort or whatever, you could theoretically do that – but the place that just abounds in oracles is the financial system Patrick I think it's worth saying that not merely can you theoretically buy weather insurance contracts because of the existence of the NOAA and available data there; in fact, we did bring that insurance product to market. The dominant way to buy weather insurance is not – well, maybe not the dominant way. There is a huge market for crop insurance which doesn't depend on going out to a field in Kansas and seeing whether there was hail damage on those individual stocks, it just says, “pays out if the amount of precipitation in this zip code is above X in a data source that we mutually think is acceptable.” (Sorry, digression over.) Byrne Yeah, I think parametric insurance for everything is a really exciting use case. Crypto does have the problem that actually a lot of emerging markets do, where as it grows, the financial speculative part of it grows faster, and that is just where more of the wealth accrues. And this happens to a lot of countries as they get rich, if they don't have really strict currency controls (and sometimes even if they do) – over time, more and more of their wealth is just from money pouring in and there being transaction costs and fees and things associated with that, and when there's more liquidity, asset prices go up. So just being a levered buyer of whatever the relevant asset is tends to pay better than doing things that actually create wealth. Then what happens after a while is, that just crowds out all of the legitimate stuff. Like I suspect that, in 2021, it was probably a lot harder to raise money for a crypto remittance startup vs. raising money for a crypto gambling kind of product, but then that kind of thing runs through a cycle and eventually the market gets really overheated; people realize that, not only are the fundamentals bad, but there basically aren't any fundamentals other than just, “money is flowing in, people like gambling on this.” Eventually, yeah, something breaks, and then at that point we do start to see, okay, how many frauds are there? How big are they? How egregious are they? How easy would they have been to spot if anyone had actually been paying attention? But then, we do end up with more infrastructure. There are more people who have either a crypto wallet they actually have custody of, or at least have access to crypto, so a more real-world facing use case is actually more practical in the wake of that bubble. Then I think the crypto-skeptic view that I do agree with is, “Okay, well, show me.” I remain cautiously optimistic about that. But I think there's also a dynamic where the fiat system does actually improve over time, and certainly the user-facing experience of that fiat system does improve. There are a lot of things like, “okay, I want to transfer money overseas because I want to pay my supplier, I'd like this to settle instantly because my supplier (like me) is a small business that's cash constrained, but I just took delivery of this product and can sell it right away, so I do have the cash to do this,” et cetera. We can get all that stuff to move faster – even if the actual flow of money is still slow and there are still fees associated with it, the user experience of that flow of money can be fast, like regardless of what the underlying rails are, as long as there's someone who is willing to take some credit risk. You as a financial services company can provide instantaneous transfers to anyone who has an account with you just by updating the number in their account, but now you're taking the risk that the money that you told them they had is not money that you have, and if you are regulated by any sensible regulator, they're just going to say, “you eat that loss.” Having a bearer instrument does mean that you can just underwrite more of those transactions, and that is especially valuable in cases where the KYC/AML regime is kind of weak, where contract enforcement is kind of weak, but you at least know – if the address you control got this amount of this particular stablecoin from this other address that you know is associated with this person – you know they have actually paid you. It does provide that layer of certainty in places where there's less trust. So there is almost this race between the fiat system to create just a more trustworthy world, and then the crypto system to create a more user-friendly world for things that are trust-agnostic, for better or for worse. And that's an exciting situation. I like the fiat system a lot, and the vast majority of my financial interactions are with fiat and not with crypto; I do think that even if you are a fiat bull, if you are fiat-pilled, that to your point earlier, crypto has spurred people to think a lot more about this plumbing, how it works, why it works, how it could be improved – I think that this is actually a case where, like, if the end result of crypto is that there are more CEOs of banks who know what a database is and they know what could go wrong if you store the amount of money in someone's account as a float rather than as something more defined precision, etc., that is actually a better world. Certainly a world where everyone has mobile apps that allow them to transact with anyone they're legally allowed to transact with, have that transaction settle as instantaneously as it needs to according to the perception of each person involved. That is a really good world. If what undergirds that world is every country having their strategic Bitcoin reserve and every country constantly minting new pseudo-currencies for different purposes and creating all kinds of smart contracts, that's great; if it is a world where it's actually dollars, not Bitcoin underlying everything, but people still have access to those dollars, not Bitcoin, and that it functions a bit better because there is that competitive impetus, then that's also a win. Patrick Yeah. Central bank digital currencies were all the rage for a while and there was a question of, “will each country have its own blockchain?” Of course, the consulting companies are willing to say, “Yes, absolutely, we will create all the blockchains for you.” [But] it turns out that when you dug into it, what a number of economies really wanted was, “No, we just want a universally deployed domestic payment rail that is as good as can be possibly made de novo in 202X – what can you deliver?” In Brazil that's Pix; in India, it’s UPI. There are many product dimensions to a payment rail, but those are pretty good payment rails and are extremely widely adopted in both nations – and as far as I know don't have a blockchain within spitting distance of them. But I will [give crypto] credit, there might be something there: if we successfully draw the right set of brains and the right distributed set of actors into enthusiasm about fundamental infrastructure with the story, and then the infrastructure that results is not exactly the one that the story was told about – is that a net positive for society? I can credit that a little bit. To use that thought for an investment thesis we are seeing spring up before our very eyes – and I'm being careful in wording this because it is very not obvious to me that LLMs/AI, etc. are a bubble or a mania in the way that we usually think about bubbles and a mania; some of the smartest people we know think that this is the most important technology ever and/or the last technology ever. (I don't think I endorse either of those things, I think it's only as important as the internet is – as someone who’s said that the internet is the human race’s crowning achievement to date.) [But] it turns out there is an entire supply chain implicated by AI, regardless of whether we reach some future where either all value in the lightcone is expunged by our new robot overlords or we create the most valuable technology ever. The supply chain includes, among other things, a tremendous amount of electricity being required to run data centers, which is bringing a lot of enthusiasm in our parts of the world from people who are pretty smart and accomplished to this question of, “If power wasn't simply utilities that you never thought about, or something that you told the specialized engineering team, ‘yeah figure out how many megawatts we need for the new data center,’ but was actually something you were extremely interested in on a day-to-day basis, what would you do?” The answers to that bouncing around the industry right now are many and varied but quite interesting. There's a continued walk down the almost-miraculous solar cost curve which itself a little bit lives in the shadow of a solar bubble, and maybe an “irrational” – it was certainly at the time described as “irrational” – amount of investment by China to solar manufacturing, and them being non-salubrious in “dumping” that low cost solar on the United States; now we have these machines that a previous guest, Casey Handmer, described as just spitting out wealth that you can plop down anywhere where the local environment supports that. Another kind of post-investment bubble thing that is getting pulled forward by the AI boom is, there's a great amount of interest in geothermal, next generation geothermal specifically, at the moment. Next generation geothermal turns out to be on the tech tree behind fracking, because the technologies that enabled fracking also enable you to drill 10 kilometers down and then take a 45 degree bend and attempt to extract heat from the rocks where the heat actually exists, which is basically everywhere if you go 10 kilometers down. That's an irrational amount of depth to ever dig to before we were digging to it routinely for fracking purposes for discovery of shale. Then the things they do after they dig to that depth rhyme very heavily with fracking and subsurface engineering and similar. So it's kind of like the classic developmental economics flying geese curve, where these these booms give us those unlocks on the tech tree which future booms end up building off on – and again not doomcasting or endorsing the perspective that the AI boom will live up to the 90th percentile claims being made about it; in either event, it is very likely to lead to a great number of smart people spending a number of years of their life thinking seriously around what the supply chain for that looks like, and regardless of what happens with respect to near-human cognition being available from an Nvidia chip in 20 years, abundant amounts of free green energy seems like something that we will not regret having. So that'll be nice. Byrne Right. Yeah, the fracking boom was a really interesting one to write about, because it is just such an extraordinary technological achievement, that we drill [so] incredibly deep below the Earth's surface. We used to drill for oil on easy mode, and this was really fun history to read: you go to places where either there are oil leakages on the ground, or you get to do the Indiana Jones thing of just wandering around the Middle East and looking for fire cults. Wherever there is a fire cult, there's probably a temple where there has been a flame burning for all of human history, and it's probably because there's natural gas seeping out. So you drill near there and you find oil and you find your fortune. That’s not trivial, and certainly not trivial with late 19th to early 20th century technology – but fracking is unbelievably cool because what you do with fracking is you drill down deep and then you get to rocks where there is oil in the rocks, but it's trapped between the rocks. There's oil and gas in there, but it's not actually going to just flow up in a friendly way. So what you do is you blast this slurry down there and you crack open the rocks, a bunch of fluid flows through them and then as the rocks settle again, the cracks are held open by tiny bits of sand. Just the fact that someone is manipulating a sand-width size crack, a mile below the surface, miles below the surface, in order to get the gas to flow through molecule-by-molecule is just really cool to me. Then, once you've done that in one place, you move the pipe along, turn it a little bit and continue the procedure. And it has just all kinds of effects on how people think about the energy supply, because now you actually have a lot more control over how much energy gets produced and at what pace it does. Fracking has this notoriously much steeper depletion curve, but that also means that the overall returns are just a lot more predictable: if you are fracking, you aren't spending money now in order to get oil and gas over the next 20 years. It's like, you're spending money now, you get oil and gas pretty much right now over the next few years. From a sort of accounting, economic standpoint, one of the things it does is it just totally changes the supply elasticity: it means that oil and gas prices do not have to be as volatile, particularly within the US where, as we are aware, this fracking mostly happens. That has a lot of geopolitical implications. Often extreme swings in oil prices in particular cause trouble for the world, and so if those swings are less extreme, then that's an improvement. And yeah, it gave us a lot more of an understanding of what's going on in the earth's crust. To your point, we realized there is this other energy source, which you don't have to set on fire, you don't have to emit a lot of CO2 and you also don't have to keep finding more of it. Like, it's just there. So yeah, geothermal is a really interesting category, and it is a category that is particularly interesting because we have made this enormous investment in both the capital equipment and the skills that are necessary to do most of the steps of this. I think it is interesting to imagine a world where the big fracking companies of today are still big energy companies, but they have negative net emissions because they use a little bit of their power for CO2 capture or something like that, direct capture or other things like that. Patrick Yeah. My finger to the wind is that, if we ever get to the happy world where geothermal is tens of percent of the US energy mix, many of the companies that benefit from that will be the current oil services majors – Halliburton, et cetera – because they have the machines and have the know-how for digging the holes. In a previous conversation with my buddy Austin on the podcast about fracking specifically, one the things I loved the most was, seen from the macroeconomic perspective, the cost curve and the learning curve associated with every new technology just looks like this magical up and to the right kind of thing. That's the sum over very many hours and years of labor by very smart people – and in the fracking case, what Austin has called redneck engineering: there was no tablet that someone could discover on the side of a mountain that said, “PS, the right diameter of sand is X in this region,” it was just people trying a lot of mixtures of the various fracking fluids and like, “Okay, we're going to do it in multiple stages, we’ll measure the output and see how it goes,” etc., learning progressively more and more and more about the shape of reality with respect to what sands work. Also incredible is that you can do a certain amount of it in like an actual laboratory, but given that the laboratory is highly pressurized 10 kilometers under the surface of the earth and deeply inhospitable to people (and also deeply inhospitable to cameras) getting information from it is somewhat difficult; the physical model of the world is one in which we are making best guesses as to what the fluid is doing when it goes down the hole. I will retell an anecdote because I love it so much: it's so hard to get a wire down the hole in a fashion that the wire survives, that rather than using wires to communicate with the drill bit that is 10 kilometers down, they use modems. The modem doesn't make acoustic sound over the wire or an electrical sound that approximates an acoustic sound at any point – no, they send an acoustic signal vibrating through the mud in the hole and then listen for that acoustic signal on the other end, and use a decoupler to turn that back into a waveform to get a very small number of bits per second of signal from the machine that is deep down the hole and which they are attempting not to break. A bit of wild, mad science involved there. So this is one of the other boom things: we discover wild, mad science to make some of the curves work. Byrne Yeah, it is crazy to think of the bandwidth of mud being one of the constraints on human progress. But it’s true. That is part of what bubbles are doing: people start out really, really optimistic and they do run into problems, they do run into cases where the math just does not pencil out the way they imagine it would. But that means that we are always engaged in this long exploration process of finding out what the fundamental constraints really are, and sometimes a breakthrough in one area means that we've revised our list of fundamental constraints somewhere else. With computing, there have been a couple different cases of scaling breakthroughs. One of them was the discovery of the vacuum tube, where you actually have a device that can do fairly simple logical operations such that you can implement it in a machine. Then we ran into this problem of, the vacuum tubes are mechanical, they do break, and so the bigger your machine, the more likely it is that it breaks; the more complicated your algorithm is, the more likely it is that something breaks down. So you have one of those dynamics where you're scaling your inputs a lot faster than you're scaling your outputs and you're doing things less and less efficiently over time. Then transistors do not actually have moving parts, so they don't have that particular problem – but they run into their own scaling obstacle. It's really fun to read about the early days of this: one of the books that I cite in Boom has an excerpt from, not Scientific American but a magazine of that type in the 50s, where it's speculating that perhaps in the future computers could be the size of a small house, and that's how much we could shrink them. But people ran into this problem with transistors, where the more of them that you connect – and you need all of them to be connected and working for that particular cluster of them to do anything useful – the more of them you connect, the more likely it is that you have one little issue somewhere that makes the whole thing not work. Then it turned out that there was a way around that too, which is that you don't actually plug together individual discrete devices, you actually etch the entire set of connections chemically, and now with many other things – but yeah, you etch it, a one shot [process] where you create one solid thing. That turned out to be a much more scalable architecture. In the intersection of the hardware and software industry, we just continuously run into [patterns like this]. A lot of things are defined by finding some process that works, scaling it up 10x and then it breaking in ways that you did not realize things could break. So we are always learning just what the limitations are on storing and moving and processing data, and there's always a new set of them at each level of scaling – but we are always learning just what those fundamental limits are, and the more advanced these things get, the more we do actually get to, like, “this is what is physically possible in this reality,” and we get pretty close to just being able to demonstrate that. Patrick Yeah, I've [heard it said] – and [keep in mind that I am] firmly from the software side of the house, not the hardware side of the house – that there might not be many more nanometers to drop, simply because we're getting close to, like, physics rearing its ugly head and saying, “heat dissipation just isn't going to get any better for you.” But on the other hand, a reason for renewed optimism about things is that there have been many times in human history where people have been quite convinced – and had good reasons for being convinced – on, like, “Hey, we're running up against physical limits, like peak oil, that's all we're going to do about it,” etc., and then human ingenuity and economics get unleashed in ways that sounded very improbable at the time. So I guess that is as good a place as any to cut this off, although I am certain we could continue conversing for many, many hours. Byrne, what's the name of the book again? Byrne It is Boom, and the subtitle is Bubbles and the end of Stagnation. You can check out stripe.press/boom for a really nice looking, very slick site telling you a bit more about it. Patrick As long as I have you on, Tammy who runs Stripe Press is a buddy of mine from my Stripe time. What was the experience of being an author like for you? Byrne It was excellent. They were incredibly helpful. They were very patient, and there were many times where they were waiting on me and doing a really nice job about it. It was a really, really incredible experience. Also, they really like the book, and like talking about it – it was nice to really feel like we're aligned, we're a team, we're actually working on this together. It's a shared goal, a shared objective. Yes, highly recommend working with Stripe Press. They are wonderful people, and I think I've read a bunch of the Stripe Press books – I've never gone wrong picking up a new Stripe Press book. I like them a lot. Patrick So, there's our mutual testimony for working with Tammy and the team at Stripe Press. Alright, Byrne, thanks very much for coming on today. For those of you who don't read The Diff already, I'll drop a link to The Diff in the show notes too, for when you want to get Byrne in less-than-book-length doses. But he is one of my favorite writers and thinkers on these subjects and I wish more people would discover him. And for the rest of you, I will see you next week. Byrne Thank you so much.