The One-Big-Side-Bet Model of Wealth Creation

In this issue:

• The One-Big-Side-Bet Model of Wealth Creation—A common pattern in economic history is that above-average growth is driven by some specific technology, but the concentrated upside comes less from direct bets on it than from indirect bets on its impact.
• Building Japan's Financial Ecosystem—Japan has a surplus of tiny publicly-traded growth companies as a result of a shortage of VCs. But that shortage is slowly being addressed.
• The New Balance of Trade—China's auto exports get a temporary boost from the uneven global rollout of EVs.
• Where Return on Investment Goes—An asset-light business is often adjacent to someone else's assets, and when growth slows down those sorts of assets move onto the growth company's balance sheet.
• Broken ETFs—Leverage isn't infinite.
• Swipe Fees and Accounting—In a sufficiently complex economy, nobody knows whether they're getting a good deal.

The One-Big-Side-Bet Model of Wealth Creation

A pattern that shows up at the intersection of broad economic growth and individual wealth accumulation looks like this: find an important, generational trend. And then don't bet directly on the trend, but on the immediate consequences. Consider:

• Railroads were a major contributor to global economic growth in the 19th century, especially in a relatively low-density country like the US (as of 1875, population density was 14x higher in France and 27x higher in the UK).^[1] Railroads were so capital-intensive that a buy-and-hold investor had to start with a substantial stake in order to make money from them, but their stocks were liquid and volatile enough, and securities laws were loose enough, that there was a lot of money in turning control of a railroad into trading opportunities. Meanwhile, even bigger fortunes were made from supplying those railroads with steel (Carnegie) or using them to transport oil (Rockefeller).
• Electrification briefly made some direct fortunes, but most of those were done in by leverage in the early 30s and by regulation a bit later on. It's hard to disentangle this from the impact of oil—the deployment of one energy source that's cheap and reliable and another that's cheap and easy to move was a broad dividend for the economy, and many of the big fortunes of the 50s were the result of money that was made earlier but that compounded because of economic growth (or, in the case of the DuPonts, compound interest further compounded by a decline in the real cost of their industry's primary feedstock). There were plenty of rich people in oil, but it was also a lumpy, hit-or-miss proposition; The Big Rich does a good job of illustrating that some oil fortunes were from a series of canny positive-edge bets, and others were from putting down more chips than anyone else was willing to.
• The PC revolution was a case where participants were quite aware that they were riding a wave. Bill Gates cited the declining cost of hardware, and the attendant rise in the importance of software, as a persistent tailwind,but also a constant risk—every hardware refresh cycle meant a chance for competing software businesses to get ahead.
• There was a fairly smooth transition from a world where cheaper processing power and memory were major drivers of wealth creation to one where accumulating data was the winning move. Just as software founders in the 70s and 80s instinctively built for a future generation of hardware that was both better and more widely-distributed, early online businesses were full of people who insisted on logging every possible user interaction, and casting a wide net for picking up other datapoints that could help them target ads. One of the qualitative differences here was that better data didn't just improve products, but improved companies' strategies, too. The nature of their business meant that competitive threats showed up in search volume and in social media posts, and, particularly in social, meant that companies could map the social graph even better (when social network X takes off because invitation links are being spammed to network Y, it's a threat to Y, but it also means that Y has richer data on which of its users are early adopters and which of those early adopters influence their friends).^[2] And it's worth noting that, as above, participants in the labor- and capital-intensive part of this strategy mostly didn't make as much money as the entities directly adjacent to them; Google would have been a much worse business if Larry and Sergey had to make a webpage for every search result, or if Zuck had to send a photographer to every college party in order to get pictures. Fortunately, these resources were contributed for free.

What each of these large-scale booms has in common is that it starts with the capital-intensive growth of some universal complement—faster transportation, cheaper energy, quicker computation, more information—and then figuring out which adjacent business has durable competitive advantages. This can show up in surprising ways: one possibility is that the high historical returns from systematic value investing, and its subsequent collapse, reflect the possibility that non-qualitative value investing was really a bet that some day "find all the stocks with a return on equity above 10% over the last five years, but that are currently trading below book value," would go from a full-time job to a couple clicks.

Which raises the question: what are some of the complements to intelligence-on-demand, which is an assumption to pencil in even if models don't improve much. (In fact, a glut of GPUs coupled with a plateau in model performance means even cheaper inference than in a counterfactual where models improve.) There are some obvious answers, mostly starting with the consultancies that get described as body shops; the Accenture employee who delivers a presentation on outsourcing customer service operations probably stays human, but a lot more of those customer service operations will move from call centers to datacenters.

For now, it's much easier for an individual contributor to figure out which parts of their jobs to automate than for their boss's boss's boss to do this; the important information is granular details in workflows, which makes AI a fun economic paradox: at the level of providing models, it's highly centralized because of the fixed cost of hardware and talent, and because of the knowledge institutions accumulate after repeatedly training models. But at the end user level, it's radically decentralized; the component of a task that's most amenable to a drop-in replacement is whatever part is usually done with a few seconds to minutes of quiet thinking, so individual contributors have a temporary monopoly on the most critical information. For some functions, that doesn't keep them safe, because some jobs are just a way to convert complicated natural language queries into a more structured format, whether that's customer service or taking orders at a drive-through. But one tier up in complexity, there are roles where the mix of fairly-rote and high-touch is opaque, even to the people doing the job. (It was a surprise even to model developers when AI models turned out to be able to write working code. Even if they had a sense that some things they wrote were clever and some were boilerplate, the boundary between the two wasn't sharply delineated until people figured out how much of their job Copilot could do, pretty much instantly, once they wrote the name of a function.)

If those past trends are any guide, there's a long deployment period ahead of us. Even if AI gets deployed faster than the steam engine, there's a lot of experimentation to figure out the right applications—steam engines were used to pump water out of flooded mines in 1712, and used to pull trains along tracks in 1804. It took a bit more than a century after that to fully deploy railroads, and their economic impact in the form of cheaper transportation persists to this day (not to mention other things we largely owe to railroads: national consumer brands, accurate clocks, time zones, org charts, early communications infrastructure, fresh beef, etc.). If you're looking at the proximate impact of a trend, it's possible to be too late. But if you're betting on second-order effects, it's very early.

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• A company building the new pension of the 21st century and enabling universal basic capital is looking for a general counsel to help lead capital markets and regulatory efforts. (NYC)
• A hyper-growth startup that’s turning customers’ sales and marketing data into revenue is looking for a product engineer with a track record of building, shipping, and owning customer delivery at high velocity. (NYC)
• A growing pod at a multi-manager platform is looking for new quantitative researchers, no prior finance experience necessary, 250k+ (NYC)
• An AI startup building tools to help automate compliance for companies in highly regulated industries is looking for a director of information security and compliance with 5+ years of info sec related experience at a software company. Experience with HIPAA, FedRAMP a plus. (NYC)
• Ex-Ramp founder and team are hiring a high energy, junior full-stack engineer to help build the automation layer for the US healthcare payor-provider eco-system. (NYC)

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.

Elsewhere

Building Japan's Financial Ecosystem

Japan has one of the weirder equity markets in the world, because it can be divided into:

1. Fairly normal-looking companies like the kind you find in any country's market—some global exporters, some big banks and financials, the usual mix.
2. Small companies that trade at a low single-digit multiple of stable cash flows once you back out their cash on hand (but that trade that cheaply because they keep piling up cash, neither reinvesting in the business nor returning capital to shareholders).
3. Microscopic growth companies that are basically at the Series A or B stage, but are publicly-traded.

This is the result of a few oddities of Japanese markets, most notably a shortage of venture capital. Venture ecosystems have a feedback loop, where one generation of tech companies spawns the angels and VCs who back the next one, so they're hard to kickstart. But Neuberger Berman is trying, by launching a fund, targeting institutions, that will invest in a mix of public and private opportunities ($, Nikkei). One feature of the Japanese ecosystem today is that when a new company starts to grow, that growth and attendant margins are visible to its peers. Some of the alpha in VC comes from letting companies keep the magnitude of their success secret for just a little longer.

The New Balance of Trade

A broadly useful truism is that the countries that most want to avoid dependence on the dollar are exporters who need to do business with the US in order to have someone to sell something to. There's a correlation with having stable institutions and getting permission from the global economy to run chronic trade deficits, and on the other side of this there are countries whose domestic politics are only tenable because of their exports. This limits everyone's options; when the US tries to use dollar dominance in a serious way, it has an immediate negative impact on consumers, and when other countries try not to deal with the US, they don't have anyone to deal with—usually. A fun exception to this is one of the details in this story about how China became the world's #1 car exporter: they've done well selling cheap electric vehicles, but in Russia they found a market where gasoline is cheap, environmental issues have basically no political pull, and it's hard to import cars from the rest of the world. So it's where more of their internal combustion vehicles end up. Cars are a weirdly common case for this kind of strange trade relationship, like how strict inspections on three-year-old cars in Japan made slightly-used Priuses popular in Mongolia ($, Economist).

Where Return on Investment Goes

When companies earn a high return on investment, it's often because there's an existing asset that was worth building the first time, but is worth even more when repurposed, and the company that repurposes it gets to capture that upside without the cost and uncertainty of building it. This happened with media content libraries with the rise of cable, and then happened all over again with streaming, and may happen yet again with AI-generated video. ($50 for a five-minute conversation with an AI-generated Mickey Mouse or Elsa is objectively expensive, but a lot cheaper than Disney World—and it's in the category of gifts you can buy at the last minute if the planned birthday/Christmas gifts aren't big hits). It also happened with housing and Airbnb, or cars and Uber. But the inevitable next step of this is that the platform that connects all that spare capacity and already-built infrastructure grows fast enough to hit limits, at which point it needs to build some of its own. That's a point Meta is reaching with bandwidth; the company is in the early stages of planning a wholly-owned undersea cable that could cost up to $10bn, though the initial cost is lower. There's a scale at which bandwidth is free, a scale at which it's a product you buy in a standardized way, a scale at which it's a bespoke deal, and, as it turns out, a scale where you need to build the physical connections to move those high-value bits. This is naturally less attractive than an asset-light business like selling ads, but it's also the way those asset-light businesses evolve from being the first to exploit preexisting opportunities to sole owners of some new opportunity space.

Disclosure: Long META.

Broken ETFs

In modern markets, any asset that gets volatile because there's a lot of investor interest will spawn a levered version of the same asset. Individual traders could supply their own leverage, but they're sometimes irrationally reluctant to do so (this shows up across asset classes in the "lottery ticket" effect, where the most volatile slice underperforms, and within credit—the lowest-rated investment-grade debt underperforms on a risk-adjusted basis, and the highest-rated junk bonds outperform, too). One practical reason for that reluctance is that an unlevered strategy can be on autopilot, but maintaining a constant amount of leverage requires rebalancing, and patient counterparties. And for some ETFs there simply aren't enough counterparties to go around, so they're not able to track the assets they're proxies for ($, WSJ). Prime brokers don't have absurdly long memories, but many of them can probably recall cases where levered investment vehicles lost value fast enough that their equity was wiped out, leaving the lender on the hook. And that's something they'd like to avoid.

Swipe Fees and Accounting

The popularity of credit cards among consumers is easy to explain: people who reliably pay off their statement balance each month like getting an interest-free loan plus rewards, and people who don't regularly pay them down are grateful that there's a way for them to borrow. For merchants, the argument was trickier, but there were two good cases: first, some stores already extended credit, but banks had economies of scale and were willing to take on the credit risk. And second, early adopters of credit cards spent more frequently, and spent more per transaction; accepting Diner's Club or Amex was basically a way to target the top 10-20% of earners, and, especially at establishments where revenue per customer varied substantially, getting the biggest spenders was a meaningful win.

That describes the first few points of credit card penetration in payments. But the last few points, as stores and customers start to drop cash entirely, look a bit different: swipe fees are an increasingly material expense to small businesses, and one they don't have any good way to reduce. This is part of the same centralization that helped credit cards replace store credit in the first place: the banks have better information, and when they're collecting, they're essentially doing it on behalf of a bigger set of customers. But those banks, and credit card networks themselves, also have a much better view of customers' and merchants' price sensitivity. Giving an intermediary high share of some information-generating business activity means giving them the means to continuously tweak their pricing to their own advantage. And at that point, the problem is not just that the cost is higher, but that merchants can't know precisely what tradeoffs they're making, and the companies that do know are not interested in giving up their information advantage.