This is the once-a-week free edition of The Diff, the newsletter about inflections in finance and technology. The free edition goes out to 10,551 subscribers, up 185 week-over-week. This week’s subscribers-only posts:
- Moonshots, Uncertainty, and Certainty describes the peculiar risk-management approach the Apollo Program used to reduce the odds of failure for a project when testing was economically infeasible and in some cases practically impossible.
- Malls: Competitors and Complements looks at the economics of malls, and why Facebook rather than Amazon is their biggest Big Tech threat.
- The Big Rich and the Transience of Risk-Driven Wealth: In the mid-20th century, Texas oil money was synonymous with wealth. The big Texas oil fortunes have declined in importance for much the same reason they existed in the first place.
- Every Investor is a Market-Maker: It’s always smart for a business owner to ask where their profits come from. In this piece, I flip around the traditional view of why investors make money, with implications for both the size and structure of the investment management industry.
In this issue:
- Newsletters and New Media Economics
- Coming Attractions for Next Week’s Diff
- Fortnite, Apple, and Platform Wars
- Apple’s Other App Problem
- Theaters Restart the Two-Sided Network
- Singapore, Reconsidered
- Wealth Taxes and Peak California
- A16Z Revisits AI Economics
- Bullet Trains and China Stimulus
Newsletters and New Media Economics
Paid newsletters are popular enough that the business is getting its fair share of is-this-a-bubble-or-what? pushback. One view is that newsletters are quickly becoming a saturated market. Sure, Andrew Sullivan can quit New York Magazine and instantly have a six- or seven-figure subscription income (the Substack leaderboard rounds the numbers a bit, but that range makes sense). But if every writer at NYMag quits, and they all charge $5/month, the cost of recreating the magazine is multiples of what it costs to subscribe. To get a year of Andrew Sullivan, you pay $50/year. To get a year of everyone-left-now-that-he’s-quit, you pay… $20/year for your first year, and $60 thereafter.
The math doesn’t add up—at least as long as consumption patterns are constant. And clearly they’re not. Paid media is subject to three pressures, which work in tension with each other to determine the optimal structure of a media company—whether it’s, at one extreme, a giant company with a flat cost for every conceivable piece of information, or at the other, a series of one-person companies with highly variable pricing. Those forces:
- The convexity of knowledge
Bundling Consolidates Media
There are dozens of good explanations of bundling. Here’s a very relevant one, from the creator of the successful Everything Bundle on Substack. Bundling works when customers have heterogeneous tastes and the cost of creating one more copy of the product is low. A bundle essentially lets a group of newsletter-writers dynamically price-discriminate: most readers are subscribing because one or two components of the bundle are great and the rest are nice-to-have, so Everything’s $20/month sticker price is implicitly charging something like $15 for one newsletter in the bundle, $1 for another, $0 for another—but which newsletter is the premium product within the bundle varies from subscriber to subscriber.
As Baschez puts it:
The thing that makes bundles work is they eliminate waste. What waste? The wasted demand of all the people who want access to each newsletter a little bit, but not enough to pay the market price.
Most high-circulation publications are bundles. Some people read The Wall Street Journal for news about markets, others for news about deals, others because they like the editorials. Bloomberg is an incredibly broad bundle, and part of its utility is that any investor who spends all their time on one kind of asset occasionally needs to take a peek at another; an equity investor sometimes needs to know what’s going on with bonds, currencies, and commodities, but doesn’t want to spend much on a dedicated product just to track them. An algorithmic trader might get 99% of their Bloomberg utility from data feeds, but 1% could glance at its news aggregation service 1% of the time to understand what headline made the algorithm break that day.
And that points to another benefit of bundles: they lower customer acquisition cost, especially when selling a product to additional users at the same company, because it’s much easier to convince a company to go from 1 user to 10 than to go from 0 to 1. When customer acquisition costs are high, the bundle naturally expands to meet every need the biggest customers have.
But this raises a natural question: why is the Everything Bundle not literally everything? Why doesn’t it have a higher price point, and include all the latest developments in sports, celebrity gossip, astrophysics, and limericks? Because bundles benefit from clusters of adjacent readers. The readers within the cluster can have heterogeneous preferences, but they need preferences for the same approximate kinds of things, or it’s too expensive and time-consuming to attract them. There are plenty of publications that bundle content, but it’s adjacent content: beauty, fashion, and celebrity news; programming, venture deals, and career advice; politics, business news, and very well-written obituaries; etc.
One way to look at it is that a bundle has to appeal to the lowest common denominator for some very specific definition of “common.” When local newspapers were monopolies, their bundle was everything the average person in a given area would care about: local happenings, local sports, the weather, obituaries again, and a sprinkling of national and international updates. But the exact contents of that bundle only worked for a specific locale; telling people about the weather “on the east coast” was not exactly valuable, while today’s weather in Boston was important to Boston Globe readers.
Bundles tend to grow until they reach a highly profitable mature state—at which point any change in the underlying audience, or the availability of competing products, seriously weakens their economics. The bigger a bundle gets, the more likely it is that a subset of users are all paying for basically one piece of the bundle, which could be sold separately at a better price. And as soon as a bundle is partially unbundled, there are two options: stop offering the part of the bundle that now has a competing single-purpose product, at which point the bundle switches from optimally-priced to overpriced, or keep offering it and accept lower margins. Bundles grow gracefully and shrink painfully.
Another driver of aggregation is distribution: what’s the optimal way to get a given piece of content to the audience most likely to pay for it? Media companies have complained that Facebook and Twitter took over their distribution: instead of selling an entire edition, media companies' outputs get deconstructed into their individual components and distributed to the audience that wants exactly that.
For commodity news—a summary of what Biden or Trump just said, or what happened at a game, or what Apple put in a press release, that’s especially brutal. There aren’t many people with a competitive advantage in interpreting every piece of news in a given category. (Nobody hears a quote from Trump and says “I’m going to wait for an article by so-and-so before I decide whether or not I agree with this.”) But for stories that require some editorial judgment—when the problem is choosing what to write, not being the first to publish—social media’s effects on distribution are a bit different.
In that case, when distribution is through social media and peer recommendations, it belongs to the writer, not the publication. When you read the Wall Street Journal because someone throws it on your lawn every morning, a Journal writer can’t go independent. But when you see the same articles because they make it into your Facebook news feed, you may end up following the writer directly—they start to take control of their distribution. Media companies have encouraged writers to promote and discuss their work on Twitter, which has a short-term positive payoff in traffic. It’s a bit like outsourcing: companies used to outsource the most annoying, labor-intensive part of their supply chains, but the companies getting outsourced to moved up the value chain until they could build a competing product and sell it. Xiaomi and Huawei’s snazzy smartphones exist in part because Apple and Samsung didn’t want to directly manage hundreds of thousands of people working on assembly lines. (Edit: a reader correctly points out that this example would be more compelling if it were actually true. A more accurate example: Samsung assembled smartphones for Sony and Toshiba before building its own brand; Xiaomi and Huawei benefited from the growth of the phone assembly business in Shenzhen, but were not themselves contract manufacturers.)
Controlling distribution for a product with high fixed costs and low marginal costs is powerful, because it means not just getting content to anyone who’s interested, but restricting it to people who express their interest in dollars. For a large media company aiming at a broad audience, deciding on price is an exacting science, and there are all sorts of price-discrimination tricks for getting users at the right time. (For example, everyone from Amazon Prime to The Economist to Netflix tries to get college students in at low prices—in Netflix’s case, a price of $0 through a shared login—in order to raise the price on them later.) Fiddling with the pricing and offer dials to maximize conversions pays off at a large scale, but simple approaches work fine at a small scale, because individual writers can have a monopoly on their output, and monopolists can afford mistakes.
The Convexity of Knowledge
The writer-as-monopolist is an important economic fact. One thing the most popular tech/finance newsletter writers have in common (and here I’m thinking of Ben Thompson and Matt Levine) is a set of obsessions they keep coming back to, mental models they keep applying, and themes they keep revisiting. If Facebook tries to make inroads on Google’s revenue, you know Ben will have an Aggregation Theory lens for it; if a company gets sued for doing something clever with credit default swaps, Matt Levine is guaranteed to walk through the math and incentives at play. It’s very hard to outsource this kind of thing! Knowledge has compounding returns, but mostly when it’s contained in the same skull; two people with half the expertise yield well under half the insight.
This is well-understood in programming: there’s a whole book on the fact that when you add programmers to a team, the time spent explaining the project to them can more than offset the extra time they spend working on it. Distributed computing is hard enough with actual computers; it’s far harder with human beings.
This implies that in niches where there’s too much information for any one person to absorb, the most economically efficient outcome is for media coverage of that niche to be dominated by exactly one person, who works fairly hard and has more comprehensive knowledge of the topic than anyone else. If distribution costs were high, the result would be a specialty publisher that hires all these experts and then flogs the results of their work, but as distribution costs decline, it makes more sense for them to go independent.
Patrick McKenzie has suggested this in one domain:
But it can apply outside of finance. It will probably start in finance—that’s where the money is—but there’s no reason it can’t move to other topics, too. Everyone who needs to have an opinion on CFIUS, or needs to know every major development in mRNA vaccine research, or needs to know how GPT-3 is being used in production, can afford to pay a lot to be sure they’re not missing anything.
And this is a good counterpoint to the newsletter-skeptic argument about newsletter fatigue. Yes, many people spend entirely too much time reading newsletters—but that means there’s a market for compressing their two hours of somewhat indiscriminate, redundant reading into ten minutes of tightly-focused reading instead. For sub-enterprise price points (i.e. under about $500 a head), the cost of a newsletter in time is bigger than its cost in dollars. To put a little more math on it: paid readers open my newsletter about 75% of the time. At 10 minutes per newsletter, 250 issues a year, and a 75% open rate, that’s about 31 hours of reading per year. At the current US minimum wage of $7.25, the opportunity cost in time is almost exactly in line with the price of a subscription to The Diff.
This is all very good news, for writers and for readers. Filtering news and adding useful commentary is a nontrivial task, it’s hard to scale, and scaling it in one domain doesn’t imply skill in doing it somewhere else (if I switched places with someone who wrote a sports newsletter, we’d both lose all our readers). But the subscription newsletter model encourages people to identify niches where the signal/noise ratio is out of whack, and charge a premium price for a higher-signal publication.
Bundling reacts to differentiated desires by creating a less differentiated publication that’s fairly valuable to everyone. But as the cost of the reader’s time rises, focus pays off. And the subscription newsletter model makes it easier than it’s ever been to profitably focus on exactly one topic, and build a one-person monopoly.
 This piece focuses on one piece of the media business: paying money to get content. There’s another motivation, though: paying as a form of patronage. At low price points, patronage dominates: “This is nice, and I’d pay the price of a latte every month to support it,” is a motivating factor for some purchases. At price points beyond that, patronage only tends to work if it’s paid in exchange for either a) recognition, or b) additional hard-to-scale services. Call it the Tote Bag Line: past about $60, donations need to be paid back with something that advertises the donor’s generosity.
 I suspect that this is one reason the correlation between income per hour and total hours worked has been rising. The best specialist in any given domain gets a higher incremental return on one more hour of experience or learning than the next-best, because they can fit new information into more models, find more analogies, and make more connections that encourage further research.
A Word From Our Sponsors
Versett creates, builds and scales technology companies.
Each month, millions of people use the platforms, apps and tools we’ve developed for clients like American Express, TD Bank, and Lincoln Financial. But it’s not just the technology. Most of the value of new digital initiatives accrues after launch—so we help you build internal teams to operate and scale productively. The result? Successful launches, clear roadmaps, pragmatic hands-on help.
Versett is different. Take a deeper look at how we work at http://versett.com/thesis
Next week’s newsletter will be a series, inspired by a question from Andrew Walker of Yet Another Value Blog: historically, some of the top companies by market cap saw their value seriously impaired—to zero in some cases, down 80% or more in others, “dead money” for a decade in others. It would be very odd if 2020 were the first time investors in the very biggest companies didn’t see at least one of them underperform. So, one day a week next week we’ll be writing a hypothetical obituary for a large-cap high-growth name, looking back from 2030 at what went wrong.
And at the end of next week, I’ll be trying something new: a Zoom chat with paying subscribers for open-ended Q&A. Bring questions!
Fortnite, Apple, and Platform Wars
Epic, publisher of the wildly popular Fortnite (with 350m registered users as of May) gave players a novel offer for in-app purchases: they could buy in-game currency for $9.99 through Apple, or buy the same amount directly from Epic for $7.99. This is strictly against Apple’s rules: Apple requires developers to route payments for most digital goods through the App Store, giving Apple a 30% cut. So, in keeping with the company’s hard-line approach to off-platform transactions, Apple promptly removed Fortnite from the app store.
And then the fun started.
Within minutes, Fortnite had launched a response on two fronts, legal and PR.
- On the legal side, a 65-page complaint laying out their argument that Apple is a monopoly, and that its pricing is unjustified.
- For PR, they produced a parody of Apple’s famous “1984” commercial. It’s not bad.
What’s admirable about this is that usually Apple is the company a few steps ahead of the competition. They like to buy up the global supply of key components before they launch a new feature, so their competitors take longer to catch up; they strike early deals for new products that turn out to be lucrative for them later; they’ve adapted to the maturity of the smartphone market by selling pricier phones and adding more services revenue. And this time, Epic was the better-prepared company.
It will be hard for Epic to press their advantage. While they call Apple’s policies abusive, and say that they stem from a monopoly, Epic gave users the same payment option on Android, and were promptly removed from the Google Play store, too. It’s hard to say that a company’s behavior is uniquely bad when it’s not unique, although I suppose they could argue that Apple and Google are colluding.
And, paradoxically, their PR strategy won’t keep its momentum for long because Apple’s response doesn’t especially inconvenience current users. Apple removed the game from the App Store, but existing users can still play it, so instead of millions of fans being annoyed that they can’t access their game, Apple is inconveniencing future Fortnite fans, and the small number of users who buy an additional device and want to use it for Fortnite. So Epic has one day of momentum, but it will diminish. Fortnite is valuable to Apple and Google, but Apple and Google are far more valuable to Epic, so ultimately their bet rests entirely on their legal case—whose arguments have been under consideration for some time, with no effects so far.
Apple’s Other App Problem
The looming TikTok ban has gotten a lot of attention in the US because of the app’s popularity, but a WeChat ban could be more significant. The prospective ban is vague, but if it prevents Apple from offering WeChat on phones in China, it will threaten the company’s $44bn in China sales. Apple has an interesting protectionist argument here: since iPhones assembled and sold in China still use US-designed software and hardware, they’re positive for the US trade surplus, so a wide-ranging WeChat ban would run counter to the economic nationalist argument for a more limited US-only ban.
Theaters Restart the Two-Sided Network
I’ve been writing for a while about the double-bind movie theaters are in: studios will consider a theatrical launch rather than a streaming launch if movie theaters start getting visitors again, but visitors come to watch movies. AMC’s solution is to move very far along the demand curve: offering 15 cent movie tickets for re-releases of prior hits. This is actually a clever way for theaters to show that they can still fill seats, and it’s a good way to simplify the consumer decision when Tenet finally comes out: instead of asking “Do I want to see this movie?” and “Do I feel safe coming back to theaters?” they can just ask the first question.
There’s a great piece in Palladium on some of the complexities of the Singapore Story. Singapore’s development has many lessons for other emerging markets, but they can’t be applied blindly, and even Singapore’s hyper-competent civil service did not have a perfect record:
History provides us with a natural experiment. In 1994, Lee Kuan Yew and Chinese Vice Premier Li Lanqing signed the “Agreement on the Joint Development of Suzhou Industrial Park.” Under the agreement, Singapore would maintain a 65% ownership stake in the project and develop the city of Suzhou into a modern industrial powerhouse—all running on Singapore’s public-administration and industrial development expertise.
Yet by 1999, Lee had failed in Suzhou. Five years into the project’s 20-year development plan, Suzhou Industrial Park had only attracted $754 million dollars of investment out of target of $20 billion, 5,000 residents out of target of 600,000 and 14,000 employees out of a target of 360,000.
Yet competition for foreign direct investment from nearby Suzhou New District—a smaller, older, and less-supported development that Singapore previously dismissed—proved too fierce. Singapore’s elite group of civil servants simply could not navigate China’s multi-level government and apply the Singapore model at scale.
The Singapore Story, as it turns out, was very Singapore-specific. The blueprint can’t be copied blindly, any more than Lee Kuan Yew could have made a list of everything Renaissance Venice or medieval Hamburg did, without adapting it to modern circumstances.
Wealth Taxes and Peak California
California is considering a tax on wealthy residents. The tax, they promise, would only affect 30,000 people out of a population of just under 40m. But it would have an important impact on the state’s growth. California’s tech economy is driven by a network effect, where people who made their fortune in one generation of technology invest it in the next generation. Discouraging this process would not have a terrible impact immediately, but it would lower the odds that the next great tech company was founded in California. Further, the tax exempts directly-held real estate, so it would push California’s housing prices even higher—which acts as an implicit subsidy for companies that can generate cash flow and pay high taxes, relative to startups that burn capital and pay employees in illiquid equity. This proposal accelerates every piece of bad news from my Peak California piece last year.
A16Z Revisits AI Economics
Earlier this year, A16Z put out an important piece explaining why AI companies have worse unit economics than other startups. The cost of compute and training models remains high as a company grows, so gross margins don’t hit the levels other software companies have come to depend on. Now they have a follow-up, on how AI companies can achieve better economics. The Straussian reading is that the companies that are best able to profit from AI are the ones that a) have many lines of business that allow them to share functionality across domains, and b) have an established user/customer base, so they can pick which long-tail problems to address based on where returns will be favorable. A16Z, in other words, is treating Big Tech as the natural beneficiary to AI.
Bullet Trains and China Stimulus
China plans to double its high-speed rail network by 2035 ($). (Note that this article appears in Nikkei, Japan’s most read business publication, but somehow avoids being bitter even though China’s high speed rail economics are driven by amortizing the cost of borrowing Japanese IP in this area.)
In past crises, China has stabilized its economy by borrowing to build infrastructure when there’s a demand shortfall. When China lagged other countries, this was a reasonable model, but on a road- and rail-mile per dollar of GDP basis, China is already at rich-world levels. So the only way to justify large capital projects is to leapfrog ahead of other countries. Two side effect to watch:
- High speed rail affects real estate prices, by expanding a city’s commuter catchment area. That would tend to depress real estate prices close to workplaces and increase them elsewhere. Since China’s banking system lends so much against real estate, this could have unintended negative consequences for the banking sector.
- High speed rail is also a substitute for flying. China has tried, for decades, to build a domestic aerospace industry, but it hasn’t quite clicked; the country’s commercial aircraft are still far behind Boeing and Airbus. China’s domestic airlines started recovering from Covid before other countries, so this could have been an opportunity for China to push ahead in aviation. By focusing on rail instead, they’re tacitly conceding that catching up to established aviation companies won’t happen any time soon.