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 19,531 subscribers, up 110 since the last edition.

In this issue:

• An Airline IPO? In This Economy?

• Facebook vs Bullying

• Scale and AI

• Chip Shortage Updates

• Decarbonization

• Supermicro, Redux

• The Value of Safety: A Natural Experiment

An Airline IPO? In This Economy?

Airlines are a fun industry to track as a proxy for the broader economy. They're exposed to consumer sentiment (from leisure travel), corporate sentiment (business travel), financialization and Internet-mediated price-discrimination, commodity costs (fuel is the biggest source of expense variability), the bargaining strength of labor, and the cost of capital (a plane lasts for decades). There's something naturally cool about working with heavy machinery, and a plane is the largest and most expensive piece of capital equipment the average person in the developed world will personally encounter, so airlines seem to attract a share of talented managers all out of proportion to their historically mediocre and volatile returns.

2020 was an interesting year for airlines, but a tough year for analysts; instead of looking at the effects of new planes, new routes, and new tricks to salami-slice the flight experience into discrete components that can all be priced separately, understanding the industry was mostly about tracking infections, government policy, and vaccination rates.

But things are starting to get interesting again. After a year in which the industry produced a collective net profit of negative $370bn, there is—an upcoming airline IPO! Sun Country, owned by PE firm Apollo Global Management, plans to go public this year.

Skift, the indispensable travel site, describes the company's backstory here: it was a family-run airline, run by two brothers who engaged in such non-shareholder-value-maximizing behaviors as keeping first class because "they liked owning a premium airline," and giving away elite status in its mileage program to friends. The company occupies a niche compared to Delta, with which it shares many routes: Delta is expensive, and offers first-class service; Sun Country is much cheaper on direct routes, but doesn't have the same kind of loyalty program or the same level of service. Skift speculates that Delta would rather keep it around as a nonthreatening competitor than face off against a better-capitalized and more aggressive airline. (Competition from bigger carriers has another perk: when Sun Country got CARES Act funds, which were supposed to be conditional on keeping its routes, it was able to shut some down anyway because they were already served by other carriers.)

Today, Sun Country is, technically, a profitable company: it reported operating income of $21m in the first nine months of 2020, on revenue of $293m. But that profit includes $64m in CARES Act grants; without them, Sun Country would still be unprofitable. Of course, it would be a miracle for an airline to turn a profit during a pandemic; Sun Country's performance is surprisingly good, but the focus on what comes next.

Most financial disclosures don't fit into a single literary genre, but an airline S-1 right now does: it's retrofuturism, a glorious vision of the possible derived by extrapolating from a now-lost golden age in the past. To buy the stock is to bet on normal airline economics, but to analyze those economics means looking back at 2019.

And Sun Country circa 2019 was a very interesting little airline. Unlike most carriers that focus on scheduled service, the company has a large charter operation (around a third of 2019 revenue), flying sports teams, gamblers, and members of the military. As they point out, these are cycle-resistant; sports haven't been cancelled due to a recession in a long time, and the military is also famously recession-resistant. That leaves Vegas: 38% of Sun Country's charter revenue in 2019 came from long-term contracts with casinos, but most of that vaporized last year: visits to Vegas were still down 64% last month. But Vegas seems like a very strong candidate for the "revenge travel" thesis, that people will use up their accumulated vacation days, and their need for a vacation, in a splurge of roaring-20s excess. Nothing says "I'm glad I'm not locked in my house all day and forced to obsess over distancing and masks" like going to a city where basically everything is legal.

Their traditional scheduled service business wasn't doing too badly, either. After getting acquired by Apollo in 2017, scheduled service passenger count grew 43% in two years, and their ex-fuel unit costs dropped 20%. It's hard to build a long-term advantage as a low-cost carrier, because the typical pattern is that unit costs a) decline with scale, but b) rise over time.

Costs decline with scale because, while a plane is a fixed cost, an organization for managing bookings, marketing the airline, keeping planes maintained, hiring people, etc., is also close to a fixed cost. Low-cost airlines often specialize in just a single plane, which reduces training costs, makes it easier to keep parts on hand, otherwise keeps things simple and low-cost. The problems with this model are that a) different planes are economical for different routes, and b) every so often, the airline has to go through the very exciting process of either giving up on the single-plane model or switching to a different plane. The alternative is to specialize in being adaptable to all sorts of different planes. This is most cost-effective for network carriers; they can use smaller planes on low-traffic routes, and feed that traffic into busier routes. Delta, for example, flies or has on order no fewer than 22 different kinds of planes, and has gotten good enough at maintaining them that it performs outsourced maintenance and parts purchasing for other airlines—including Sun Country.

The reason costs rise over time is that unions get stronger. The economics of unions are a lot like the economics of sticky enterprise software products like ERP. At first, they under-monetize, but over time, they develop a keen sense for exactly how much they can ask for. Meanwhile, they tend to grow over time. And they often ask for seniority-based pay, which means that slower expansion in headcount leads to higher pay per employee. As a result, new airlines tend to have better-looking economics than mature ones, because they outgrow their unions. As they mature, the union turns into a partner, with a stake in the upside (higher wages when times are good) and less of a stake in the downside (wage gains stick around, unless the airline goes bankrupt).

The result of these two scaling factors is that new low-cost carriers grow at a breakneck pace when they're young, taking risks that range from their operations to their advertising. And, eventually, the pace slows down, margins get volatile, and the airline has to move at a more sedate pace. Unfortunately, its younger competitors also have a growth mandate, and haven't hit their growing pains.1

Sun Country has one other difference with major airlines: its Amazon deal. Once again, Skift has the story: in early 2019, Sun Country tried to sell Amazon a used plane. Amazon turned them down, but made a counteroffer: would they be willing to retrofit some of their planes to carry cargo for Amazon? This deal was signed on December 17th, (coincidentally, the day after the first documented Covid-19 hospitalization in Wuhan). And while it didn't save the company—the CARES Act did that—it does make Sun Country more flexible in its business model. When cargo is doing well, they'll do more cargo; when passenger travel is doing well, they'll haul more passengers—and they can shift planes between charter and scheduled service depending on exactly what kind of travel people want. (Amazon seems to like the model, too. Part of their transaction with Sun Country involves a warrant to buy 1% of the company, and that warrant can cover up to 15% depending on cargo volume.)

Sun Country was not an airline designed to deal with the pandemic and its aftermath, but it ended up developing a structure that's well-suited to an uncertain travel market. Perhaps the best way to view them is that they're a company with a comparative advantage at buying planes: unlike a pure-play cargo, scheduled service, or charter airline, they're not locked into just one use case. And unlike a leasing company, they're not buying a plane and then turning around and finding an operator for it, who might end up capturing most of the economic upside. 2021 is a strange time for an airline to go public, but a strange airline like Sun Country is the kind that could make it work.

Share

A Word From Our Sponsors

ExitUp is a weekly newsletter that delivers a curated list of fresh job openings across PE, VC, strategy/biz ops, finance, marketing, and product management right to your inbox. You'll also get useful insights to help you navigate your job search so that when the time comes, you'll be ready to nail the interview. Subscribe for free today, and find your exit opportunity with ExitUp.

Subscribe to ExitUp

Elsewhere

I have a new piece in Works in Progress, on how the SPAC boom represents a more optimistic outlook. I've argued before that bubbles are a way to coordinate building different parts of the future—the dot-com bubble encouraged the infrastructure necessary to make the sites work, and encouraged the sites to build in anticipation of infrastructure. One way to look at this is that it's a business cluster in time rather than space; just like hedge funds all clumped together in Midtown and movie companies in Hollywood, all the commercial spaceflight companies are clustering together in the timeline where space travel is cheap and popular.

Facebook vs Bullying

When I wrote about Facebook and superlinear risk scaling ($), one thing that stood out in Facebook's transparency reports was how hard it was to detect bullying. For other kinds of content they block, 95% or more was caught automatically, without any human intervention. For bullying and harassment, Facebook's algorithms at the time caught just 13%, while users flagged the rest. In the company's latest transparency report, bullying and harassment now get caught automatically roughly half the time. This is a good look at the arms race between humans and AI moderation. Even when the edge cases are tricky, at sufficient scale the algorithms catch up.

Scale and AI

In a very related story, The Next Platform sketches out how complex language models could end up costing $1bn+ in the next five years. That is the cost of building the model, not a marginal cost from running it, but it's striking. (One estimate put the cost of training GPT-3 at $4.6m.) If language models can be human-like, but only at huge-company scale, they'll be another avenue for the largest tech companies to achieve returns from scale. And the companies that can afford that kind of investment are also companies with potential access to a larger corpus of up-to-the-second training data. Bloomberg has been using automated fill-in-the-blank headlines and copy for decades; perhaps in a few years, you'll be able to have a natural-language chat about a 400-page SEC filing, starting half a second after it drops.

Chip Shortage Updates

• The Biden administration plans "aggressive steps" to mitigate the undersupply of chips, with an emphasis on identifying other parts of the supply chain that could be fragile.

• The EU wants to build cutting-edge foundries in Europe so they won't be caught short next time.

• China's government has suggested that chipmakers send more chips to Chinese automakers. China has had some difficulty building its own supply chain; one $20bn chip fab project in Wuhan, where construction continued even at the height of the pandemic, turns out to be partly fraudulent ($, Nikkei).

I've argued before that the world is moving to having two separate chip supply chains, although China has a great deal of catching up to do there. It's possible that two is an underestimate, and that the next big shortage will be in semiconductor equipment instead. (It's also interesting that, before Covid was widespread outside of China, there was a brief flurry of discussion about supply chain durability, which I participated in. That got set aside once the pandemic was a global problem, but it's something to focus on. A common pattern for technology is that as it advances, the number of use cases goes up, but the number of competent manufacturers goes down, as the minimum skill and capital waterline rises. This gives many assets the same trait oil has, where it turns local problems into global ones once supplies look like they're at risk.)

Decarbonization

• Russia's largest gold mining company is switching from coal power to hydro ($, FT) for most of its electricity needs. This is especially notable given that, as the article points out, Russia has an easy time complying with the Paris agreement: Russian emissions dropped precipitously after the collapse of the USSR, so the country is already on target.

• Shell plans to get to net zero by 2050 ($, FT), and is setting a stricter standard than other oil majors: it's including emissions from oil and gas that Shell markets but doesn't produce. The emissions competition between oil majors is partly about the timing of net zero emissions, and partly about the definition.

Supermicro, Redux

In October 2018, Bloomberg published an investigation of Supermicro, alleging that the company embedded additional chips in its products that sent information to China. Amazon and Apple, both cited as victims, denied the report, but Bloomberg did not retract it. They're back with another report claiming even more extensive infiltration. This one takes pains to explain why some of the participants might have denied it:

Without a fix on China’s ultimate purpose, U.S. leaders decided in 2013 to keep the discovery secret and let the attack run, according to three officials who were informed of the plan. , then-director of the National Security Agency, played a central role in the decision, the officials said. The Pentagon devised undetectable countermeasures to protect its networks, two of them said.

One thing this story illustrates is that "trusting experts" is not a complete solution, because a) you need expertise to evaluate experts, unless you catch them lying, and b) even if you do catch them lying, in matters of national security they might be required to lie. So the story remains ambiguous; Bloomberg is once again staking its reputation on these claims being true, and given how many people they've spoken to, it's hard to imagine the story being entirely fabricated. On the other hand, they're still getting on-the-record denials from Supermicro, the NSA, and others. The preponderance of evidence suggests that Bloomberg is reporting on an investigation that was in progress until approximately the moment they published the story.

The Value of Safety: A Natural Experiment

Tyler Cowen points out that Delta's empty-middle-seat policy is a way to measure how much people care about Covid-19 safety, and indicates that—at least among people willing to fly—money and convenience trump disease risk. These people might be wrong, of course, but it's a good benchmark for the tradeoffs involved in pandemic prevention measures.