It’s easy to get distracted by Amazon’s stock price and forget that, on paper, they’re a financial train wreck. Historically, being a software provider for a high-growth sector is a great way to report 40%+ operating margins with minimal capital expenditures. Amazon… does not report stats like this. Their operating margin is about 3.4%. Because, in addition to being in the business of building software for growth industries, they’re in the business of building warehouses, buying planes and trucks, and hiring an army of pickers, packers, and sorters to operate it all.

Why does Amazon choose to run this business? And why don’t investors penalize them?

They’re not alone:

• Why does Netflix run their own studio, rather than outsourcing the low-ROI and lumpy studio business to companies that have been doing it longer?
• Why does YouTube pay stars, and pay to promote them, instead of using free home videos and cheap licensed videos? And why does Google build a Pixel instead of letting Samsung and Xiaomi fight over who will own the Android handset market?
• Why does Uber, an app and logistics company, want to be in the crowded and competitive auto financing business?

And what about the companies that don’t behave this way? If Amazon, Netflix, Google, and Uber are so smart, why doesn’t Priceline own a chain of hotels? Why does Amazon have ~45x the owned and leased square footage of EBAY, despite selling an estimated ~4.5x as much merchandise each year?

There are two good answers, one driven by financial and tactical constraints and one more strategic.

Headcount Constraints and Capital Constraints

As companies grow, things break. If their growth means accreting features onto a software product, they break a lot faster . Software isn’t just a way to tell machines what to do: it’s a way to tell humans what the software is doing. There’s a whole cultural edifice you need for a large number of people to contribute to the same project. When you read about what it’s like to work at companies like Facebook and Google, you often read about an obsession with standards and best practices. And these companies like to hire weirdos, not boring nitpickers. They care about process because they have to.

But too much process will strangle a small company. At some point, a company has to evolve from “everyone has permission to change anything” to “please refer to section 42(b)7,” but doing it too fast or too slow is fatal. This sets a natural speed limit on how fast a tech company can scale: if they’re doubling their headcount each year, then half of their employees have been at the company for less than a year. They either have to hire for a very specific cultural fit (which slows growth down a lot as they grow), accept a short-term hit to productivity as they train every new hire to contribute in exactly the right way, or build an unmaintainable tower of babel.

Other businesses don’t run into the same kind of human scaling problem. A burger joint or a clothing retailer can afford gung-ho expansion: in the worst-case scenario where someone doesn’t live up to the ideals of Hamburger U, you can replace them without having to rewrite all of their code.

When cash piles up on a tech company’s balance sheet, they reach a point where they can’t add new people to a project. But they can often find something adjacent to what they’re already doing.

Let’s suppose that markets are fairly efficient in the aggregate. At least in the sense that if people are running warehouses, movie studios, and handset factories, it’s because a dollar invested in such a business generates a pretty acceptable return (or it would be invested somewhere else). Moreover, a movie studio that could spy on Reed Hasting’s strategy meetings, and snoop through Netflix’s viewership data, would have even better returns. A handset company that knows what will be in the next Android release is a handset company that will avoid costly mistakes. Clearly, if an adjacent business on its own is a decent business, and adjacent business with the information advantage of being owned by a disruptive technology company is a great business.

And why not return capital? Because capital, for a successful tech company, is cheap to raise and expensive to give back. Raising capital means either borrowing at favorable rates (Netflix still pays interest as if its credit rating were a notch higher) or issuing high-multiple stock. Returning capital either means paying a high multiple or paying a taxable dividend, and either way signals to investors that growth is over.

Oddly enough, the investor who best explains this is Warren Buffett. Back when Berskhire was small and middle-market companies weren’t getting bids from PE shops, Buffett was able to accumulate a portfolio of small growth companies that required little to no capital to keep on growing. Now, two of Berkshire’s biggest businesses are railroads and regulated utilities.

Railroads and regulated utilities can grow, but it usually involves buying trains, building power plants, etc. In 2015, for example, Berkshire Hathaway’s railroad capital expenditures were $5.8bn, and their energy expenditures were another $5.8bn. The guy profiled in Roger Lowenstein’s excellent bio was not the kind of guy who would shell out $11bn a year to expand his company’s presence in a commodity business. On the other hand, all the really great non-commodity companies tend to trade at high multiples. They were cheap back when Buffett was buying them, but now he’s bought them (or KKR has, or Blackstone, or 3G).

Buffett has more cash than he knows what to do with, and fewer great businesses left to buy. (Moreover, it’s a law of the universe that every New York Times bestseller about an investor eliminates about half of the expected return from investing with him.) But Berkshire Hathaway isn’t cheap enough for a buyback. Investing in slow-growth, high-capex businesses is the best of a batch of subpar options. In the event that Berkshire’s stock does get cheap enough to be worth buying back, those companies can also support a lot of leverage.

And this explains why some tech companies, like eBay, Priceline, and Expedia, don’t start writing giant checks for fixed assets: their stocks are too cheap. Ironically, a cheap stock tends to cause a business to stay high-quality.

BATNA

If you can build a better, cheaper studio, you don’t just have more cash on hand: you have a BATNA with every other studio. When Netflix’s coopetition with Hollywood leans less towards co-op and more towards -tition, they can always call up their bankers and say that this year, they’re putting some real money into originals.

(It’s a fun exercise to parse management’s behavior in light of this. They’ve said a few times that they plan on ramping originals time up this year, from 600 hours to 1,000. How much of this is because they’ve refined the formula for making hits, and how much is because they’re gravely disappointed in studios for cooperating with Amazon and Hulu?)

It’s impossible to analyze strategy in game-theoretic terms, because the optimal behavior is always to spend just a dollar more than the “did this CEO just read an article about applied Game Theory and decide to take it for a spin?” threshold.

So it could be the case that Jeff Bezos never thinks about how Fedex and UPS view his logistics empire, or that Sundar Pichai signed off on the Nexus 7 without really considering how much Samsung would have to step up their game. It’s possible that Uber’s captive fleet of drivers and vehicles won’t influence auto companies when they try to offer self-driving cars.

Possible.

On the off chance that these CEOs give any thought whatsoever to strategy, there are two implications: one, that the accounting ROI on high-capex ancillary businesses is lower than the economic ROI. And two, that you never want to invest in someone who is part of the supply chain of a tech company that’s throwing off lots of cash.

What About Facebook?

Facebook is enormously profitable, scaling their core business as fast as they can, and yet they don’t fit my theory: they haven’t found some capital heat-sink that can absorb the $12bn or so in free cash flow they’ll generate this year. And it’s completely unclear to me why. Maybe there isn’t an adjacent business they can really invest in. What’s adjacent to human connection? Maybe they’re more disciplined — they never raised as much money as they could, so maybe cheap capital isn’t as tempting to them as it is to others.

Or there might be a meta-rule: a heuristic for judging technology companies is useful if and only if there’s at least one glaring exception. If tech companies really are cheap because the founders know something investors don’t, a sufficiently comprehensive theory is suspiciously good.