Economics 101 gives us lots of models that are simple, compelling, and not borne out by real-world observation. And that's what it's for: the whole point is to show that by making a small set of fairly reasonable assumptions, we can construct elaborate abstractions—and we do this because, when the abstractions don't match reality, it gives us few places to start looking for deviations. It's a way of re-enchanting the world, by giving us a rigorous framework for thinking about the complexities of human behavior.

For example: why do economic profits exist?

In accounting, profit is just revenue minus costs.¹ But in an economic sense, profit is when revenue minus costs exceeds the cost of capital, where cost of capital is a function of the risk involved. It's very difficult to come up with an exact metric for cost of capital, but in general it's higher for riskier projects and lower for safer ones.²

So if the cost of capital for a restaurant is, say, 10%, and the restaurant costs $1m to start and earns an annual accounting profit of $100k, its economic profit is nil: it's earning its cost of capital, but no more. And this model is useful, because it roughly describes the real world: if everyone who started restaurants earned a 20% return on capital, more people would start them, but since those people are dividing a roughly finite number of nightly diners over more and more establishments, profits would quickly converge on the cost of capital. Or they might go below that cost of capital, at which point some restaurants that shut down would be converted into other establishments until the economy reached a rough equilibrium.

This doesn't happen everywhere, though. Backing out their cash on hand, Google's return on capital, minus the debt fraction, is 40%. And Apple's is 146%, even without backing out cash (if you remove their cash and long-term investments, and just focus on the equity needed to run the business, you get a nonsensical number, because excluding those Apple's equity is negative). Clearly, in an efficient market people would rush to start new search engines and smartphone companies until the situation reached equilibrium, and Apple and Google had the same return on equity as a car company, airline, or bank.

The reason this doesn't happen is, of course, that these companies have intangible assets that don't show up on the balance sheet but do contribute to returns. And, crucially, these assets need to be hard to duplicate—for Google's economics to be sustainable, it needs to be impossible for someone who builds half of Google, paying full price for those intangibles, to get half the profits Google generates.

The usual way this works is some combination of positive feedback loops, compounding value for intangibles, and asymmetric views of long-run economics:

• The biggest tech companies benefit from network effects, sometimes with overlapping feedback loops: search gets better data and higher bid density with more users, and that funds more frequent crawls, better spam detection, and new search features. The iPhone has a wide range of apps, attracting more users—and those apps exist because of the users Apple was initially able to attract. It's possible to duplicate some kinds of network effects merely by spending money, as ride-hailing, food delivery, and in an earlier era group buying all demonstrated. But it's cheaper to be the first to figure the loop out. Positive feedback loops also apply to employees: workers in tech have a good sense of which startups are the prestigious ones to defect to, and once those defections start happening, they tend to improve the target company's mystique further.

• A related phenomenon is that some intangible assets compound in value. Amazon (disclosure: long) built its pro-consumer reputation early, through things like easy returns and allowing reviews on the site. (It's easy to forget that, to publishers, it was controversial to say the least that someone could put up a one-star review of a book, and that Amazon would literally show this to someone shopping for the book in question. It will be interesting to see if Twitter's decision to allow community notes on ads—assuming this was a decision and not a bug—will have a similar effect.) The value of data has also compounded at a surprisingly strong pace, and some products that start out non-strategic turn out to have useful ties to other parts of a company's business.

• The other source of long-run economics is an asymmetric view about the long-run value of investments, whether they’re tangible or not. One reason railroads have done well over the last decade is that it's effectively impossible to build new ones in the US, so the existing railroad network is a monopoly by default. But if you look back at cases where a startup successfully topples an incumbent, there's almost always a point where the incumbent suddenly becomes aware of the threat; they realize that with a big enough investment, the startup could kill the incumbent. But as long as the startup has better information about what threat it represents than the incumbent, it can seamlessly pass from "too small to worry about" to "too big to deal with."

One important takeaway here is that big tech companies' economics are partly the result of survivorship bias: the amount of money invested by companies trying to compete with Google, directly or indirectly, exceeds what's on Google's balance sheet. In winner-take-all industries, the equilibrium state is not for the surviving company to have typical returns on capital, but for the capital invested in that company, and invested in efforts to compete with that company, to have typical returns.

The trick is always to build something that's no longer worth replicating, and one way to have decent odds of this, but decent odds of failure, is to be one of a dozen companies trying to build the same thing, and then to be the one that figures it out first.

Read More in The Diff

The Diff spends a lot of time talking about competition, and compounding competitive dynamics. Some case studies:

• This profile of Russian search engine Yandex is also a look at why search is one of the best businesses of all time.

• This piece covers "bet-the-company" ideas, which are one way to build a durable advantage (or to lose everything).

• "Dueling over platforms" looks at cases where companies fight one another for network effects.

A Word From Our Sponsors

Every day Refind picks 5 articles that make you smarter, tailored to your interests. Loved by 200,000+ curious minds.

Subscribe to get 5 links per day!

1. "Just" is doing a lot of work there, since revenue and costs both require some estimates, and these estimates can be wrong. There are some companies whose economic performance is, from quarter to quarter, exactly the opposite of what their costs are—LNG companies, for example, sign long-term contracts that are partly tied to prices, and then they partially hedge those contracts with futures. Accounting rules require them to mark their futures positions to market each quarter, but they don't allow them to book profits on changes in the value of their contracts in the same way. So in a given quarter a company might book, say, $500m in profits from hedges, but also face a $600m decline in the net present value of the contract those hedges affect. So for LNG companies, at least, that $100m of "good” earnings are bad news.

2. Why is this so difficult? Because the real cost of capital is not just a function of the risk of the project itself, but how that risk correlates with other risks. A gold mine might have a lower cost of capital than a copper mine with similar expected margins, because copper prices generally move with the economic cycle, while gold prices rise when investors are worried. So even if they get the same returns over time, the gold mine owner tends to have more money when other assets have lost value, which makes their portfolio more stable. Or, more to the point, the gold mine owner has relatively more money when other assets they might want to buy are cheaper.

Help us learn more about you!

We're building a media kit and need some anonymous aggregated subscriber data. If you work in finance and have time for a 3-question survey, that would be very helpful:

Join the discussion!