Thinking about the technology industry, it seems like there are two kinds of successful tech companies:

– Companies with a nonzero marginal cost of production
– Companies with a monopoly in their market

Virtually every famous startup fits in one of these two categories. Apple? Nonzero marginal cost (I’ll abbreviate NMC). Microsoft? Monopoly. AirBnB? NMC. Facebook? Monopoly. Dropbox? NMC. Google? Monopoly. (There’s Bing, but Bing loses money hand-over-fist; it couldn’t exist on its own.)

Everyone knows about monopolies, but marginal cost isn’t discussed much. Simply put, marginal cost is how much the company has to pay for each customer it acquires, or each unit it ships. Because of Moore’s Law, many tech products have trivial marginal cost. The price of bandwidth and memory is so low that it costs about as much to serve a website to ten thousand as one thousand. A site like Facebook is a zero marginal cost product, because the cost to Facebook of yet another user is basically nothing.

On the other hand, a company like Heroku has high marginal costs, because every user adds significant expense (in the form of buying more servers). They charge more than they spend, which is how they make a profit, but they don’t pocket most of their revenue. It’s worth noting that Heroku, Dropbox and AirBnB, at one time considered the three most successful Y Combinator companies, are all NMC, even though most YC startups aren’t.

I suspect the reason for this pattern is that zero marginal cost is deadly to most companies. Virtually every business has competition – even if the product is totally new, someone will soon be along to copy it. In a competitive market, a product’s market price is pushed down to near its marginal cost. If the marginal cost is zero, then the price inevitably gets pushed to zero, which is why so much Internet stuff is free. (There’s no way most people would use a paid version of, say, Google Calendar, even if it was better.) And it’s damn hard to make money by giving stuff away. To a first approximation, zero price equals zero revenue.

The main exception to this failure mode is when the company has a monopoly. A monopoly protects you from competition, so you can charge something for a product that costs nothing to make and get away with it. A Microsoft CD costs about $1 to manufacture, but it can sell for hundreds of dollars if the customer has no practical alternative. You can also put up lots of annoying ads, or do something else that’s equivalent to charging money.

(Every now and then, someone proposes a micropayment solution to the zero-cost failure mode. The way micropayments are supposed to work is that, if the real market price of seeing a web page is a tenth of a cent, you pay “a tenth of a cent” instead of “nothing”. The company can then make money on the tenth-cents it collects from everyone. It never works, because there’s a big psychological difference between “free” and “not free” – the gap between $0 and $0.001 is a lot bigger than the gap between $0.001 and $1. Ads are effectively a micropayment system, which do work a little, but see The Online Advertising Fallacy for why every web business can’t just run ads.)

The lesson for individual startups is that, if you’re within striking distance of an NMC product, focus on selling an NMC product. Stay close to the money. For a famous example, take AirBnB, one of the most successful Y Combinator companies. They sell an NMC product – a place to stay – and are wildly profitable. But one can easily imagine a similar site that doesn’t make any real money. Suppose that, instead of selling rooms themselves, AirBnB set up a forum for other people to sell space, like how Craigslist works. Even though the user’s experience might be almost identical, this site would now sell a product with no marginal cost (web hosting), and it probably couldn’t get any significant fraction of the total money spent on rooms. It would have to be a monopoly to work (like Craigslist is), and even then it would likely be way less profitable.

Even without a monopoly, could one charge for a zero margical cost product? History doesn’t seem to support the idea. For another famous example, Loopt was once called the most successful of the original eight Y Combinator companies. Their product – a location-based mobile app – was zero marginal cost, and used to charge for it. But as smartphones exploded and more competition arose, they eventually had to give it away for free, and they were recently bought out for far less than their investor valuation. They were the first major player in the market, and were run by someone extremely determined (Sam Altman), but without a monopoly it didn’t hold up.

To illustrate the principle of NMC, here’s a quick thought experiment. These numbers are made up, but are comparable to real numbers that various friends have seen running web businesses. Suppose your business starts a recipe site. People post recipes for free, and you put ads on the site to make money. A typical price for online ads might be $2 CPM, which means you get paid $2 per thousand people who see the ad. This means that for every person who visits the site, the net revenue (gross revenue minus cost of goods sold) is $0.002.

Suppose, on the other hand, you started a site that took people’s online recipes, made the food, and delivered it to people on demand. A visitor to the site might spend $20 on food, but a random visitor probably isn’t going to pay (suppose 10% do), and the food costs something to prepare and deliver (suppose 10% margins). Even so, for every person who visits the site, your net revenue is now $0.20. That’s two orders of magnitude greater – a whopping huge difference. That’s the difference between $100 and $10,000, or between a lifestyle business and a publicly traded company.

Orders of magnitude matter. Net revenue per user matters. A lot. A great deal of the emphasis in Silicon Valley is on “eyeballs”, but eyeballs don’t mean anything if revenue per user is a tiny fraction of a cent. One might well just ask Moot, the founder and CEO of 4chan – he runs one of the most popular websites in the world, one which is arguably a major cultural force, and he complains about not being able to sell it for anything substantial. It’s hard to know exactly what revenue per user or profit per user will be, but it’s easy to get a decent estimate from the business plan. How many zeroes are there? Is there a related business plan with fewer zeroes?

The lesson for investors is that there might be easy money on the table by only investing in NMC startups. Judging people is hard, judging markets is hard, judging monopoly potential is hard… but judging whether a startup is NMC is easy, since one can just look at what the company sells. That might not find the next Google, but everyone is already looking for the next Google – there’s probably money to be made finding the next Etsy.