One of the nicest kinds of companies to own is the sort that acts as the intermediary between fragmented producers and consumers. W.W. Grainger, Nintendo, YouTube, Airbnb, and Roblox all fit this model. They can outsource operationally complex tasks to their suppliers, who benefit in turn from all the demand that the platforms control. At first, the optimal move for those platforms is to make participation as frictionless as possible—few ads to interrupt the users, and cheap ads so advertisers have room to experiment. As the platform matures, it might be able to capture a huge amount of value (as in the examples above), though in some cases the nature of the platform, or the scale required, precludes this—Snap and Pinterest both have great products and loyal users, but they don't have enough users (or the right ones) to get critical mass for their ad targeting, so their revenue per user is a fraction of someone like Meta.¹

These companies have a complex set of incentives. Part of their business model is to let other people think about all the day-to-day business questions—if someone's running ads on Facebook, but they don't have enough inventory to fulfill orders, this is only slightly Meta's problem. This hurts Meta about as much as it hurts a restaurant if someone orders dessert before they realize they're actually too full to enjoy it.

But in the long run, the incentives start to shift. Platform companies do care if their users make careless errors, even if those errors generate more revenue right now, because they cut revenue growth in the future. And these platforms can't be hands-off, both because the platform will get polluted with low-quality offerings if it isn't policed and because that pollution generally functions as a substitute for ads, just one that's a lot worse for users.

So their thinking starts to look like that of a monopolistic real estate owner. If you own all the land in a given area, whether it's because you're starting a planned community like Celebration, Florida, launching a charter city, or you happen to be the ruler of a city-state like Dubai or Singapore. In all of those cases, your economic incentive is not to maximize present rents, but to maximize the net present value of future rents—accepting a 1% lower profit (edit: a 1 point lower return on investment) today in exchange for 1% incremental growth over time is valuation-neutral, and if you can trade current rents for higher growth, that's accretive. Auction-based ad pricing means that, over time, advertisers bid ever-closer to the value they derive from the platform (and if they can get away with less, it's perfectly fair—it just means that they have an advantage over all competitors bidding for the same audience).

In that sense, the business model mostly takes care of itself, but only if the platform takes care of every other detail. This explains part of why the biggest platform companies tend to be founded by people who obsess over the product (Google, Meta) rather than the business (though there are plenty of examples of that, too; Jeff Bezos seems like an avid reader, but he probably didn't dream of selling as many SPARXIBAZL-brand kitchen goods and UGONOWAMI backpacks and the like. And Reed Hastings seems deeply interested in movies mostly in the sense that they have a high fixed cost and uncertain payoff, making them a better and better business with scale).

Part of what the platform/real estate parallel shows is that big tech is an abstract implementation of Georgist economics. Everyone on the platform is either paying rent or is, statistically, sending some revenue to people who do. The Georgist tax policy isn't to soak the landlords for all they've got, and it also isn't to run some mercantilist-adjacent policy of piling up financial claims for the state rather than using the money. Instead, the optimal approach is to set the ad tax, or the sales commission tax, transaction fee or whatever other tax there is, at a level that keeps ROI high enough that people will bet their business on this distribution channel—and then to raise it to the point that they can't afford to leave but aren't making much money when they stay. And this process is not the direct result of platform-level policies, but the indirect result of allowing competitive bidding for attention.

Like some of the real-world examples above, these places are pretty bland ones to live in. Your master planned community doesn't have guerilla graffiti artists, so it won't produce a Keith Haring, and while you can debate the artistic merit of Haring's work, there's no debate about whether it's a pain to scrub spraypaint off your storefront for the umpteenth time. Platforms work very hard to keep users from imposing negative externalities on one another, except when those users deliberately opt in, but there are plenty of load-bearing negative externalities—learning that you're wrong about something is, in the very short term, more embarrassing than informative, unless you have naturally peculiar emotional responses or deliberately train yourself. So platforms will probably encourage filter bubbles, ideally at the level of communities (hence Facebook's groups) but sometimes at the whole-platform level (Bluesky is just a different world from Twitter).² This is what most people opt into, even at a cost, in most aspects of their lives. For a given commuting distance into a metro area, basically every major city makes you pay a premium for boring residential real estate over the interesting kind. So, if the real market is any guide, this is utility maximizing.

Of course, that only goes so far. The Internet is a space where the commuting distance is always zero, so while some quasi-geographic constraints still hold, others don't. You can still sustain interesting, quirky, not-commercially-optimal communities online. But the economic pressure is always towards homogeneity, and in an efficient market the effort used to keep a given online community weird will be roughly proportionate to the value of making them a little more homogeneous.

Disclosure: long META, GOOGL.

The Diff writes a lot about platforms, and how they capture value (and comparatively less about real estate). 

• This model helps explain why the durability of big tech’s growth stands out ($).

• Understanding platforms means understanding why ads are the least-bad way to monetize many services.

• We profiled DoorDash at the time of its IPO ($).

• And Roblox arguably takes this model further than anyone else ($).

Join the discussion!

1  Meta’s blended ARPU globally is ~$52.80, Snapchat is ~$12.40, and Pinterest brings in the rear at ~$6.90. Note that Snapchat and Meta’s numbers are calculated using DAUs, whereas Pinterest is calculated using MAUs because the company doesn’t report DAUs (less important for their engagement model). In reality, Pinterest’s numbers likely look similar to Snap’s when using DAUs. These are all for the LTM period as of Q2 2025 for Meta and Snap, and as of Q1 2025 for Pinterest.

2  For many platforms, the marshmallow test is whether they'll encourage whatever the equivalent of the quote-tweet-dunk is, whether it's response videos on YouTube, parodies on Netflix, fisking in the blogosphere, or literal quote-tweets. This can be done in a healthy, constructive way; the visual format of a quote-tweet is convergent evolution with how the Talmud is presented. But in the degenerate case, a quote-retweet is designed to strip away whatever context would make the original content less dumb or crazy, and since there's effectively unlimited demand for information about how The Other Side is very crazy and dumb, that's the typical direction in which things go.