Benevolent Monopolists, Ruthless Monopsonists

Plus! Liquid Content; Rapid Vesting; All-Time Highs; Tariffs and Predictions; Moving Down the Funnel

Byrne Hobart Byrne Hobart

In this issue:

audio-thumbnail
The Diff December 15th 2025
0:00
/1033.508571

Benevolent Monopolists, Ruthless Monopsonists

An odd feature of the modern economy is that if you use your smartphone to search for a flight and hotel, you're interacting with a big stack of monopolies and consolidated markets—ASML, Zeiss, ARM, Broadcom, TSMC, one of Cadence/Synopsys/Siemens EDA, one of Shin-Etsu and SUMCO, one of Linde, Air Liquide, and Air Products, etc. If you're doing a reasonable amount of shopping on your phone, it's probably a phone made by Apple, which is not a monopoly but which sure produces financial statements appropriate for one.

And, if there's a part of this transaction that leaves you feeling vaguely ripped-off, it probably involves how much money you paid for the hotel, of which you had many options, or the airline, where, again, you probably had a few different choices.

It's a deeply weird feature of the modern economy that there's a whole set of transactions routed through a group of monopolists with something like half a trillion to a trillion dollars in cumulative capital investment and R&D, and what they produce is mostly consumer surplus. Chips, phones, and search engines are all a great deal for consumers, and the inputs into them tend to be great deals for the companies that provide that consumer surplus. This should be surprising: if you were sketching things out based on economics 101 models, and you assumed that an economy had a mix of competitive and monopolistic businesses, you wouldn't be surprised that the monopolists have high returns on capital, but you would be surprised by where it shows up—Meta could put up a paywall between you and your friends' pictures, Google would be a good deal if you paid $100/month for search, and if we lived in a world where every smartphone was 10x as expensive, spending $1k for a cheap Android handset with similar specs to the iPhone of a decade ago would still feel like an okay deal for a handy gadget. You'd be able to use it for browsing, email, messaging, and shopping. It wouldn't run many modern games, but who'd use such a nice machine for gaming, anyway? That would be a privilege reserved for people rich enough to buy a $10k top-of-the-line iPhone, a luxury product.

One view is that this is a matter of differing discount rates: companies will undercharge in order to lock their users in, and then raise prices once their customers are addicts. This has happened in some places: back when retail was the main growth business, Amazon tended to launch new categories before they were viable as a standalone business, so Amazon would already have the customers by the time someone else realized that you could sell DVD players or diapers or clothes online. They'd had lots of competitors in that field in 1999 and 2000, when companies like Pets.com did deep discounts to acquire customers (though that story is more complicated than it looks: their then-CEO did a retrospective on the business noting that this was a temporary promotion, and that their competitors were doing it, too.[1]) Lyft and Uber realized that they could engineer a change in customer behavior if they priced their product at "a little more than the bus," rather than "a little less than ordering a black car," and ran into the problem that their investors were so taken with the logic of this plan that they underwrote similar models in all the geographies that the original companies had hoped to enter once they'd locked down a
profitable monopoly and could finally expand. Tesla Robotaxi is now doing the same, pricing a ride as “a little more than a subway ride” instead of “a little less than ordering an Uber”. DoorDash infamously offered food from restaurants that hadn't signed up, at a discount to menu pricing, meaning that restaurant owners could literally get paid to order their own food.

That pattern does show up reasonably often, especially when venture markets are frothy. It works well when the product starts out highly-substitutable, and has to have a compelling price to get customer attention, and then gets less so. When Netflix started streaming, its customers already had a device that could play DVDs, many of them probably had substantial DVD libraries of their own (on-demand!), and many households had slow connections. So streaming got tacked on as a free benefit for DVD renters. When Uber started, people were implicitly structuring their plans around the assumption that they'd need to take care of transportation or have a fairly consumer-unfriendly experience hailing a cab or ordering a car service. And early delivery platforms couldn't coast on the growth of a phone-addicted but talking-on-the-phone-phobic audience—if they were going to compete against the easy dinner trifecta of pizza, Chinese, or leftovers, a lower cost was a good way to make this a no-brainer for consumers.

And there's plenty of evidence for this! Uber's mobility take rate was 22% when they went public, and is 30.6% today; DoorDash's has gone from 11% to 14% since their IPO. But a swing of this magnitude looks less like a category of companies that set money on fire in order to smoke the competition out of the market and then jacked up prices, and more like the natural evolution of a business that gets better margins as it matures. This can happen in part because it gets better at capturing available revenue and in part because a denser logistics network with (in the case of meal delivery) more options for end consumers is also a business that's responsible for more of the value creation relative to restaurants and drivers. Every delivery service needs both, but when DoorDash is implicitly giving every restaurant access to years of purchase data, customer lock-in by way of loyalty programs, and comprehensive data on which drivers will pick up an order with six wontons and deliver only four of them (and, for delivery drivers, a blacklist of customers who did or said weird stuff when the order arrived). They're charging more than they used to, and also offering a more valuable service than they used to, and the ratio between the two is one of those dollars-to-subjective-quality comparisons where the only real measure is what the market will bear. If they were going for a monopoly, it didn't work, because Uber has its own rising-take-rate food delivery business and companies like Wonder are on the ascent, acquiring Grubhub late last year, and charging, you guessed it, $0 delivery fees.

It also implies a low discount rate, which is one reason this thesis was so popular in the 2010s. But having a very low discount rate isn't strictly an advantage. If your discount rate is too low, you'll keep everything in t-bills—they hit your hurdle rate, will tend to pay a bit more than inflation, and have basically zero volatility. At the level of a business, it might mean losing a million dollars this year in order to earn $1.1m a decade in the future. The time value of money matters, and the big losses these companies are willing to take in the present require extraordinarily big profits in the future.

But if Meta and Google have spent the last few decades getting us addicted to their services only so that the next time you use their apps, you get sent to a page that lets you pick a monthly subscription plan and enter their payment information, they've been remarkably patient. Or maybe busy running ad platforms that collectively produce about half a trillion dollars in revenue.

The ad business is a good entry point for understanding what's really going on, which is that the winning strategy for a modern monopolist is to systematically under-exploit so they'll have a captive audience they can charge for access to. One of the classic ways to measure how consolidated an industry is is to look at the Herfindahl-Hirschman index, which is just the sum of the squares of each industry participant's market shares. So a completely monopolized industry has an index of 1.0, a market split between four participants has a score of 4 * .25*.25, i.e. .25, and an industry with 100 companies that each have 1% market share has an HHI of 0.01. It would describe nobody's explicit thought process and many companies' behaviors if you model their approach to consumers as maximizing net present HHI points.

All understandable if your biggest motivation in life is some combination of bragging rights and the unbeatable rush of periodically explaining your business to hostile octogenarian legislators. But these companies are big businesses because they use that lock on consumers as a way to ruthlessly profit from the advertisers or sellers who want to reach those customers. They hire smart auction design theorists to design pricing mechanisms that, as closely as possible, mimic the effect of picking up a merchant or advertiser by the feet and shaking them vigorously until their wallet falls out. This model even applies in a B2B context: a cloud platform is a way to maximize the utilization of hardware, which means that what it implicitly sells hardware vendors is access to all the demand they couldn't quite reach if everything were on-prem.

When you're looking at a situation like this, there's an obvious source to mine for wisdom: early 20th century Italian Marxist Antonio Gramsci. It's basically corporate Caesarism, where a logjam among elites is broken when one faction of them aligns itself with the broader populace in order to form a coalition that can beat any competitor. Amazon has more economic heft than any one of its suppliers, and every time it makes shipping a little faster, or commissions another big-budget Prime Video series, or Amazon Basics-ifies some product with a higher-than-usual attach rate to high-margin purchases, it's giving up some consumer-facing margin in order to capture much more supplier-facing money.

This takes advantage of consumers' implicit last-touch attribution model for which brands they think about and care about. To the extent that they care at all, they'll think about the company they're interacting with: they might wonder why Netflix greenlit a show, or how Shein could possibly make money selling them shorts for so little, but they don't think about things further down the supply chain. (Warren Buffett once lamented the economics of the suit-lining company he'd taken control over: "Nobody had ever gone into a men’s clothing store and asked for a pin striped suit with a Hathaway lining." And we certainly don't browse the racks while wondering how our purchase affects these suppliers.

But these big platforms don't want to kill off their suppliers. If they did that, they'd have to get into the much grubbier, less-scalable, and lower-margin business of whatever it was that those suppliers did. They operate at a scale where they don't have to worry about whether suppliers get too much leverage, except in a handful of manageable cases—if you search Amazon for Lego, an extremely recognizable brand and an $8bn company, you'll see plenty of ads for definitely-not-Lego competitors.[2] What the platforms want is for their suppliers to exist in something approximating perfect competition, where any temporary margin advantage gets bid away by competitors, and where cost of capital is relentlessly tracked because so much of it stored in space rented from Amazon.[3] Amazon wants selling on Amazon to be a mediocre business that's easy for new entrants and that doesn't lend itself to margins that are durably higher than peers. Google does not want to be an OTA. They don't want to extract every last dollar from suppliers, just every dollar above their cost of capital.

Sometimes, they can also create a net perceived surplus just by how they label transfers. Travis Kalanick recently noted that customers treat a $1 tip as spending a bit less than $1, and drivers treat a $1 tip like a bit more than $1 when they receive it. On one hand, this is exploiting various cognitive biases for personal profit. On the other hand: if there's something you buy for a dollar that makes you feel like you spent a little less (maybe a handmade gift from a Christmas fair instead of an equivalent-quality mass-manufactured product that's sold at a big markup), and there's some way that someone can make money and feel like they got an extra psychic benefit on top of it that makes that dollar worth more than ordinary income (think of a token payment for some volunteer work, or putting off full-price work to do a project for someone you like priced at the friends-and-family rate), you're both happier for it, even if the transaction involves the spending and receipt of the same dollar.

It's a bit disconcerting to think that some trillion-dollar company's net economic relationship with you might be that it made you poorer in a way you're more than okay with. But those are pretty marginal effects. What these companies end up doing is mostly exploiting a long tail of smaller companies—but also offering those companies a level playing field. It's exactly what that Econ 101 perfect competition model would imply in the absence of monopolies: companies that have to price their products at what the market tells them to. It makes almost every part of the economy functionally less monopolistic, at the cost of toleranting a few big monopolies.

Disclosure: Long AMZN, GOOGL, META, TSM.

  1. If you're unlucky to be in an industry where everyone's selling at a loss to get market share, and you simply can't bring yourself to pivot, the next best move is to raise so much money that you suck all the oxygen out of the room, do a massive PR blitz that makes you synonymous with the category, and hope that you've raised enough that you have more runway than your competition—which is especially hard as you get to the point where you're selling at a negative contribution margin, because then every point of market share you gain increases your cash burn. ↩︎

  2. I have many fond childhood memories of building Legos, and an extensive collection that belongs to and is being continuously augmented and reconfigured by my kids. But the single event that contributed the most to my future Lego-buying was a time when I accidentally ended up with a knockoff, and came to deeply appreciate Lego's quality control and the accuracy of their instructions. I'm sure at least some of the knockoff companies are getting pretty good by this point, but it can take literally a generation for their brand name to catch up if they do, at least unless Lego makes a big mistake. ↩︎

  3. Even though on balance Amazon is dealing pretty sharply with merchants, this is a case where they're making those merchants' lives better. It's hard to have a natural conception of the cost of carrying lots of inventory, but that's a real cost and a drag on business' returns. Renting storage in small increments rather than buying it wholesale just pushes their accounting profits closer to their economic profit by turning more of the cost of capital into a GAAP cost. ↩︎

You're on the free list for The Diff! Last week, paying subscribers read about why media mergers are uniquely hard to get right ($), why the variance in travel patterns over the course of a day makes it hard for autonomous vehicles to replace human-driven ones ($), and how media starts out social and then ends up isolating ($). Upgrade today for full access.

Upgrade Today

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

Find a Role

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.

Find Talent with Diff Jobs

Elsewhere

Liquid Content

Amazon is letting Kindle readers ask an LLM questions about the book they're reading. There are some decent ill-formed criticisms of this (is asking an LLM that different from asking a friend who's already read it, or asking Reddit to explain a confusing part of the plot?). One interesting feature of it is that it's one more way LLMs make the Internet more egalitarian, for better or for worse. There are some books that are written in a deliberately impenetrable style, or with dense references to a cluster of works that readers need to be familiar with to interpret them. Historically, this meant that tackling a given work involved having mastered lots of time-consuming prerequisites, but if you can use an LLM, you can just keep asking clarifying questions until you get an explanation. That still requires attentive reading—you won't get answers about the allusions so subtle you missed them—but this kind of feature means that readers will be able to explore more niches and that authors can be less afraid to write something niche.

Rapid Vesting

Some of the ways companies formally and informally structure hiring is that there's an initial trial period where the company's still trying to confirm that it didn't make a mistake. It's common for employee options to vest monthly—but with the first chunk of vesting delayed a year just in case. This was basically the Silicon Valley Noncompete, keeping labor markets a little less liquid and average tenures a little longer than they otherwise would be. But OpenAI is now eliminating that cliff and having employees options vest continuously ($, The Information). This is very slightly bearish for AI in two ways. First, it means that OpenAI has a slight compensation advantage that other labs will have to match in some way, but second, it implies that as far as OpenAI is concerned, AI talent will be constantly hopping from one company to another, instead of staying at one clear winner.

All-Time Highs

Cisco has finally surpassed its early 2000 share price record ($, FT). (In case you're wondering: up 50.7%, or 1.6% compounded including dividends, still down 20% adjusted for inflation.) Cisco was a cyclical company that went through such a great cycle in the late 90s that investors started to extrapolate it forward indefinitely. They got the big-picture bet exactly right, but a small haircut to very high growth expectations will produce a big shift in the terminal value of the business, and that's going to hit market value even harder if the multiple rerates because investors realize it won't grow monotonically. Cisco ends up being a different kind of capital allocation success story: they're a company that took advantage of amazing growth opportunities, hit the point where it couldn't reinvest much, and, over the last two decades, bought back a third of its shares outstanding.

Tariffs and Predictions

If you look at a log-scale chart of the S&P 500, you can point to plenty of times in history where everyone was panicked and it was a great time to buy. And you can tell yourself that you would have bought then. Some people do, whether it's because they have a list of things they'll buy when they get cheap enough and they buy when everything's cheap enough, or because they're truly committed to buying when there's blood on the streets. But this retrospective on the economic and market impact of tariffs shows why that's so hard ($, WSJ). There was a wild disconnect between what CEOs and economists were saying and what Trump was saying, and it turned out that both sides were wrong and tha tthe market powered higher in part because so many products were exempt from tariffs and in part because the AI trade kept working. If you buy because people are panicking, the real bet you're making is probably not that they're wrong to panic, but that they're focused on the worst headline and that something else will catch people's attention soon enough.

Moving Down the Funnel

Google is testing out property listings directly in search pages (Zillow is currently down ~12%). Real estate is a very search-friendly market, in that there's a huge amount of heterogeneous supply. It's also a market that's good for vertical search, because the right format for a search page is so different from general-purpose queries—at some point you need to add a dropdown for number of bedrooms and bathrooms, and data quality is very different when every search result needs to be an active listing. So there's a big lift both from a frontend perspective and in terms of aggregating and updating data—not to mention the strategic question of how much this will hurt their revenue from real estate advertisers and from vertical search in general. It'll be interesting to watch how Google handles this, but it's easy to over-index on small tests.