Subscribe
Share
Share
Embed
The Diff is a newsletter exploring the technologies, companies, and trends that are making the future high-variance. Posts range from in-depth company profiles, applied financial theory, strategy breakdowns and macroeconomics.
## Ad-Supported Platforms are a Unique Antitrust Problem
To most of the people who interact with the big ad-supported platforms, they're basically an infinite free-money glitch. These companies' employees are some of the best-paid people on earth, their investors have been well-rewarded for backing them, and the typical user experiences either an affordable online ordering experience or free video, social media, or search (often all three!).
Something doesn't add up. Where is all that money coming from!?
The answer is that these companies are an ideal test case for an increasingly common business meta-model, where companies try to create a consumer surplus at one end in order to maximize their negotiating leverage for capturing the producer surplus everywhere else in their supply chain.
This is a peculiar model. Typically the way you'd think about a monopoly is to imagine a dynamic market and then take away some of the dynamism. Shrink the number of sellers in a given market, and you’d eventually have a number small enough that they can coordinate to raise prices, or, with a single seller, set whatever price they want. And the big platforms have some room to do this: people have been complaining that Google has been getting worse for over a decade, and to the extent that they're right it hasn't gotten so much worse that a meaningful number of people have switched. Similarly, lots of people know YouTube is full of ads because they spend a lot of time watching videos on YouTube and getting interrupted by the ads. Amazon is in the paradoxical position where it's ostensibly cluttered with junk from brands unpronounceable in any human language, but also the first place you'd shop across many different categories. So, in theory these companies could try to monetize their lock on certain consumers by turning to you, the consumer, and demanding whatever price they can get away with.
Instead, these monopolists are cynically looking for opportunities to give you more: Amazon wants to offer the Universal Bundle of convenient access to consumer goods plus a decent streaming service and some music; Google decided not to let AI exist as a separate category from search, when they'd already been converging for a while; Meta conceded that short video was a meaningful category and that they had no choice but to compete in it (and also dropped a few tens of billions on making the metaverse the next big category).
The predominant way these companies use their monopolistic power is at the other end of the business: all that generosity to the consumer is monetized through ruthless exploitation of their business counterparties. It makes sense: businesses respond to marketing based on return on investment, but consumers just don't model the world that way most of the time. And even when they do, it's rare for consumers to find a (healthy, non-vice) category where, once they've sampled it a bit, they'll happily increase their spending. But for a business, that's the whole idea: figure out which kind of spending leads to the most contribution margin bang for the buck, and max it out. If a consumer is spending 50% more on the same category each year, something is going wrong; if a business is spending 50% more on something every year, the usual bet is that things are going great.
This dynamic gives platform operators a few imperatives:
1. They need to care about users, generally some mix of the average existing user and the theoretical marginal user.
2. They want to support as broad a mix of activities as they can over time. The more points of contact they have with their users, the better they understand them, and the better the ad targeting will be.
3. They want advertisers' businesses to be as commoditized as possible. For them, the ideal business is one that's high-margin, but with minimal competitive differentiation, such that the main way competitors distinguish themselves from one another is how much they're willing to bid for an audience (e.g, personal injury, insurance, DTC, increasingly sports betting, etc.). This means that they want to make things easy for new entrants and harder for incumbents.
4. On a related note, the platforms want to avoid letting people go direct, for whatever definition might apply: if someone is clicking a search ad or tapping a social one, that interaction is worth more if they've already made up their mind, and worth even more if there's no chance that looking for the product somewhere else would reveal that it has a lower price.
The best way to understand the result of this is to consider the funniest section of Tim Wu's The Age of Extraction. He talks about the experience of someone who started selling pomade on Amazon, got great results, ultimately depended on Amazon for 90% of his business, and then found that Amazon kept raising fees and cutting into his margins, while competitors kept competing on price and bidding up ads. Wu is a major figure in the neo-Brandeisian movement, which wants to limit how much companies can take advantage of pricing power, and whose members tend to be progressive on other issues. And here he is, looking at a market where incumbents can't slack off, where everyone competes on a level playing field, and where the taxes the platform levies are precisely calibrated based on ability to pay—and hates it! Amazon basically implemented the neo-Brandeisian wishlist within their platform, and the neo-Brandeisians aren't having it!
Some of this is the result of deliberate planning, but a lot of it is the accumulation of a natural evolutionary process. You don't have to set out to create the closest approximation to perfect competition and optimal taxation. All you have to do is ask what the best way to increase ad revenue is, and think a few steps ahead.
It's no wonder it's so frustrating to regulate these companies. You can talk a good game about powerful, unaccountable businesses, but it rings a little hollow when they're all in the business of providing an accountability layer: they spend a lot of effort on policing the ads they run, and in an ad-driven business, spam is not just a bad user experience but direct competition. These businesses have accidentally ended up running a kind of populist strategy: 99% of people experience Amazon as increasingly convenient and reasonably cheap, and a small fraction of them are small business owners who struggle to survive in Amazon's world of high taxes and perfect competition.
This model leaves plenty of unanswered questions. For one thing, the cost of ads is economically similar to a tax from the perspective of the business buying the ads, but it's equivalent to a tax in the sense of how it's spent. Right now, those taxes are invested in building datacenters; if your preference is for more economic redistribution to the elderly rather than better and cheaper AI, you might prefer that someone other than big platforms aggressively tax small businesses. On the other hand, suppose you take the view that these companies have been biding their time for the last few decades, only pretending to vigorously compete against one another, relentlessly attempt to commoditize one another's products, and dump vast sums into speculative new products, etc.—one of these days, the thinking goes, they'll start taking advantage of consumers and raking in profits. But if they actually do that, instead of stacking up capex that will give them a big depreciation shield, they'll be paying taxes! (And even if they spend enough to reduce their tax burden, that shows up as tax revenue somewhere else—an incremental GPU sale has about a 75% gross margin, so those retained earnings are still generating tax revenue somewhere.
The other unanswered question is how far this can go. A broad platform that puts itself between millions of businesses and billions of consumers wields enormous influence, albeit in a subtler way than earlier generations of maximum-breadth businesses like monopoly newspapers or the big three TV networks. Consolidated media companies have a lot of discretion about what does and doesn't count as news, and thus perform an agenda-setting function for legislature. Search, social, and e-commerce probably have more influence on people's lives, in the aggregate, but it's through subtle differences in their behavior, like how they weight aggregate popularity against individual preferences in product recommendations, or which kinds of content they're willing to host but not willing to make easy to find. But this is also more diffused internally. When the big three networks were dominant, it took three people saying "that's not news" for a story not to be news. But now, things are less autocratic, and the collectively influential decisions are happening more frequently and lower down the org chart.
This is a historically unusual setup. There are surprisingly powerful businesses that devote a great deal of their energy to either making consumers better-off in order to exploit more businesses, and trying to kneecap one another's efforts to do this. The companies in question are a bigger share of the economy every year, and a bigger share of effective state capacity. Which means that if there existence is, or will lead to, a problem, it needs to get fixed—but also, their size and disproportionate share of economic growth means that botched regulations make the world a lot poorer.
----
Disclosure: long GOOGL, META, AMZN.
## Elsewhere
### Brittle Buyers
One of the interesting features of levered ETFs is that buying them expresses a preference for more volatility, and actually creates that volatility, too, because maintaining their leverage means buying when the price rises and selling when it falls. Less mechanically, they add volatility in the sense that some of their holders will be using a bit more leverage on top of that, and adding to positions as they're able to. So it's interesting to see that retail investors have been buying the semiconductor dip using levered ETFs. THe memory trade was already volatile—cyclicals in a supercycle tend to be that way—and now it will be more volatile still.
### Rates
Global interest rates are at pre-Great Recession levels. It's sometimes astonishing how few years of heavy borrowing it takes to earn a decade of subpar growth. But as The Diff argued shortly after the release of ChatGPT, Low rates make speculative research a relatively better deal because it shrinks the gap between cash flows in 2030 and cash flows in 2040. Which means that the converse is probably true, too: a higher-rates environment implies more deployment of existing technologies and fewer big new developments. There are still plenty of things that germinated in a low-rates environment and then reached profitability before capital got expensive. But scaling existing technologies is easier to underwrite than developing new ones. Fortunately, the current crop of ready-for-scale technologies is unusually promising, with lots of complementarities between more compute, cheaper energy to power it, and more robots to put it into action. In fact, there’s been some interesting work that argues the higher interest rates go, the closer the market believes we are to superintelligence or truly “transformative AI” rapidly diffusing across the economy. (Though for now, AI borrowing is a single-digit percentage of the world’s annual net increase in debt, though it’s rising fast.)
### Substitution Effects
Claude Fable isn't seeing heavy adoption among corporate spenders. In one sense, that's perfectly understandable, since it's an unusually expensive model. On the other hand, the typical pattern has been that users promptly upgrade to the best available model, at least for their most valuable tasks. But at some point, the goal of any big lab will be to have something at every price/quality tradeoff, with the best models partly serving specialized uses where they're worth it, and partly functioning as a tool for employee and customer retention. Customers aiming for either best or cheapest are going to be picky, but the ones who know they can upgrade or downgrade without switching vendors will tend to be happy.
### Demand
Nvidia plans to raise prices of Blackwell and Vera Rubin by 17%. It's revealing that the GPU shortage has lasted as long as it has, since Nvidia could have solved it at any moment by raising prices. Which implies that, to Nvidia, the cost of letting the high bidder get as much compute as they want is higher than the revenue they'd forgo by not charging a market-clearing price. But if there are more competitive hardware products for inference, and if the providers of inference are fragmented, Nvidia might conclude that their ability to control the market has eroded but that the risk of a single centralized winner is lower.
Disclosure: long NVDA.
### Trades
Citadel is mostly out of the Situational Awareness trade. One of the advantages big funds have is their balance sheet, specifically their ability to get good terms from prime brokers, allowing them to periodically warehouse a significant amount of risk. It's possible for a fund to run some strategies that have an indirect payoff, where they aren't great on their own, but generate significant commissions, which encourages their brokers to offer them credit on more generous terms, which they can then use to execute financing-driven trades like the treasury basis trade. And you never know which market will be the one where the willingness to wire a huge sum overnight and then accept a few days of elevated volatility pays off. So the incentive for these funds is to diversify their strategies a little more than they otherwise would, and to support some subpar businesses longer than those businesses can justify on a standalone basis. A wide range of strategies is a great complement to a balance sheet that can periodically swing a lot of capital into whatever trade presents the best immediate opportunity.