|
In this issue: - In Defense of the Ad Economy—Ads are annoying, except as compared to the alternative. And this is true at many levels. Not only do ads elegantly solve some pricing problems, they actually create a continuous economic incentive to hunt down spammers and even to care about the quality rather than quantity of time spent.
- Wage Elasticity—If wages are showing the impact of lower immigration, they aren't doing so in a way that validates anyone's story.
- Coopetition—Being direct competitors doesn't preclude a fruitful supplier/customer relationship.
- Capital Structure—Every time a new company borrows to buy GPUs, the argument for lending against GPUs gets a little weaker.
- Market Breadth—The revenge of the S&P 493.
- Playing Chicken—Trump is negotiating with Canada, and Canada is negotiating with the stock market.
In Defense of the Ad Economy
If you're going to enjoy the fruits of someone else's labor and capital, you're going to get it in one of three ways:
- It's a genuine labor of love, a passion project, a friendly gesture, etc.
- You're paying with money.
- You're paying with something else, often (and increasingly) a combination of your attention and the data that allows that attention to be sold to the highest bidder.
Of these models, #1 is of course wildly popular, but hard to measure and hard to scale. The second model, a straightforward transaction where you pay for something and get frictionless access to it, is theoretically popular, but in practice there's often low willingness to pay: in the EU, Meta started offering a paid, ad-free version of their app, but fewer than 1% of users signed up. That version was priced at €10-13/month, or around twice their revenue per active user, but presumably the users who are most annoyed by ads are the ones who place a high value on their time. (The price has since been cut to €6-8/month.) Snapchat’s story is a bit more encouraging, with 16M subscribers to Snapchat+ which ranges from $4 to $14 a month, and ad free is only available at the higher tiers. That’s 3% of their DAUs, and the most successful social subscription offering out there (for reference, X Premium has ~700k subscribers). And when Netflix launched an ad-supported tier, they said it actually produced more revenue per user than the paid tier, at least in the US, though, to be fair, they've since stopped saying that and now talk more about signups, with the revenue contribution happening later.
It's a bit strange that, particularly in the developed world where money is more abundant than time, it's easier to get people to pay for some services with time and attention rather than money. There are plenty of exceptions, of course, even in media. Movie theaters don't monetize blockbusters by interrupting them every few scenes to show you exciting new shampoos and car insurance. Publications targeting a business audience can either a) expect their buyers to expense the purchase, or b) let them anticipate an ROI. (There are times when failing to read the Wall Street Journal is expensive.) Obviously, most physical goods with high marginal costs can't be monetized with ads, and since ads only produce a return on investment when they lead to a purchase, large swathes of the economy have to have a model other than ads if the ads are going to exist at all.
But there are a few reasons ads end up being fairly economically efficient:
They embed automatic price discrimination, at least if you flex the definition of that term to cover revenue per user instead of revenue from those users. The people watching TV, searching, or scrolling their feeds all vary in their propensity to spend and in how responsive they are to advertising. Their incomes vary, and the way they dispose of those incomes also affects how valuable it is to advertise to them. For some of these users, the platform in question may have identified them as being in the market for something (a new car, a vacation, an upgraded phone, etc.) at which point they're much more valuable to an advertiser. And if there's not much specific information about them from their browsing habits or other data sources, there's the implicit information that they're looking at social media right now—someone who's doing that is clearly looking for a way to burn up a little free time, so casual games end up being the universal reserve bidder for otherwise-untargetable inventory.
The people who tap on ads and click "buy now," and the aggregate populations whose in-store purchases rise in response to geo-targeted online brand ads, are essentially subsidizing the service for everyone else. If it's a user-generated content platform, this is one of many forms of cross-subsidization—the small fraction of users who create content, and the even smaller fraction whose content is responsible for the majority of the viewing hours, are basically subsidizing the entire setup by contributing their labor. The parts of the audience that advertisers want to reach subsidize it financially. And then everyone else gets a site that can afford more features, a snappier app, more content, and all the rest.
Ad-based models are also a form of usage-based pricing: use the product more, and you see more ads; show a demonstrated tendency not to curtail your usage in response to higher ad load, and you see even more ads. And, as you spend time on the site, you're also giving them information that increases the revenue those ads generate, through better targeting. Usage-based pricing is always more popular in theory than in practice, because there's a cognitive load to asking "is reading the next article in the New York Times worth 73 cents to me?" Whereas the cognitive load of just clicking on the article, and scrolling past the ad that says—wait a sec, I'd save how much on car insurance? Hold on...
The usual cost structure of ad-supported businesses is that the incremental profit from one more ad view is quite high, so it's not as if they literally have to ration viewing. But those marginal economics are the result of high fixed costs, and the level of fixed costs a business undertakes is partly a function of how much high-margin incremental activity that fixed cost investment will create. At launch, YouTube videos were 320×240 pixels, and going from that to full HD, 60 frames-per-second video is not cheap from a storage or bandwidth perspective. But it's worth it to YouTube if those quality improvements lead to more usage, especially if that usage directly feeds into more revenue.
The last big reason that ad-supported models win, especially for communications tools, is that they always exist. The only question is who's taking advantage of them and what their incentives are. Any platform that aggregates human attention will be attractive as a way to manipulate that attention. This doesn't always happen for commercial purposes—lots of people go into journalism in order to have an impact on what their audience will believe, not for the money, and while this is not a commercial interest, it's analogous to an ad in that it's some third party's effort to hijack your attention to change your beliefs and behavior.
Online, the most common form this takes is spam. Yishan Wong has a great thread on this dynamic, where he points out that if there's an audience, marketers will try to reach it either by paying directly to be put in front of their audience, or by paying indirectly for product placement, buying old accounts and using them to shill products, creating upvote farms, etc. There's a wonderful two-sidedness to this: ads price out the lowest-return form of spam, and then they make the rest of the spam more expensive to the platform because that spam is both replacing paid revenue and degrading the quality of the service. Spammers don't really ask themselves what would happen if they posted so much garbage they made a site literally unusable, and if they did ask, the answer would be: that site will die, someone else will start another site, and we'll spam that one instead. So the spammers implicitly have a higher discount rate: there's a revenue-to-future-usage tradeoff that they'll take, but that the platform owner wouldn't take. Even if there's a revenue cost to cracking down on aggressively spammy actors, there's a present value of future revenue gain from doing it well. It was a headwind to Facebook when they decided that Zynga, a material source of revenue, was also cluttering everyone's newsfeed with messages about Farmville strawberries and the like. So they reduced the reach of that kind of message, a financial blow from which Zynga never recovered. There is an optimal level of this kind of status update, but it's much lower than the rate that's optimal for Zynga.
So, if you broaden the definition of "advertising" from the content that an advertiser pays a publisher to show you (at the expense of the content), and say that it's anything that anyone wants you to see in order to manipulate you, ads on platforms actually lower the ad load, and improve ad quality. The ads you see are like the unpleasant pinprick from getting a vaccination—a little unpleasantness now that spares you from more later on.
In one sense, this is a perfect alignment of incentives: they want the service to be better, so you'll use it more, so they'll make more money. On the other hand, it assumes that people really want the service to be better, and the ad-supported model is already at least partly based on the cognitive limitation that we're more willing to give up time than money even though time can be converted into money and there are other things we might prefer to do. Especially if a service hijacks attention, and turns "let me just check one thing really quick" into a marathon viewing/doomscrolling/gaming session. The alignment isn't perfect, because while the market is very good at giving you what you want, it doesn't have a good way to help you choose what you want, at least until we invent Ozempic for Attention.
The platforms can reasonably ask: are you not entertained? You're choosing to look, and in some sense choosing to keep looking. But they also can, and do, ask themselves about the quality of different kinds of engagement. Employees of these platforms are paid heavily in equity, and there's a rough correlation between how important the decisions they can make and how much their compensation is a function of the stock going up over long periods. So the optimum they approach—slowly, and they're far from it now—is that your attention is their asset, and while hyperbolic discounting might lead you to spend time in ways you regret, their equity is ultimately valued based on discounted cash flows which will continue into eternity. Advertising, then, is a way to give someone a stake in annoying you just enough that you become part of one of the world's great gushers of wealth, but not so much that you quit—and, in the bargain, shielding you from other commercial or otherwise manipulative messages unless they get a cut of the action.
Disclosure: Long META, GOOGL.
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - Ex-Citadel/D.E. Shaw team building AI-native infrastructure to turn lots of insurance data—structured and unstructured—into decision-grade plumbing that helps casualty risk and insurance liabilities move is looking for a data scientist with classical and generative/agentic ML experience. You will develop, refine, and productionize the company’s core models. (NYC, Boston)
- An AI startup building regulatory agents to help automate compliance for companies in highly regulated industries is looking for an applied AI research engineer to work with large amounts of data and build LLM powered enterprise workflows. 4+ years as a software, ML, or data engineer/scientist required. (NYC)
- A Google Ventures-backed startup founded by SpaceX engineers that’s building data infrastructure and tooling for hardware companies is looking for a software engineering manager with 7+ years experience building large scale distributed systems. (LA, Hybrid)
- Thiel fellow/ex-Ramp founder and team are hiring a high energy, full-stack engineer to help build the automation layer for the US healthcare payor-provider eco-system. (NYC)
- An OpenAI backed startup that’s applying advanced reasoning techniques to reinvent investment analysis from first principles and build the IDE for financial research is looking for software engineers and a fundamental analyst. Experience at a Tiger Cub a plus. (NYC)
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. 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. Elsewhere
Wage Elasticity
The Economist has a piece on the labor market impact of lower immigration ($, The Economist), which is oddly definitive-sounding given that what it really brings up is a mystery: in jobs where a high percentage of the workers are immigrants, wage growth was higher than average, but since then, it's been lower. If immigrants are crowding out native workers, a sudden shortage of them should lead to higher wages for those same workers. There hasn't been much time for a big effect, and there's other noise in the data—jobs that recently experienced rising wages might be more sensitive to slight economic slowdowns, since hiring is somewhat forward-looking. Even if jobs are off-the-books, there should still be a labor market impact. If someone is getting paid in cash while using an account that belongs to someone with work authorization, for example, they still leave their delivery network a driver short if they stop working. The data so far aren't consistent with an immigration-suppresses-wages story, but they're also inconsistent with an immigration-raises-wages story. In that model, too, wages for immigration-exposed careers would go up relative to the rest. The usual argument is that even unskilled immigration is beneficial, because many unskilled jobs save the labor of skilled ones (meal delivery, lawn care, etc.). If that's true, large-scale deportation would show up economically as a decrease in wage growth for the non-immigration-sensitive sectors (who use immigrant labor to give themselves more time for higher-wage labor), and an increase in demand for immigration-sensitive jobs (the people who decide to replace this labor now find it's scarcer than before).
So while the wage numbers are interesting, they don't tell anyone's story particularly well at this point.
Coopetition
One of Google's advantages in AI is that it can afford riskier buildouts than its peers, because it has so many more reserve bidders for infrastructure than its peers. It can monetize directly through subscription AI products, use AI summaries to improve search revenue, or rent it out to someone else through their cloud platform. "Someone else" turns out to include OpenAI ($, The Information). That's a compelling symbol of Google's freedom to take risks on capex: a company that's increasingly their direct competitor is willing to give them additional revenue. Google just has to be less desperate than that to have the upper hand.
Capital Structure
Meta plans to raise $3bn in equity and $23bn in debt to fund datacenters ($, FT). Lenders are surprisingly eager to treat carefully-air-conditioned buildings full of GPUs as excellent collateral, which has been very good to the founders of CoreWeave among others. But one reason for this is that there are so many potential buyers for this capacity. Every time they lend to another GPU buyer, though, they're making the average GPU buyer more financially encumbered. That's part of the process of finding a supply/demand equilibrium for some kind of capital asset that can be used by many buyers. Leverage speeds up the deployment process, but also means that there can be bigger air pockets in demand, and that the buyers who gave those GPUs value as collateral already own what they need.
Market Breadth
Technical analysts sometimes talk about "market breadth" as a sign of the sustainability of a rally. Like many things pure technical analysts care about, a) the historical data on it mattering is fairly weak, but b) there are actually existing traders who use heuristics like this to make anecdotally good calls. One economic intuition around it is that high market breadth, i.e. most stocks and industries are up, implies that the economy overall is doing well. Low breadth and a rising market implies that there's a particular industry or cluster that's experiencing most of the growth. And if that's happening, it means that the economy is being somehow restructured—in the 70s and early 80s, that meant an economy being restructured around more expensive oil, which was great for the energy sector and a pain for everyone else. In the mid-2000s, it meant an economy being restructured around more levered balance sheets and more global flows of capital, which did not work out especially well. And in the post-2022 rally, the big gais were largely from AI-exposed companies, and anyone who owned the "S&P 493," i.e. big companies other than big tech, was behind. But, during the post-tariff rally the biggest winners have been non-tech companies, with dollar stores as a standout. If you're betting on very long-term growth, you're probably more excited to see AI leading the market than deep-discount retailers. But if you're worried that that AI growth is going to lead to 20% unemployment for white-collar workers, ever-higher energy prices, and crazy permutations in org charts and supply chains, it's a relief to know that boring retailers still have it.
Playing Chicken
On Friday afternoon, the US ended trade talks with Canada because of their digital services tax. The tax was promptly nixed over the weekend. It's an interesting update on the Trump-always-chickens-out thesis; the first announcement happened intraday, and while stocks dipped briefly, they rallied into the close. The weekend meant that there wasn't much demand for temporarily market-moving rumors, so there was enough time for Canada's government to decide that the tax wasn't worth it. Trump is correct to see a digital services tax as, in effect, a tariff that mostly hits American companies, but one that's more efficient from a revenue-collection standpoint because the companies probably won't stop providing their services in foreign markets, even if it's less profitable than before. And Canada's probably correct that, if Trump names a single sticking point, fixing that problem makes it a lot easier to get a net beneficial deal.
|
