Ad-Based Platforms Have Mostly Solved Optimal Taxation

In this issue:

• Ad-Based Platforms Have Mostly Solved Optimal Taxation—A consumption tax with a tax credit for producing positive externalities, all used to invest in making services widely available for free, describes both a pretty ideal tax system and the business model of some of the world's most profitable companies.
• The Infrastructure Put Option—It's okay for some labs to overshoot, as long as the industry as a whole undershoots.
• Monetization—A first-class plane ticket is gradually becoming something people actually buy.
• Headless SaaS—Separating what Salesforce is from what it does.
• Gazumped!—If you create a liquid market, expect hedge funds to start making opportunistic trades there, even if that market is in hedge fund talent.
• Bits and Barrels—When you own an appreciated asset whose price will be lower in a world where you can actually sell it.

The Diff April 20th 2026

0:00

/829.622857

1×

Chat with this post on Read.Haus.

Ad-Based Platforms Have Mostly Solved Optimal Taxation

Globally, around thirty cents of every dollar of output gets gets captured and redistributed through the tax system, and of course everyone on earth either feels that they'd get more of what they want if either their taxes were lower or someone else's were higher, so a lot of energy goes into discussing the ideal tax system. What economists usually say is that the best system—or, depending on their beliefs and their general temperament, the least-bad system—is one that minimally distorts behavior, except perhaps by distorting it in a prosocial direction by taxing vices and annoyances.^[1] The party line among many economists is that the rough rank-ordering is that you want to tax property, then consumption, then income, then corporate profits and capital gains.^[2]

Optimal tax policy tends to get hashed out the usual way such issues are: lots of small interest groups lobbying for their preferred policies, compromises designed to make the current budget impact look better while sticking some future politician with the problem of either raising taxes again or justifying a bigger-than-expected deficit. A technocratic approach would look very different, and would be wildly unpopular. Sometimes, the argument for a tax's efficiency is exactly the one that explains why it's unpopular.

For example, economists tend to like property taxes. Homeowners and other real estate owners passively benefit from economic activity that takes place near them. San Francisco homeowners, for example, have made a lot of money thanks to the efforts of the tech industry (mostly staffed by renters) who make the city such a desirable place to live, and have their very own property tax rule that has the same distorting effects as a capital gains tax, i.e. it's a financial incentive to stay in a house that isn't the best fit for you, but lets you continue to pay minimal property taxes.

The unfairness isn't a huge deal, here. Plenty of other people passively benefit from economic growth in a roughly similar way; geographically-constrained service providers, from barbers to wedding planners to neurosurgeons, also have more pricing power in places where everyone's well-off. The problem is that low property taxes basically disguise the social opportunity cost of not having the best match between people and places. If you're a talented screenwriter, and you don't live in LA, that's a big negative for your career and at least a small one for society overall. And if there's someone in LA who doesn't specialize in the same industry that LA itself specializes in, that, too, is suboptimal. But if the message property taxes send you is "you should probably move back to your hometown and get a normal job," or worse "you should probably leave your hometown to make room for someone doing something that doesn't sound like a real job but that pays vastly better than what you make," that message isn't welcome.

And the US can afford to operate like this. We're a very rich country, and will remain so even if the geographic allocation of talent isn't optimal. But it would be a good idea to keep track. One fairly simple way to price this is to estimate some optimal property tax rate, and then book the actual revenue as a tax-plus-rebate, so there's a line item in the budget for how much we're subsidizing people to stay living somewhere. This exercise can be a little more rigorous if you do it for rent control, and treat the gap between what people pay and the market rent as a subsidy. ChatGPT estimates that if everyone paid market rent, and the city government wrote them a check for the difference between their rent-controlled rent and market rent, housing subsidies would be about 5% of NYC's budget and 20% of SF's.

You could make the property tax even more unpopular with a Harberger tax. That's a policy that addresses two problems at once:

1. All houses are different, so while you can estimate their value based on comparable homes, there will always be slight gaps based on qualitative factors.
2. The value of a single house also varies depending on who is valuing it. Homes have sentimental value, and that should be reflected in the decisions we make about them.

The specific implementation Harberger suggested was that everyone gets to value their home for tax purposes, and must accept an offer to buy the house at that price. In other words, sentimental value can have a dollar value attached. This has the advantage that if someone lives in a $200k home, and likes it, they can just agree to be taxed as if it's worth $250k. And as soon as you put this into practice, you end up telling people that the more happy memories they have, the more they'll be taxed, and you're implementing this possibility by letting institutional single-family landlords—or, as politicians put it, "BlackRock,"—buy them out.^[3] It's hard to construct a less popular political platform.

A savvy property tax regime might take this one step further, and adjust the valuation of homes in a given neighborhood based on who's a good neighbor. Now we're in even more unpopular territory—Blackstone/Rock gets to buy your house because you don't smile when you say hi to your neighbor! But this is a property tax-maximizing approach: people like living in a neighborhood where their neighbors are friendly, packages don't get stolen, there are baby- and pet-sitters a few doors down, etc. Once they like some resource that's scarce, there's room to price it accordingly.

This is fun as a thought experiment, though it would be basically impossible to implement. But there is an environment that basically functions this way, where the equivalent of real estate is always offered to whoever is some combination of the highest bidder and the most responsible user: that's how online advertising works! In fact, if you described online ads as a tax system, it would sound pretty ideal:

• The main thing that gets taxed is consumption, rather than income; there are low-margin dropshippers in competitive categories who are constantly trying to find new tricks for getting their ads clicked, and there are brand advertisers who know that a little nudge can convince someone to upgrade their iPhone or drink a Diet Coke with lunch.
• Platforms are profit-maximizers, but they're trying to maximize the present value of future profits, which doesn't just mean that they have an incentive to avoid juicing revenue temporarily, but that they have an incentive to cut their discount rate by occasionally taking a revenue hit (e.g. Reels) or a COGS headwind (generative AI search results) to stay relevant.^[4]
• Ads are part of the content, and are judged accordingly. The more a given ad fits in with the rest of your Instagram feed, the less that ad detracts from the experience, and the less the advertiser will need to pay. This has been true in an approximate way for decades, ever since the introduction of the quality score in search. Over the ensuing decades, platforms have accumulated a huge amount of data on the impact of different ads on long-term user behavior. A quality score serves the same function as policies like child tax credits or opportunity zones: if $1 of forgone taxes produces at least $1 worth of the kinds of things tax dollars are spent on, that's a win.

As more of the economy is mediated^[5] through large, ad-supported platforms, run by profitable, tax-paying companies, we're slowly imposing an economically sophisticated and technically impressive tax collection system on a growing share of the economy, with these outsourced tax collectors keeping much of the upside by kicking a bit back to the US government.

The US government captures a small percentage of the upside from this, at least in direct terms. But that's because the platforms use ad revenue to fund public goods. A powerful search tool that can access scanned copies of basically every public-domain book, a tool for updating all of the loose ties in your life about important life updates, a video service so profitable that there's no particular reason for them not to let you use their platform to host stuff like a video of your kid's talent show performance. All of this is the kind of thing that could plausibly be provided by governments, which already support information-sharing systems like libraries, universities, and schools. In fact, one of the first big hits in the consumer-facing networking category was a government project. But we've arrived at a split where it's so straightforward to collect good information on taxpayers' willingness to pay, and it's so profitable to make them pay it, that this part of government is a private-sector initiative.

Disclosure: long GOOGL, META

1. A growing share of that 30% of economic activity involves using taxes to pay for healthcare, which means that the fiscal case for those Pigouvian taxes is stronger. If you're responsible for your healthcare, it doesn't cost your neighbors anything if you're sedentary. But as healthcare-for-other-people by way of taxes becomes one of the biggest line items in everyone's budgets, that behavior becomes their business. ↩︎

2. It's more a matter of convenience and bookkeeping that we treat capital gains as a taxable event. A capital gain is when the present value of the (after-tax) net present value of an asset is higher than it was when you bought it, in nominal terms. But even if you hold on to equities for a long time to defer capital gains, you can treat the deferred taxes as an interest-free loan. But in that case, we're basically saying that we assess the corporate income tax twice, and that one of those ways involves providing subsidized credit to people only on the condition that they hold on to investments they otherwise consider suboptimal. So if we reduced capital gains taxes and raised corporate income taxes so we were taxing the same share of corporate profits, we'd have a system that was simpler and better-aligned. ↩︎

3. There is a fun case to be made, of the sort that makes people hate you, that this is socially optimal. Society's tolerance for inequality is mostly a function of what can be redistributed from the haves to have-nots. So wealth inequality is a big deal, education inequality is popular, but nobody gets worked up about athletic or beauty inequality. Is it fair that some people have many cherished memories associated with their home, and for other people a home is just a place to sleep and watch TV? It's impossible to say. But in terms of memorable pleasant experiences, this tax does reduce the inequality between the had and the had-not. ↩︎

4. The biggest platforms used to be less purely fixated on maximum profits, both because they had a long monetization runway and because their founders tended to be more excited about product and engineering questions than selling ads. But now, those older applications are mostly a source of cash for AI bets, so they'll probably accept more sources of profit that they would have objected to in the past. ↩︎

5. Meta’s economic contribution team found that Meta Ads are linked to ~$548B of US economic activity annually, which is, as of last year, ~2% of US GDP. ↩︎

Diff Jobs

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

• Well-funded, frontier AI neolab working on video pretraining and computer action models as the path to general intelligence is looking for researchers who are excited about creating machines that learn from experience, not text. Ideally you have zero-to-one pre-training experience and/or are a high-slope generalist who’s frustrated that the big labs aren't doing this. (SF)
• High-growth startup building dev tools to help highly technical organizations autonomously test/debug complex codebases is looking for a senior design engineer to own their design system and build the visual abstractions customers rely on to simulate their software systems, find bugs, and quickly remediate them. A compelling portfolio, a rare blend of design and engineering chops, and a deep understanding of how the internet and browsers work required. (D.C.)
• Series A startup building multi-agent simulations to predict the behavior of hard to sample human populations is looking for researchers and engineers (ML, platform, infrastructure, etc.) to improve simulation fidelity and scale the platform to hundreds of millions of simulation requests. Problem-solving and genuine interest in simulation matter more than pedigree. Experience working with languages with an algebraic type system is a plus. (NYC)
• A Fortune 500 cybersecurity company with decades of proprietary security data is running an internal incubation with a pre-seed startup mentality and a mandate to build something new in AI. They are looking for a founding engineer who can ship fast, an engineer with a security background who’d be excited to contribute to OpenClaw’s security efforts, an AI researcher, and a generalist (ex-banking/consulting/PE background preferred) who wants to wear a bunch of different hats. Comp is FAANG+ and cash heavy. If you want to build something new in AI, but also need runway, this is for you. (SF/Peninsula) 
• A leading AI transformation & PE investment firm (think private equity meets Palantir) that’s been focused on investing in and transforming businesses with AI long before ChatGPT (100+ successful portfolio company AI transformations since 2019) is hiring experienced forward deployed AI engineers to design, implement, test, and maintain cutting edge AI products that solve complex problems in a variety of sector areas. If you have 3+ years of experience across the development lifecycle and enjoy working with clients to solve concrete problems please reach out. Experience managing engineering teams is a plus.

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.

And: we're now actively deploying capital into early-stage companies through Anomaly. Our focus is on defense, logistics, robotics, and energy. If you'd like to chat, please reach out.

Elsewhere

The Infrastructure Put Option

The capital allocation challenge for AI labs is that they tend to need an order of magnitude or so more compute each year, and have to buy it without quite knowing what they'll need it for once it's available. This would lead to the classic cyclical dynamic where slow-feedback supply overshoots demand, prices come down, the weakest players in the industry shut down, and behavior gets more rational. The only problem is that so far, it's been hard to actually buy more compute that can be put to good use. The closest you might be able to get today is xAI renting some of its unused capacity to Cursor for model-training purposes. With the other labs, whoever has the best product has to scramble for compute and has a hard time training the next one, but in xAI's case they lost another valuable complement, researchers, and ended up being a lab that doesn't have a competitive model, still has good distribution, but has resources that assumed that this distribution would lead to more token demand than it actually did. At least for now, a lab that doesn’t quite make it can gracefully retire into being a neocloud instead. But every time that happens, it means they're counting on demand from a smaller set of survivors.

Monetization

Airlines were pioneers of the loot box business model: buy a business-class ticket and, especially if you're a loyal customer of that airline, there's a chance that you'll get bumped up to a class that costs literally an order of magnitude more! In theory, people bought these seats, but in practice they weren't priced in a way that would sell many of them. This has completely changed, and now first class seats are cheaper and are actually getting bought. One reason for this is that airlines have so many other premium categories, so they can actually make reasonable estimates about the true market-clearing price of a nicer in-flight experience. But another reason is that, by initially pricing first class seats out of reach, they ensured that every price cut meant more customers who were thrilled that something they hadn't ever imagined buying was, somehow, arguably affordable.

Headless SaaS

Making it cheaper to deliver code has forced enterprise software companies to come up with a new taxonomy for what they do. Their three basic services are:

1. Provide a system of record in some category, and an API for accessing it.
2. Build a nice frontend on that system of record so non-coders can use it.
3. Identify tasks that are common across customers and adjacent to the original product, and turn them into features.

The middle one is what users think of when they think of the product; from the user's perspective, the interface defines the tool. But now, it's orders of magnitude cheaper to build such interfaces, and they don't require coding experience. So existing software companies have to either bet on #1—be the best system of record, with the best-documented API, so everybody builds whatever they need on top of that backend—or to double downo on the third by having their own engineers build custom tools and integrations for customers. Salesforce is making an interesting bet on the first version, by exposing everything its products can do as an API or an interface for agents. If some customers are asking whether it makes sense to build the subset of Salesforce they want or to pay for the whole thing, Salesforce can answer that question for them by letting them build what they want on top of what they already have, and avoid any messy questions about data migration. But Salesforce probably has a comparative advantage at building those customizations, too; the evolutionary pressure right now is for more SaaS companies to evolve into their sector's Palantir.

Gazumped!

One of the minor mysteries of online advertising is that if you buy something in some infrequent-purchase category, like refrigerators or wallets or something, for the next few weeks you'll see a lot of ads for—the same kind of thing you just bought! Which makes a big more sense if the model of purchasing decisions is that the single time when you're most likely to make a generally-infrequent purchase is that you just made it and ended up slightly dissatisfied. If your new wallet is slightly bulkier than you expected, or your smart fridge turns out to be more annoying than convenient, you'll figure that out fast and make another purchase.

Roughly the same dynamic shows up in another infrequent-transaction category, hedge fund portfolio managers switching jobs. The new trend is funds poaching people who've accepted an offer somewhere else. Hedge fund recruiting is competitive enough that hiring a star manager can be a multi-year campaign of keeping in touch, commiserating, etc., before making an opportunistic offer. Hedge funds design their compensation around making it expensive for the stars to leave, through a combination of deferred compensation and noncompetes. But once someone's made that jump, their talent is temporarily a liquid asset, and it's going to get bid up to the equilibrium price.

Bits and Barrels

The war in Iran makes oil more expensive, which, all else equal, is good for oil exporters. But all else is very much unequal for exporters whose shipments pass through the Strait of Hormuz, so the UAE is asking the US for emergency access to dollar funding ($, WSJ). This is a wonderful illustration of why futures prices aren't like other prices. The UAE has a lot of oil, and that oil is more valuable right now. But the world in which that oil can be shipped and sold is a world where the strait is open and prices are lower. Meanwhile, the US retains its comparative advantage in handling global disorder: America can set tough terms on emergency funding for oil exporters near Iran, and also produce more oil domestically to take advantage of high prices.