What We Periodically Relearn About Rent Control

Plus! Backstops; Complements and Substitutes; Personality Hire; Commoditization; IPOs

Byrne Hobart Byrne Hobart

In this issue:

audio-thumbnail
The Diff July 27th 2026
0:00
/1089.645714

Talk to this post on Read.Haus.

What We Periodically Relearn About Rent Control

When I talk about policies, I like to focus on second-order effects. That's a lot of syllables, so if you're in a hurry, just say "effects," because most of the impact of a law like this is not its upfront redistributive impact—tenants pay a little less, landlords earn a little less—but on how labor and capital get reallocated in response. Rent control:

  1. Lowers the return on owning residential real estate.
  2. Raises the uncertainty of that return; once there's some rent control, it's easier to extend it to other units, or to adjust its parameters. And since it disincentivizes maintenance—the landlord doesn't capture the upside any more—it leads to later regulations, which may or may not match what tenants were willing to pay for.
  3. It also means that a portfolio of properties in a given city is less diversified than it used to be, because there's a common factor for returns. Someone who owns an apartment building in a city is making a bet on that city's macroeconomy, but they're also making a smaller bet on which neighborhoods will become or remain trendy. Property owners can diversify that risk by owning buildings in different places, but if there's a common factor that affects returns citywide, they miss that.

Which doesn't mean you can't do it. Environmental regulations have a similar effect on the chemical industry, adding a new core competency they need to be good at and making their results more correlated because they're sensitive to changes in laws, which adds up to a higher cost of capital and a lower supply of chemicals. Non-carcinogenic drinking water is a pretty good trade for that.

The existence of market rent is not that kind of externality, though. Rent is a price signal about the opportunity cost of living in one place rather than somewhere else; it's a way to tell you that if you insist on taking up some of the very finite stock of housing supply that's within walking distance of, say, OpenAI's headquarters, then you'd better get busy doing something valuable enough that you're the high bidder for that real estate. And, by the same token, that if you don't have some very short list of places where you can do whatever it is that you want to do, you can optimize your real estate spending more for cost and local amenities rather than optimizing for location and then somehow solving for cost.

There is a potential negative externality here: many of the things that make cities interesting places to live aren't all that lucrative. The existence of trendy, expensive brunch places implies (for now) the need for people to bus the tables, and it's also a bit perverse that the cities with creative scenes are also the cities where you can earn the median American household income and still pretty much qualify as a starving artist. This tends to make expensive cities a bit spikier, bland in some ways but incredibly interesting if you happen to care about whatever industry drives up those rents. One reason you overhear so much about GPUs and RSUs and the like in SF is that many of the people who work in the city and don't have those interests do have long commutes, so there are fewer hours in the day where they're ambient.

In theory, a partially rent-controlled city can solve for this: maybe the bankers and lawyers and traders in New York pay a little more rent, and the novelists and indie rockers pay a little less. But in practice, it's incredibly hard to target these people with low rent. San Francisco has a live/work building program, whose original intent was exactly this: a building gets exempt from some zoning restrictions, but the tenants need to be artists. I actually lived in one briefly, and the way the building manager explained it was that the definition of artist had been loosened to basically anyone who had any kind of business that was remotely creative-adjacent—and that the way this business was audited was that the city would sometimes check in and see if everyone had bought a business license ($50), but would give them a grace period in which to get the license even if they didn't have it. If you zoomed in on the building in Google Maps, you'd see a little confetti of non-operating dance studios, LLCs for ostensible freelance writers who had day jobs at Dropbox, etc. San Francisco wanted to be welcoming to artists, but did not want city hall employees spending their days interrogating the question of What Is Art, so what they ended up doing instead was very moderately increasing the housing supply in exchange for making everyone participate in a slightly grubby ritual of pretending to be an artist while working in tech.

In a way, a city that has a creative scene but prices out creative types is just doing its job; the goal is to enforce an up-or-out model where, if you're going to insist on not just being a novelist but being a novelist in Brooklyn, you'd better get cracking and publish something good or, sooner or later, you're moving back home (sometimes, this means an artist who couldn’t make it in Brooklyn and is forced to move all the way back home to the Upper East Side) It's just Baumol's cost disease, for rent rather than wages. And the market in creative talent keeps getting more efficient, because there are so many online channels now; people can and do Substack their way to a book deal. And then they can move to the big city and hang out with the other successful artists.[1]

Maintaining a city's character is one macro argument for rent control. The micro one is basically that, but instead of being at the level of career paths, it's at the level of households. It is, in some sense, unfair that if you rent in a city, and your income goes up but doesn't keep pace with the cost of living, you'll eventually have to leave. It would be nice to have some protection against that.

All true. Not only is it true, but we can quantify just how nice it would be by looking at the down payment and mortgage payment for buying rather than renting. There is no rule preventing anyone from getting the exact bargain rent control offers—a fixed price, and certainty that your home will be yours. But, especially in places with rent control, it's murderously expensive. Slowly giving a subset of leases the economic characteristics of a mortgage is basically piecemeal redistribution of a down payment's worth of economic value from the landlord to the tenant. And if tenants either didn't prefer this in the first place (or they would have bought) or treat it as a default expectation, this is wealth-destroying in the sense that we're taking some resources that society has produced (in this case, the right to stay somewhere at a given price even if market rent rises) and redistributing it to someone who doesn't want it. It's wealth-destroying in exactly the same way that it would be if a restaurant switched your order with another customer's; you're getting something, but if you'd wanted it, you would have ordered it. And it’s also hard to ensure that the recipients aren’t themselves well-off but lucky ($, WSJ).

But: landlords are not all that sympathetic, as a class. Some of this is just the result of the usual economic luck where, like IRS agents or divorce attorneys, the average person is mostly encountering them in an already-unpleasant situation, and it's often their job to deliver news that makes that situation worse. But also, there's the question of what they actually do. You can divide your rent by your after-tax pay and multiply that by 31 to figure out how many days of work you did for your landlord; over that period, your landlord might have answered a text or two, maybe sent over a plumber or something, but mostly cashed a check.

One of the ways people attack rent control is to argue that the landlord has an actual job, and that it's hardly passive income. Sometimes, tenants need things at inconvenient times, and sometimes, there are expensive repairs that wipe out months of income. All of this is true, but it's just not a great argument; publicly-traded residential REITs spend 15-20% of their revenue on repairs, property management, and recurring capex. That service is just not the main thing renters are paying them to do.

What renters are paying for, what they're actually renting, is deferred consumption. For the landlord to own a building, they need to have bought it instead of spending that money on something else, and for them to rent it out, they need to have already taken care of their housing. The entire system of owning real estate that you don't personally use is a system that allows some people to consume less than they earn over time, and to convert that deferred present consumption into future consumption. The renter is, implicitly, someone who doesn't want to defer a big slug of consumption by buying property, either because they can't afford to or because they'd rather put the money to work somewhere else. It's unavoidable as long as there's population growth that leads to demand for housing, especially if people also raise their standards of how much square footage they want. In cities where housing supply is constrained, whether by geography or politics, prices will be high, and landlords will capture some of that upside.[2]

When landlords buy and sell existing properties, the main economic activity they're engaged in is changing the form that their savings take. This doesn't have much of a direct effect on tenants; a more liquid property market is generally a pricier one, since it means that there's an incentive to buy a property that charges lower rent than it could and then raise those rents. But it's neutral in the sense that if a building is sold, exactly as much money is going out of property speculation as is coming into it. The more interesting effect is on the actual changes in supply. If the present value of the rent that could be earned from turning an empty lot into an apartment building, or turning a small one into a bigger one, is higher than the cost of building it, then supply gets added. But "Present value" is doing a lot of work, here, because one of the big inputs is the discount rate, and the discount rate that buyers demand is a function of what risks they're taking and how much they can hedge those risks. Owning five units rather than one provides some diversification, and, as CAPM tells us, you don't get paid for taking risks that you could trivially diversify your way out of. But if there's a force that makes the cash flows from all of those properties move in lockstep sometimes, then the inverse holds true: adding new units adds less diversification, so the market-clearing price of those units is lower, which means that fewer potential properties clear the return hurdle necessary for them to actually get built.[3]

So, from a purely practical standpoint, rent control is bad news because it breaks the transmission mechanism whereby a city that's a desirable place to live becomes a city that can fit more people. But it does so in an indirect way that's hard to measure, by lowering the expected return and raising the hurdle rate for buying buildings, which, years later, leads to a counterfactually lower supply. This dynamic means that it's also a political perpetual motion machine:

  1. If people look back on some stricter rent control policy imposed a year ago, they'll see that tenants are paying less, and that housing supply is unaffected—even if new supply isn't exempt from the rules, it can still be worth finishing a project in progress even if its overall return will be lower than expected.[4]
  2. Supply constraints keep prices high, which means that housing affordability remains a live issue. And if voters remember that the world didn't suddenly end the last time there was a push for rent control, they might decide that it's a decent solution this time. Things eventually reach a breaking point—in the 1970s, an empty lot full of rubble was worth more than a rent-controlled building on top of it, especially if that building was insured, so landlords burnt them down.. (That's yet another case where rent control is an autocatalytic policy. Landlords are a lot less popular when some of them are committing deadly crimes. Even if those crimes are the direct result of policies that also made potential arsonists the high bidders for those properties, the connection is indirect enough that it's easy to miss.)

You can shove this whole debate aside by arguing that housing is not a normal kind of product and shouldn't be subject to the same market rules as iPhones and sneakers. After all, people need housing. But this ignores the location-location-location rule: when people talk about a right to "housing," they're implicitly assuming some quality baseline, and one part of that quality is where the house is. But there can't be a general right to live in a particular dwelling, so the housing-as-a-human-right argument collapses into saying that actually, there are around 8.3bn distinct rights to housing, all of which are rights to slightly different homes. If people have a right to housing-of-some-kind rather than particular kinds of housing, Universal Basic Used RV would be fine. Housing is an absolutely enormous asset class, and it basically requires a setup where most of the cost of building a home is paid by someone other than the person who lives there, because the alternative is for people to be homeless, or to live with their parents, until they've saved up enough to buy a house for cash. If people are going to get access to appealing kinds of fixed capital whose cost is wildly front-loaded relative to the consumption they enable, it takes a pretty complex system and a lot of coordination to actually make that fixed capital available in the quantities people want. And if you're going to tweak that system, you need to be absolutely sure that you understand how it works, or you'll have a lot of relearning ahead of you.

  1. One downside to increasing efficiency in talent markets is that there isn't as much good art about the experience of being poor. Even the best natural talents produce some pretty uneven early work, but it takes cosmically bad luck to be both very talented and still broke after years of perfecting your craft. So, relative to previous generations, we'll have a lot more creative work that comes from a middle-class-on-up perspective. We may just be stuck with The Meadowlands as the last great work about being broke. Which is a pretty good consolation prize. ↩︎

  2. Those same restrictions on housing stock, or haphazard affordability programs that randomly deliver benefits to a subset of renters, actually make these cities even more focused on whichever industries they're best at; there are people who get subsidized housing (from the state, from mom and dad, or from a romantic partner) and can work in lower-paying fields, there are people who work in the local moneymaker industry, but it's harder to survive in the middle. If that goes on long enough, it makes a stark class division, and there's resentment on both sides—the biglaw associate with roommates is annoyed that someone a few blocks away is paying a third of market; that person doesn't like that many of their neighbors would really prefer that they move to Staten Island or something and create a vacancy in Manhattan. ↩︎

  3. This should not be read as a claim that mom and pop landlords are running Monte Carlo simulations of different asset allocation approaches in order to achieve the maximum ex ante sharpe ratio at a given vol target. They're pretty far behind the curve, sometimes even on the basics (my favorite landlord of all time bought a building in a kind of gritty, working-class neighborhood in Brooklyn, moved to Florida, and apparently visited rarely enough that he didn't notice that Williamsburg was not so gritty and that he could have been charging a lot more). But over time, competitive markets gradually approximate theoretically optimal behavior because the market participants who behave more optimally end up compounding their wealth faster. If you assume landlords vary widely in their ability to pay smart prices for property, you end up assuming that on average, property sales between landlords are transferring cheap properties to smart landlords and vice-versa. ↩︎

  4. As a simplistic example, if a new rent control law lowers the present value of some prospective building by 15%, but the developer has already paid 20% of the cost of getting it built, it's still worth finishing even though in retrospect it wasn't worth starting. ↩︎

You're on the free list for The Diff. Last week, paying subscribers read about the Baumol Bounce, the possibility that we'll see widespread productivity growth in historically unproductive sectors ($), thoughts on rogue models ($), and a reminder that the market generally doesn't sell off because of excessive capex ($). 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

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.

Contact Anomaly

Elsewhere

Backstops

Speaking of capital: if you take a log scale chart of aggregate investment into various categories, and extrapolate the line that's going up fastest—it's worked perfectly so far!—then if you go out far enough into the 2030s, you can imagine a world where AI datacenters are a bigger asset class than residential real estate. Housing finance has had about a century of diligent effort to direct capital towards loans that have a term long enough that the payments are manageable. The AI industry has had to spin up new financing models on a more breakneck schedule. Google has been guaranteeing leases for datacenters that use its TPUs ($, The Information). Google seems to be pretty senior in the stack, here, given that they've guaranteed $44bn of leases but mark the expected amount they'll lose to $815m, implying credit losses in the low teens of basis points per year for a fifteen-year lease. Meanwhile, Nvidia is in talks to backstop $250bn of OpenAI spending ($, WSJ). There's a nice symmetry here: one reason equity market fundamentals look as good as they do is that the biggest net cash consumers, OpenAI and Anthropic, are both private, but most of their spending promptly converts into free cash flow and GAAP earnings for companies that are public. Nvidia, at least, is in a great position to throw around a lot of cash, especially if it keeps them the standard longer. For Google, the cash picture is trickier, but the market clearly puts a high value on their various AI-related business lines, and that gives them a kind of economic asset that they can make more tangible by converting that value into guarantees.

Both of these deals could turn out terribly, of course. But it's striking that both companies are eager to make the same bet: that the worst-case scenario for AI capex is much rosier than other market participants think.

Disclosure: long NVDA, GOOGL.

Complements and Substitutes

The Uber/Waymo partnership was always a clever one. On the demand side, Uber has distribution, and on the supply side, Uber has a flexible, price-sensitive backfill that ensures that Waymo fleet is valuable even if it can't cover everyone's commuting needs during rush hour. But that partnership seems to be fraying ($, FT). Competition is part of the reason, but it was pretty obvious from the beginning that they were competitors—Travis Kalanick had talked about autonomous driving as Uber's endgame years before. Regulation is a trickier piece, now that Uber drivers are a big enough constituency that fighting to protect their jobs is a way to get votes. Since Waymo is incredibly safe, they, too, have a marketing angle involving saved lives. And it's hard for two companies to partner when Uber is credibly accusing Waymo of threatening people's jobs, and Waymo can shoot back that by lobbying against AVs, Uber is threatening people's lives.

Personality Hire

Cognition was one of the earliest companies to make a serious bet that AI could be a drop-in replacement for the kind of task that could be assigned at work, rather than for microtasks that it would be embarrassing to delegate. What they build is a tool, but the closer it gets to humanlike performance, the more that term can get its pejorative connotation; if we're going to spend a lot of our day interacting with AI agents, then, just like coworkers, we'll be keenly aware of their innumerable personality flaws and verbal tics. Cognition is buying AI assistant Poke, apparently for its more conversational tone. It's an interesting development in corporate history that the same thing that happens routinely at the level of hiring is happening by M&A instead: a product-first company realizes that they should probably hire someone who's a little less technical but can actually get along with customers.

Commoditization

In addition to backstopping OpenAI, Nvidia is partnering with Ilya Sutskver's Safe Superintelligence ($, WSJ). Nvidia's tactical goal is to sell as many GPUs as possible, which they've done a fine job of. But strategically, their main goal is to ensure that there's more than one company with a viable shot at a frontier model, and that these companies rely on or at least use Nvidia's products. So Nvidia has two incentives: first, to help whoever is in second place, and second, to make a few high-variance bets that could lead to another lab becoming one of the top contenders.

IPOs

Chinese memory chip maker CXMT went public and rose 466%, which meant that they were briefly the most valuable listed company in China ($, FT). The Chinese government has used an interesting model where there's significant state support for the AI industry in the aggregate, but they rely on market signals to determine how that support gets distributed. But China's equity markets don't have the depth of US markets, and even though American capital markets over the last few years haven't exactly been a paragon of bland efficiency, they're at least a bit better at either giving people a way to take the other side of a bet or encouraging companies to quickly turn a rising market cap into rising capex. If the market's too thin, or there's more money rushing in than liquidity providers are used to, prices can get far out of whack before there's any corrective force.