The City Consumes Itself: Filtering, the Redevelopment Clock, and What Happens When the Dirt Runs Out
There is a garden apartment complex I drive past a few times a month. It was built in 1974 and you can still see, under the third repaint and the mismatched stair rails, that it was somebody's flagship. Brick veneer on the street elevations, a leasing office with a mansard roof, a pool that was clearly the centerfold of the original brochure. In 1974 this was Class A. It had the best rents in the submarket, the newest appliances, and tenants who chose it over everything else on offer.
Today it is Class C by every definition anyone uses. Nothing dramatic happened to it. It was maintained about as well as the average, renovated once in the 1990s and once in the 2010s, and it simply got lapped, decade after decade, by newer product built farther out on cheaper dirt. Its rents sit at roughly 55 percent of the newest deliveries two miles away. It is, functionally, the affordable housing of its submarket, and no subsidy program ever touched it.
Here is the thing that essay after essay about "the housing market" misses. That property is not at the end of its story. It is at the midpoint of a cycle that is as regular, as measurable, and as tradeable as the interest rate cycle, and much slower, which is exactly why almost nobody positions for it. The complex was built on farmland. It aged from A to B to C over fifty years. And at some computable date in the next few decades, the land under it will be worth more than the building on it, somebody will tear it down, and the parcel will become Class A again, at five times the density.
That loop, greenfield to Class A to Class C to teardown to Class A, is the metabolic cycle of a city. For most of American history it ran quietly in the background because there was always more farmland, so the industry could grow by expansion instead of digestion. The argument of this paper is that the expansion era is ending, metro by metro, on observable schedules, and that the industry that replaces it, a redevelopment industry, obeys different math, rewards different skills, and is legible in advance to anyone willing to model it at the parcel level.
The paper runs in eight parts. Part I explains what "running out of land" actually means, because it is never about acreage. Part II builds the filtering clock: the measured rate at which Class A becomes Class C, and why it takes about a quarter century. Part III is the core math, the redevelopment option, including a worked example that computes the year a real garden apartment complex gets demolished. Part IV dates the cycle and names the vintage cohorts coming due between now and 2050. Part V covers the collision with affordability, which is the political risk at the center of all of this. Part VI describes what the redevelopment era rewards and who it ruins. Part VII is about world models, both the kind you carry in your head and the kind you build in software, and why this cycle is the single best application for them in real estate. Part VIII is what would prove me wrong.
A note on the numbers. Filtering rates come from the academic literature, principally Stuart Rosenthal's 2014 work in the American Economic Review and the follow-on studies using American Housing Survey panels. Depreciation figures follow Bureau of Economic Analysis conventions for structures. Land share estimates draw on the Federal Housing Finance Agency and Lincoln Institute land price series. Stock vintage data come from the American Community Survey, Census construction series, and the standard commercial data vendors. International demolition-age comparisons come from Japanese ministry data and UK housing stock surveys. Where a number is well established I cite the range. Where I am reasoning from structure rather than measurement, I say so. The worked examples use round, defensible assumptions and I show every step so you can rerun them with your own.
Part I: You Never Run Out of Land. You Run Out of Three Other Things.
Start by killing the naive version of the thesis, because the naive version is wrong and the real version is better.
The United States is not running out of land. The continental US has about 1.9 billion acres and all urban development combined sits on roughly 3 percent of it. Metro Houston alone could physically absorb decades of national housing demand. If land were the constraint in the way water is a constraint in the desert, this paper would not need to exist.
What a metro runs out of is three specific things, in a specific order.
First, the commute shed. People carry a travel budget of roughly an hour a day, a regularity so stable across centuries and continents that it has a name, Marchetti's constant. Romans, Victorians, and Atlantans all tolerate about thirty minutes each way, and the size of a city is set by how far thirty minutes gets you. Every transport revolution, streetcar, automobile, freeway, minted new land by extending the radius. The freeway era was the biggest land-printing event in history: raise average speed from 20 to 55 miles per hour and the reachable area grows with the square of the radius, roughly seven and a half times more dirt. That is where the postwar suburb came from. But the printing press has stopped. There is no next speed increment on rubber tires; congestion has been pulling effective radii inward in most large metros since the 1990s, and remote work, the one genuine radius extension of the last half century, applies to a minority of workers and appears to have delivered most of its one-time shift. When the radius stops growing, the greenfield at the edge stops being a substitute for land in the middle, and the middle starts to reprice.
Second, entitlement. Zoning is synthetic scarcity layered on physical scarcity, and in the expensive metros it is the binding layer. The true supply of "land" in an economic sense is not acres, it is entitled floor area, acreage times allowable FAR, and a city can print land by upzoning exactly the way a central bank prints money. San Jose has plenty of dirt; it has almost no entitled floor area relative to demand. This matters for the thesis in both directions. It means "running out of land" arrives decades earlier than the geography suggests, because the legal supply exhausts before the physical supply. And it means the exhaustion is partially reversible by politics, which is one of the failure modes in Part VIII.
Third, infrastructure. A parcel is not developable because it is empty. It is developable because it has water, sewer, road capacity, school seats, and, increasingly, electricity. I wrote a separate paper on power as the new binding constraint and will not repeat it here, except to note that infrastructure exhaustion is the quietest of the three and the one that converts fastest from "abundant" to "five-year queue."
Now, the evidence that exhaustion is real and already priced, unevenly, across the system. The cleanest single measure is the land share of property value: what fraction of a home's price is the dirt rather than the structure. In the FHFA and Lincoln Institute series, land share in coastal California metros runs 60 to 75 percent. In New York and Boston, 50 to 65. In the growth Sunbelt, Dallas, Houston, Atlanta, Phoenix, it historically ran 15 to 25 percent, which is the signature of a market that can still expand: when you can always make more served, entitled land at the edge, land trades near its production cost and the structure is most of the value.
Read the land share as a fuel gauge. Above 60 percent, the metro has already exhausted; every development decision there is implicitly a redevelopment decision and has been for years. Below 25 percent, the expansion machine still works. The interesting story of the 2020s is the middle of the gauge moving. Land shares in Sunbelt metros have been grinding upward as the first-ring suburbs of the 1970s and 1980s find themselves twenty minutes from downtown with no vacant parcels left, and as the exurban fringe pushes past the Marchetti boundary into drive-until-you-qualify territory that only clears in cheap-gas, low-rate conditions. Austin's land share today looks like coastal metros' land shares in the early 1990s. The gauge moves slowly, then all at once, because land value at the margin is a residual, and residuals are levered.
The government wildcard. Before leaving the supply side, acknowledge the largest landowner in the country, because any paper about running out of land has to reckon with the fact that one owner is sitting on most of it. The federal government owns roughly 640 million acres, about 28 percent of the United States, and adding state trust lands and other public holdings pushes the public share past a third. The instinctive objection is the right one: almost all of it is in the wrong place. The federal estate is concentrated in the intermountain West and Alaska, and east of the Rockies the federal share of most states is in the low single digits. You cannot solve a Boston land shortage with a BLM section in central Nevada, and no act of Congress moves land into a commute shed.
But the objection proves too much if you stop there, because a specific and important set of metros sits with public land not at the horizon but at the city limits. Las Vegas, Phoenix, Boise, Reno, Salt Lake, Albuquerque, and St. George are effectively ringed by federal ownership; Nevada is about 80 percent federal, Utah over 60. For those markets, "exhaustion" is not geographic, it is administrative, and it is reversible by statute. There is even a working precedent: the Southern Nevada Public Land Management Act has had the BLM auctioning land around Las Vegas since 1998, several billion dollars of it, and every auction is the government printing developable land at the metro edge. Both parties floated federal-land-for-housing programs in the mid-2020s, and the idea keeps returning because it is one of the few housing levers Washington actually holds. So read the federal estate as a regional overlay on this paper's thesis: in a handful of intermountain metros, a serious disposal program could extend the expansion era by a decade or two and push redevelopment exercise dates out accordingly, and the tell is acreage actually conveyed, not bills introduced. Everywhere else, the estate is irrelevant to the clock.
There is also a second, quieter form of government land that cuts the other way: the parcels the public sector already owns inside the commute shed. Closed military bases, decommissioned airports, rail yards, obsolete civic complexes. Denver's Stapleton and Austin's Mueller were both airports; the Presidio was a base; every BRAC round has seeded a redevelopment district. Government land inside a metro is a redevelopment cohort of its own, with the special property that its exercise date is set by politics rather than by the Part III inequality, which makes it lumpy, slow, and occasionally enormous. I have written a full paper on public land and private development and will not rerun it here, except to log the asymmetry: remote public land can delay the redevelopment era where it borders a metro, while embedded public land tends to accelerate it, by dropping hundred-acre sites into markets that have none.
One more framing before the clock. In an exhausted metro, new Class A supply has exactly one source: sites that already have something on them. Which means the raw material of the next development cycle is the output of the last one, aged fifty years. The industry's inventory of future sites is not a map of empty parcels. It is a map of old buildings, sorted by vintage, and the sorting is the subject of Part II.
Part II: The Filtering Clock
"Class A" is a relative term, and everything interesting about it follows from the relativity.
A building is Class A because it is better than the alternatives available to the highest-paying tenants in its market, newest systems, current floor plates, the amenity set of its moment. The moment newer product delivers nearby, the old building's relative position slips, whether or not anything about it has physically changed. Aging in real estate is competitive before it is physical. This process, an asset sliding down the quality distribution as new supply enters above it, is what housing economists call filtering, and it is one of the most measured phenomena in the field and one of the least used by practitioners.
The canonical estimate comes from Rosenthal's 2014 study, which tracked the incomes of successive occupants of the same units across decades of American Housing Survey panels. Rental housing filters downward at roughly 2.5 percent per year in real terms: each year, the typical rental unit serves a tenant population about 2.5 percent further down the income distribution than the year before. Owner-occupied stock filters at less than half that rate, which is one honest reason single-family neighborhoods age more gracefully than apartment corridors. The rate varies with local conditions in an intuitive way: filtering is fast where new supply is abundant (the new stock pushes the old down the ladder) and slow to nonexistent where supply is constrained (in San Francisco, a 1920s building can filter up, which is what happens when the ladder itself is rising faster than the building is sliding down it).
Run the arithmetic. Put a new building's relative rent position at 100 percent of the market frontier and let it slide at 2.5 percent per year against that frontier. Using 85 percent of frontier as the floor of Class A and 60 percent as the ceiling of Class C:
Years to leave Class A: ln(0.85) / ln(0.975) ≈ 6.4 years Years to reach Class C: ln(0.60) / ln(0.975) ≈ 20.2 years
Six or seven years of genuine top-of-market positioning, and Class C by the early twenties of a building's life. That matches the lived texture of the business better than people expect: ask any leasing team how long a "new" building stays the building in its submarket, and the answer is one lease cycle plus a renewal, call it five to seven years, exactly what the math says.
In practice the slide is slower than the raw rate because owners fight it. A serious mid-life renovation, new interiors, amenities, systems, buys back position; the industry's whole value-add fund complex is a machine for purchasing ten-year reprieves from the filtering clock. So the realistic path for a well-run asset is: Class A for its first decade, a renovation, Class B through its twenties and thirties, a second and cheaper renovation with diminishing returns, and Class C sometime between year thirty and forty-five. My 1974 complex ran precisely this course.
Why can renovation only ever buy time rather than reverse the clock? Because a building's obsolescence has three layers and money only reaches the first two.
| Layer | What it is | Fixable? | Typical horizon |
|---|---|---|---|
| Physical | Worn finishes, aging systems, tired facades | Fully, with capex | Systems on 25 to 30 year cycles |
| Functional | Ceiling heights, floor plates, unit sizes, parking ratios, riser capacity, things set in the structure | Partially, expensively, sometimes not at all | Set at pour; drifts obsolete over 30 to 50 years as norms change |
| Locational and economic | What the land under the building now wants to be | Not at all from inside the building | The land's clock, not the building's |
Physical obsolescence is a maintenance problem. Functional obsolescence is a design-vintage problem: no renovation gives a 1985 office tower a modern floor plate, or a 1974 garden apartment nine-foot ceilings, or a 2005 data center the power density of a 2026 one. And the third layer is not about the building at all. It is the land underneath appreciating toward a higher use, and it is the layer that ends the story, which brings us to the option.
Before it does, one observation worth the price of the whole section. Filtering is where affordable housing comes from. The overwhelming majority of low-cost housing in America was never built as affordable housing; it was built as the Class A of 1974 and filtered down. The subsidized stock, LIHTC and public housing together, is a rounding error next to the naturally occurring affordable stock, which is simply the accumulated output of past filtering. Hold that thought for Part V, because a redevelopment era, by definition, runs on consuming exactly that stock.
Part III: The Redevelopment Option, Worked
Every parcel of improved land carries an embedded option: the right, at any time, to demolish the improvements and build the highest use the entitlement allows. Most of the time the option is far out of the money and everyone correctly ignores it. The entire redevelopment era can be stated in one sentence: it is what a market looks like when that option drifts into the money across thousands of parcels at once.
The exercise condition is one inequality. Redevelop when:
Residual land value under the new use > value of the standing asset + demolition + friction
where residual land value is what a developer of the new use could pay for the site: stabilized value of the new project, minus all costs to build it, minus the developer's required profit. Everything in this paper reduces to watching the left side of that inequality grow against the right side, so let us actually compute one, with numbers you can argue with.
The standing asset. Take a 1979 garden apartment complex, 220 units on 14 acres in a first-ring Sunbelt suburb. Average rent $1,350 a month, honest Class C. Gross potential rent is $1,350 × 12 × 220 ≈ $3.56 million; at a 52 percent NOI margin, NOI is about $1.85 million; at a 5.75 percent cap rate the standing asset is worth $32 million, about $146,000 a unit. Demolition and abatement on 14 acres of 1979 construction, call it $2.5 million. So the right side of the inequality is roughly $35 million.
The new use. Suppose the parcel is rezoned for five-over-one podium at 80 units an acre, 1,120 units. Development cost excluding land at $330,000 a unit is $370 million. Stabilized value: average rent $2,150 on smaller units gives gross potential rent of $28.9 million; at a 60 percent margin, NOI of $17.3 million; at a 5.0 percent cap, value of $346 million. Require a 15 percent profit margin on cost, $55 million.
Residual land value today = 346 − 370 − 55 = negative $79 million.
The option is deeply out of the money. This is not a defect in the example; it is a description of 2026, a market in which almost nothing pencils, and it is why the correct response to this paper is not to run out and buy teardowns. The interesting question is never whether the option is in the money today. It is when it goes in the money, and that has an answer, because the two sides of the inequality grow at different rates.
Let project value grow with rents at 3 percent a year and construction costs grow at 2 percent a year, and hold the standing Class C value roughly flat (rent growth offset by aging, a kind assumption to the old building). Then the residual at year t is 346(1.03)ᵗ − 425(1.02)ᵗ, where 425 bundles cost plus profit. Solve for the year the residual clears the $35 million standing value:
At t = 20: 625 − 632 = −7. Not yet. At t = 26: 746 − 711 = +35. Exercise.
The parcel redevelops around 2052, when the building is 73 years old. That number, seventy-some years, is not an artifact of my assumptions; it keeps falling out of this calculation across reasonable inputs, and it matches the observed age at demolition of American multifamily remarkably well.
Now the part that should change how you underwrite. Rerun it with rent growth at 4 percent instead of 3, one point higher:
At t = 14: 599 − 561 = +38. Exercise.
One point of rent growth pulls the demolition date forward twelve years. The redevelopment clock is not driven by the level of rents or costs. It is driven by the spread between rent growth and cost growth, compounded, and it is violently convex in that spread. This is why Tokyo rebuilds every 30 to 40 years and Cleveland never does, why the same 1979 garden product gets scraped in Scottsdale and merely repainted in Memphis. It also identifies the single most powerful lever in the whole system: density. Rezone the example from 80 to 120 units an acre and the residual scales with it; the exercise date jumps forward a decade without rents moving at all. An upzoning is a gift of option value to every standing owner in the district, which is precisely why owners of parking lots lobby for height and owners of buildings lobby against it.
Three corollaries worth their own paragraphs.
The covered land play is the option, purchased deliberately. Buying a functioning Class C asset at its income value, operating it for cash flow, and waiting for the exercise date is the cleanest expression of this whole paper. The grocery-anchored strip on nine acres, the aging motel on a hard corner, the 1980s office park at a future rail stop: in each case you are paid a going-in yield to hold a long-dated call on land appreciation. The craft is in buying assets whose income floor is durable enough to service debt for the full waiting period, and whose exercise date your model says is inside your fund life rather than outside it.
Japan and Britain are the two ends of the dial. The average Japanese house is demolished at roughly 32 to 38 years; scrap-and-build is the culture, tax depreciation schedules write structures to zero in 22 to 47 years, and buildings are underwritten as consumables sitting on permanent land. Britain is the opposite pole: median dwelling age near 80 years, a retrofit culture, and a planning system that makes the option nearly impossible to exercise. The United States has historically sat in between and is now sliding along the dial, toward Japan in its high-growth metros and toward Britain in its preservation-minded ones. Knowing where your market sits on that dial is knowing its metabolism.
Where the option never goes in the money, filtering has no floor. In a no-growth market the left side of the inequality never catches the right side, and the building's story does not end in demolition; it just keeps filtering, past Class C, into disinvestment, abandonment, and eventually negative value. That is the Detroit case, and it is the proof that redevelopment is not a natural right of old buildings. It is a claim on future growth, and only growth pays it.
Part IV: The Cycle, Dated
Assemble the pieces into one clock. A building delivers as Class A. Filtering takes it to Class B inside a decade and Class C by roughly year 25 to 35, renovations included. It then operates as Class C, which is to say as naturally occurring affordable housing, for another two to four decades while the land beneath it appreciates. Somewhere between year 50 and year 80, in a growing market, the redevelopment option goes in the money and the parcel resets to Class A at higher density. Call it, for American conditions, a 60 to 80 year full cycle, with the filtering half fast and visible and the land half slow and silent.
The cycle is not a smooth conveyor. It moves in waves, because construction itself moves in waves, and every construction boom writes a demographic bulge into the building stock that ages together and comes due together. American development has boomed four times at scale: the 1920s, the postwar surge of 1946 to 1973, the Sunbelt buildout of 1978 to 1990, and the 2000s and 2010s urban wave. Each cohort hits its redevelopment window on schedule, roughly sixty to eighty years after delivery. The 1920s stock came due in urban renewal (with all its brutality, it was this cycle, exercised by the state). The early postwar stock has been redeveloping since the 2000s. Which tells you what the next thirty years look like: the 1960s to 1980s cohort is the largest vintage bulge in American history, and its window opens now.
Name the cohorts specifically, because this is the actionable map.
| Cohort | Built | Scale | Why it comes due | Window |
|---|---|---|---|---|
| Garden apartments, first-ring suburbs | 1965 to 1988 | Roughly half the US apartment stock predates 1980 | 10 to 20 units per acre on land that now wants 60 to 120; huge parcels, single ownership, weak functional position | 2030 to 2055 |
| Suburban office parks | 1978 to 1992 | The bulk of obsolete office by area is suburban, not downtown | Deep floor plates, parking fields, structurally stranded by hybrid work; the land is often superb | Now to 2045 |
| Enclosed malls | 1965 to 1995 | Call it a thousand-odd survivors, a third of them zombies | 60 to 120 acres, one owner, freeway-adjacent, 80 percent parking by area; the best-located large parcels in suburban America | Now to 2040 |
| Big-box and power centers | 1985 to 2008 | Thousands of sites | Parking ratios built for a retail model that peaked in 2006; effectively land banks with a rent stream | 2032 to 2050 |
| First-generation data centers | 1998 to 2015 | Modest by acreage, instructive by clock speed | 1 to 5 megawatt facilities functionally obsolete against modern density; the fastest filtering asset class ever created, A to C in under fifteen years | Now |
Two readings of the table. The obvious one is a target list. The less obvious one is a supply forecast: these cohorts are where the next generation of Class A gets built, which means the geography of new supply in 2035 to 2050 is largely already determined, parcel by parcel, and visible today to anyone who sorts the stock by vintage, density, and land residual. The future map of American cities is not unknown. It is merely uncompiled.
A word on the Sunbelt, because it is the interesting case. Coastal metros have been in the redevelopment regime for decades; nothing in this paper surprises a Los Angeles developer. The regime change is in Dallas, Atlanta, Phoenix, Nashville, Austin, Charlotte: metros that have never in their institutional memory had to grow by digestion, whose entire developer ecosystem, brokerage culture, and lender playbook is calibrated to expansion, and whose first-ring 1970s and 1980s product is now sitting on the most valuable relative locations in the metro. Those markets will cross into redevelopment economics during the working careers of everyone reading this, and the local industry is, with a handful of exceptions, not built for it.
Part V: The Collision Nobody Has Priced
Here is the uncomfortable arithmetic at the center of the redevelopment era, and I want to state it plainly rather than bury it.
Filtering is the source of nearly all unsubsidized affordable housing. The redevelopment cycle consumes filtered stock. Therefore the redevelopment era, by its nature, demolishes affordable housing to build expensive housing, unit by unit, at exactly the moment the affordability crisis is the dominant housing politics of the country. The 220 units in my worked example rent for $1,350. The 1,120 units that replace them will open north of $2,100. The neighborhood gets five times the housing and loses every unit a $45,000 household could afford, and both facts are true at once.
The aggregate defense is real: the 1,120 new units filter too, and the metro-level evidence is solid that new market-rate supply reduces rents down the ladder, the "musical chairs" result. But the aggregate defense has never once won a zoning hearing, because the people in the room are the 220 households, not the metro. So the redevelopment era arrives wrapped in political risk of a specific, forecastable kind, and you can already read the responses in the code books of the cities furthest along the curve: demolition-replacement ordinances, right-to-return requirements, tenant opportunity-to-purchase acts, inclusionary percentages that scale with units demolished. Underwrite them as a tax on option exercise, because that is what they are, and note the perverse equilibrium available if the tax is set too high: the option stays out of the money, the Class C stock keeps aging with no exit, and the city gets neither the new supply nor the preserved affordability, just decay. Several California municipalities have been running that experiment for a generation.
For the practitioner the collision is also, cold-bloodedly, a timing instrument. The political cost of redevelopment is lowest early in a market's transition, before the losses accumulate into organized opposition, and rises from there. First movers in a newly-transitioning Sunbelt metro will exercise options under 2020s-era rules; late movers will exercise under whatever those rules become after a decade of demolitions. The spread between those two regulatory regimes may be worth more than the spread in land prices.
Part VI: What the Redevelopment Era Rewards
Expansion-era development is a logistics business wearing a real estate costume: buy entitled dirt at the edge, execute a known product at known cost, sell into a known buyer pool, repeat. Its core competencies are speed, cost control, and relationships with land brokers. Redevelopment is a different profession that happens to share a license.
What the new era pays for, roughly in order:
Entitlement craft. When land is made by rezoning rather than found by driving, the highest-alpha act in the industry is the one that changes what a parcel is allowed to be. The developer who can take a mall through a two-year specific-plan fight is doing for land what a refinery does for crude. Expansion-era firms treat entitlement as a cost center staffed by consultants. Redevelopment-era firms treat it as the product.
The waiting trade. Covered land plays reward balance-sheet patience and punish fund-life impatience. The natural owners of pre-exercise assets are family offices, open-end vehicles, and operators with permanent capital; the natural sellers are closed-end funds hitting year eight. That mismatch is a durable source of mispriced assets, and it will get bigger as more of the market's value migrates into option value that quarterly marks cannot see.
Structural and phasing skill. Demolition is the simple case. Much of the cohort in Part IV will be surgically rebuilt rather than scraped: mall department stores carved into housing over retained parking structures, office park pads densified around standing buildings, garden apartment sites rebuilt in phases with tenants in place. That is a construction management discipline an order of magnitude harder than podium-on-dirt, and the firms that have it, mostly forged in coastal infill, hold a skill that the transitioning Sunbelt will have to import or grow.
Tax and subsidy fluency. Historic credits, LIHTC rehab, property tax abatement structures tied to replacement housing, C-PACE on repositionings. In expansion these are exotic. In redevelopment they are load-bearing, because they are how the inequality in Part III gets pushed into the money a few years early, legally.
Assemblage. The expansion era's parcels came pre-sized by the farm grid. The redevelopment era's best sites are owned in fragments, and patient, quiet, sequenced acquisition, the land assemblage game I have written about elsewhere, returns to the center of the business.
And the losers: pure greenfield merchant builders in exhausting metros, whose raw material is going away; lenders whose credit models price buildings and ignore land options; and, bluntly, any operator whose plan for a fifty-year-old asset is a fourth cosmetic renovation, fighting a clock that money at that dosage does not stop.
The business model has to change, not just the skill set
It is tempting to read the list above as a hiring memo: same firm, new departments. That underestimates what is happening. The expansion-era development company is a specific machine, and the redevelopment era breaks it at the level of the model, not the org chart. Five rethinks, each of which is already visible in the firms furthest along the curve.
From velocity to duration. The merchant model is a velocity business: buy dirt, build, sell in 36 months, recycle the promote, and the firm's value is its deal treadmill. Redevelopment value accrues to whoever owns the parcel while the option ripens, which can be a decade, and a treadmill cannot hold anything for a decade. The model that fits is buy-operate-wait-build: acquire the covered land play at its income value, run it competently for cash flow, and be the developer standing on the site when the inequality flips. That is a completely different P&L, income-led with a development kicker, and firms will have to decide whether they are developers who operate or operators who develop, because straddling with expansion-era overhead kills both.
From fund life to permanent capital. The ten-year closed-end fund is the expansion era's native vehicle and it is structurally wrong for assets whose payoff date is year twelve. Expect the serious redevelopment players to migrate to evergreen vehicles, family and insurance balance sheets, listed or open-end structures, and long-dated separate accounts, and expect the closed-end funds to become their favorite sellers. A firm that cannot change its capital duration will systematically hand its best exercise dates to whoever can.
From product companies to place companies. Expansion firms scale by repeating a product across metros: the same garden deal in eight states. Redevelopment alpha is local and relational: entitlement outcomes, council coalitions, neighborhood trust, and assemblage intelligence compound within a submarket and transfer poorly across them. The winning shape is deep and narrow, a firm that owns the fifty best-dated parcels in three corridors and knows every council agenda by heart, not shallow and wide. This inverts twenty years of institutional pressure toward geographic diversification, and the institutions will be the last to notice.
From fees on building to margins on entitlement. In the expansion model the money is in the development fee and promote on vertical construction. In the redevelopment model the largest single value-creation event is frequently the rezoning, which can multiply a parcel's residual before a shovel moves. That argues for business lines the old model never had: entitlement-and-exit as a standalone trade, ground leases that let the firm keep the land and sell the building, and structured deals with incumbent owners, the families and small operators who control the aging cohort, where the firm contributes entitlement craft and capital against the incumbent's basis. The developer as land refiner, paid on the grade change, not the pour.
From gut to model. When the game is timing on assets everyone can see, the parcel-level modeling of Part VII stops being a research toy and becomes the acquisition engine, as central to the redevelopment firm as the land brokerage network was to the expansion firm. The firms that treat it as overhead will be buying the exercise dates the modeled firms already sold.
None of this requires prophecy; every element exists today in some coastal infill shop, Japanese developer, or family office. What the transition does is take a set of adaptations that were niche and make them the industry's median business model, one Sunbelt metro at a time.
Part VII: World Models, or How to See a Slow Cycle
Everything to this point has been about the cycle. This section is about the instrument, because a cycle this slow, this parcel-specific, and this legible is almost a laboratory demonstration of what world models are for, and I mean that term in both of its senses.
The first sense is the old one: the model in your head. Charlie Munger's latticework, the trader's regime map. The expansion era ran on a simple mental world model, "growth moves outward along the freeways," and an entire industry could navigate with it. The redevelopment era's mental model is the contents of this paper: land as a fuel gauge, filtering as a clock, every parcel as an option, politics as a tax on exercise. Practitioners who internalize that model will see the same city everyone else sees and read it differently, the way a geologist reads a road cut.
The second sense is the one from AI research, and it is the one I find genuinely exciting. A world model, in that usage, is a system that maintains an internal state of an environment and simulates it forward: not a prediction bolted onto data, but a persistent, updatable representation of how the world evolves, checked against reality and corrected when wrong. I run a small one of these against financial markets, signals in, dated predictions out, every prediction auto-verified against what actually happened, and the discipline it enforces, the humiliation of a written scorecard, has taught me more than the predictions themselves.
Real estate's redevelopment cycle is close to the ideal environment for such a system, for reasons worth spelling out:
The state space is enumerable. A metro is a finite list of parcels, each with vintage, structure type, density, entitlement, assessed improvements-to-land ratio, rent position, and ownership. All of it is public or purchasable. You can literally hold the whole state in a database, which you cannot do for an economy or a battlefield.
The dynamics are slow and lawful. Filtering runs at 2.5 percent a year. Land residuals evolve with rents and costs. Entitlements change through processes with public dockets and statutory clocks. Nothing important happens in milliseconds; everything important compounds over quarters. A model that is merely decent gets years of warning.
The verification loop closes. Every demolition permit, rezoning approval, and land trade is a ground-truth event against which the model's parcel-level predictions can be scored. A world model that predicted "this corridor's garden stock begins trading as land by 2031" gets marked right or wrong, publicly, on a knowable schedule.
So the concrete system, buildable today with existing data and existing models: maintain state per parcel; compute each parcel's standing value and residual land value under current and plausible entitlements, which is the Part III inequality, run at scale; simulate forward under scenario paths for rents, costs, rates, and zoning; and emit a ranked surface of exercise dates, this parcel in 2031, that corridor in 2038, updated as every new permit, sale, and council agenda item lands. Layer on the signal families that leak the future early, and here the anchors matter, because a Costco land purchase or a new nonstop is exactly the kind of third-party, capital-backed forecast that moves a corridor's rent-growth prior, and I have written a whole paper on reading those. The output is not a mystical oracle. It is the covered-land acquisition pipeline, the entitlement lobbying map, and the disposition warning system, generated by machine and audited by scorecard.
Two honest limits, because world models are having a fashionable moment and fashion breeds sloppiness. First, the model's weakest module will be politics, the zoning outcomes that swing exercise dates by decades, and no simulation substitutes for knowing whether the council member's coalition survives November. Encode politics as scenario branches with hand-set probabilities and update them like a forecaster, not an engineer. Second, a world model is only as honest as its verification loop. A system whose predictions are never scored decays into a justification engine for deals someone already wanted to do. The scorecard is not a feature of the system. It is the system.
The broader point stands above the tooling. The expansion era rewarded being somewhere first, and information advantage meant hearing about a land listing early. The redevelopment era rewards being right about timing on assets everyone can already see, and that is an analytical advantage, the kind that compounds to whoever builds the better model of the world and submits it to the discipline of being checked. For the first time, the technology to do that at parcel scale exists, is cheap, and is mostly unused by an industry still running the old mental model. That gap will not stay open for a generation. It might stay open for a decade.
Part VIII: Three Scenarios, Dated
Structure first, numbers second, and everything in this section is a scenario, not a fact.
Scenario one, the grind (55 percent). No policy revolution, no macro rupture. Rent growth runs 2.5 to 3.5 percent, cost growth 2 to 3, and the spread stays thin, so exercise dates arrive on the slow schedule. Coastal metros continue as they are. Two or three Sunbelt metros, Austin and Nashville are my candidates, cross visibly into redevelopment economics by 2032, marked by the first institutional-scale garden apartment teardown wave in their history. Suburban office and mall redevelopment proceeds steadily all decade. By 2040 the industry's org charts have quietly rebuilt around entitlement and phasing talent, and nobody remembers deciding to change. This is the base case because sixty-year cycles mostly just continue.
Scenario two, the unlock (25 percent). The state-level zoning reform wave of the early 2020s reaches escape velocity, and legal land supply expands faster than any point since the freeways: broad upzoning of commercial corridors and first-ring residential across a dozen states by the early 2030s. Perversely, this accelerates the redevelopment era rather than delaying it, because upzoning does not create dirt, it deepens the option value on already-improved parcels; the residuals in Part III jump discontinuously and exercise dates lurch forward a decade. Demolition volumes surprise everyone. So does the political backlash by the mid-2030s, which caps the wave. Land prices are the tell: watch improvement-to-land ratios collapse in newly upzoned corridors within eighteen months of passage.
Scenario three, the freeze (20 percent). Construction costs inflate persistently faster than rents, driven by labor scarcity, tariffed materials, and competition from the data center and power buildout for the same trades, the collision I documented in the gigawatts paper. The spread goes negative, residuals fall, and the redevelopment option drifts out of the money across whole metros. America turns British: the existing stock is repaired forever because replacing it never pencils, Class C ages in place without an exit, filtering with no floor spreads beyond the traditional no-growth markets, and the affordability crisis worsens through decay rather than displacement. This is the scenario nobody positions for, and the hedge against it is owning assets whose standing income justifies them without any option value at all.
Part IX: What Would Prove This Wrong
The section every thesis owes its readers.
Remote work resumes extending the radius. If distributed work moves from its current plateau to a second structural leg, the commute shed grows again, the edge becomes a substitute for the middle, and exhaustion recedes by a decade or more. Watch the share of fully remote job postings and inter-metro migration to exurban counties; both flattened by 2024 and would have to re-accelerate.
Autonomy prints land. Full self-driving at scale is a transport revolution of exactly the kind that has always minted new land, by raising tolerable commute distance and by releasing parking, which is 25 to 30 percent of urbanized land, into the developable supply. My skepticism is about timeline, not physics. If robotaxi service is operating without safety drivers across the broad geography of ten American metros by 2030, shorten every land-scarcity claim in this paper.
Construction productivity finally breaks its fifty-year stagnation. Factory-built modular at 20 to 30 percent below site-built cost changes the Part III inequality from the cost side and would, oddly, accelerate redevelopment rather than refute the thesis; but it would also revive greenfield economics at the exurban edge. Watch modular's share of multifamily starts, which has been stuck in the low single digits for decades; a sustained move through 10 percent is the tell.
Congress opens the federal estate. A disposal program at real scale, hundreds of thousands of acres conveyed around the intermountain metros rather than announced, would restart the expansion machine in exactly the markets where this paper predicts the Sunbelt-style transition. The Las Vegas precedent shows the mechanism works when the land is adjacent and the politics align. Watch acres patented and auctioned per year, not legislation introduced; the ratio of the two has historically been brutal.
The filtering rate itself shifts. If durable-goods-quality construction (or, less happily, permanent undersupply) slows filtering broadly, the clock in Part II stretches and the cohort schedule in Part IV slips a decade. The American Housing Survey panels update; someone will rerun Rosenthal every few years, and I will be reading.
And the honest one: the cycle is real but untradeable. It is possible that everything here is true and the rents to it accrue entirely to incumbents, holders who never sell, municipalities that tax the option value away at exercise, and communities that price the political cost fully. In that world the paper is correct as description and worthless as strategy. I do not believe it, the covered-land trades already on the books argue otherwise, but a thesis that cannot name its own null result is a sales document.
Cities have always consumed themselves. Rome is forty feet deep in previous Romes; the London that exists is the fifth or sixth London on the site; Manhattan's building stock has turned over almost entirely twice since the Civil War, and nobody mourned the version that was lost because the version that replaced it was the one they lived in. What is ending in America is not land. It is the two-century anomaly in which we could grow without digesting, the frontier subsidy that let an entire industry mistake expansion for development.
The complex from 1974 does not know any of this. It will stand there, collecting its $1,350 rents, filtering another notch each year, while the land beneath it quietly appreciates toward the day the inequality flips. Somewhere, on somebody's parcel-level model, its exercise date already has an estimate attached. The only question this paper really asks is whether that model is yours.