The Permit Pulse: What 5-Unit-and-Up Filings Say About the Next Cycle

• 

  Five-storey apartment building, the mid-rise product that dominates US 5-unit-and-up permitting

Los Angeles nearly doubled its apartment permitting last spring. San Antonio lost three quarters of its own. The national number, which averaged them together, said almost nothing worth knowing.

For four years the story of American apartment construction had a simple shape and a simple geography. Capital went south. Cranes went up over Austin, Phoenix, Nashville, Charlotte and Tampa. The metros that had spent the 2010s absorbing transplants spent the 2020s trying to house them, and the permit counts followed, peaking in 2022 at 679,898 units authorized in buildings of five units or more, the highest total in nearly four decades.

Then it stopped. By 2023 the figure had fallen to 561,369, down more than seventeen percent in a single year. Rates had moved, construction lending had tightened, and the developers who had been racing each other to the same suburban arterials stopped filing.

Here is what almost nobody noticed. The slump ended. Full-year 2025 multifamily permitting came in at 516,886 units, up 5.6 percent over 2024, and by December the seasonally adjusted annual rate had climbed back to roughly 515,000 to 523,000, the strongest reading since the summer of 2023.

And here is the part that matters more. The recovery did not happen where the boom did.

The map turned inside out

Run the year-over-year comparisons metro by metro and the picture stops looking like a national cycle and starts looking like a migration.

Austin, the poster child of the boom, was down between 33 and 39 percent across successive readings from 2025 into 2026. Phoenix fell about 34 percent. Atlanta gave up somewhere between 17 and 21 percent depending on the window. San Antonio, in one May 2026 reading, was down nearly 77 percent from the prior year. Charlotte, Nashville, Tampa, Fort Worth and the Raleigh-Durham corridor all contracted, several by thousands of units.

Meanwhile Los Angeles jumped 93.9 percent year over year. Columbus posted the largest single-metro gain in the country at one point in 2025. Detroit, Chicago, Newark, San Jose, Denver, Madison and Bridgeport-Stamford all added meaningfully. So did a scattering of smaller markets: Savannah, Fayetteville-Springdale-Rogers in Arkansas.

Metro	Direction	Magnitude	Reading window
Los Angeles	Up sharply	+93.9%	Year ending May 2026
Denver	Up	+3,321 units	Year ending April 2026
Madison, WI	Up	+2,639 units	2025
San Jose	Up	+2,417 units	2025
Bridgeport-Stamford-Danbury	Up	+2,141 units	2025
Detroit	Up	+1,781 units	Year ending August 2025
Chicago	Up	+1,459 units	Year ending August 2025
Columbus, OH	Up	+1,288 units	Year ending May 2025
Atlanta	Down	-17% to -21%	2025
Phoenix	Down	-33.7%	Year ending April 2025
Austin	Down	-33% to -39%	2025 into 2026
San Antonio	Down sharply	-76.8%	Year ending May 2026
Raleigh-Durham	Down	-2,805 units	Year ending August 2025
Washington, DC	Down	-2,687 to -5,322 units	2025 into 2026

Aggregate analysis of the same data found that 80 of the 100 largest American metros moved by at least ten percent in one direction or the other during 2025, while the national total moved 2.6 percent. That single sentence is the case for reading this data at the metro level and ignoring the headline entirely.

A caution before anyone builds a thesis on the biggest percentage movers. Jackson, Mississippi posted something like a 677 percent increase, which sounds spectacular until you learn it went from five units to a hundred and eight. Syracuse and Baton Rouge produced similar-looking numbers off similarly small bases. Percentage change on a tiny denominator is a headline, not a finding. Absolute unit counts and per-capita screens are the honest filters.

Where the number comes from, and why it is worth trusting

The data underneath all of this is free, and it has been collected continuously since 1959.

The Census Bureau's Building Permits Survey asks permit-issuing jurisdictions to report new privately-owned housing units authorized, broken into four structure categories: one unit, two units, three to four units, and five or more. That last bucket is the standard multifamily proxy. It is published at the national, regional, state, metropolitan and micropolitan, county, and individual place level. The universe runs to roughly 20,100 permit-issuing jurisdictions, covering all 3,143 counties, and more than 99 percent of privately-owned residential building in this country happens inside a permit-issuing place.

Response is voluntary. But because building permits are public records, the survey carries an exception to the usual confidentiality rules, which is why local-level data can be published at all. That exception is the entire reason a study like this one is possible.

Three mechanical things to know before using it.

Imputation. Not every jurisdiction reports every month. Missing values are either pulled from another source or imputed on the assumption that the ratio of current-month to year-ago authorizations is the same for reporters and non-reporters. County and place products distinguish between reported-only figures and estimates including imputation. Since January 2025 the imputation factors have been calculated at the division level, with Florida, Texas and California handled individually rather than folded into a region.

The 2022 break. Beginning with January 2022 data, the monthly survey moved from a representative sample to a cutoff sample. Collection is now attempted monthly on the roughly 8,600 jurisdictions that averaged more than five new units annually between 2018 and 2020; the remaining 11,400 or so are imputed monthly and surveyed annually. The consequence is good news and a trap at once. From 2022 forward, monthly metro and county data provide complete coverage, which is what makes metro-level monthly analysis viable. But it also means pre-2022 monthly metro series are not cleanly comparable to what came after. Anyone charting a metro's monthly permit history across that boundary is charting a methodology change as much as a market.

Revisions. Each monthly release revises the prior month. The annual benchmarking every May can move things further. Single months are noise; trailing twelve-month sums are signal.

One more thing worth stating plainly, because it has affected every figure in this article. The lapse in federal appropriations that began on October 1, 2025 disrupted the release schedule badly. September and October 2025 data came out together on January 9, 2026. November and December followed together on February 18. February and March 2026 were pushed to a combined release on April 29. The 2025 annual data, normally out on the first workday of May, landed on May 14, 2026. Collection itself was affected, not just publication. Expect larger revisions than usual across that stretch.

The national picture, in four series

The headline permit number is only one of four related series in the monthly New Residential Construction release, and reading them together is the only way to understand where the cycle actually sits.

Permits. June 2026 came in at a 445,000 annual rate for buildings of five units or more, down 4.9 percent from May and 6.3 percent from a year earlier. That is a step back from December's 515,000, and it is why anyone declaring a recovery on a single month deserves suspicion.

Starts. June 2026 multifamily starts ran at a 513,000 annual rate, up 76.3 percent month over month and 19.3 percent year over year. A seventy-six percent monthly jump is not a market event. It is a small-sample artifact of a series that counts excavation, and it reversed an equally implausible drop the month before. Full-year 2025 starts were up 17.4 percent over 2024, which had itself fallen roughly a quarter. The National Association of Home Builders projects multifamily starts declining about five percent in 2026 to a 392,000 pace and another six percent in 2027 to 367,000.

Completions. 2024 was the flood. Multifamily completions hit 608,000 units, the highest since 1986, with 54 percent of them in buildings of fifty units or more and 95 percent built specifically for rent. The South took 292,000 of them, just under half. Then it drained fast. Net deliveries ran roughly 695,000 in 2024, about 531,000 in 2025, and forecasts for 2026 cluster between 382,000 and, on the more conservative reads, closer to 300,000. RealPage counted just 75,205 units delivered in the first quarter of 2026, the thinnest quarter in four years.

Put the four together and the shape is clear. The wave that flooded 2024 was permitted in 2021 and 2022. The pipeline being permitted now will not reach a rent roll until 2027 or 2028. And the trough between those two waves is arriving right about now.

Why the Sun Belt stopped

The explanation is not mysterious, but it is more interesting than "oversupply."

The oversupplied markets are, in fact, oversupplied. San Antonio's vacancy sat near sixteen percent in the second quarter of 2026, Austin in the low thirteens, Oklahoma City around twelve. Austin had posted negative rent growth for eleven consecutive quarters through the middle of 2026. Dallas-Fort Worth had been negative since the third quarter of 2023. National vacancy hovered above eight percent, with asking rent growth running near half a percentage point annually and Yardi Matrix projecting roughly 0.5 percent for 2026, one percent for 2027, and 2.3 percent for 2028 before conditions normalize.

But demand did not disappear. Austin absorbed more than 20,000 units in the first quarter of 2026 against roughly 14,900 delivered. Phoenix led every market in the country with about 5,800 units absorbed in the same quarter. These are not markets where nobody wants to live. They are markets where too much arrived at once and rents had to break to clear it.

What stopped new filings was the underwriting. Multifamily cap rates rose to roughly 5.6 to 5.8 percent by 2026 from a 2021 low near four percent. Insurance became a genuine value driver rather than a line item: CBRE calculated that rising insurance costs had cut multifamily values 3.6 percent nationally since late 2019, with the South-Central region down 7.8 percent and Florida down 6.8 percent. Construction lending remained expensive, with bank facilities running roughly 5.5 to 8.75 percent and debt funds well into double digits at 60 to 75 percent of cost, meaning 25 to 40 percent equity. A large tranche of 2021 and 2022 loans is refinancing from an average around 4.8 percent into something closer to 6.2.

Against that, the gateway and Midwest markets look different not because they got better but because they never got overbuilt. Los Angeles did not add a decade of supply in three years. Neither did Detroit, Chicago or Columbus. When the arithmetic tightened, the markets with the least new competition were the ones where a new project could still be underwritten.

There is also policy in the mix. Federal housing finance regulators set the 2026 government-sponsored enterprise purchase caps at 88 billion dollars each for Fannie Mae and Freddie Mac, 176 billion combined, a twenty percent increase over 2025. Cheap agency permanent debt, quoted in the low fives for the best credits, does not rescue a project in a market with negative rent growth. It does rescue one in a market with none.

The finding: a permit is a rear-view mirror

Now the part that reframes everything above.

Almost every treatment of building permit data calls it a leading indicator. That is true in the narrowest sense. Permits lead starts by about a month and completions by twelve to eighteen. The Conference Board includes them in its Leading Economic Index for exactly that reason.

But a permit is only leading if you are asking about construction. If you are asking about the development industry, it is one of the most backward-looking numbers published.

Consider what has to happen before an application reaches a permit counter. Site selection. Land control. Survey. Concept planning and test fits. Civil engineering. Then entitlement, which in most American jurisdictions runs three to eighteen months on its own, with a variance or conditional use permit adding three to four months and a rezoning adding four to six on top of that. In Seattle, entitlement plus permitting on a mid-rise routinely consumes 24 to 36 months. California's Legislative Analyst has reported roughly two and a half years as the average approval time for projects requiring a rezoning, and a 2024 UCLA working paper found that Los Angeles projects requiring a full environmental impact report took 504 additional days on average.

Even in the fast markets the lag is real. Dallas can move a by-right multifamily project from application to permit in three to six months, but add a planned development rezoning and another six to twelve months appears.

So when the June 2026 permit data shows Los Angeles up ninety-four percent, that is not a signal about the Los Angeles market in June 2026. It is a receipt for site planning, engineering and entitlement work that was commissioned and paid for somewhere between the middle of 2024 and the middle of 2025. The capital decision came earlier still.

The metro rotation visible in today's permit data is a picture of where developers decided to go roughly two years ago. It is genuinely useful, in the same way a company's earnings report is useful. It is also, unavoidably, history.

The backlog makes this worse, and more interesting. Census tracks a series most people never look at: units authorized but not started. For buildings of five or more units, that figure stood around 110,000 in January 2026, down from a February 2023 peak near 160,000 but still far above the pre-2021 norm of roughly 30,000 to 95,000, and multiples of the 27,000 to 37,000 lows of the post-crisis years.

A hundred and ten thousand permitted apartments sitting unbuilt is a lot of paid-for design work waiting on a capital stack. It also means the conversion rate from permit to start is currently worse than the historical relationship implies. For multifamily, roughly a third of projects break ground in the same month the permit issues and about eighty percent within two months, which is meaningfully slower than single-family, where nearly half start immediately and more than ninety percent within two months. When financing seizes, that distribution stretches, and a permit becomes what it always technically was: an option to build, not a commitment to build. Permits expire. Extensions get filed. Some projects never come back.

That is the counterintuitive claim, stated plainly. The 2022 permit peak overstated how much would get built, because a chunk of it never started. The 2025 permit recovery understates how much design work is happening, because the design work that will produce 2028's permits is being done right now, invisibly, in markets that will not show

up in this data for another two years.

What to watch instead

If permits are a lagging read on development activity, the leading read has to come from earlier in the chain.

For anyone trying to anticipate the next rotation rather than confirm the last one, the signals sit upstream: land transaction volume, pre-development site acquisition, entitlement application filings at the planning department rather than the building department, and rezoning petitions. None of that is aggregated in a federal dataset. Some of it is available locally, in the same municipal open data portals that publish permits, and it moves twelve to twenty-four months ahead of the number everyone quotes.

The second thing to watch is the supply cliff. Under construction is down by nearly half from its peak. Deliveries fall through 2026 and probably 2027. In markets that spent three years absorbing everything thrown at them, the rent line and the new-supply line are going to cross, and the metros where that happens first are the ones where developers who were early into entitlement will have permits in hand and everyone else will be starting a two-year process from scratch.

That is not a prediction about rents. It is an observation about queue position.

What is being designed differently

Three structural shifts in the data bear directly on how these projects get laid out, and they are worth reading alongside the permit counts.

Bigger buildings. In 2024, 54 percent of multifamily completions were in buildings of fifty units or more, up from 25 percent in 2004. Over the same span, buildings of ten to nineteen units fell from 24 percent of completions to four percent. The middle of the market has largely stopped being built. That has consequences for parcel size, structured versus surface parking, fire access, and everything else that follows from density.

Parking reform is now mainstream, not marginal. More than 3,700 cities worldwide have reduced or eliminated parking requirements, and over a hundred have removed minimums entirely. Austin repealed its minimums citywide in 2023. Colorado barred them for transit-area multifamily as of mid-2025. In Minneapolis, average parking provided per new unit fell from about 1.0 to under 0.75. A 2025 federal analysis drawing on Sightline Institute research estimated that removing Colorado's minimums would yield 71 percent more homes in transit-oriented areas and 41 percent more overall. Every one of those changes is a site plan question before it is a policy question.

Build-to-rent cooled off after running very hot. Roughly 39,000 single-family rental homes were delivered in 2024, a 455 percent increase over 2019. But starts fell to about 68,000 in 2025 from roughly 84,000 in 2024, and first-quarter 2026 single-family built-for-rent starts came in around 14,000 against 19,000 a year earlier. More than 110,000 units remain under construction, with Texas at roughly 22,000 and Arizona and Florida near 14,000 each, and 82 percent of the pipeline concentrated in the Sun Belt. Horizontal multifamily is a fundamentally different land planning problem than a five-over-one, and the geography of it is precisely the geography that is currently slowing down.

The pulse, read correctly

The national multifamily permit number bottomed and turned. That is real and it is worth knowing.

But the useful version of this data is not the national number, and it is not even the metro ranking. It is the recognition that this series measures the output of a development process that started years earlier, and that the interesting question is therefore not what the permit count says about next quarter but what it reveals about decisions already made.

By that reading, 2026's data says something fairly specific. Capital rotated out of the Sun Belt sometime in 2024. The markets absorbing that rotation are the ones that spent the boom being called boring. And the projects that will define 2028 are sitting on drawing boards right now, in metros that will not appear in a permit table for another two years.

We will run this series again when the next annual file drops. The rotation is the story worth tracking.

Sources:

• Primary federal dataUS Census Bureau Building Permits Survey, including the CBSA and county files and the January 2022 methodology change.

• Census and HUD monthly New Residential Construction release for permits, starts, units under construction and completions.

• Federal Reserve Economic Data series for units authorized but not started in buildings of five units or more.

• National Association of Home Builders and its Eye on Housing arm for annual permit and start totals and forecasts.

• RealPage, Arbor, GlobeSt and the National Apartment Association for metro-level permit movement.

• CBRE for the effect of insurance costs on values. Newmark and Yardi Matrix for cap rates.