A borrower brings a lease-up deal. The broker pulls comps. The comps show concessions, because the submarket absorbed a lot of new units in the last eighteen months. Everyone underwrites to the comps, prices in the concessions, and the deal either works or it does not.
Nobody in that chain asked the question that decides it: how many units are still coming.
Because the answer changed, and it changed quietly.
What the pipeline actually looks like now
Multifamily. Starts and deliveries are running roughly a third below the 2023 and 2024 cycle highs (Yardi Matrix, August 2026). The units causing today's concessions were financed in a different rate environment and are finishing now. What follows them is a materially thinner set.
Office. This is the one nobody believes until they read it. CBRE has office under construction at 15.4 million square feet, down about 87 percent from the second-quarter 2020 peak, with only 2.2 million square feet completed in the quarter, the lowest first half since CBRE began tracking in 1990 (CBRE, Q2 2026).
Retail. Vacancy sits at 4.4 percent and has held there, with more than half of tracked markets under 5 percent. Mall vacancy fell 30 basis points to 8.5 percent, and roughly 4,130 store openings were announced in the first half, slightly exceeding closures. Retail effectively stopped building a decade ago and is now living with the consequences in both directions.
Three property types, one pattern. The construction decision was made two or three years ago, in a market that no longer exists, and nobody re-made it.
Why the numbers you can see say the opposite
Here is the part that makes this hard to act on. The visible indicators are still soft, and they will stay soft for a while, because they are reporting on the wave rather than on what follows it.
Roughly a quarter of apartments are still offering concessions (Zumper, 25 August 2026). Austin is down 16.6 percent year over year, Houston 14.6, Dallas 13.0. Those are real numbers describing real lease-up pain in specific submarkets.
But underneath, the direction already turned. National occupancy is 95.5 percent, up 90 basis points since the start of 2026, with same-store effective asking rents up 0.9 percent year over year, an eighth consecutive monthly increase (RealPage, August 2026). Yardi has the average advertised rent at $1,773, up 0.4 percent annually, the strongest yearly reading in nearly a year.
And on the office side, the indicator moved further than anyone expected: vacancy fell 30 basis points to 18.3 percent, the largest quarterly decline since 2015, on a ninth consecutive quarter of positive net absorption and leasing up 16 percent year over year (CBRE, Q2 2026).
Concessions are the last thing to go, not the first. A landlord holding a building that filled up on concessions keeps offering them until they are sure they can stop. So concession data lags occupancy, which lags absorption, which lags the delivery schedule. By the time the comps look good, the trade is priced.
The honest counterweight
Two things cut against this, and leaving them out would make the argument worse rather than stronger.
Demand is not booming. Trailing twelve-month multifamily demand is running around 271,300 units against a decade average nearer 340,000 (RealPage, August 2026). Absorption is positive and improving, not strong. A thinner pipeline meeting soft demand produces stabilisation, not a squeeze.
The regional spread is enormous. The Midwest leads at 2.0 percent annual rent growth. The South is the only region posting declines and the only one below 95 percent occupancy, with San Antonio down 3.7 percent at 93.1 percent occupied (RealPage, August 2026). "The supply wave is over" is a national statement, and nobody underwrites a national property. In parts of Texas and Florida the wave is genuinely still washing through.
So the claim is narrow and it is this: the pipeline is no longer the dominant risk in most markets, and a lot of underwriting still treats it as though it were.
What to do with it, concretely
Underwrite against the delivery schedule, not against the comps. For any lease-up or value-add deal, pull units under construction and permitted in the submarket, with expected delivery dates, and set your absorption assumption against that rather than against what the neighbouring property had to offer in 2024. Those are different questions and only one of them is about your hold period.
Treat concessions as a burn-off assumption with a date on it. If the submarket has thin deliveries ahead, concessions are a temporary cost with an end, and modelling them as permanent understates the deal. If the submarket still has a lot coming, they are not temporary and you should say so. Either way it is an assumption that deserves a line in the memo instead of being inherited from a comp set.
Get the timing asymmetry the right way round. A project breaking ground now delivers into 2028 and 2029, into the thinnest pipeline in years. A project bought stabilised today is competing against units delivering through the back half of this cycle. Those are opposite exposures to the same fact.
Then check whether the capital agrees with you. This is where an argument becomes a deal. If your read on a submarket is genuinely differentiated, the lenders who share it are not necessarily the ones you already call. Build the deal memo with the underwriting and the comps in it, drop your lending preferences, and run competing quotes rather than taking the first one from the relationship that happens to be closest.
Before any of that, run the coverage. A thesis about supply does not survive a deal that does not clear a debt service test, and you can check the numbers in four fields before you build the model.
The short version
Comps describe the past. In a stable market that is close enough. At a turn it is the wrong instrument, and this is a turn.
The supply that produced today's concessions has been delivered. Multifamily starts are down about a third from the peak, office construction is down roughly 87 percent, and retail stopped building years ago. Meanwhile occupancy is climbing, absorption has been positive for nine straight quarters in office, and rents have risen for eight consecutive months in apartments.
That does not make every deal work. Demand is below its own average and the South is still absorbing. But it does mean the risk most models are pricing most heavily is the one that is retiring, and the assumption worth arguing about is not what the comps did. It is what is still coming out of the ground.