Everyone is still reading the apartment market off the glut. The glut is what already got built.

Per Multifamily Dive's report on CoStar data, developers started roughly 55,000 apartment units nationwide in Q1 2026 — about 73% below the early-2022 quarterly peak, and the lowest quarterly figure since 2011.

Then it fell further. Per Inman's reporting on Census Bureau data, multifamily starts dropped 41.6% in a single month in May 2026, from a 486,000 to a 284,000 seasonally adjusted annual rate. That is a different unit from the quarterly figure above — an annualized pace, not units actually started in the quarter — but both series point the same direction, and the annualized one fell off a shelf.

Here is why that matters more than any rent forecast you'll read this quarter: a large apartment project takes roughly 18 to 24 months from groundbreaking to delivery. The buildings available to rent in 2028 are being started — or not started — right now. That isn't a projection. It's a calendar entry.

We wrote in edition 034 that the Sunbelt apartment glut had peaked. It did. Peaking is where the story starts, not where it ends.

The units delivering now came from starts made in 2024. The construction that would fill 2028 and 2029 is the construction that just collapsed. So the market spends the next stretch absorbing the tail of a building boom while the pipeline behind it runs dry — and most broker pro formas and most lender underwriting are calibrated to the tail rather than the gap.

Be precise about what this does and does not prove. Falling starts are a supply fact, and the lag arithmetic is arithmetic. Whether that becomes a shortage depends on demand in 2028, which nobody can source today. What is settled is the supply side: there is no version of the next two years that contains meaningful new inventory.

This is where it stops being a construction story.

Per FHFA's November 24, 2025 announcement, the 2026 multifamily loan purchase caps are $88 billion for each Enterprise — $176 billion combined, with at least 50% required to be mission-driven, affordable housing.

Now hold the two facts next to each other.

Federal housing policy has authorized $176 billion of capacity to finance apartments in 2026. Apartment construction fell to a fifteen-year low in the same window.

The lever being pulled is debt availability. The thing that stopped is feasibility — construction costs, insurance, taxes, and interest against rents that stopped rising fast enough to cover them. You cannot fix a feasibility problem with a lending cap, because a cap is permission to lend, not a reason to build. Developers aren't sitting out because nobody would fund them. They're sitting out because the math doesn't clear.

That's the whole disconnect in one sentence: policy is pushing harder on the one input that was never binding.

And there's a second-order version specific to the affordable mandate. Half of that $176 billion must go to mission-driven housing — the segment with the thinnest margins, the most fragile feasibility, and therefore the most sensitive to exactly the cost pressures that stopped the starts. The capital most tightly targeted is aimed at the deals hardest to make pencil.

Watch three numbers, not the headlines:

Construction cost trajectory. Feasibility returns when the spread between development cost per unit and stabilized value per unit reopens. That happens through costs falling or rents rising, and right now only one of those is plausibly moving.

Rent growth in the markets that already absorbed their supply. The metros that worked through their deliveries first are the leading indicator; when the overbuilt Sun Belt markets follow them, feasibility comes back. Watch the quarterly rent prints in your own submarket rather than the national average, which blends two markets moving in opposite directions.

The lag itself. Even if every constraint cleared tomorrow, the first new units are 18 to 24 months out.

If you own apartments and your loan matures in 2026 or 2027: your refinance is being underwritten against today's soft comps, and the fundamentals that fix your rent roll arrive after your maturity date. The whole game is bridging to 2028 — extensions, a rate cap you can actually afford, partial paydown, preferred equity. Every month of runway you buy is a month closer to a market with no new competition in it.

If you're buying: you're buying into the last soft window before the delivery gap. Underwrite the exit against a 2028–2029 delivery environment, not against the concession-heavy comps you're seeing today. But be honest about the middle — the tail of this delivery wave still has to clear, and the operator who runs out of liquidity in 2027 doesn't collect anything in 2028.

If you're developing: the projects that break ground in the next twelve months deliver into the thinnest competitive environment in a decade. Entitlements are the cheap thing to hold right now. Being shovel-ready when feasibility returns is worth more than being right about when it returns.

If you're a renter or an employer thinking about workforce housing: the concessions you're seeing this year are not a new normal. They're the tail of a wave that has already crested, in markets getting no meaningful new inventory until at least 2028.

Manufacturing solved this in 1984 and gave it a law.

Eliyahu Goldratt's theory of constraints states it in one line: the throughput of a system equals its most binding constraint, and improving any non-binding resource improves nothing. A plant with a slow oven does not make more bread by buying more flour. It makes more flour.

Apartment production is a system with several constraints stacked in series — capital, land, entitlement, labor, materials, and the rent the finished unit can charge against what it cost to build. Output is set by whichever of those binds hardest. Nothing else in the chain matters until that one moves.

For most of the last decade capital genuinely was the binding constraint, and expanding it genuinely did produce units. That is no longer true, and the starts data is what tells you so. Developers are not sitting out for lack of a lender. They are sitting out because the finished unit does not clear its cost.

So the $176 billion is flour. It is a real increase in a real resource, and it is aimed at the part of the chain that already had slack.

There's a second constraint underneath, and it's the reason this can't be fixed quickly even once feasibility returns: the 18-to-24-month build time. Time is a constraint you cannot buy your way past, and it is the one that turns today's decision into 2028's inventory whether anyone likes it or not.

Which points at where the actual lever sits, and it isn't federal. Entitlement speed, permitting time, and pre-approved plans are municipal. They are unglamorous, nobody announces them from a podium, and they are the only inputs that touch a constraint that is currently binding.

The tell for any policy response, in any market: ask which constraint it relieves. If the honest answer is "one that already had slack," you are watching flour arrive at a bakery with one oven.

There is an objection to all of this that I cannot rule out, and leaving it unsaid would be the kind of thing this newsletter exists to catch.

The argument above assumes that supply arriving late into a market whose demand has moved on produces a glut. That is what happened in the Sunbelt. But it is not what a long lead time necessarily does — it is what a long lead time does when the demand that justified the building turns out to have been temporary.

Run the same structure with durable demand and the outcome inverts. A long lag feeding into household formation that is genuinely still there produces a ramp toward a higher equilibrium, and the late supply gets absorbed rather than dumped. Same lag, same capital, opposite result. The variable that decides it is not in any construction dataset.

Which narrows this piece to something more useful than its headline. The supply gap is settled — that part is a calendar, and no forecast is required. What is not settled is which side of that gap the demand lands on, and anybody telling you they know is selling something.

So the question to carry into your own underwriting is the one the starts data cannot answer: is the household formation waiting on the other side of the lag durable, or was it borrowed from a pulse that has already passed? I am working on that number. When I have it from a primary source, it runs here.

Claim: Quarterly U.S. multifamily construction starts will not return to their 2021-2022 quarterly average through Q1 2027 — and at least one additional quarter between now and then will print below 100,000 units nationally, even as agency multifamily lending capacity remains at or above the 2026 cap level.

Stated confidence: 71% · Verification date: March 31, 2027 · Status: OPEN

This newsletter is the free version of what I do: take a number everyone is repeating, find out whether it holds, and show the work.

The paid version is that same check on your numbers. For this week's question, send me the market or the deal and I'll come back with the units actually under construction in that submarket, when they deliver, what's left in the pipeline behind them, and what your rent assumption has to be to survive the gap between now and 2028.

You get a written answer with the sources, the arithmetic, and an explicit list of what I could not verify. Same discipline as the tracked prediction above: a claim, the evidence, and no confidence I haven't earned.

cokas.io — written scope back in 48 hours. No sales call, ever.

Every prediction in The Pattern Brief carries a verification date and is revisited in a future edition — including this one.