False Ceilings
March 2026

Multifamily demand analysis · 2025–2035

The Last
Tailwind
The New
Normal?
After the
Surge
The Era That
Taught Us Wrong
326K

Multifamily's decade of structural growth was powered by forces that are now decelerating simultaneously. A data model of where demand lands, and how wide the range of outcomes really is.

Scope: 5+ unit professionally managed apartments, US national Model: 6-driver structural + Monte Carlo (2,000 runs per scenario) Horizon: 2025–2035 Benchmarked against: JCHS, NAA/NMHC, CBO
What this model measures

This is a demand forecast. It estimates how many households will want to live in a multifamily apartment over the next decade, based on demographics, immigration, and structural economic forces.

It is not a supply forecast. It does not model how many units get permitted, built, or delivered. It does not model absorption, vacancy, or rents. Those are downstream consequences of demand meeting (or not meeting) supply. This model sits upstream of all of them: how many people will want apartments, before we ask whether anyone builds them.

The diagram below shows where this model fits in the broader multifamily cycle. Everything to the right of the highlighted box is out of scope.

Demographics
Population growth, cohort sizes, immigration, aging
This model
Household formation demand
How many households want a 5+ unit apartment
Effective demand
Filtered by income, affordability, credit access
Supply response
Permits, starts, completions, capital availability
Market outcomes
Absorption, vacancy, rent growth, NOI
Why this matters: Most industry analysis starts at supply response or market outcomes and works backward. This model starts at the beginning: the structural forces that determine how many people will need apartments. If demand is structurally lower than what the industry has been underwriting to, no amount of supply-side analysis changes that.
Model controls
Structural scenarios
Immigration recovery speed
Slow Fast CBO baseline
AI deployment lag
0 yr 8 yr 4 yr lag
Show uncertainty bands

Estimated multifamily households (millions), 5+ unit buildings. Historical: Census HVS/ACS 2000–2024 (2024 = 22.0M, ACS official). Projections: Monte Carlo P10–P90 bands per scenario. Immigration slider adjusts the recovery path from 2026 onward. All scenarios share the same 2025 trough from the current immigration policy shock.

Net new multifamily households per year (000s). The dashed line marks the 2011–2017 surge era average (326K/yr), the benchmark baked into a generation of operator underwriting. No scenario returns to it. The 2025 trough reflects the immigration cliff. The 2029–2030 peak reflects immigration normalisation colliding with the Boomer 65+ peak. The post-2031 decline reflects thinner birth cohorts born during the post-2008 fertility collapse entering prime renting age.

View

Plateau scenario only. Immigration share varies by scenario (15% in Erosion, 42% in Plateau, 50% in Resilience). This decomposition is not representative of the other paths. Under the Plateau scenario, immigration accounts for approximately 42% of gross renter HH demand variation year-to-year. It is the dominant single driver. Native cohort (blue, Gen Z peak) declines structurally from 2030 as post-2008 thin birth cohorts arrive. 55+ renter offset (red) peaks as all Boomers pass 65 by 2030, then tapers as smaller Gen X replaces them.

Average annual MF HH demand (000s/yr) over the 2025–2035 projection window. Our three scenario bars update live with the immigration and AI lag sliders above. Expert benchmarks are fixed point estimates. The first bar (326K/yr) is the 2011–2017 surge era average, the operator underwriting benchmark, not an all-time mean. JCHS figures apply 47% MF capture rate to total renter household projections. NAA/NMHC 2022 is natively in apartment units.

Source Annual MF demand Year Alignment Key caveat
JCHS Harvard — Base scenario ~140K/yr 2025 Aligned Middle immigration path; approximates our Plateau scenario at CBO baseline slider position
JCHS — High-immigration scenario ~246K/yr 2025 Aligned High immigration recovery assumed; approximates our Resilience scenario with slider at 75–100
JCHS — Low-immigration addendum ~100K/yr Dec 2025 Aligned Published Dec 2025 in response to current immigration policy; approximates our Erosion scenario with slider at 0–25
NAA/NMHC (Hoyt/Eigen10) 266K/yr 2022 Context-dependent Pre-immigration shock baseline (2022); shifts relative to current slider position
CBO Demographic Outlook Jan 2026 Input to model CBO immigration path used directly in our sub-model

Every industry argument about multifamily demand implicitly picks a reference era. Pick the wrong one and the math lies to you. The chart below applies structural break detection to the full 2000–2035 series to identify epochs of genuinely distinct demand regimes, and to show where the current forecast sits relative to each of them. The finding: the surge era that shaped every underwriting assumption in this industry was an anomaly, not a baseline. The central scenario sits at 61% of it. Even the optimistic scenario doesn't reach 80%. The industry has been pricing for a world that ended in 2017.

Net new multifamily households per year (000s), 2000–2024 historical and 2025–2035 projected (Plateau scenario, adjusted by current slider settings). Vertical dashed lines mark structural break points detected via PELT algorithm (L2 penalty). Coloured bands indicate distinct demand regimes. The horizontal reference line at 326K marks the post-GFC surge era average (2011–2017), the benchmark that shaped a generation of operator underwriting.

Period Avg demand vs surge era Character
2000–2004 Hist −176K/yr −154% Dot-com bust and 9/11 shock. Household formation collapses as unemployment spikes and consumer confidence craters. MF stock actually shrinks. The era operators have wiped from memory.
2005–2007 Hist +117K/yr −64% Credit-fuelled ownership boom. Loose mortgage standards pull young renters into ownership at record rates, suppressing MF demand even as the economy roars. Homeownership hits 69%, an all-time high. The bubble is inflating and everyone in apartments is losing residents to it.
2008–2010 Hist +240K/yr −26% Mortgage crisis and foreclosure wave. Six million homeowners lose their homes between 2007–2012. They don't disappear. They become renters. MF formation surges precisely because the financial system is imploding. The worst economic crisis in 80 years is multifamily's best lead-generation event.
2011–2017 Hist +326K/yr Baseline Post-crisis renter wave. Millennials enter peak renting age, homeownership stigma runs high, and credit tightening keeps ownership out of reach. The demographic and cultural stars align simultaneously. Every underwriting model, cap rate assumption, and development pipeline gets calibrated to this era. That is exactly the problem for what comes next.
2018–2020 Hist +240K/yr −26% Supply glut meets affordability ceiling. Record apartment completions hit a market where rent-to-income ratios have stretched to breaking point. New supply competes for a shrinking pool of affordable renters. The cycle was already rolling over when COVID arrived to confuse the read entirely.
2021–2023 Hist +297K/yr −9% Pandemic distortions unwind. Remote work triggers household splitting (roommates separating, young adults fleeing parents), stimulus cash funds deposits, and two years of suppressed formation catches up at once. The industry mistakes a one-time rebound for renewed structural strength. By late 2023, absorption collapses and the illusion evaporates.
2024–2027 Proj +166K/yr −49% Three shocks arriving simultaneously. Executive Order restrictions slash net immigration to near-zero. Post-COVID household formation normalises downward. AI-driven hiring freezes begin suppressing early-career formation among the 25–34 cohort. Forces that arrive together are more damaging than sequential ones. Each removes a pillar the others could have offset.
2028–2035 Proj +210K/yr −36% Demographic maturity, not recovery. The thin cohorts born after 2008 enter prime renting age. Immigration partially normalises under CBO baseline assumptions. But the surge-era conditions: a once-in-a-generation demographic bulge, homeownership stigma, and loose credit. They do not return. This is the new structural floor, not a launchpad.
The industry's reference era is 2011–2017: 326K/yr average. That is the number baked into cap rates, development pipelines, and absorption assumptions across most of the institutional market.

Against that benchmark, this model's central forecast is not a continuation. It is a structural step-down. Even the optimistic scenario (Resilience, 252K/yr average) runs at 77% of surge-era norms. The central scenario (Plateau, 199K/yr) runs at 61%. The pessimistic scenario (Erosion, 142K/yr) runs at 44%.

This is what "the last tailwind" means. Not that demand disappears. It doesn't. But the era that taught the industry what normal looks like is over. The new normal is lower, more volatile, and far more sensitive to immigration policy and AI-driven labour market shifts than anything operators have priced for.

Note on conservatism: These scenario averages do not incorporate the homeownership affordability constraint (see Methodology tab), which structurally reduces household exits from rentals into ownership. That omission likely understates MF demand by 15–30K/yr across all scenarios. The step-down argument holds even after adjusting upward. Surge era avg (2011–2017): 326K/yr  |  Plateau projection avg: 199K/yr  |  Implied step-down: −39% (before affordability upward adjustment)

What this model is — and isn't

This is a structural demand model, not an econometric forecast. It is designed to be transparent about how its numbers were derived and honest about where genuine uncertainty lies. Every assumption is either grounded in cited data, inferred from the literature, or explicitly flagged as a modelled construct.

Throughout this tab, assumptions are labelled using three epistemic categories:

Grounded Directly sourced from cited empirical data. Inferred Derived from literature but not directly observed. Modelled Constructed assumption. Reasonable but uncertain; needs verification.

Driver 1 — Native cohort formation (Gen Z peak years 2025–2032)

GroundedBase annual demand ~265K–285K units (Resilience), ~240–265K (Plateau), ~220–250K (Erosion), derived from JCHS McCue et al. 2025 cohort projections and Census ACS headship data.
GroundedThe 2025–2032 projection window coincides with Gen Z (born 1997–2012) entering peak household formation ages. The oldest Gen Zers turn 28 in 2025; the core of the cohort moves through 25–34 peak renting ages through the early 2030s. Pew Research documents the share of 24–25-year-olds living with parents declining from 20% to 18% in 2024–2025, signalling accelerating formation. Demand decline from 2030 onward reflects the post-2008 birth cohort trough. Thin years born during the fertility collapse following the GFC enter prime renting ages ~2029–2033. CDC National Vital Statistics confirms this.
InferredMF capture rate held at 47% throughout. ACS/NMHC 2022 shows 47% of renter households in buildings of 5+ units. We infer this rate is structurally stable absent major supply-side shifts.

Driver 2 — Immigration contribution

GroundedImmigration baseline: CBO Demographic Outlook, January 2026. Trough ~400K net in 2025, recovery to ~1.1M/yr by 2030. This is the most current official projection available.
InferredImmigration share of multifamily demand: 15% (Erosion), 42% (Plateau), 50% (Resilience). Derived from JCHS and Pew Research on immigrant household formation and renting propensity. Not directly measured.
InferredSlider maps 0→0.4× and 100→1.5× of the immigration component. Range bounds are illustrative of plausible policy scenarios; they are not probability-weighted.

Driver 3 — Headship rate effect

Grounded25–34 headship rate fell ~10pp since 1980 peak (Treasury Department, JCHS). This suppressed household formation is an established structural fact, not disputed.
InferredAnnual headship drag: –5K units/yr central case, σ ±3K. This reflects continued affordability constraint. The direction is well-evidenced; the magnitude is calibrated, not measured.
ModelledAI displacement sub-channel (see AI Deployment Lag below) feeds into this driver by suppressing headship among displaced 25–34 workers. This is the most speculative component of the model.

Driver 4 — Racial/ethnic composition shift

GroundedJCHS (McCue et al. 2025): 98% of net household growth 2025–2035 will come from non-white households. Hispanic and Black renter rates ~55% vs ~27% for white non-Hispanic households (ACS 2022).
InferredAnnual demand add from composition shift: ~8K units/yr (σ ±4K). Inferred from the JCHS growth split combined with differential renter rates. Direction is well-grounded; exact magnitude is a back-of-envelope calculation.

Driver 5 — 55+ renter offset

Grounded55+ renter households grew 43% from 2009–2019 (Harvard JCHS). All Baby Boomers will be 65+ by 2030 — a demographic certainty.
InferredAnnual demand add: ~15K units/yr (σ ±8K), peaking 2029–2031, tapering thereafter as smaller Gen X cohorts replace Boomers. Taper timing and magnitude are modelled extrapolations.

Capture rate sensitivity — homeownership affordability constraint

GroundedThe median monthly mortgage payment on a median-priced home is approximately $1,200 more than average apartment rent as of 2025 (Freddie Mac). Only 28% of US households qualify for a mortgage on a median-priced home. The median age of first-time homebuyers reached 40 in 2025, the highest on record (Viking Capital/NAR). These are structural, not cyclical, constraints.
InferredThis model holds the MF capture rate constant at 47% (share of renter households in 5+ unit buildings). In practice, sustained homeownership unaffordability reduces the outflow of households from rentals into ownership, effectively raising the MF-relevant share of total renters above 47%. The direction of this effect is well-evidenced; its magnitude is not modelled. A capture rate of 50–52% under high-affordability-constraint conditions would add approximately 15–30K/yr to MF demand across all scenarios. This is a meaningful upward bias the model does not capture.
ModelledThis driver is intentionally excluded from the structural model to maintain comparability with JCHS and NAA/NMHC benchmarks, which also use fixed capture rates. It is named here as a known upward bias: the model's outputs are likely conservative by 15–30K/yr on this dimension alone.

AI deployment lag parameter

GroundedAI capability benchmarks show rapid saturation velocity. Stanford HAI AI Index 2025: GPQA (PhD-level reasoning) gained 48.9pp in a single year (2023–2024) and saturated by Nov 2025. SWE-bench (real software engineering) went from 4.4% to 71.7% in 2024. MMLU saturated Sep 2024.
InferredBenchmark saturation is used as a leading indicator of white-collar role displacement risk, not a direct measure of employment impact. The gap between capability and deployment is real but closing. Anthropic "Labor Market Impacts of AI" (March 2026) finds no systematic unemployment increase yet, but documents hiring slowdowns in AI-exposed early-career roles (Stanford: –16% employment in exposed occupations aged 22–25 since ChatGPT launch). The inference chain from benchmark saturation to headship suppression runs: capability arrival → hiring contraction in entry-level white-collar roles → income suppression for 25–34 cohort → delayed household formation. Each step is plausible; the compounded inference is speculative. Treat the AI channel as a directional sensitivity, not a point estimate.
ModelledDeployment lag slider (0–8 years, default 4): captures the gap between AI capability arrival and actual workforce penetration, driven by regulation, organisational inertia, adoption friction, and reskilling. Default of 4 years is consistent with historical technology adoption lags (electrification, computing). The range bounds are illustrative.
ModelledDisplacement effect on headship rates: –2 to –4pp for 25–34 cohort at peak under aggressive scenario, translating to ~150–300K fewer renter households. This is a modelled construct derived from Brookings (2026) adaptive capacity analysis and Anthropic labor paper. Treat as order-of-magnitude, not a point estimate. Erosion scenario is most sensitive; Resilience least (productivity gains partially offset).

Monte Carlo uncertainty bands

ModelledP10/P90 bands from 2,000-run Monte Carlo per scenario (6,000 total, seed=42). Six uncertain parameters drawn per run: native formation (σ ±35K/yr), immigration by scenario (σ ±20–50K/yr), headship drag (σ ±3K/yr), racial composition shift (σ ±4K/yr), 55+ renter offset (σ ±8K/yr), MF capture rate (σ ±3pp). Year-level economic shock ±20K/yr also applied. Important caveat: the ±20K/yr shock reflects normal business-cycle variation. A genuine recession, particularly one without a foreclosure wave, could suppress household formation by 60–120K/yr for 2–3 consecutive years, which would push outcomes below the P10 band. No explicit recession scenario is modelled; the P10 boundary should not be treated as a recession floor.
Modelled2035 P10–P90 spreads: Erosion 0.63M, Plateau 0.76M, Resilience 0.91M. Wider for higher-demand scenarios because immigration uncertainty compounds more. These are not confidence intervals in a frequentist sense. They reflect the range of outcomes given the parameter assumptions, which themselves carry uncertainty.

What the model does not capture

Supply-side constraints (construction costs, permitting, labour). Rent elasticity effects on household formation. Geographic concentration of demand. Credit market conditions affecting homeownership transition. Homeownership rate dynamics (discussed in detail in its own section below, and a candidate for inclusion as a sixth structural driver in a future version). Second-order productivity effects of AI on household income. Single-family rental substitution: the model holds MF capture rate at 47% but institutional SFR has been gaining share since 2012 and a declining capture rate would reduce MF-specific demand even if total renter demand holds. Supply-side feedback on demand: when units are unavailable, some household formation is suppressed or delayed; this model does not capture latent demand. Tariff-driven construction cost increases (estimated at $12,800–$25,500 per unit by John Burns Research, 2025) reduce new supply and indirectly support demand absorption, a positive feedback loop this model ignores. All of these matter and none are in scope here. This model addresses the demand side only.

One structural limitation worth naming explicitly: the 55+ renter offset is a gross addition. It counts new senior renter households forming, but does not net out household dissolution as Boomers age into assisted living or die. The net contribution of the senior cohort is therefore somewhat smaller than the model implies, particularly after 2030. This is why the stock floor never falls below the 2024 baseline even in the Erosion scenario. Annual net formation remains positive throughout, but the Erosion slope through 2029–2032 is likely modestly overstated.

The unmodelled driver: homeownership rate

GroundedThe US homeownership rate has ranged from 63.4% (2016 trough, post-GFC) to 69.2% (2004 peak, credit bubble) over the past 25 years, settling at approximately 65.7% as of Q4 2024 (Census Bureau HVS). It is not stable — it moves by multiple percentage points across economic cycles and responds to affordability conditions, credit standards, and demographic composition shifts.
GroundedThis model does not include homeownership rate as an explicit driver. It is the most significant structural variable absent from the model. The MF capture rate (held fixed at 47%) operates downstream of the rent-own split — it measures what share of renters live in 5+ unit buildings, not whether a household rents at all. The upstream question of how many households are renters in the first place is determined by homeownership rate, and that assumption is buried rather than named.
InferredDirectional sensitivity: each 1 percentage point shift in the homeownership rate moves approximately 1.3 million households between owner and renter status at current household counts (~134M total). Applied against the 47% MF capture rate, that translates to roughly 600K MF households. However, this figure almost certainly overstates the true MF impact — households on the ownership margin (those who might tip from renting to owning, or vice versa) are disproportionately older and more affluent, and more likely to occupy single-family rentals than 5+ unit apartments if they cannot buy. A more conservative MF capture rate for marginal renters — closer to 30–35% — would imply 390–460K MF households per 1pp ownership shift. The direction is clear; the magnitude is contested.
ModelledThe model implicitly assumes homeownership rate stays roughly constant across all three scenarios through 2035. This is a strong assumption: the Erosion scenario (sustained immigration suppression, AI-driven income uncertainty, persistent affordability stress) is structurally more consistent with continued ownership suppression, which would add to renter demand and partially offset the immigration headwind. The Resilience scenario (strong economic recovery) is more consistent with ownership recovering toward 67–68%, which would reduce the renter pool and act as a partial drag on MF demand. These scenario-consistent ownership paths are not captured here. Their omission means Erosion demand is modestly understated and Resilience demand is modestly overstated.
ModelledThree forces drive homeownership rate historically and are absent from this model's explicit architecture: (1) Mortgage affordability — the combined effect of rate, price, and income on monthly payment burden, currently at historically elevated levels with 30-year rates near 6.1% and median home prices requiring monthly payments roughly $1,200 above average apartment rent; (2) Credit access — lending standards that determine whether would-be buyers can qualify regardless of desire; (3) Demographic composition — because Hispanic and Black homeownership rates run 25–30pp below white non-Hispanic rates (ACS 2022), the racial composition shift already modelled in Driver 4 has a homeownership corollary that structurally suppresses aggregate ownership rates independent of macro conditions. This third force means the racial composition driver partially captures the homeownership story, but not explicitly and not completely.
PlannedHomeownership rate could be incorporated as a sixth structural driver in a future version of this model. Proper implementation requires: establishing scenario-specific ownership rate paths anchored to FRED/Census HVS historical data; empirically estimating the MF capture rate for marginal renters (likely requires ACS microdata); and adding ownership rate uncertainty to the Monte Carlo parameter set. That is a Claude Code session with primary data work — not appropriate to rush for initial publication. The omission is named here rather than hidden.

Primary sources

CBO Demographic Outlook, January 2026 — Immigration baseline and population projections
JCHS Harvard, McCue et al. 2025 — Household formation, racial composition projections, 55+ renter trends
Anthropic, "Labor Market Impacts of AI," March 2026 — AI observed exposure measures, employment trends
Stanford HAI AI Index 2025 — Benchmark progression data (MMLU, GPQA, SWE-bench saturation dates)
Brookings, "Measuring US Workers' Capacity to Adapt to AI-Driven Job Displacement," Feb 2026 — Adaptive capacity by occupation
Census Bureau HVS/ACS 2022 — Historical MF stock, headship rates, renter household composition
NAA/NMHC, Hoyt/Eigen10 2022 — Expert demand benchmarks (Expert benchmarks tab)
Pew Research Center — Immigrant household formation and renting propensity
FRED/Federal Reserve, CDC National Vital Statistics — Historical macroeconomic series, birth cohort data
2035 outcomes by scenario
2024 baseline
22.0M
MF households (ACS 2024)
Erosion median 2035
23.6M
+142K/yr avg · immigration near-zero
Plateau median 2035
24.1M
+199K/yr avg · immigration normalises
Resilience median 2035
24.6M
+252K/yr avg · immigration rebounds
Erosion
Key assumptionImmig stays near-zero
Net new 2025–2035+1.64M
Avg annual~142K/yr
P10 / P90 203523.34 / 23.96M
JCHS equivalentLow-immigration addendum
Plateau
Key assumptionImmig → ~920K/yr
Net new 2025–2035+2.07M
Avg annual~199K/yr
P10 / P90 203523.70 / 24.47M
JCHS equivalentBetween base and low-traj
Resilience
Key assumptionImmig rebounds to 1.1M+/yr
Net new 2025–2035+2.58M
Avg annual~252K/yr
P10 / P90 203524.14 / 25.05M
JCHS equivalentNear low-trajectory scenario
Model architecture

How this model works

This is an illustrative structural demand model, not an econometric forecast. Its purpose is to identify the dominant forces shaping multifamily demand over the next decade and show how sensitive the outcome is to each. Economists can and do build more rigorous models. This one is designed to be transparent, reproducible, and honest about its limitations.

The model has two layers: (1) a structural layer that projects annual multifamily household demand from five demographic and policy drivers, and (2) a Monte Carlo layer that applies historically-calibrated economic shocks on top of each structural path. 2,000 simulations per scenario, 6,000 total.

Native cohort formation
Gen Z and millennial age progression. Shaped by actual birth cohort sizes by year. Declines from 2030 as post-2008 thin birth years enter prime renting ages.
Immigration contribution
CBO January 2026 net immigration forecast. Trough in 2025 (~400K), recovery toward ~1.1M/yr by 2030. Dominant swing factor across all scenarios.
Headship rate effect
Share of adults forming own households vs doubling up. Treasury data: 25–34 headship fell 10pp since 1980. High rents suppress; easing releases latent demand.
Racial/ethnic composition
JCHS: 98% of net HH growth 2025–2035 will be non-white. Hispanic and Black households rent at ~55% vs ~27% for white households. Structural pro-renting offset.
55+ renter offset
All boomers 65+ by 2030 — known inflection point. 55+ renter HH grew 43% from 2009–2019. Peaks 2029–2031, then tapers as smaller Gen X replaces boomers.
Monte Carlo shocks
Six shock types calibrated on post-1980 history: mild/severe recession, immigration shock, rate spike, policy tailwind, pandemic. Applied to each structural path.

MF capture rate held at 47% (NMHC/ACS 2022). Historical series: Census 2000 decennial, HVS annual, ACS, Pew Research. All MF = 5+ unit professionally managed buildings. Note: the 55+ renter driver is a gross formation figure and does not net out senior household dissolution; the Erosion scenario stock line may be modestly overstated in the 2029–2032 window as a result. See Methodology tab for full discussion of model limitations.