How many empty homes are in the United States?
Executive summary
Official datasets and major analyses converge on roughly 14.8–16 million vacant housing units in the United States — roughly 9–10% of the housing stock — but the meaning of “empty” depends on definitions: many are seasonal, held off market, uninhabitable, or otherwise unavailable to people seeking homes [1] [2] [3] [4]. More narrowly defined tallies — vacant year‑round units, investor‑held unoccupied properties, or vacant units in large metros — produce very different headline numbers, so the short answer is “about 15 million, with important caveats” [5] [6] [7].
1. The headline: roughly 15 million vacant homes, by Census accounting
Multiple outlets that rely on U.S. Census Bureau vacancy counts put the total at roughly 14.8–16 million vacant housing units nationwide, a share that translates to about 9–10% of all housing stock [1] [2] [8] [4]. These are the broad “vacant housing unit” figures used in the Census Bureau’s Housing Vacancies and Homeownership release and reproduced in datasets such as FRED, which tracks vacant units and related subcategories [9] [5].
2. Why the range exists: definitions and subcategories matter
The Census definition driving those headline totals treats a unit as vacant if no one is living in it at interview time, but that includes seasonal or recreational units, properties “held off the market,” and units temporarily occupied by people with another usual residence, while excluding dwellings exposed to the elements or condemned — so the composition of the vacant pool is heterogeneous by design [9] [10]. LendingTree and other studies emphasize that a big slice of vacant units are seasonal/recreational, which reduces the practical supply available to year‑round home seekers [2] [3].
3. Different lenses give different counts: metros, investor holdings, and “available” homes
When analysts narrow the field — for example to vacant units within the largest metros or to properties actively available for sale or rent — the numbers fall sharply: a LendingTree analysis focusing on 50 major metros reported about 5.6 million vacant properties across those markets, and ATTOM’s property data identified roughly 1.4 million residential properties with no occupants at a recent quarter, of which investors owned more than 882,300 [7] [6]. Those narrower measures are closer to “supply that could immediately relieve local demand,” but they are not what the broad Census vacancy number counts.
4. Context and trends: peak after 2008, then decline, and persistent mismatch
Vacancy totals peaked near 19 million in the wake of the 2008 housing crash and declined more than 20% by the mid‑2020s as homeownership recovered, but vacancy remains large in raw terms even as household formation and demand have put upward pressure on prices [3] [4]. Analysts and journalists repeatedly note the seeming paradox — millions of vacant homes alongside affordability crises — and point out that the mismatch is largely structural: many vacant units are in low‑demand areas, seasonal markets, off‑market, or in poor condition, so they don’t translate into instantaneous affordable supply [1] [4].
5. What the numbers don’t tell — and why that matters for policy
Public datasets are blunt instruments: they record occupancy at interview time and classify units into subcategories, but they do not tell whether a vacant house is habitable, legally available, subject to local vacancy taxes, entangled in probate, or economically feasible to retrofit — factors that determine whether an empty home can address homelessness or active housing shortages [9] [10]. Reporting based on property data firms (e.g., ATTOM) or market slices (large metros, investor portfolios) supplements Census numbers, but each source carries different incentives and blind spots, so policymakers should triangulate among Census vacancy totals (~15 million), investor‑level counts (~882k investor‑held vacant houses within ~1.4M unoccupied residential properties), and market‑specific vacancy measures when designing interventions [6] [7] [9].