The Census Bureau and HUD reported that apartment completions ran at a 302,000 annual rate in August, down 35.7% from a year earlier. That decline clears the survey's own margin of error. In the same release, permits for buildings with five or more units rose 9.4% year over year to 467,000. The number driving the headlines — a 22.5% monthly plunge in multifamily starts — is one Census cannot distinguish from zero.
What happened
The September 17 release, CB26-147, covers August and is jointly issued by Census and HUD. For buildings with five units or more, all figures seasonally adjusted annual rates:
- Completions: 302,000, down from 359,000 in July and 470,000 in August 2025. The annual decline of 35.7% carries a 90% confidence interval of plus or minus 19.6 points. That range does not include zero, so by Census's own standard the drop is statistically significant.
- Starts: 344,000, against 444,000 in July and 407,000 a year ago. The monthly change of -22.5% carries an interval of plus or minus 24.9 points, and the annual change of -15.5% carries plus or minus 27.5. Both ranges include zero. Census's explanatory note states that when a range contains zero, the change is not statistically significant and it is uncertain whether there was an increase or decrease.
- Permits: 467,000, up 9.4% from 427,000 in August 2025, though down 3.1% from July. Permits come from Table 1, which Census builds from a non-probability sample that is not subject to sampling error — no confidence interval is attached.
Across all building types, housing starts came in at 1,275,000, down 2.6% from the revised July estimate, while single-family starts rose 7.6% to 918,000. Those two monthly changes also carry intervals that include zero. Regionally, overall Midwest starts were down 10.8% year over year and overall Western starts up 5.2%.
Why it matters for property managers
The delivery wave really is ending. Completions tell you how many new units will compete with yours next year, and it is the series moving with statistical confidence. RealPage counted roughly 340,200 units completed nationwide in the year ending in the second quarter, after deliveries peaked near 588,000 units in late 2024. The two series cover different universes, but they point the same way: August's 302,000 annual rate extends that decline rather than starting it.
The pipeline is not closing, though. A 9.4% annual increase in permits is not what developers walking away looks like, and permits are the one series in the release with no sampling error — though Census notes Table 1 is still subject to nonsampling error, on a 75.8% total quantity response rate. A unit permitted this month does not compete with you for years. If your 2028 underwriting assumes a supply drought, the permit data does not yet support it.
Relief comes with a lag, and it has not landed. RealPage put same-store effective asking rents up 0.9% year over year in August, with occupancy at 95.5%. CoStar's Apartments.com data had the national average asking rent at $1,751 in August, about a dollar below July. Several Sun Belt markets are still cutting: RealPage showed San Antonio down 3.7%, with Charlotte, Tampa and Houston near -2%. Falling completions improve next year's competitive picture, not this month's renewal offers.
One month of starts is not a forecast. Census states plainly that it may take six months to establish an underlying trend for total starts. The August starts figure is preliminary and July's was already revised.
What to do this week
This is general information, not financial or legal advice. Confirm local rules and your own numbers with your advisors and counsel.
- Pull your submarket delivery schedule, not the national number. National completions falling 35.7% means nothing if two towers deliver eight blocks from you in March. List every competing project within your trade area with its expected delivery quarter.
- Split permits from completions in the 2027 budget. Use completions to set next year's vacancy and concession assumptions, and treat the 9.4% permit gain as a 2028-and-later input. Conflating the two is how supply forecasts go wrong.
- Re-time concessions to the delivery calendar. Where nothing new delivers in the next four quarters, test trimming a half-month of free rent on new leases before touching renewal pricing.
- Underwrite lease-ups on current absorption, not the supply drop. RealPage recorded demand of about 271,300 units in the year ending in the second quarter, below its decade average. Fewer competitors do not fill units on their own.
- Re-check renewals against local comps. With national rents up 0.9% and some Sun Belt markets negative, price to your submarket. Check your state and local notice requirements before sending any increase.
- Give owners the signal-versus-noise version. Completions down with confidence, permits up, starts unreadable for now. Owners who read only the starts headline may push for decisions the data does not support.
What we're watching
- The October 20 release. Census lists it as the next scheduled report, and it will revise August.
- Whether permits hold. A second and third month above last year's pace would confirm the pipeline is refilling.
- Absorption against thinner supply. Demand below the decade average is the constraint on pricing power, not construction.
- Sun Belt concessions. Markets like San Antonio at -3.7% should stabilize first if the completions decline is real.
- Revisions to the starts series. The August figure is preliminary, and Census says preliminary seasonally adjusted estimates of total permits, starts and completions are revised 3.8 percent or less on average. The plus-or-minus 24.9-point interval measures sampling variability, not the size of the coming revision.