Research
How long HVAC equipment actually lasts
The "15 to 20 years" everyone quotes has one solid source, and it is fifteen years old. We re-ran its method on current federal survey data.
Written by HYRE HVAC Research Desk Primary-source research, data analysis and fact checking
The finding
Half of central furnaces are still running at 22.3 years, and half of central air conditioners and heat pumps at 18.2 years. Ducted heat pumps: 15.2. Window units: 10.0. These are fitted to EIA’s RECS 2020 survey.
One assumption we cannot observe moves the furnace figure by 6.6 years, so read each as a range. Dividing installed units by shipments gives 9.8 years, and that is wrong.
Why is HVAC equipment lifespan so hard to measure?
Because nobody records when equipment fails. There is no registry of American furnaces.
No agency records when one is scrapped, no warranty database is public, and no survey follows a unit from installation to failure.
The number everybody wants, the spread of ages at which units are retired, has never been measured directly for US home HVAC equipment.
What is measured is the installed base. Every five years or so EIA runs the Residential Energy Consumption Survey and asks, among several hundred other things, how old the main heating equipment and the main cooling equipment are.
The 2020 cycle got answers from 18,496 households, weighted to represent 123.5 million occupied homes, with 60 replicate weights supplied so that the uncertainty on any derived share can be computed rather than guessed.
That is an age distribution of survivors, not of deaths, and the two are linked by arithmetic.
If shipments were flat and every unit faced the same survival function, the share of the stock at each age would be proportional to the probability of surviving to that age.
Invert that relationship and the survival function falls out of the survey. This is the method Lutz and colleagues published in 2011 and that DOE has used in furnace rulemakings since.
We are not proposing it; we are re-running it on data fifteen years newer.
HYRE analysis: The reason to re-run it is that a lot changed in between.
SEER2 took effect on January 1, 2023, the A2L refrigerant transition on January 1, 2025, and the heat pump share of shipments moved substantially.
A lifetime estimate fitted to 1990–2005 equipment is being quoted about equipment that did not exist when it was fitted.
How long do furnaces, air conditioners and heat pumps last?
A median of 22.3 years for a central furnace and 18.2 for central cooling. Fitting a two-parameter Weibull curve to each class gives the lines below. Read them as "what share of units installed on day one are still in service at age a". The median is where the curve crosses 50%.
Central furnaces last longest of the ducted equipment, at a fitted median of 22.3 years. Central air conditioners and heat pumps taken together sit at 18.2.
Ducted heat pumps on their own come in shorter, at 15.2. That is what you would expect from a machine that runs in both seasons, though RECS cannot show that cause and we will not pretend it can.
Window units are a different category at 10.0 years: they are appliances, not installations, and people replace them like appliances.
The shape parameter matters as much as the median. Every fitted shape is above 1, between 2.03 and 5.02, which means the hazard rises with age: equipment does not fail at random, it wears out.
That is why the curves bend instead of decaying evenly, and why a 22-year-old furnace is a truly different risk from a 12-year-old one. The equipment lifespan estimator places your own unit on these curves.
The assumption we cannot observe: The inversion needs to know how fast the installed base has been growing, because a growing stock looks younger than its survival function alone implies and would otherwise be mistaken for equipment that dies early.
We fitted each class at five growth rates from −1.0% to 2.0% a year. Every headline above is quoted at 1.5%.
The full sensitivity is in the table: the central furnace median runs from 17.0 to 23.6 years across that range and the central cooling median from 14.5 to 19.1.
That spread is larger than the difference between most of the numbers being argued about in this subject, and any page quoting a single lifespan to the year is quoting a precision the data does not contain.
How much does the answer move for each equipment class?
| Equipment class | Homes | Sample n | Median at 1.5% growth | Range across growth assumptions | Weibull shape | Fit residual | Lutz et al. 2011 |
|---|---|---|---|---|---|---|---|
| Central furnace (all fuels) | 74.4m | 11,511 | 22.3 yr | 17.0–23.6 yr | 2.76 | 0.59 pp | 22.6 yr (gas furnaces) |
| Central air conditioner or central heat pump | 82.7m | 12,211 | 18.2 yr | 14.5–19.1 yr | 2.54 | 0.87 pp | 18.0 yr (central air conditioners) |
| Ducted central heat pump | 16.1m | 2,218 | 15.2 yr | 12.0–16.0 yr | 2.03 | 1.03 pp | 14.6 yr (heat pumps) |
| Steam or hot-water boiler | 9.3m | 1,510 | 36.6 yr | 26.9–38.7 yr | 5.02 | 2.11 pp | 17.5 yr (gas boilers) |
| Window or wall air conditioner | 21.4m | 3,045 | 10.0 yr | 8.6–10.3 yr | 2.05 | 2.58 pp | 11.1 yr (room air conditioners) |
| Gas storage water heater | 55.4m | 8,108 | 15.6 yr | 13.2–16.1 yr | 2.65 | 1.93 pp | 12.0 yr (gas storage water heaters) |
| Electric storage water heater | 54.6m | 7,944 | 15.5 yr | 13.2–16.0 yr | 2.71 | 1.92 pp | 9.7 yr (electric storage water heaters) |
Fitted median service life by equipment class, RECS 2020. "Range" is the span of the fitted median across stock growth rates of −1.0%, 0%, 1.0%, 1.5% and 2.0% a year. "Fit residual" is the root-mean-square error between the age-bin shares the fitted function implies and the shares RECS observed, in percentage points.
Lutz et al. class names differ slightly from RECS class names and are given as published.
Their central air conditioner figure carries a stated standard error of about 0.3 years, but the authors note that the uncertainty from assuming what share of the stock is residential rather than commercial is larger than that: 16.5 years at 100% residential, 19.7 years at 90%.
Can you work out a lifespan from shipment data?
No. It is a tempting shortcut. AHRI publishes monthly US shipments of central air conditioners, heat pumps and furnaces, and RECS gives the installed base. Divide one by the other and you seem to get an average lifetime: the years it would take shipments to replace the stock.
This information was received from AHRI: Their reported unitary air conditioner and heat pump shipments were 8,656,674 in 2023, 9,681,770 in 2024 and 7,749,529 in 2025.
AHRI’s own definition puts residential capacity at "BTUHs of 64.9 and below", and the sub-65 bands account for about 96.5% of the total, so the residential flow averages 8,398,771 units a year across the three years.
Set against the 82.7 million households RECS records with central cooling equipment, that gives 9.8 years.
HYRE analysis: that figure is not a lifespan, and it is wrong by roughly a factor of 1.9. Three reasons, and they compound in the same direction.
First, shipments are not installations in occupied homes: they include new construction, second systems, commercial units inside the residential capacity band, and units that sit in distribution.
Second, RECS counts households with the equipment, not equipment: a home with two systems appears once.
Third, and most importantly, stock over flow only equals mean lifetime when the stock is stationary, and it plainly is not: the same AHRI series fell 20.0% between 2024 and 2025 alone, straddling the A2L refrigerant transition of January 1, 2025.
We publish the shortcut so it can be dismissed with a number attached.
The right use of the shipment series is as the flow input to the same inversion the survival fit uses, not as a divisor, and as a check on whether the growth assumption above is plausible.
It is not a lifespan, and no page should present it as one.
How do these results compare with published estimates?
Mostly close. The point of repeating a method is that it can disagree with the original. Ours mostly does not, and where it does, the direction tells you something.
Central air conditioners: 18.2 years here against 18.0 published in 2011. That is agreement, and it is the most useful result on this page: the number the industry quotes for cooling equipment survives being re-estimated on a completely different survey cycle.
Furnaces: 22.3 against 22.6. Also close, and both sit inside the range DOE adopted for its consumer furnace rulemaking, which put the average product lifetime at 21.4 years (22.5 in the North and 20.2 elsewhere) and reported a literature range of 16 to 23.6 years.
Three independent estimates landing in the same decade is about as much corroboration as this subject offers.
Heat pumps: 15.2 against 14.6. Close on the number, but the 2011 authors flagged that no single realistic survival function fitted the whole heat pump shipment and stock history, and restricted their fit to the 2001–2007 surveys.
Our fit inherits that difficulty: the ducted heat pump has the largest residual of the ducted classes at 1.03 percentage points, and the class has been growing fast enough that the growth term is doing more work than we would like.
Treat the heat pump figure as the weakest number on this page.
Boilers and storage water heaters: substantially longer than published. Our boiler fit is 36.6 years against 17.5 for gas boilers in 2011, and both storage water heater classes fit around 15.6 years against 12.0 and 9.7.
We do not present these as corrections. The boiler stock is old, shrinking and concentrated in the Northeast, which violates the growth assumption in the direction that inflates the fitted life, and the boiler fit has the worst residual in the set at 2.11 percentage points.
HYRE analysis: the boiler and water heater medians here should be read as upper bounds on a declining stock, not as service life expectations.
What equipment ages did the survey record?
| Equipment class | Under 2 years | 2 to 4 years | 5 to 9 years | 10 to 14 years | 15 to 19 years | 20 years or more |
|---|---|---|---|---|---|---|
| Central furnace (all fuels) | 10.8% ± 0.3 | 14.2% ± 0.4 | 23.2% ± 0.4 | 19.9% ± 0.4 | 13.6% ± 0.4 | 18.2% ± 0.5 |
| Central air conditioner or central heat pump | 13.0% ± 0.3 | 16.3% ± 0.4 | 26.6% ± 0.4 | 20.7% ± 0.5 | 11.9% ± 0.3 | 11.5% ± 0.4 |
| Ducted central heat pump | 15.3% ± 0.9 | 17.9% ± 0.8 | 28.1% ± 1.0 | 18.8% ± 1.0 | 10.3% ± 0.8 | 9.6% ± 0.7 |
| Steam or hot-water boiler | 6.0% ± 0.7 | 6.8% ± 0.8 | 17.7% ± 1.2 | 18.2% ± 1.2 | 13.1% ± 1.1 | 38.1% ± 1.7 |
| Window or wall air conditioner | 16.1% ± 0.9 | 31.0% ± 1.0 | 31.2% ± 0.9 | 13.5% ± 0.7 | 4.6% ± 0.5 | 3.7% ± 0.4 |
| Gas storage water heater | 13.9% ± 0.4 | 17.7% ± 0.5 | 31.8% ± 0.6 | 20.6% ± 0.5 | 8.7% ± 0.4 | 7.3% ± 0.3 |
| Electric storage water heater | 13.4% ± 0.5 | 18.4% ± 0.5 | 32.1% ± 0.5 | 20.8% ± 0.5 | 8.4% ± 0.3 | 6.9% ± 0.3 |
Age distribution of the main installed unit, RECS 2020, as reported by the household. Every cell is a survey estimate with its standard error, computed from the 60 replicate weights EIA supplies. These six numbers per class are the entire empirical input to the fits above.
The "20 years or more" bin is open-ended. It anchors how much mass sits in the tail; it carries no information about the shape of the tail, which is why the fitted curves beyond about 25 years are extrapolation and are drawn as such.
Who has the oldest HVAC equipment?
Renters, homes heated with fuel oil, and large apartment buildings. The survey supports one more cut that the fitted medians hide: the age of central equipment differs by group, and the differences are large enough to see past the standard errors.
Rented homes run older equipment: 13.9% of owner-occupied homes with central equipment have a unit under two years old, against 10.4% of rented ones. The gap is the wrong way round from what a naive story about maintenance budgets would predict, and it is consistent across the young bins.
Mobile homes replace fastest: 21.5% of mobile homes with central equipment have a unit under two years old. That is the highest of any housing type, well above single-family detached at 14.1%. Large apartment buildings replace slowest, at 7.4%.
Fuel oil furnaces are the oldest heating equipment in the country. Only 6.3% are under two years old against 10.4% of gas furnaces, and the fuel oil stock carries the heaviest 20-plus tail.
HYRE analysis, stated as a limitation rather than a finding: These are differences in the age of installed equipment, which is a product of replacement behavior as much as of failure.
A group that replaces early will show a young stock without its equipment being any more durable.
We have deliberately not fitted separate survival functions to these subgroups: the inversion assumes the survival function is independent of household characteristics, and using it to measure a difference between household characteristics would be circular.
How did we fit the survival curves?
EIA Residential Energy Consumption Survey 2020, public use microdata file v7. 18,496 responding households representing 123.5 million occupied primary housing units.
Age of the main unit is reported by the occupant in six bins: Under 2 years, 2 to 4 years, 5 to 9 years, 10 to 14 years, 15 to 19 years, 20 years or more.
60 replicate weights, EIA’s own documented estimator: the plain sum of squared replicate deviations. Every share printed on this page carries its standard error, because these are survey estimates and not counts.
Stationary-stock inversion of surveyed equipment-age distributions with an explicit stock-growth term. A two-parameter Weibull survival function is fitted by least squares so that the stock age distribution it implies reproduces the age distribution RECS actually observes.
Under stationary shipments the age density of the stock is proportional to the survival function. Where the installed base grows at rate g the density is proportional to S(a)·e^(−ga), so a growing stock looks younger than its survival function alone implies. Each class is fitted at five growth assumptions.
Each fit reports the root-mean-square error between its implied age-bin shares and the observed ones, in percentage points. They range from 0.59 to 2.58 pp. The two worst, boiler and window unit, are flagged in the text rather than averaged into a headline.
December 2025 U.S. Heating and Cooling Equipment Shipment Data, released 2026-02-13, with the June 2026 data, released 2026-08-14 checked for currency. Public, no membership required.
This information was received from AHRI, which is the attribution AHRI asks for.
AHRI publishes on the second Friday of the month for shipments made two months earlier, so the newest available data is structurally about two months old.
Every figure in the prose, both tables and both charts is computed from lifespan.json by script, and the CSV linked from this page is written from the same file at build time. A number in a sentence and the same number in the table below it cannot disagree.
What are the limits of this estimate?
This is the fundamental one and it applies to every published lifespan figure for this equipment, ours included. RECS records the age of installed equipment, not retirements.
The survival function is inferred from the shape of the surviving stock. If the inference is wrong, the number is wrong, and no amount of precision in the arithmetic fixes that.
Fitted at 1.5% because that is the middle of the plausible range, but the central furnace median runs 17.0 to 23.6 years across −1.0% to 2.0%. Any single-number lifespan claim, including one made from this page, is quoting a point from a range.
A furnace installed in 1998 and one installed in 2018 are assumed to face the same hazard at the same age. Manufacturing, refrigerants, controls and installation practice have all changed. This is the same assumption the published literature makes, which makes it conventional rather than correct.
The oldest bin is "20 years or more" with no upper edge.
It fixes how much of the stock is old; it says nothing about whether that tail runs to 25 years or 45.
The curves are drawn to 30 years because the shape is fitted, but past roughly 25 the line is the model talking, not the survey.
Both break the stationary-growth assumption badly (the boiler stock is shrinking and regionally concentrated), and the boiler fit carries the worst residual in the set at 2.11 percentage points. We publish them because suppressing an inconvenient fit is worse than labeling it, not because we would defend the numbers.
Vacant homes, seasonal homes and non-residential buildings are outside RECS entirely. A home with two central systems is counted once, at the age of the main one. Second systems are invisible throughout.
A median is a statement about a population. Half of all central furnaces outlive the median and half do not, and which half a particular machine is in depends on installation quality, sizing, run hours and luck, none of which RECS observes.
Start by dating your own unit with the system age lookup, then weigh a big repair with the repair-or-replace threshold.
Questions
How long does a furnace last?
How long does a central air conditioner last?
Do heat pumps have a shorter lifespan than air conditioners?
Can you work out equipment lifespan from AHRI shipment data?
Why does every website say 15 to 20 years?
Does maintenance make equipment last longer?
Should I replace equipment when it reaches the median age?
Written and audited by
HYRE HVAC Research Desk
Primary-source research, data analysis and fact checking
We are a research desk, not a sales floor. We read the federal microdata file, the statute or the manufacturer data sheet ourselves, and we publish the figure with the document it came from and the date we retrieved it.
Where a number cannot be traced to a primary source, we publish the shorter page and say what we could not verify.
The counts below are generated from the published pages themselves, last counted September 28, 2026, and they are what we have actually published rather than what we intend to.
- 13
- studies published
- 12
- federal sources read and cited
- 8
- studies published with their full dataset as CSV
- 51
- jurisdictions reproduced against EIA’s own tables
How this desk works
- Primary sources only. Federal data comes from the agency that collects it, in the file that agency publishes. We do not cite an article that cites a source; we download the source and compute the figure ourselves.
- We validate against the agency before we publish. First, we use each federal microdata file to reproduce the agency’s own published tables. Our cooling research reproduces EIA’s state estimates and standard errors for all 51 jurisdictions. That check caught a variance formula that was off by a factor of four.
- Every estimate carries its uncertainty. These are survey figures, not counts. Standard errors are computed from the replicate weights the federal file supplies and printed beside the estimate. An estimate too imprecise to publish is reported as such rather than printed.
- Nothing is typed by hand. Prose, tables and charts all read from one dataset built by script, so a number in a sentence and the same number in the table below it cannot disagree.
- We publish the data, not just the conclusion. 8 of our 13 studies offer the full computed table as a CSV download on the page, so you can check the analysis or disagree with it. A study without a row-level dataset gets no download link and claims none in its structured data.
- We correct in public. Where we have published a figure wrongly we fix the figure, rewrite any analysis that rested on it rather than patching the number underneath it, and leave a dated correction note on the page.
- We do not install or sell HVAC equipment, and we take no payment for placement, ranking or a favorable mention. Nobody buys a position on this site.
Data as of EIA RECS 2020 public use microdata v7; AHRI shipments through December 2025. Authorship on this site is organizational: the analysis belongs to the desk rather than to a named individual, and we do not publish credentials we do not hold.
Our editorial policy sets out how we source, date and correct what we publish.
The data behind this page
Every figure on this page is computed from one file, and that file is published here so the analysis can be checked, disagreed with, or reused.
Sources & retrieval dates
Find out how old your system actually is
None of this applies until you know the age of the machine in front of you. The serial number carries it, and the lookup decodes the major manufacturers’ formats.