Where do heat pumps actually save money?

This analysis models the annual cost of heating a 2,000 sq ft home with an air-source heat pump against a new 92% AFUE natural gas furnace, using state retail energy prices from the EIA and population-weighted heating degree days from NOAA.

A note on denominators, because it changes every figure below. 51 jurisdictions are covered: 50 states and the District of Columbia. Of those, 50 have modelled heating demand and are the population every heating comparison on this page is stated out of. Hawaii falls below the model’s heating threshold of 500 heating degree days, so there is no heat-pump-versus-furnace comparison to make there and it is reported as N/A rather than folded into a denominator of 51.

The heat pump is cheaper to run in 17 of the 50 jurisdictions with modelled heating demand. The median of those 50 costs $336 a year more to heat with a heat pump.

The result is not driven by climate. It is decided by whether a state’s electricity-to-gas price ratio falls below its own efficiency ratio, COP divided by AFUE — an exact algebraic condition that holds in every jurisdiction tested, and which is set out under why results vary.

A note on that median, because two figures on this site look like they disagree and do not. The median here is taken across the 50 jurisdictions with modelled heating demand, an even count, so it is the average of the two middle values. The research index reports the median across all 51 jurisdictions including Hawaii, which is a single middle value. Both are correct for the population named alongside them.

This is a narrow question: modelled operating economics against natural gas, under one stated baseline. It is not an assessment of whether a heat pump is a good idea. See what this analysis does not measure.

A fourth figure is worth stating plainly because it is the one most easily misread: 48 of 50 jurisdictions with heating demand do not recover the premium within 15 years. That is the 33 with no operating saving at all, plus 15 that save money but too slowly. Separately, Hawaii is N/A.

Ranked by annual operating difference. Positive favours the heat pump. Payback is the incremental $4,000 premium divided by the first-year operating saving. A payback of none means no payback under the static baseline: there is no operating saving to recover the premium from, so it is not recovered at any horizon while prices hold. N/A means the jurisdiction has no modelled heating demand and is not compared.

Only 2 of the 50 jurisdictions with heating demand recover the $4,000 premium inside 15 years:

• Arkansas — saves $321 a year, recovering the premium in 12.4 years. Electricity 13.28¢/kWh against gas at $2/therm, 3,523 HDD.
• Georgia — saves $273 a year, recovering the premium in 14.7 years. Electricity 14.9¢/kWh against gas at $2/therm, 2,897 HDD.

The next 15 jurisdictions save money every year but not fast enough to clear the premium inside the horizon. The pattern in the winners is mild winters combined with a low electricity-to-gas price ratio — not one or the other.

The five widest gaps. None has a payback under the static baseline, because there is no annual saving to accumulate:

• Alaska — $3,942 a year more than gas. Break-even would require electricity at 8.34¢/kWh; the state average is 26.68¢.
• Maine — $1,874 a year more than gas. Break-even would require electricity at 14.94¢/kWh; the state average is 29.68¢.
• New Hampshire — $1,351 a year more than gas. Break-even would require electricity at 15.21¢/kWh; the state average is 26.8¢.
• Vermont — $1,285 a year more than gas. Break-even would require electricity at 14.05¢/kWh; the state average is 23.9¢.
• Wisconsin — $1,248 a year more than gas. Break-even would require electricity at 8.63¢/kWh; the state average is 18.87¢.

Alaska is an outlier driven by both terms at once: 11,003 heating degree days push seasonal COP down to 1.80, the floor of the model’s range, while electricity costs 26.68¢/kWh.

The same delivered heat Q costs Q/COP × P_e on a heat pump and Q/AFUE × P_g on a furnace, where P_e and P_g are prices per kWh of input energy. Setting one below the other, Q cancels:

The table above lists electricity in ¢/kWh and gas in $/therm, which are not comparable as printed. One therm carries 100,000 BTU and one kWh carries 3,412 BTU, so a therm is 29.3083 kWh of fuel energy. Normalising gas to the same units:

Worked through for Massachusetts, using the values in its row of the table above:

Floor area, insulation and degree days all cancel out of this comparison. They determine how much a state wins or loses, but not which side it lands on. Climate re-enters only through COP, because seasonal efficiency falls as it gets colder.

Tested against the engine, this condition predicts the outcome correctly in 50 of 50 jurisdictions with modelled heating demand, with no exceptions. That is not a fitted result — it is the same algebra the model runs, so agreement is a consistency check rather than evidence.

The switching ratio is not a single number. Because COP varies with climate, it ranges from 1.96 to 3.91 across the dataset, with a median of 2.95.

A flat threshold is a shortcut, not a law

Because the median switching ratio is close to 3, “a price ratio below 3.0 favours the heat pump” looks like a rule. It is not one, and we do not publish it as one. Tested across the dataset, a flat 3.0 cut misclassifies 3 jurisdictions (Virginia, Kentucky, Nevada). Jurisdictions where the heat pump wins have price ratios from 1.84 to 3.25; those where it loses run from 3.09 to 6.26. Those ranges overlap, so no single fixed cut can separate them.

The exact condition costs nothing extra to state and is always right, so there is no reason to publish the approximation instead.

A multi-state memorandum of understanding coordinated by NESCAUM (Northeast States for Coordinated Air Use Management) targets heat pumps at 65% of residential-scale heating, air conditioning and water heating sales by 2030, and 90% by 2040, across participating states.

Participation was staged, and the dates matter. NESCAUM’s announcement of 7 February 2024 was headlined “Nine States Pledge Joint Action” and named nine states: California, Colorado, Maine, Maryland, Massachusetts, New Jersey, New York, Oregon, Rhode Island. The District of Columbia appears in the participant list later in 2024. Washington joined in September 2025, which NESCAUM announced alongside the agreement’s action plan, describing the agreement as “initiated in February 2024”. That brings current participation to 11 jurisdictions.

Under this baseline, 9 of those 11 participating jurisdictions currently show a higher heat pump operating cost than gas. The exceptions are Oregon and Washington.

Read this carefully, because it is easy to overstate. The participant list is a documented fact with dates; it is not a ranking of incentive generosity, and this analysis does not attempt one. The observation is that in 9 of these 11 jurisdictions the price ratio currently sits above the switching ratio — a statement about tariffs, not about whether the commitment is sound. Purchase incentives reduce the installation premium; they do not change the annual operating comparison, which is what this table shows. Sources: NESCAUM press release, 7 February 2024; NESCAUM press release, 16 September 2025 (Washington joins).

Published heat pump payback estimates vary widely — commonly quoted ranges run from about three to twelve years, and shorter figures appear regularly. Those numbers and the ones here are usually not measuring the same thing. The differences that matter most:

• The fuel being replaced. This analysis compares against natural gas, the cheapest common US heating fuel. Estimates built on replacing heating oil, propane or electric resistance heat start from a far more expensive baseline and produce much shorter paybacks. As an illustration from this same model, delivering the same heat by electric resistance would cost $6,793 a year in Massachusetts against $2,697 for the heat pump — a large saving, in a state where the comparison against gas is negative.
• Incremental versus full cost. This model prices only the $4,000 premium over replacing a furnace and air conditioner. An estimate that divides savings into a full installed price is answering a different question and will produce a longer payback; one that assumes the furnace did not need replacing anyway will produce a shorter one.
• Incentives. No rebate, tax credit or utility incentive is applied here. Applying them reduces the premium and shortens payback proportionally — a 30% reduction in the premium shortens every payback in the table by 30%.
• Equipment and geography. Some published figures describe ground-source or geothermal systems, or new construction rather than retrofit, or a single favourable climate zone rather than a national spread.

We are not claiming other published figures are wrong. Most are defensible given their own stated baseline, and where a baseline is not stated the number simply cannot be compared. The useful comparison is like-for-like: against a new gas furnace, with no incentives, on incremental cost, this is what the arithmetic gives.

Annual delivered heat is a degree-day heat-loss calculation:

With a 2,000 sq ft home and a heat-loss coefficient of 0.25 BTU/hr per sq ft per °F, representing a moderately insulated house. A draughty older house runs closer to 0.40 and a well-sealed new build closer to 0.15.

Seasonal coefficient of performance is derived from climate rather than assumed constant:

This is a modelled relationship, not a measured one. It is a deliberate simplification of manufacturer performance curves, and it is the single assumption most likely to move the results. It gives 3.60 in Florida (698 HDD) and 1.80 in Alaska (11,003 HDD). Only one jurisdiction sits at each bound.

Costs then follow directly:

Payback is simple and undiscounted: the incremental installation premium divided by the first-year operating saving. Where there is no operating saving there is nothing to recover the premium from, so no payback exists at any horizon and the position worsens each year. The table records this as none, meaning no payback under the static baseline — prices held flat in real terms, no incentives, no escalation. It is a statement about this baseline, not a prediction that a heat pump can never pay back in that state: a change in the electricity-to-gas price ratio, or an incentive applied to the premium, moves it.

• Residential electricity price — US Energy Information Administration, Electric Power Monthly, Table 5.6.B (year-to-date through May 2026).
• Residential natural gas price — US Energy Information Administration, Natural Gas Residential Price, latest published year per state.
• Heating degree days — NOAA Climate Prediction Center, population-weighted state normals, base 65°F.
• Multi-state MOU participation — NESCAUM (Northeast States for Coordinated Air Use Management), Multi-state MOU: Accelerating the Transition to Zero-Emission Residential Buildings, initiated February 2024; participation dates from NESCAUM’s releases of 7 February 2024 and 16 September 2025.

Efficiency figures (COP, AFUE), the installation premium, the home model and the heat-loss coefficient are modelled assumptions, not sourced data. Which is which for every field on this site is recorded on the data sources page.

• The $4,000 premium is a single modelled figure. Real incremental cost varies with existing ductwork, electrical service capacity, equipment tier and local labour, and can plausibly range from under $2,000 to well over $8,000. Payback scales linearly with it: halve the premium and every payback period halves.
• The COP model is a simplification. A linear function of heating degree days cannot capture equipment tier, cold-climate inverter performance, backup heat strategy, or distribution losses. A modern cold-climate unit will beat this model in a cold state; an undersized or poorly commissioned one will fall short of it anywhere.
• State-average energy prices hide enormous variation. Utility territories within a single state can differ by more than the difference between states, and time-of-use, seasonal and tiered tariffs are not modelled at all.
• One house stands in for every house. Floor area, insulation, air sealing, thermostat behaviour and occupancy all move the result, and none are state-specific here.
• Gas fixed charges are excluded. A household that removes gas entirely also sheds a standing monthly charge, which this model does not credit. That omission works against the heat pump, and in states with high fixed charges it is material.
• Incentives are excluded. Rebates and tax credits reduce the premium and shorten payback proportionally. They do not change the annual operating comparison.
• Maintenance and equipment life differences are not modelled.
• Prices are held flat. No escalation or divergence between electricity and gas is assumed. Since the outcome is decided by the ratio of the two, a sustained change in that ratio would move states across the switching line — this is a snapshot, not a forecast.

The question here is deliberately narrow: under one stated baseline, what are the modelled operating economics against natural gas? A household decision properly includes several things this analysis does not price, and a negative number in the table above does not by itself make a heat pump the wrong choice.

• Cooling. A heat pump replaces an air conditioner as well as a furnace, which can change the equipment comparison entirely.
• Emissions, which fall in most US grids regardless of cost.
• Removing combustion equipment from the home, and the associated safety and air-quality considerations.
• Comfort and control, which owners frequently rate above cost.
• Avoiding a gas connection altogether in new construction, and the standing charge that goes with it.
• Future rate and policy change, which this snapshot cannot anticipate.
• Incentives, which are real money and are excluded here by design.
• Building-specific performance, which can differ substantially from any state average.

Payback Math takes no position on whether households should install heat pumps. It publishes the arithmetic under stated assumptions, including when the arithmetic is unwelcome.

Both files are generated from the same engine that produces this page, so they cannot disagree with it. Derived Payback Math values may be reused with attribution. Underlying EIA and NOAA data are US federal government works and are not redistributed here in raw form.

• CSV — all 51 jurisdictions (51 rows, one per jurisdiction)
• JSON — rows plus full baseline, headline figures and threshold tests

Payback Math, “Heat Pump vs Gas Furnace Economics by State — 2026”, model version 2026.08.6, data updated August 2026. https://paybackmath.com/research/heat-pump-vs-gas-by-state (accessed 2026-09-05).

Payback Math is an independent modelling project, not an academic institution, and this analysis has not been peer reviewed. If you find a methodological error we will correct it and record the change in the changelog; corrections are logged rather than quietly edited.

• Calculation model 2026.08.6
• Input data updated August 2026
• Jurisdictions covered: 50 (50 states and the District of Columbia)
• Full model and constants: methodology
• Per-state calculator: heat pump vs gas by state
• Other findings: research index