Federal gasoline tax fixed at 18.3 cents per gallon since 1993 lost about 73 percent of its purchasing power by fiscal 2023 — the Highway Account collected USD 31.70 billion against USD 39.62 billion paid out in the first nine months of fiscal 2026

What is happening?

The Highway Trust Fund pays out more than it collects every year, and the difference is closed by moving money out of the General Fund. The Highway Account closed June 2026 holding about USD 47.07 billion, after collecting USD 31.70 billion and paying out USD 39.62 billion in the first nine months of fiscal year 2026. The Mass Transit Account closed the same month at about USD 13.60 billion, against USD 4.69 billion collected and USD 10.42 billion paid out over the same nine months. On the CBO February 2026 baseline the Highway Account ends fiscal 2028 at minus USD 8.8 billion.

The taxes that feed the fund have not been raised since 1993. Gasoline and gasohol are taxed at 18.3 cents per gallon to the trust fund and diesel at 24.3 cents, both flat nominal rates, with 2.86 cents of each routed to the Mass Transit Account. The rate went from 3 cents in 1956 to 4 cents in 1959, 9 cents in 1982, 14 cents in 1990 and 18.3 cents in 1993, and there it stopped. Between fiscal 1993 and fiscal 2023 the gasoline tax lost about 73 percent of its purchasing power to inflation.

Since 2008 Congress has closed the gap by appropriating General Fund money into the fund. The Congressional Research Service counts eleven such transfers through 2021, totaling USD 213.785 billion into the Highway Account and USD 273.968 billion into both accounts, as of 2025-06-03. The largest was USD 118 billion in the 2021 infrastructure law, and none of that amount carried a budgetary offset. The current surface transportation authorization expires 2026-09-30. The record opened for this document runs to 2026-07-24, at which point the House reauthorization bill had stalled without floor action, the Senate had advanced nothing comparable, and an extension attached to a continuing resolution was the expected path. What happened in the two weeks between that date and the stamp on this document is not in the record read here.

Whose problem is this?

The people who pay the fuel tax and the people who pay for the transfers are no longer the same group, and no vote was ever taken to make that swap. It happened one appropriation at a time.

Where does this problem end?

The boundary here is fiscal rather than physical. What is at stake is how the road program is paid for, not whether roads are built.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a solvency standard for the Highway Trust Fund. Congress has never enacted one, there is no trustee report and no statutory adequacy test, so nothing states what balance, what coverage ratio or what date would count as fixed. The Eno analysis computes what solvency would require — a total increase of about 17.8 cents per gallon phased from 2026-10-01 through 2036, or a USD 120 annual registration fee on every vehicle, or roughly 2.4 cents per mile driven — but those are analyst options offered for comparison, not a legislated goal.

How big is it?

Between 239.8 million and 339.1 million people, as of 2024. The low bound is 239,824,944 licensed drivers, the people who can pay the federal fuel tax directly at the pump and who use the federally aided road system. The high bound is the 339,058,370 United States resident population, because freight prices, bus and rail transit funded from the Mass Transit Account, and passenger travel reach people who hold no license and buy no fuel. Both figures come from the same FHWA table, published in February 2026.

The width of that range is the width of a question no source answers. Counting who pays the tax and counting who rides on what it buys give different totals, and nothing opened here chooses between them. A vehicle count would be larger still, and a vehicle is not a person: 297,525,836 motor vehicles were registered in 2024, of which 97,413,321 were automobiles and 189,754,931 were trucks.

What this range does not measure is harm. No opened source assigns any per-person magnitude to the shortfall, because the effect of a shortfall is deferred project delivery, slower reimbursement to states and a smaller future program, and no document converts those into a per-person figure. The one per-person number in the record runs the other way — an average driver pays about USD 92 a year in federal fuel taxes, on an assumed 24 miles per gallon and 12,000 miles, with no distribution stated around that average.

Under what conditions does it arise?

1. The instrument decays while nobody decides anything. A per-gallon rate fixed in nominal terms loses value every year through inaction alone. Between fiscal 1993 and fiscal 2023 the gasoline tax lost about 73 percent of its purchasing power. That figure has not been recomputed by any opened source since, and three further years of inflation have passed.

2. The base shrinks because other policies succeeded. Average fuel economy rose from 13.1 miles per gallon in 1975 to 27.1 in 2023, with new vehicles expected at 49 miles per gallon in 2026. Fuel burned per 100 miles by an average passenger car fell from about 7 gallons in the mid-1960s to about 4 today. Growth in vehicle miles traveled fell from about 4.5 percent a year in the 1950s and 1960s to about 0.8 percent since 2002, and FHWA forecasts about 0.6 percent a year over the next 30 years. CRS gives a different series with different break years — 3.8 percent annually from 1957 to 1988, 1.4 percent from 1989 to 2022, and a projected 0.5 percent through 2050 — and the two series agree on direction.

3. Spending was authorized above receipts and stayed there. New contract authority equaled 97 percent of new trust fund receipts and interest in fiscal 1997. It was 126 percent in fiscal 2002, 149 percent in fiscal 2009, 130 percent in fiscal 2014, 137 percent in fiscal 2020 and 169 percent in fiscal 2026. The gap is therefore not only a revenue story.

4. These are capital programs, so cuts land slowly. CBO estimates a first-year outlay rate of 25 percent for federal-aid highways and 10 percent for transit formula grants. Cutting outlays by USD 5 billion in the first year requires cutting USD 20 billion of new highway authority or USD 50 billion of new transit authority, and the second is arithmetically impossible against roughly USD 15 billion of new transit authority a year.

5. A transfer is one sentence and a rate increase is a recorded tax vote. The operative text of the 2008 transfer says only that out of money in the Treasury not otherwise appropriated there is appropriated to the Highway Trust Fund 8,017,000,000 dollars. Nothing further is required, and contract authority financed that way is mandatory budget authority exempt from sequestration.

What has been tried?

Two directions have been tried and only one of them has ever been carried through. Nine or eleven times, depending on how the laws are counted, the shortfall was closed by transfer. Not once since 1993 has it been closed by raising the dedicated rate.

What was found?

Why is it still unsolved?

Incentive inversion — the response that fixes nothing is cheap and has been getting cheaper, while the response that would fix it costs a recorded tax vote.

Movement one. The instrument decays while nobody decides anything. A per-gallon rate fixed in nominal terms loses value every year without a decision being taken, and the base under it shrinks for reasons that count as policy successes elsewhere. Fuel economy roughly doubled between 1975 and 2023. Growth in miles driven fell from about 4.5 percent a year to about 0.8 percent. The result is a revenue line the February 2026 baseline projects nearly flat in nominal dollars from fiscal 2026 to fiscal 2036, USD 28.2 billion to USD 29.3 billion for gasoline, which is a real decline of roughly a quarter at any ordinary rate of inflation. Nobody has to vote for that decline and nobody can be recorded as having caused it.

Movement two. The two available responses carry opposite political prices. Raising the rate is a recorded tax vote. Appropriating a transfer is one sentence in an appropriations bill. The Eno analysis states the mechanism directly rather than leaving it to inference — the transfer route is the path of least resistance precisely because it requires no change to how funds are distributed and does not run afoul of anti-tax pledges. The price differential has widened rather than narrowed. Between 2012 and 2015 House budget leadership insisted that transfers be offset, and USD 99.6 billion of the USD 272.8 billion total carries an offset from that period. One of those offsets was pension smoothing, which covered USD 8.7 billion of an USD 18 billion transfer by reducing federal spending inside the ten-year enforcement window and increasing it by more afterward. When that leadership changed, the norm went with it. None of the USD 118 billion transfer in 2021 was offset, and as of 2026 no offset is required by statute or by the rules of either chamber.

The cheap option became cheaper while the expensive option stayed exactly as expensive as it was.

Movement three. The deadline has already arrived repeatedly and nothing broke. Since 2008 the fund has reached the wall again and again, and each time the wall was moved — 2008, 2009, 2010, 2012, 2014, 2015 twice, 2020, 2021. Obligations were never cut, and no opened source documents a failure to pay. Because the transfer works, fiscal 2028 does not function as a deadline; it functions as the date of the next transfer. The enforcement mechanism that would make it real runs the wrong way as well. If reimbursement is withheld, states have sued the Department of Transportation in the Court of Federal Claims and won, and the judgment is paid either from the trust fund or by the court from the Permanent Judgment Appropriation at 31 U.S.C. 1304, which is General Fund money with interest attached. The discipline the trust fund was supposed to impose can be bypassed from both ends.

What might break the pattern, and why this document does not predict it. Three things differ in 2026. The Mass Transit Account crosses zero first, in 2027, inside the next authorization window rather than after it. The House bill that advanced in May 2026 contains no General Fund advance appropriations, which would make it the first reauthorization since 2008 not built on a transfer. And the Eno analysis notes that a generational budget fight over Social Security insolvency in 2032 may leave little appetite for routine trust fund bailouts. None of the three has happened. Through 2026-07-24, the newest date in the opened record, the House bill was stalled, the Senate had no bill, and an extension attached to a continuing resolution was the expected path.

Types considered and set aside. An institutional gap does not fit, because the mechanism to fix this exists and has been used eleven times; it is simply used in the direction that avoids the vote. Measurement absence does not fit either, because CBO, Treasury and FHWA all measure this monthly and annually, and the correction of August 2025 shows that apparatus working in public. A cost-structure reading would describe construction costs, which are real but are not what the opened sources place at the center.

What observation would mean it is solved?

Candidates — (a) Congress enacts dedicated revenue, whether a rate increase, a registration fee or a mileage charge, sufficient to cover trust fund outlays (b) a reauthorization passes with no General Fund advance appropriation (c) the ratio of new contract authority to new trust fund receipts falls back toward 100 percent.

(a) alone repeats the original defect. A rate fixed in nominal terms begins decaying the day it is enacted, which is exactly how the current position was reached. The Eno sizing of a minimum fix phases 17.8 cents per gallon in over ten years, so a bill could be passed, celebrated and already behind schedule at the same time. Indexation is what would distinguish a fix from a repeat, and no source opened here reports an indexation proposal that has advanced.

(b) alone can be satisfied by cutting. A reauthorization with no transfer can be produced by shrinking the program to fit the revenue rather than by fixing the revenue. Making the whole fund solvent through spending cuts alone would require blending the accounts and cutting both by 48 percent, a USD 30.1 billion first-year highway cut and a USD 7.0 billion first-year transit cut. That satisfies the observation while leaving the instrument exactly as decayed as before.

(c) alone counts a ratio rather than solvency. The ratio can be driven to 100 percent from either side, and it says nothing about the accumulated deficit already in the accounts. It also moves with the reauthorization cycle rather than with any decision, since the figure is taken in the last or peak year of each act. The three have to be read together, and (a) has to be read against whether the rate is indexed.

What is it connected to?

Fills with researchstate transportation revenue systems, the transit agency operating cliffs that CRS names for Philadelphia, the Bay Area and Pittsburgh, the 2032 Social Security insolvency date that the Eno analysis links to the future politics of trust fund bailouts, and construction cost inflation on the spending side. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• No source names a target. Congress has never enacted a solvency standard for the Highway Trust Fund, and nothing opened here states what balance, what coverage ratio or what date would count as fixed. The absence is structural rather than accidental — unlike Social Security, this fund has no trustee report and no statutory adequacy test, so there is no document whose job it would be to say.
• No source states how many people are affected, and none states a per-person magnitude. Every figure in the record is a dollar amount of account balance, revenue or outlay. FHWA publishes licensed drivers and registered vehicles in an entirely different report, and no document connects the two. The single per-person figure available runs the other way, an average payment of about USD 92 a year, with no distribution around it.
• The cumulative transfer total is not stated the same way twice. CRS gives USD 213.785 billion to the Highway Account and USD 273.968 billion combined. Eno gives USD 211.7 billion and USD 271.8 billion in one table and USD 272.8 billion in another, differing on sequestration treatment and on whether the 2012 law counts as one transfer or three. No source tells a reader which basis to use. A fourth figure, USD 197 billion, was carried into this round as the working number for the transfer total and could not be reproduced from any source opened here, so it is not used.
• What happens at depletion is described three incompatible ways and nobody reconciles them. CRFB reports an automatic cut to match revenue, about 40 percent for highways. CRS says the fund cannot carry a negative balance and describes slowed reimbursements and reduced apportionments as administrative responses. The federal baseline meanwhile prints negative balances from fiscal 2028 through fiscal 2032, because a baseline must assume scheduled spending continues. The official projection therefore prints a number that the same body of law says cannot exist.
• The federal mileage pilot authorized in 2021 was never implemented and no source names who decided that, when, or why. A program Congress created and funded simply did not happen, and the record holds the fact without the agent.
• No source says why the rate has not been raised since 1993. Every document records that it has not been. The one piece that reaches for a reason argues about how funds are distributed to localities, which is a different question from why the vote is never taken.
• The record opened here stops on 2026-07-24 while this document is stamped 2026-08-08. Nothing found covers the intervening two weeks, and the authorization expires 53 days after the stamp. This document cannot say what the position was on the day it was written.

See the evidence

No projection in this document was read from the body that produced it. The Congressional Budget Office baseline is the source of every forward balance and revenue figure here, and every cbo.gov address attempted returned HTTP 403, so those tables reach this document through the Eno Center reproduction, which states the February 2026 baseline as its own source. What was read directly is the measured side — FHWA Table FE-1 for actual balances and flows through June 2026, FHWA Highway Statistics for drivers, residents and vehicles, and two Congressional Research Service reports for the rates, the rate history and the transfer table. Where sources disagree the disagreement is left visible rather than resolved. The cumulative transfer total is USD 273.968 billion for both accounts on the CRS count and USD 271.8 billion or USD 272.8 billion on the two Eno tables, which differ on sequestration treatment and on whether the 2012 law is one transfer or three; a fourth figure carried into this round as a working number could not be reproduced from any of the three and is not used. The Mass Transit Account exhaustion date is late 2027 after the August 2025 correction and fiscal 2028 in a CRS report published thirteen days later, and neither document mentions the other. The consequence of depletion is given three incompatible ways by three sources. Two vehicle-miles series with different break years are reported side by side rather than merged. One row is headline-level only, and the conclusion that no federal electric vehicle fee was enacted is labeled an inference throughout rather than a finding. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 239,824,944–339,058,370 As of 2024

Sensitivity What the width means: the gap between 239.8 million and 339.1 million is the gap between counting who pays the tax and counting who rides on what it buys. No opened source picks between the two definitions, and none offers a figure in between. For reference, the same table gives 237,655,885 licensed drivers against a resident population of 334,914,895 for 2023, so both bounds move slowly and the width is stable. A vehicle-based count would be larger than the driver count and is not a count of people at all: 297,525,836 motor vehicles were registered in 2024. What this number fails to count: it counts exposure, not harm. No opened source assigns any per-person magnitude to the shortfall, because the effect of the shortfall is deferred project delivery, slower reimbursement to state governments and a smaller future program, none of which the record converts to a per-person figure. The single per-person figure available runs in the opposite direction, an average federal fuel tax payment of about USD 92 a year on an assumed 24 miles per gallon and 12,000 miles, with no distribution stated around that average. The range also treats every counted person identically, although the federal share of road spending varies sharply by state, with the average state formula apportionment running at 206 percent of what that state paid into the fund in 2024. The limit in the opposite direction: the low bound understates payers rather than bounding them, since indirect payment through freight prices reaches people with no license, and the high bound understates reach in a different way because it counts only people inside the United States while the fund also finances the freight network that other economies use.

Regional breakdown No opened source gives a regional or state-level count of people affected by the shortfall. State-level shares of trust fund tax payments and of vehicle miles traveled do exist in the Eno tables sourced to FHWA, but those are shares of tax payment and of miles driven, not shares of an affected population, and converting them would be proportional allocation of a national figure. The federal share of road spending also varies sharply across states, so any split by population share would be wrong in both directions at once.