USD 37.4 billion to rehabilitate the non-federal high-hazard dams of the United States as of 2025-03 — about 65 percent of dams in the country are privately owned and owners are frequently unable to fund the repair

What is happening?

Rehabilitating the non-federal dams of the United States would cost USD 165.2 billion across 88,957 structures, and the high-hazard-potential subset alone — 15,412 dams, the ones where failure is judged likely to cost at least one life — would cost USD 37.4 billion. Both figures were published by the national association of state dam safety regulators in March 2025, applied to 2024 National Inventory of Dams counts.

The federal grant program built for precisely this purpose is offering USD 11.4 million in fiscal year 2026. The Rehabilitation of High Hazard Potential Dams program posted its notice of funding opportunity on 2026-08-05, anticipates 50 awards in a range of USD 100,000 to USD 7.5 million, and closes applications on 2026-08-31. It is authorized at USD 60 million a year and it received no federal appropriation at all in fiscal years 2023 and 2024. Two dates then sit on top of each other — awards under that notice are due by 2026-09-30, and the legislative authority for the program is set to lapse in September 2026 unless a reauthorization now moving through both chambers arrives first.

The bill itself lands on the owner. About 65 percent of dams in the United States are privately owned, and the Congressional Research Service states without euphemism that responsibility for rehabilitation and repair lies with dam owners while many non-federal owners are not willing or able to fund those costs. The hazard class that creates the obligation is frequently not something the owner did. FEMA calls it hazard creep — houses arrive below a farm pond built decades earlier, the classification rises, and the spillway requirement rises with it.

Whose problem is this?

The party that must pay for the repair is almost never the party that would be harmed by the failure, and no federal instrument opened here connects the two.

Where does this problem end?

The boundary matters here because the repair cost is measured and republished every few years while the population it protects is not counted anywhere. What is missing is not an estimate of the work but a payer assigned to it.

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

Now

Needs a new measurementthe target state: no source opened here states a year by which the backlog should be closed, an annual figure that would close it, or an acceptable residual. The closest thing to a target in the record is a recommendation to fully fund two programs whose combined authorized ceiling, added here from the two published authorizations, is USD 73.9 million a year against a USD 37.4 billion high-hazard need.

How big is it?

The scale of this problem is published in dollars and in dams. It is not published in people, and that is a finding rather than an omission. The quantity a reader wants is the number of people inside the failure inundation zone of a non-federal high-hazard dam. It cannot be assembled from anything opened here, and the obstruction is documented. The National Inventory of Dams carries no downstream population figure. The inundation maps that would yield one are withheld by federal regulators on post-9/11 security grounds — the Federal Energy Regulatory Commission for roughly 1,800 power-producing dams, all non-federal, and the Bureau of Reclamation for 430, which said it would begin sharing in 2025 in a process that could take more than eight years. No source opened in this round records whether that sharing began. Standard FEMA flood maps exclude dam-failure risk by design. Where the maps do exist for state-regulated dams they sit inside emergency action plans, which are generally not public, and about 18 percent of state-regulated high-hazard dams have no such plan at all. Hazard classification is in any case a qualitative judgment about probable loss of life rather than a count, so the figure of 15,412 carries no population weight even in principle.

Two anchors exist as illustration and must not be read as a national total. In Michigan, reporting published 2026-07-16 puts more than 100,000 residents below high-hazard or significant-hazard dams, a figure the reporting itself calls conservative because it covers only 219 state-regulated structures. In California in 2017, a large state-owned dam that did not fail triggered the precautionary evacuation of about 200,000 people and more than USD 1.1 billion in emergency response and repair.

A scaling chain from the Michigan figure to a national one was attempted during research and rejected. Its denominator mixes high-hazard and significant-hazard dams while the multiplicand is high-hazard only, and the per-dam cost methodology behind the USD 37.4 billion is a function of height, age and condition rating with no population term anywhere in it. Two quantities that share no mechanism cannot be multiplied into a third.

Needs a new measurementthe affected population is not derivable. No source opened here publishes how many people stand below a non-federal dam, and the maps that would produce that number are withheld by the federal agencies holding them.

Under what conditions does it arise?

Four conditions hold this state in place, and nothing opened here shows any of them changing.

1. The asset earns nothing. A non-federal dam is, in the majority of cases, a farm pond, a millpond, a lake behind a subdivision or a recreational reservoir. It has no rate base, no toll and no tariff. The published per-dam cost runs from USD 2.87 million for the smallest old dam in poor condition to USD 95.3 million for the tallest, in 2022 dollars, and nothing in law converts the downstream benefit into revenue the owner can borrow against. The watershed portfolio alone is credited with USD 2.2 billion a year in avoided flood and erosion damage, which accrues to people who do not receive the invoice.

2. The obligation is created by somebody else. Development in the failure inundation zone raises the hazard class, and reclassification triggers stricter spillway capacity, structural upgrades, more frequent inspection and an emergency action plan. GAO documents the pure case — about 12,000 watershed dams built between 1954 and 1980 in farmland, classified low hazard, then reclassified as houses arrived. The local government that approved the subdivision bears none of the resulting cost.

3. Federal oversight ends exactly when the structure is oldest. Under the watershed program the federal interest ceases when a dam reaches the end of its evaluated life and the contractual agreement expires. Nearly 60 percent of that portfolio passed that point by the end of 2025. GAO also found about 400 of those dams listing a private individual as sponsor because the original sponsoring entity no longer exists, and such individuals lack the eminent domain and taxing authority the regulation requires of a sponsor.

4. The state layer is thin and carries written holes. States regulate about 70 percent of the inventory with roughly 455 full-time staff nationally, which works out at about 190 dams per inspector on the reckoning of the engineering society. The holes are statutory rather than accidental — Delaware exempts dams owned by private individuals and entities, Missouri exempts agricultural dams and all dams under 35 feet regardless of storage or hazard, and Texas exempts privately owned significant-hazard and low-hazard dams storing under 500 acre-feet in less populated counties. Six states had not given their regulators authority to require an emergency action plan even for a high-hazard dam.

What has been tried?

Two directions were tried and a third was proposed and dropped. The grant channel was built and then funded far below its own ceiling, the loan channel was drafted in 2003 and never enacted, and the audit channel produces recommendations about management rather than money.

What was found?

Why is it still unsolved?

Cost structure — the cost is measured, published and undisputed, and no party is assigned to it.

The first movement is that the bill lands on an asset that produces no income. A private impoundment has no ratepayers to bill and no tariff to raise, and the smallest old dam in poor condition already costs USD 2.87 million to rehabilitate on the published schedule. The Congressional Research Service says the responsibility lies with the owner and that many owners are not willing or able to meet it, which puts the mechanism in plain words in a document written for the body that would have to change it. Nothing in law turns the avoided flood damage downstream into money the owner can borrow against, even where a federal agency has priced that benefit at USD 2.2 billion a year.

The second movement is that the obligation frequently originates in a decision the owner did not make. Hazard creep is the term FEMA uses for it, and GAO documents the clean version — dams built in farmland and classified low hazard, then reclassified because a subdivision arrived downstream. This is where an inverted incentive genuinely lives, and it is real as a second-order component. It explains why the bill appears on a particular desk. It does not explain why the bill goes unpaid, because reclassifying nothing would still leave the same owner unable to fund a spillway.

The third movement is that every channel built to close the gap is sized for a different problem. Our own arithmetic on the two published figures puts the FY2026 grant at about 0.03 percent of the stated high-hazard need. That channel also requires a 35 percent non-federal match from parties defined by their inability to pay, a state-level and a local-level federally approved hazard mitigation plan already in hand, participation in the national flood insurance program in good standing, and a benefit-cost ratio approaching four to one, which structurally favours dams standing above dense and valuable development. And it will not pass money to a private for-profit owner at all, which is the ownership class holding roughly 65 percent of the dams. That exclusion is the strongest available reading for a different failure type and it should not be minimised. It is not the generating one, because removing it would redistribute a pot of USD 11.4 million rather than change the size of the pot.

Everybody knows the cost and nobody is assigned to it. The parallel channel makes the same point arithmetically — our own division of the two published figures shows the watershed program has rehabilitated about 1.5 percent of its portfolio in twenty-two years, while nearly 60 percent of that portfolio passed the end of its evaluated life by the end of 2025.

What observation would mean it is solved?

Candidates — (a) federal funding for non-federal dam rehabilitation rises to a level tied to a stated schedule for retiring the backlog (b) the count of high-hazard dams in poor or unsatisfactory condition falls for several consecutive years (c) every non-federal high-hazard dam carries a current condition assessment, a current inspection and an emergency action plan.

(a) alone is weaker than it looks. There is no schedule anywhere to be met, so there is nothing for a funding level to be tied to. Money already appropriated to this purpose has been redirected, and the record does not permit anyone to say how much of it survived. And the largest federal channel excludes the ownership class that holds most of the dams, so enlarging that channel does not reach them.

(b) alone is worse. That count moves when the method moves. Dams with no condition rating were assigned one in proportion to the national distribution of rated dams when the cost estimate was built, and 84 percent of high-hazard dams in the watershed inventory have no condition value on file at all. A number that responds to reclassification and to imputation cannot be read as concrete being poured.

(c) alone counts paper. A current plan is not a repaired spillway, and an inspection records a condition rather than changing it. Removal is also a legitimate outcome that this observation would score as a loss. The three have to be read together, and (b) has to be read against whatever the rating method was in the year it came from.

What is it connected to?

Fills with researchfloodplain development approval and local land-use authority, the national flood insurance program and its community rating system, hydroelectric relicensing at the Federal Energy Regulatory Commission, dam removal and river restoration, and the same owner-pays structure as it appears in levees and in privately held bridges. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• Who has to find the money. The state regulators association published a public and private split of the high-hazard need in 2009 and again in 2012. The 2023 methodology and its 2025 update publish no ownership decomposition at all, so the single breakdown that would say who must pay is not in the current estimate. The total has more than quadrupled since the 2003 baseline — a rise the engineering society attributes primarily to more dams entering the inventory rather than to deterioration — and no source states why the decomposition is no longer published.
• How many dams actually need rehabilitation. The USD 37.4 billion sits beside a count of all 15,412 non-federal high-hazard dams, while the per-dam costs behind it apply only to dams rated less than satisfactory — and dams with no rating were assigned one in proportion to the national distribution. The headline pairs the cost of a subset with the count of the whole population, and the size of that subset is never published.
• Where the infrastructure money stands. USD 585 million was appropriated to the high-hazard grant program and was to remain available until expended. Separately reported are more than USD 71 million granted since 2019 across 40 states and Puerto Rico, USD 364 million redirected out of about USD 3 billion, and nearly half of an USD 800 million share redirected in 2024. Not one source opened here states where the USD 585 million stands today, which years the redirections came from, or what they were redirected to.
• What happened in FY2025. FY2023 and FY2024 are documented at zero and FY2026 is USD 11.4 million. No source opened here gives the FY2025 appropriation for this program, and the FEMA page that would carry it refused every request.
• How long the backlog would take. No source performs the division. Our own arithmetic on the two published figures — USD 37.4 billion of high-hazard need against USD 11.4 million a year — puts it on the order of three thousand years. Because nobody states a target, a schedule or an acceptable residual, nobody can be behind schedule.
• How many dams have failed. The state regulators association states plainly that no one knows precisely how many dam failures have occurred in the United States. Reporting is voluntary, and CRS notes that few private or local dams are included, which is exactly the class of dam this problem is about. The apparent rise from about 3 failures a year to about 76 is therefore uninterpretable, because no source separates a real increase from the founding of the database in 2010.
• What happens to a dam whose owner cannot be found. GAO documents about 400 watershed dams listing a private individual because the original sponsor no longer exists as an entity, and notes that private individuals lack the eminent domain and taxing authority the regulation requires. What the state is supposed to do next is written nowhere in the material opened here.
• Why the FY2026 notice prioritises data centers and nuclear generation. It directs FEMA to prefer applications serving data centers, nuclear power generation, multi-state agriculture or multi-state water supply. No document opened here explains how that priority was chosen, what it displaces, or how it interacts with the benefit-cost screen applied to the same applications.

See the evidence

No dam was inspected for this document, and no count was read from the inventory itself. The National Inventory of Dams was opened and served a title with no statistics, because it is a client-rendered application — so every dam count here reaches the reader through an intermediary carrying its own refresh date, and that is the direct cause of the disagreement recorded in block 8. Read directly and in full were the March 2025 cost report from the state regulators association, the GAO audit of July 2025, the FY2026 notice of funding opportunity, two Congressional Research Service reports and two documents from the engineering society; the two wire-service investigations were read through trade-press reprints. Where sources overlap they agree — the USD 165.2 billion and USD 37.4 billion figures appear in three documents, hazard creep is named by four, and the FY2026 figure of USD 11.4 million appears both in the notice and on the grants portal listing. Where they disagree the disagreement is left visible rather than resolved: the 2003 baseline is given as USD 34 billion on one page of a report and more than USD 36 billion on another page of the same report, the high-hazard dam count appears as 14,934, more than 16,700, 16,875 and 15,412 in four documents of similar vintage with none of them stating whether the difference is scope or refresh date, the damages from the 2020 Michigan failures are USD 175 million in one source and more than USD 200 million in another with neither stating a basis, the average dam age drifts from 61 to 65 years across three documents published within twelve months, and Alabama appears as a state with no dam safety program on one page, as a state that legislated an opt-in program in 2023 and 2024 in another, and as the single excluded applicant in the FY2026 federal notice. Several figures here were computed for this document rather than published anywhere — the FY2026 grant as a share of the high-hazard need, the time the backlog would take at that rate, the share of the watershed portfolio rehabilitated in twenty-two years, and the combined authorized ceiling of the two programs — and each is labelled where it appears. Two dates are pending rather than past as of 2026-08-08: applications close 2026-08-31, and the program authority is set to lapse in September 2026, the same month its awards are due. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range Not derivable

The reason and what is missing are listed under “What is missing” below