Medicaid expansion became optional and the subsidy floor did not — 1.2 million adults below the poverty line have no coverage pathway

What is happening?

An estimated 1,208,000 uninsured adults ages 19 to 64 live below the federal poverty level in states that have not adopted the Medicaid expansion, and no coverage program is available to them at any income. The figure is a KFF estimate built on the 2024 American Community Survey 1-Year Estimates, restated as 1.2 million in a KFF brief published 2026-07-27. These adults hold too much income for the Medicaid their state offers and too little for a Marketplace premium tax credit, which by statute begins at 100 percent of the poverty level.

Ten states have not adopted the expansion, with decisions stated as of May 2026 — Alabama, Florida, Georgia, Kansas, Mississippi, South Carolina, Tennessee, Texas, Wisconsin and Wyoming. The measured gap sits in eight of them. KFF states directly that a coverage gap does not exist in Georgia or Wisconsin, because both cover adults to 100 percent of the poverty level through a Section 1115 waiver, and the KFF state table accordingly carries a number for the other eight and marks those two rows NA. The count of eight is a derivation from that table and no source opened here writes it as a sentence.

Nobody legislated this into existence. The Affordable Care Act set the premium tax credit floor at 100 percent of the poverty level because the same statute required every state to cover everyone below 138 percent through Medicaid, so the floor was written to be redundant. In NFIB v. Sebelius, decided 2012-06-28, the Supreme Court held that conditioning the existing Medicaid funding of a state on accepting the expansion was unconstitutionally coercive — the opinion weighed a threatened loss of over 10 percent of the overall budget of a state. The remedy was narrow. It struck the withholding mechanism, left the expansion standing as a state option, and did not touch the subsidy floor. The expansion became optional and the floor did not.

Whose problem is this?

The body that could close this in a single session is a state legislature, and there are eight of them. No one of them can be compelled by any of the others, and none of them is required to report on the number.

Where does this problem end?

The boundary matters because nothing here is disputed on the facts. The rules are published, the counts are published, and the two rules simply do not meet.

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

Now

Needs a new measurementthe target state: no source opened here names a target value, a target date, or a reporting body for the size of the coverage gap. Georgia publishes a projection of 18,301 active Pathways enrollees by October 2026, and that is a target for the enrollment of one state program rather than for the gap. The number itself is estimated once a year by a private foundation from a Census survey, and no agency is answerable for it.

How big is it?

Between 1,208,000 and 2,400,000 people, depending on which question is asked. Both ends come from KFF and rest on the 2024 American Community Survey 1-Year Estimates, so the two bounds do not mix vintages.

The low bound of 1,208,000 counts adults for whom no program exists at any income. The high bound of 2,400,000 counts everyone who would become newly Medicaid eligible if every state adopted the expansion — the gap population plus roughly 1,192,000 adults between 100 and 138 percent of the poverty level who hold a subsidized Marketplace option today. KFF publishes the 2,400,000 directly and does not publish that increment; the subtraction is performed for this document. The 2,400,000 is an eligibility count and not a coverage count, and no source opened here states an expected take-up rate for either half of it. The width is not statistical uncertainty. It is the distance between having nothing and having something unaffordable.

Rounding and concentration. All values are KFF estimates from the 2024 ACS 1-Year Estimates and the share column is arithmetic performed for this document. The eight state values are published to the nearest thousand and sum to 1,207,000 against a published national total of 1,208,000; the difference of 0.08 percent is rounding rather than a discrepancy, and it is stated here because a reader who recomputes the column would otherwise not know whose arithmetic to distrust. Two states hold 72.2 percent of the total and Texas alone holds half. Wyoming carries 8,000 people and Texas carries 605,000 under the same statutory arrangement, 76 times apart in scale, which is why the identical decision reads as a major fiscal question in one capital and as a rounding error in another.

Composition differs by state, and this is inference rather than a published figure. The eligibility limits imply it. Tennessee covers parents to 105 percent of the poverty level, above the subsidy floor, so its 89,000 is essentially all adults without dependent children. Texas covers parents to 15 percent, so its 605,000 includes nearly every poor parent in the state. The national statement that 78 percent of the gap population has no dependent children is an average across eight states whose composition ranges widely, and it is right nationally while being wrong about Texas by a large number of parents.

What this number cannot count. The survey is annual and point in time, so an adult who spends four months of a year in the gap after losing a job may not appear, and nothing published describes that churn. It excludes the Georgians whose coverage is on paper, discussed below. It stops at 2024 and therefore predates the removal of the federal incentive for late expansion. Whether it excludes people by immigration status is not stated on the page that publishes it. It counts adults only, so it understates how many people live in a household with someone in the gap. And no figure here measures the consequence — untreated illness, medical debt or delayed care — for which the uninsured rate difference of 14.5 percent against 8.0 percent is the nearest available proxy and is a rate rather than an outcome.

Under what conditions does it arise?

1. Two eligibility rules were written to meet and no longer do. The premium tax credit floor at 100 percent of the poverty level was drafted on the assumption that Medicaid would cover everyone below 138 percent. When the expansion became optional, the assumption failed while the floor stayed. Nobody has to defend the gap because nobody chose it.

2. Adults without dependent children are categorically ineligible at any income. In all eight gap states the Medicaid limit for that group is zero percent of the poverty level as of January 2026, which is the structure that existed before 2014 preserved intact. That single fact is why 78 percent of the gap population has no dependent children.

3. The decision belongs to eight legislatures acting separately. Each one carries the full political cost of adopting and captures only the benefit inside its own borders. In two of them the population is large enough to dominate the national figure and in one it is small enough to be politically invisible, so the same statute presents as a different magnitude of decision in each capital.

4. Cost is not what is stopping it. The federal match on the expansion population is 90 percent, which means a holdout state is declining a subsidy of nine dollars for every one of its own. Whatever holds the decision in place, the price of the decision is not it.

5. A workaround that reaches one person in ten scores the same as one that reaches everyone. The published measure is eligibility based. A waiver that makes people eligible removes them from the count whether or not the program reaches them, so the indicator cannot distinguish a program that works from a program that exists.

What has been tried?

Two directions were available and both narrowed inside the same statute. A state deciding in 2026 faces a worse offer than a state that decided in 2023, because the extra federal match built to buy out late adopters ended before those states used it, and from 2027 the population a holdout would take on carries a federal work requirement with verification machinery the state has to build. The Wisconsin arrangement is the least examined of these — no source opened here prices what covering that population at the ordinary federal match costs the state each year.

What was found?

Why is it still unsolved?

Institutional gap — neither program owns this population, and the space between them is not anyone's rule to enforce.

The first movement is definitional. Medicaid in these states excludes the population by category, since adults without dependent children are ineligible at any income and parents are cut off between 15 and 67 percent of the poverty level in seven of the eight. The Marketplace excludes them by threshold, since credits begin at 100 percent. Neither exclusion is an exemption granted to anyone and neither is a rule that someone declines to enforce. This is not a rule going unenforced, which is why the failure is not enforcement absent. It is not an exemption either, because these adults were never inside the scheme to be released from it. And the population is measured carefully every year, so it is not a measurement absence. What is missing is a rule, not the observance of one.

The second movement is that both available routes are structurally disfavored, and the price of the decision is not the obstacle. A holdout state declines a federal match of 90 percent, which no ordinary account of cost explains. The state route asks eight legislatures each to absorb the whole political cost of a decision whose benefit stops at the state line. The federal route asks a national body to appropriate for beneficiaries concentrated in eight states whose own delegations largely oppose the underlying program, so there is no natural coalition for a fix that pays out in eight places. Then the one federal instrument built precisely to make the state route cheaper for latecomers ended on 2026-01-01, before those states used it, and from 2027 the population a state would take on carries a federal community engagement requirement that the state must administer. The problem did not merely persist across that statute. The cost of solving it rose.

The third movement is the one that keeps the pattern stable, and it is about the indicator rather than the policy. Because the published measure is eligibility based, a state can move its own number to zero by making people eligible without reaching them. Georgia did exactly that: a program covering roughly one adult in ten removes an estimated 180,000 to 200,000 people from the national count, and the count carries no note that one of its zeros is a program at a tenth of its intended scale. The published 1,208,000 is therefore a floor on the number of people with no realistic path, not a ceiling. Meanwhile no document opened here names a target figure, a target date or a body that reports against it. No agency fails when this number does not move. Nobody built this, and that is the point.

What observation would mean it is solved?

Candidates — (a) the eight remaining legislatures adopt the expansion and coverage takes effect (b) the annual estimate of the coverage gap population falls toward zero (c) the uninsured rate in the non-expansion states converges on the rate in the expansion states.

(a) alone counts statutes. Adoption sets eligibility and eligibility is not coverage. From 2027 the expansion population carries a federal requirement of 80 hours a month with verification machinery each state has to build, and the Georgia experience is what adoption without reach looks like at one-tenth scale. An observation that stops at the vote would record the fix at the moment its hardest part begins.

(b) alone is the weakest of the three, for the reason the whole document turns on. The measure is eligibility based, so a waiver that makes people eligible drops them out of the count regardless of whether the program reaches them. The number can fall because more people are covered or because fewer people are counted, and nothing on the page distinguishes those two. A fall in this figure is evidence about definitions before it is evidence about people.

(c) alone imports everything else. The uninsured rate moves with employment, with Marketplace enrollment above the poverty line, and with premium levels that have nothing to do with this gap, and the gap population is only 5.2 percent of the uninsured to begin with. The series also runs two years behind, so a change would be observable long after it happened. The three have to be read together, and (b) has to be read against what the definition was doing that year.

What is it connected to?

Fills with researchuncompensated hospital care and rural hospital finance in non-expansion states, the administration and cost of work requirement verification systems, Marketplace affordability just above the poverty line, and the treatment of low income adults by immigration status. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• The KFF footnote is wrong on exactly the two rows where it matters. The state table marks Georgia and Wisconsin NA and explains NA as not applicable because the state has expanded Medicaid. Neither state has expanded Medicaid. Nothing on the table indicates that the blank is caused by a Section 1115 waiver, and a reader working from the table alone would conclude that 42 states expanded.
• No opened source writes the sentence that the coverage gap exists in eight states. The brief says ten and the table shows eight, and the two are reconciled only by a paragraph elsewhere in the brief. The number eight in this document is a derivation from the table rather than a quantity anyone published.
• The Georgia zero is an eligibility zero, and no source says so beside the number. The national figure is published by one body, the Georgia enrollment count by a second and the Georgia gap estimate by a third. No source opened here places them on the same page, so the national total carries no indication that one of its zeros is a program reaching roughly one adult in ten.
• No target exists anywhere. Not a value, not a date, not an owner. Nothing opened here states how many people should be in the gap, by when, or who reports against it.
• Everything published in 2026 describes 2024. The estimate rests on the 2024 American Community Survey. The two-year lag means no published figure covers the period after the federal incentive for late expansion ended on 2026-01-01, and no source says when a later figure will exist.
• Take-up is not reported for either half of the 2.4 million. The figure states how many would become eligible and not how many would enrol. The group between 100 and 138 percent of the poverty level is already subsidy eligible and largely uninsured anyway, which is itself evidence that eligibility and coverage are far apart, but no source opened here quantifies the distance.
• No source says what happens after 2026-12-31 in Georgia. The waiver runs to that date. Whether the state has applied to renew, and what CMS intends, appears in nothing opened here. If it lapses, roughly 18,500 people lose coverage and an estimated 180,000 to 200,000 return to the national count, which would move the national figure for a reason unrelated to any change in the circumstances of the people counted.
• The statutory text was not read. The effective dates of the two federal provisions are corroborated by a post-enactment source, but their section numbers appear only in a pre-enactment summary of committee text, so this document carries the dates and omits the numbers. Nor is the definitional boundary of the coverage gap stated on the page that publishes the number — whether it excludes people by immigration status, and how lawfully present adults below the poverty line who are subsidy eligible by a separate rule are handled, decides who is in the denominator and is not written down.

See the evidence

No primary statutory or judicial text was read in full. The Cornell Legal Information Institute page for NFIB v. Sebelius was opened directly and is the only judicial source here; a second host for the same case returned HTTP 403. The federal poverty guidelines were taken from the agency that publishes them. Everything else about the current federal statute comes from two secondary readings that agree on the effective dates and disagree in status — one is a post-enactment law firm alert and the other a pre-enactment summary of committee text — so this document carries the dates from both and omits the section numbers that only the pre-enactment document supplies. The gap counts, the eligibility limits, the state distribution and the uninsured rate difference all come from a single publisher, KFF, working from one federal survey; the two independent estimators sought as a cross-check both returned HTTP 403, and so did the federal survey release that would have tested the rate difference. Where sources part company the disagreement is left visible rather than resolved. The KFF brief says ten states and the KFF table shows eight, and only a separate paragraph reconciles them. The KFF footnote explains the two blank rows as expansion, which is not what happened in either state. The Georgia data tracker and the Georgia Budget and Policy Institute do not contradict each other on spending so much as decline to cover the same ground, since the state publishes a total and a state share while only the advocacy organization publishes the split between benefits and administration. Four further lines of inquiry — the 2024 and 2026 Mississippi sessions, the Florida ballot initiative, the fiscal cost of the Wisconsin arrangement, and any change in premium tax credit generosity — were pursued and are absent from this document entirely, because every source that would have carried them returned an error. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 1,208,000–2,400,000 As of 2024 American Community Survey 1-Year Estimates, published by KFF 2026-07-27

Sensitivity The width of the range is not statistical uncertainty and should not be read as an error bar. It is the distance between having nothing and having something unaffordable: the low bound counts adults below the subsidy floor and outside every Medicaid category, and the high bound adds adults who already hold a subsidized Marketplace option they mostly do not buy. Both ends come from one publisher working from one federal survey, so the two bounds do not mix vintages, but they also carry no independent corroboration; the two independent estimators sought in this round both returned HTTP 403. The count fails to capture several things at once. It is an annual point-in-time survey, so an adult who spends part of a year in the gap after losing a job may not appear, and nothing published describes that churn. It counts adults only, so it understates how many people live in a household with someone in the gap. It stops at data year 2024 and therefore predates the end of the extra federal match for late expansion on 2026-01-01. Whether it excludes people by immigration status is not stated on the page that publishes it. And it measures exposure rather than consequence, with the uninsured rate difference of 14.5 percent against 8.0 percent in 2024 as the nearest available proxy. The most important correction runs in one direction only and is deliberately excluded from the chain above. Because the measure is eligibility based, Georgia records a coverage gap of zero on the strength of a waiver program that had 18,527 active enrollees on 2026-06-30 against a state gap population separately estimated at 180,000 to 200,000 on 2025-10-13, which leaves roughly 161,000 to 181,000 people in the gap in fact and out of it in the statistics. Folding that into the range would produce a high bound near 2.56 to 2.58 million that is arithmetically real and methodologically incoherent, because it would mix three vintages and three publishers. It is reported here instead as what it is: a directional correction showing that the low bound is a floor rather than a ceiling, and that the understatement is caused by an indicator that cannot tell a program which reaches people from one which merely makes them eligible.

Regional breakdown KFF does publish a value for each of the eight states carrying a measured gap, and those eight values are reproduced in block 5 of the document. They are not carried here as a structured regional decomposition, because the decomposition would be incomplete in a way the structure cannot express. Georgia and Wisconsin belong to the same policy population and carry no value at all, since a Section 1115 waiver sets their eligibility gap to zero, and in the Georgia case that zero is contradicted by a separate state gap estimate of 180,000 to 200,000. The eight published values are rounded to the nearest thousand and sum to 1,207,000 against a national total of 1,208,000. The high bound of 2,400,000 has no state-level counterpart in any source opened in this round, so a decomposition of the low bound alone, presented against a range whose upper end cannot be decomposed, would invite a reader to allocate the whole range by the same shares. Splitting either bound by state population would be proportional allocation and would be wrong, because the gap is produced by state eligibility rules rather than distributed with population.