Shut the border and white America still shrinks
The "great replacement" theory blames immigration for the falling white share of America's population. A forecast for every state to 2075, with odds attached, finds that most of the fall would happen anyway, and that its size depends on who is counted as white
In 2011 Renaud Camus, a French writer, published a book whose title gave a name to an old fear. "The great replacement" holds that white populations in Europe and America are being swapped for others, and that elites are arranging it. The phrase has since travelled from the fringe to cable news and campaign debates.
The theory makes two claims. One is about numbers: the white share of the population is falling. The other is about intent: someone is doing it on purpose. No population data can test the second, and nothing here supports it. The first can be checked, state by state. We projected America's population to 2075 for all 50 states and Washington, DC, by age, sex, race and Hispanic origin, from the Census Bureau's estimates, and ran the projection through 4,000 simulated futures to put odds on the results.
What the numbers say
The fall is real. In 2000 non-Hispanic white people were 69% of Americans. In 2025 they were 56%. Their number barely moved for two decades, at about 196m, and has fallen by 3m since 2020. The rest of the population grew from 86m to 150m.
The share most likely falls below half in 2038. Four in five of our simulated futures put that year between 2035 and 2047, and 92% put it before 2050. By 2075 the median is 36%, with an 80% range of 26% to 45%. The Census Bureau's 2023 projection crosses the line in 2046, but it starts from older estimates that put today's share nearly two points higher than the Bureau now does.
The border is not the main cause
The replacement theory puts immigration at the centre, so we ran the projection with immigration switched off. If net immigration had stopped in 2025 and never resumed, the non-Hispanic white share would still fall below half, in 2045 instead of 2038. By 2075 it would be eight points higher than with immigration as it has been, but still well under half. About three-fifths of the fall happens with the border shut. The Census Bureau's own no-immigration scenario, built on older estimates, says the same.
The reason is age. White Americans are older than everyone else: they are 73% of people aged 65 and over, and 44% of children under five. Between April 2020 and July 2025, by the Bureau's count, 12.7m non-Hispanic white Americans died and 8.5m were born, a gap of 4.2m. Migration from abroad added 1.0m.
Who arrives is not in dispute. Of the 8.3m net migrants in those five years, 12% were non-Hispanic white and 49% Hispanic. The mix differs by state: white people are 18% of those arriving from abroad in Texas and New Jersey and over 60% in Montana, Vermont and West Virginia, where few arrive at all. So immigration does lower the white share. It is the smaller part of the story.
What it would take
If not the border, then births? We asked the projection a narrow question: if nothing else changed, how many children would white women need to have, starting now and for 50 years, for the non-Hispanic white population of 2075 to match today's?
For the country as a whole the answer is about 2.2 children per woman, against 1.4 today, counting children classed as white. Without that, the white population falls by 28%, from 192m to 138m. Holding the white share at 56%, not just the number, takes about 3.5.
The map asks the same question of each state on its own: how many children would white women in that state need for that state's white population to hold? The answers differ because Americans move. In seven states and Washington, DC, the white population grows with no change in births, because white Americans from elsewhere keep arriving: Idaho, South Carolina, Tennessee, Montana, Utah, North Carolina and Florida. California would need 3.7, to make up for those who leave as well as those who die.
The state numbers are not a recipe for the national one. Nationally, moves between states cancel out, so the country needs only 2.2 overall, however the births are spread. And births are one lever of several. More white immigrants would do it too, though it would take about a million more a year; fewer than 200,000 a year have come, net, since 2020. So would longer lives, more mixed families counted as white, or a wider definition, which is the largest lever of all.
Who counts as white
The definition matters more than any assumption about migration. The 36% counts only people who tick white, nothing else, and are not Hispanic. A child with one white parent and one Asian parent is not in it.
Count instead everyone who is non-Hispanic and white or part-white, and the 2075 figure is 41%. Add Hispanics whom the Bureau's estimates class as white, and it is 71%, down from 77% today. On that count the white majority is never in question.
The broad figure has a weakness of its own. The Bureau's estimates assign about four in five Hispanics to "white alone", but in the 2020 census only about a fifth of Hispanics chose that for themselves. How people see themselves lies somewhere between the lines, and it moves. Italians, Irish and Jews were always white in the census and in law, but immigration officials once recorded them as separate "races" and many Americans thought of them as lesser kinds of white; nobody sorts their grandchildren that way.
Fifty different countries
Nine states and Washington, DC, already have no white majority on the narrow count: Hawaii, California, New Mexico, Texas, Nevada, Maryland, Georgia, New Jersey and Florida. Fifteen more are likelier than not to join them by 2050, among them New York, Illinois, Virginia and Washington, and six more by 2075.
That leaves 20 states that will probably keep a white majority to 2075, among them Ohio, Michigan, Wisconsin, Kentucky and Tennessee. For Maine, Vermont, New Hampshire, Montana and West Virginia the chance of losing it is under one in ten. The national average describes few actual places.
So the numbers in the replacement theory are half right, and the story built on them is wrong. The white share is falling and will almost certainly keep falling. It falls mostly because white Americans are old and are having fewer children than are needed to replace themselves, which no one arranged. And the size of the fall, 36% or 71%, is set less by who arrives than by which box their grandchildren tick.
This article was corrected on October 8th 2026 after a check against sources. An earlier version said Italians, Irish and Jews were once "counted apart" from the white majority; they were always counted as white in the census and in law, and were treated as separate "races" in immigration records and in popular usage. It also said a 2024 federal standard "will" move people out of the white category; the standard has not yet been adopted by the Census Bureau.
Data and methods
Data
US Census Bureau population estimates by state, sex, age, race and Hispanic origin: the intercensal estimates for 2000-2010, Vintage 2020 for 2010-2020 and Vintage 2025 for 2020-2025. The national benchmark and the no-immigration scenario are the Bureau's 2023 National Population Projections. Vintage 2025 uses a newer race base than earlier vintages: it puts the non-Hispanic white share in 2020 at 58.8%, against 59.7% in Vintage 2020. No change is ever measured across vintages, and the history chart shows the step. The estimates use the Bureau's "modified race" categories, which reassign people who chose "some other race" in the census; this is why they class most Hispanics as white.
Projection
A Hamilton-Perry cohort projection in five-year steps from 2025, for 51 areas, two sexes, 18 age groups and six groups (non-Hispanic white, Black, Native American, Asian or Pacific Islander, and two or more races; and Hispanic of any race). For each cell the method takes the change in the size of a cohort over a past five-year period, which combines deaths and migration, and applies it again. Where a cohort shrank its ratio is applied; where it grew the absolute increase is added, which stops small, fast-growing groups from exploding. This follows Hauer (2019), who uses differences for county populations projected to grow and ratios for those projected to decline; we apply the rule cohort by cohort. Births are children under five per woman aged 15-49 in 2025; children of two or more races are counted per woman of any group. Cells of fewer than 50 people borrow the national figure. The central projection averages cohort change in 2015-20 and in 2020-25. The specification was written before the model was run.
The two wider definitions of white apply 2025 shares, by state and age, to the projected groups: 92% of the non-Hispanic two-or-more group are part-white, and 85% of Hispanics are classed as white alone or mixed.
Odds
A Bayesian model, fitted with PyMC, describes how much the projection's inputs moved from year to year in 2010-25: for each group, yearly cohort change net of age structure (which moves with deaths and migration) and infants per woman; and for each state, its departure from the national figure. Cohort change is modelled as reverting to a long-run level with correlated shocks across groups; birth rates as a random walk with no drift. The projection is then run 4,000 times, each under one draw of the model's parameters and one simulated future. The ranges quoted hold 80% of those futures, after the small stretch described under Tests. Yearly swings were about eight times larger for Hispanic and Asian Americans than for white Americans, which is migration.
Tests
Short test. Started in 2010 with cohort change from 2005-10, the projection missed states' non-Hispanic white share in 2020 by 1.18 percentage points on average. Assuming shares stayed as in 2010 missed by 3.56; assuming each share kept changing at its 2005-10 pace missed by 1.28. Nationally the method put the 2020 share 1.3 points too low.
Long test. On the health-statistics agency's 1990-2020 estimates, which keep race categories consistent (and have no mixed-race group), the method was started in 1995, 2000, 2005 and 2010 and run to 2020. It missed states by 2.84, 2.28, 1.45 and 1.20 points on average; a straight-line trend missed by 2.83, 2.68, 1.33 and 1.26; no change by 10.0, 7.6, 5.4 and 3.3. So the method is as good as the trend and no better. At three of the four starts it put the national white share too low, by 1.3 to 2.7 points.
Tests of the ranges. An odds model fitted to the 1990s only and launched in 2000 was too narrow: 39 of 51 states fell inside their 80% range in 2010 and 31 in 2020, and the national figure fell outside its 95% range. Its ranges were tight because its decade was calm. The published model's ranges, fitted to 2010-25, are about 60% wider at the same horizon. Set against the method's actual errors from the four starts, state by state, they would have held 40, 48, 44 and 47 of 51 states at 10, 15, 20 and 25 years, and all 14 national errors. The published ranges are stretched by the most that this second test called for (8%). Stretching them by what the first test called for gives the harsh reading in the box above: a 76% chance of falling below half by 2050, and a 2075 range of 14% to 56%.
No immigration
A separate national run of the same cohort method on the 2025 base, in three groups (non-Hispanic white, Hispanic, everyone else). Net migrants from abroad by group are the Census Bureau's own (its components of population change for 2020-25; for 2015-20 the total, split by group as in 2010-19), given the age and sex pattern of its 2023 projection, and are subtracted from each cohort's change. "None" means zero net migration from abroad after 2025. The Bureau's no-immigration scenario is stricter: no arrivals while people keep leaving. With migration left in, this simpler model puts the white share in 2075 at 34.7%, a point below the main state-level projection.
Who arrives
Net migration by group is the Bureau's, national only. State figures are from the American Community Survey's one-year tables for 2021-24: people aged one and over whose home a year earlier was abroad, by race and Hispanic origin. That is a gross count that includes Americans returning from abroad, which is why white people are 27% of it but 12% of net migration. State figures pool four years because single years are noisy.
Births needed
A what-if on the central projection: white children per woman are multiplied by one factor, from 2025 on, until the white population (or share) of 2075 equals that of 2025, with deaths, immigration, moves between states and mixed-race births left as they are. The national figure applies one factor everywhere; each state figure is solved for that state alone. Children per woman is a total fertility rate: the rate that, with the age pattern of the Bureau's 2023 projection, yields a state's non-Hispanic white infants from its non-Hispanic white women by single year of age. Nationally it reproduces the Bureau's count of white births to within 0.3%. It counts only children classed as white alone and not Hispanic, so it is below the official rate for white mothers, some of whose children are classed as mixed or Hispanic.
Not modelled
Changes in how people describe themselves, changes to official categories, rising intermarriage, and shocks that hit some ages more than others. State-level births and deaths by race are folded into the cohort ratios, so the split between ageing and migration is taken from the Census Bureau's national scenarios, not from our model.
The data behind every chart can be downloaded from the data on GitHub, free to reuse with credit.
Cite this work
Our articles and data visualisations rely on work from many sources. When citing this article, please also cite the underlying data sources, listed in Data and methods.
Signum Average (2026) - "Shut the border and white America still shrinks" Published online at signumaverage.com. Retrieved from: https://signumaverage.com/articles/america-in-2075/ [Online resource]
@article{signumaverage-2026-10-05-america-in-2075,
author = {{Signum Average}},
title = {Shut the border and white America still shrinks},
journal = {Signum Average},
year = {2026},
month = {oct},
url = {https://signumaverage.com/articles/america-in-2075/},
note = {Online resource}
}
Reuse this work freely
Text, charts and data on this page are published under the CC BY 4.0 licence: you may copy, adapt and republish them for any purpose, as long as you credit Signum Average. Third-party data keeps its original licence.