Six thousand years of the world's biggest cities, from Uruk to Dhaka. Which city was the largest at each moment, how big it was, and how the centre of the urban world wandered from Mesopotamia to China, out to Europe and back to Asia.
Around 2800 BC the largest city in the world was Uruk, in the marshy south of what is now Iraq, with perhaps 80,000 people. In 2025 it is Jakarta, with 41.9 million. In between, the title passed through the cities of Sumer and Egypt, Babylon, Alexandria and Rome, Constantinople, Chang'an and Baghdad, the capitals of Song and Ming China, London, New York and Tokyo. By the United Nations' latest projections it will belong to Dhaka by 2050.
Nobody counted most of these people. The figures before the twentieth century are the estimates of two historians who spent years on them: Tertius Chandler, whose Four Thousand Years of Urban Growth (1987) gathered city sizes from censuses, travellers and earlier scholars, and George Modelski, whose World Cities (2003) did the same for the ancient world. In 2016 Meredith Reba, Femke Reitsma and Karen Seto turned both books into one geocoded dataset. After cleaning it holds 1,654 places and 10,033 population estimates. From 1975 I switch to the UN's World Urbanization Prospects 2025, which counts cities in a new way, explained below. Read every figure before 1950 as an informed estimate, and often a round one.
Circles: every place with a population estimate at the chosen date, coloured by region, with area proportional to population (by default relative to the largest city at that date, so that the ancient world is visible; switch to one scale to compare sizes across time). Crown: the largest city, or each of the cities tied for it (within 0.5%). Dark line: the population-weighted centre of the 25 largest cities, snapshot by snapshot. Before 1975: Chandler and Modelski as compiled by Reba, Reitsma and Seto (2016), CC BY 4.0, with an average where both give a figure. From 1975: UN World Urbanization Prospects 2025, cities as defined by the Degree of Urbanisation, CC BY 3.0 IGO; cities under 300,000 are left off the map. Positions are the compilers' (before 1975) and the UN's population-weighted centres (after). Modern borders are for orientation only. Outlines: Natural Earth. Base map tiles © Esri.
Modelski's list starts in 3700 BC with a single town, Eridu, at about 6,000 people. By 3500 BC its neighbour Uruk had overtaken it, and Uruk stayed the largest city in every snapshot to 2800 BC, when Modelski puts it at 80,000. Then the crown moved around the city states of Sumer, Lagash, Mari and Girsu, to Ur, which in 2100 BC became the first city of 100,000 in Modelski's list. In 2000 BC Chandler puts Memphis, Egypt's old capital, at 60,000; he gives Ur 65,000, but Modelski gives it only 20,000, so on the average Memphis comes out on top. Babylon led in 1700 BC. Egyptian capitals, Avaris, Thebes and Pi-Ramesses, led in 7 of the 8 snapshots from 1600 to 1000 BC; the exception is 1300 BC, when Yin, the last capital of China's Shang kings, had 120,000.
How big the biggest city was, and which city it was
The largest population at each snapshot (log scale), with the bar below coloured by the region of the city on top. Dashed lines: the largest figure in each compilation on its own. The time axis is stretched towards the present. Hover or tap for each snapshot.
From 1000 BC Modelski lists only cities of at least 100,000, and for several centuries his top cities sit exactly on that line, so Babylon, Memphis and Thebes tie in 900 and 800 BC and 5 cities tie in 700 BC. The classical world brought the first giants. Linzi, capital of the state of Qi in eastern China, led in 500 and 300 BC; Alexandria in 200 and 100 BC, when Modelski gives it 1,000,000, the first city of a million in his list; and then Rome, the largest city in every snapshot from AD 1 to AD 400. Modelski puts Rome at 1,200,000 in AD 200; Chandler's highest figure for ancient Rome is 450,000.
After Rome came Constantinople (AD 500 and 600), Ctesiphon, the Persian capital on the Tigris (622, in Chandler's figures), Chang'an, the Tang capital now called Xi'an (700), and Baghdad, which ties with Chang'an in 800 and leads alone in 900 and 1000. Chandler's own peak for Baghdad is 1,100,000 in AD 932. From 1100 only Chandler's figures remain, and they keep the crown in China for most of the next seven centuries: Kaifeng, the Song capital (442,000 in 1100), briefly Merv in Central Asia (1150), Hangzhou, the southern Song capital, from 1200 to 1350 (peaking at 432,000), Nanjing in 1400, then Beijing from 1450 to 1600 and again from 1750 to 1825. In between, Ottoman Istanbul led in 1650 and 1700 with 700,000.
Every reign
Runs of consecutive snapshots with the same city on top. Ties are listed as their own rows. Click a row to see that date on the map.
London took over in the middle of the nineteenth century, New York in the twentieth, and Tokyo by 1970. In the UN's figures Tokyo stays on top until Jakarta draws level in 2010 and passes it; Jakarta is then the largest in every snapshot to 2045, and Dhaka edges past it in 2049.
From Kaifeng in 1100 to Beijing in 1825, the largest city in the world was Chinese in 13 of 16 snapshots, and until 1800 it never held more than 900,000 people.
The biggest city grew slowly for most of history and then very fast. Uruk's 80,000 of 2800 BC was not beaten until Ur in 2100 BC, 700 years later. The largest city first reached 100,000 in 2100 BC (Ur, in Modelski's list) and a million in 100 BC (Alexandria, also Modelski). Chandler is more cautious: his first city over 100,000 is Avaris in 1600 BC, and at the snapshot dates his first million city is Beijing in 1800, with 1,100,000. Between 1100 and 1750 the largest city in the combined estimates never passed 900,000.
Industrial London broke the pattern: 6,480,000 people in 1900. New York was the first city past ten million in these figures (12,463,000 in 1950) and Tokyo the first past twenty million (20,450,000 in Chandler's 1970 figure). In the UN's count Tokyo had 24,281,618 in 1975, Jakarta has 41,913,860 in 2025, and Dhaka is projected to reach 52,122,843 by 2050. The largest city of 2025 is 38.1 times the size of the largest in 1800.
How many cities passed 100,000 and a million
Places at or above each size at each snapshot (log scale). Before 1975 these count only the places in the two compilations, which are far from complete for smaller cities, so the counts are floors; from 1975 they count every UN city. Hover or tap a point.
In AD 100 the compilations hold 18 places of 100,000 or more; in 1800, 51, and just 1 city of a million, Beijing. By 1900 there were 16 million cities, 13 of them in Europe and North America, and by 1950 at least 66. The UN counts 165 in 1975, 511 in 2025 and 611 by 2050, when 37 cities will hold ten million or more (there were 8 in 1975).
To see where the big cities were, I took the 25 largest places at each snapshot and found their centre, weighted by population. Until 800 BC it never left the band between 33°E and 51°E, from the Nile to the edge of Iran: it was at 32.4°N 45.5°E in 3000 BC, close to Uruk. As Chinese and Indian cities joined the list in the first millennium BC it moved east into Central Asia, reaching 77°E by 500 BC, and Rome then pulled it back west to 45°E in AD 100. From AD 500 to 1825 it stayed between 59°E and 82°E, over Central Asia, poised between the Islamic world, India and China: 78°E in 1500 and 82°E in 1800.
Where the 25 largest cities were
Share of the population of the 25 largest places at each snapshot, by region (fewer places before about 500 BC, when the lists are short). Hover or tap.
Then it lurched west. In the nineteenth century Europe's industrial cities and then America's took over the list: by 1900 13 of the 25 largest cities were European, holding 59% of their people, and the centre sat at 60°N 4.9°W, north of Scotland, 5,102 km from where it had been a century before. Its westernmost point came in 1950, between Iceland and Greenland. After 1975 it raced back east and south: by 2025 it was at 31°N 95°E, in eastern Tibet, and by 2050 it is projected at 27°N 87°E, in eastern Nepal. In 2025, 19 of the 25 largest cities are in Asia and 1 is in Europe.
The centre of the 25 largest cities, east and west
Longitude of the population-weighted centre of the 25 largest places (solid) and of the 10 largest (thin), at each snapshot. Hover or tap.
Two cautions. A centre found this way is the average of points on a globe, so when the big cities are spread across Europe, America and Asia it falls far to the north, between them: in 1950 it is at 68°N. And part of the jump between 1970 and 1975 is the change of source, because the UN's definition makes Asian cities larger and American ones smaller (section 07).
London was 13th in Chandler's list in 1600, with 187,000 people, 4th in 1700 and 3rd in 1750. In 1800 it was second only to Beijing (861,000 against 1,100,000), and in 1825 it was 1.1% behind (1,335,000 against 1,350,000), well within the uncertainty of either figure. By 1850 it was clearly the largest city in the world, with 2,320,000, 1.41 times Beijing.
London and its rivals, 1500 to 2050
Population at each snapshot (log scale), Chandler to 1970 and the UN from 1975, when the definition changes (dashed seam). Hover or tap.
London led in every snapshot from 1850 to 1914. In 1900 it had 6,480,000, 1.53 times New York; in 1914 7,419,000 to New York's 6,700,000. By 1925 the two were level: Chandler gives New York 7,774,000 and London 7,742,000, a difference of 0.4%. By 1950 New York was well ahead. Drawing straight lines between the snapshots puts London's reign at roughly 1826 to 1925, about 99 years. In the UN's count London is a city of 10,416,420 in 2025, the 33rd largest in the world.
Chandler and Modelski often disagree, and before AD 1000 the choice of compilation can change the answer. At the 15 snapshots where both have figures for at least 8 places, they name a different largest city at 10. The starkest case is AD 1000: Modelski gives Baghdad 1,500,000, Chandler gives it 125,000 and makes Cordoba in Spain the largest city, with 450,000. Averaged, Baghdad still leads with 812,500. Chandler's highest figure for ancient Rome is 38% of Modelski's, and for Alexandria 30%.
Where the two compilations name different cities
Snapshots where both have figures for at least 8 places and their largest cities differ, with the combined result used on the map.
Most snapshots before 1000 BC rest on Modelski alone, and many later ones on Chandler alone. Some estimates are interpolated: where Chandler has no figure for a place at a snapshot date, I draw a straight line on a log scale between his two nearest figures, if they are close enough in time; 14.7% of the top ten places at the historical snapshots involve such a step. Reba and colleagues also gave each place a certainty score from 1 (most certain) to 3 for how securely it was identified and placed: 73 of the 83 crown holders before 1975 score 1, and the exceptions are Hangzhou, Memphis and Pi-Ramesses.
The UN's World Urbanization Prospects used to rely on each country's own idea of a city or urban agglomeration, which made comparisons awkward. The 2025 revision instead applies the Degree of Urbanisation, a method agreed by the UN Statistical Commission: a city is a cluster of contiguous 1 km grid cells with at least 1,500 people per square kilometre and at least 50,000 people in all, found in satellite-based population grids. It ignores administrative borders and asks only where people live densely. That changes the familiar rankings. Jakarta and Dhaka come out larger than in older lists, because their dense built-up areas sprawl across many districts; American cities come out smaller, because their suburbs are not dense enough to count; and some densely settled countrysides become very large "cities", such as Hajipur in Bihar, India, with 9,941,510 people in 2025.
The same cities in 1975, counted two ways
Chandler's 12 largest places in 1975, an urban agglomeration count, beside the UN's DEGURBA city that contains each one, and its rank among the UN's cities that year.
The seam shows clearly in 1975. Chandler gives Tokyo 23,000,000 and the UN 24,281,618, close enough. But Chandler's New York, 17,100,000, becomes 12,076,416 under the UN's definition, and Jakarta grows from 5,300,000 (20th in Chandler's list) to 11,696,369 (4th). Chandler's 1975 list counts 189 places of a million or more and the UN 165; Modelski's list of million cities in 2000 has 292, the UN 337. I have not blended the two kinds of figure: on the map, the charts and in the text, 1975 and later is the UN, and everything before is Chandler and Modelski.
The UN projects every city forward to 2050. In its figures Jakarta passed Tokyo in 2010 (they are within 0.5% of each other in that year's snapshot), Dhaka passed Tokyo in 2022, and Dhaka is projected to pass Jakarta in 2049, reaching 52.1 million to Jakarta's 51.8 million. That is a margin of 0.7%, well within the uncertainty of a projection, so the honest reading is that Dhaka and Jakarta will be the two giants of 2050. Tokyo, which peaked at 33.7 million in 2020, is projected to shrink to 30.7 million.
The largest cities of 2050, and how far they have to grow
The 15 largest cities in the UN's 2050 projection: 2025 (open circle) and 2050 (solid). Coloured by region. Hover or tap.
The biggest gains between 2025 and 2050 are in Dhaka (+15.5 million), Karachi (+11.2 million), which enters the top ten, and Jakarta (+9.9 million), then Luanda (+8.9 million) and Cairo (+6.8 million). Among cities that will have ten million or more, the fastest growing are Luanda (78%), Dar es Salaam (69%) and Addis Ababa (61%). Lagos rises from 26th to 21st. Sub-Saharan Africa has 17 of the 100 largest cities in 2025 and 24 in 2050, and takes 37% of the growth among the 2050 top hundred. 15 of them are projected to shrink, led by Tokyo and Osaka.
Data. The historical figures are Reba, Reitsma and Seto's (2016) digitisation of Chandler (1987) and Modelski (2003), on figshare: Chandler's compilation (1,597 places, 9,438 figures from 2250 BC to 1975), Modelski's ancient cities (154 places, 621 figures, 3700 BC to AD 1000) and Modelski's 293 cities of a million or more in 2000. I used the V2 files. Compared with V1 they trim stray spaces from 73 names and countries (leaving 3), correct "Instanbul" to Istanbul, move Kiev from the Russian Federation to Ukraine, and fold V1's duplicate Guadalajara row and a misspelt "Madgdeburg" row into one row each, which is why Chandler has 1,597 rows rather than 1,599; no figure changes. The files are Windows-1252 text, not UTF-8, and must be read as such (Košice, Šomron); 3 alternative names had already lost characters to question marks before the files were written. The modern figures are the UN World Urbanization Prospects 2025 files 18 to 21: the 100 largest cities at each date, the 100 largest of 2025 through time, capitals, and all 16,828 cities that reach 50,000 between 1975 and 2050, each with a population-weighted centre.
Corrections. I made 26 documented corrections, listed below. The largest is that Modelski's cities in the Americas carry, in the BC columns, exact copies of their AD figures 1,300 years later (Teotihuacan in 900 BC equals Teotihuacan in AD 400, and so on): I dropped those 14 copies. I also dropped single figures that are out by a factor of about ten from the same city's neighbours (Philadelphia in 1914 at 17.6 million, for instance), figures attached to the wrong city (Memphis, Egypt, in 1975 carries the figure for Memphis, Tennessee) and two misplaced locations.
Method. The snapshots are Modelski's dates up to AD 1000, plus Chandler's cross-sections in 1360, 650 and 430 BC and AD 361 and 622, then Chandler's benchmark years from 1100 to 1970, then the UN every five years. At each snapshot each compilation gives a place its own figure for that date if it has one. Otherwise Chandler's figure is interpolated on a log scale between his two nearest dates for that place if they are no more than 300 years apart before AD 1000, 200 years to 1800 and 100 years after; Modelski's are interpolated only between neighbouring dates of his own grid, because a gap in his grid means the place was not on his list. Nothing is extrapolated. Where both compilations have a figure, I use their average. Places within 0.5% of the largest count as tied. I matched 95 places across the two compilations by name and distance (under 30 km), and linked 1,129 historical places to the UN city that contains them, by name within 50 km. The centre of gravity averages the positions of the 25 largest places as points on a sphere, weighted by population.
Where the record is thin. The compilations are rich for Europe, the Middle East, India and China and thin elsewhere. Before AD 1000 they hold 11 places in the Americas (Modelski's are Caracol, Teotihuacan, Tiahuanaco, Tikal and Tollan) and 4 in sub-Saharan Africa. Between 1500 and 1799 they hold 241 European places, but 67 East Asian and 29 in sub-Saharan Africa. Cahokia, on the Mississippi, has 2 figures in the whole record and Great Zimbabwe 3, so the absence of a circle is not evidence that nobody lived there, and counts of cities before 1975 are floors.
How many places the compilations hold, by region and period
Places with at least one population estimate in each period (Chandler and Modelski together, after corrections).
What this is not. These are estimates, often rounded to the nearest 10,000 or 100,000, and often made by working back from city walls, house counts, food supplies or later censuses; a factor of two either way is entirely possible for an ancient city. "City" means different things in different sources: a walled town and its suburbs for Chandler, a metropolitan area in some modern counts, and a dense grid cluster for the UN. The 2025 to 2050 figures are projections. And the names are modern conveniences: I give period names where the text needs them (Chang'an, Constantinople, Edo), but the map uses one name per place.