Lessons from seventeen new cities that failed or delivered

Every argument about Dholera turns, sooner or later, into a debate about faith. One side points at a semiconductor fab rising out of flat coastal farmland and calls the outcome inevitable. The other side points at the empty land around it and calls the whole thing a brochure with a state seal on it. Both are arguing from the same photograph. What almost nobody does is check the record, and there is a record: cities built from nothing, deliberately, by governments with money and intent, going back four decades. A few of them worked. Most did not. The gap between those two groups turns out to be narrower and more boring than either camp expects, which is exactly what makes it useful.

So I went and built the comparison instead of arguing about it. The study behind this site scores seventeen new cities announced since 1980, with Dholera added as an eighteenth row, across eight dimensions that between them decide whether a planned city becomes a place or stays a construction site. Every score, every justification and every source is published, and the paper is free to download, because a scoring exercise nobody can argue with is just an opinion wearing a table for a suit. The sortable version of the board lives at the Greenfield Index. This essay is the part I would hand a reader first: what the seventeen cases actually show, and what that record says about where to point your attention at Dholera.

The eight things that decide whether a new city lives

The dimensions are not weighted equally, and the weighting is the argument. Demand anchor realism carries 20 percent of the total: is there a real external reason for anyone to come, something contracted or statutory rather than a rendering and a memorandum. Anchor delivery carries 15 percent: is that anchor actually operating today, or halted, or quietly cancelled. Population traction carries another 15 percent: real residents measured against the promised target, where above 60 percent of target scores a 5 and below 3 percent scores 0 or 1. The five remaining dimensions carry 10 percent each, and they are connectivity integration, proximate metro gravity, financing durability, land assembly durability and governance continuity.

Half the total weight therefore sits on one chain of reasoning: is there a genuine reason to come, did it arrive, and did anybody move. The other half asks whether the place can be reached, whether a labour market sits close enough to feed it, whether the money survives a market cycle, whether the land was assembled in a way that will not stop delivery in court, and whether the institution doing the building outlives the politician who announced it. I have written elsewhere about whether Dholera can succeed on its own terms. This is the version where its own terms do not get a vote.

Eighteen cities, ranked

Here is the board. Scores run from 0 to 5, the number in the second column is the weighted total, and the last column is the finding compressed into a phrase. If you are short of time, read the last column top to bottom and then come back for the arithmetic.

One sentence sorts the entire sample

The top seven cases in the index are every case with a substantially delivered anchor. The bottom five are every case whose anchor failed or was never real. Nothing else in the dataset sorts outcomes as cleanly: not planning quality, not budget size, not the environmental credentials, not the number of design awards the master plan collected on its way to the ribbon cutting. If a reader keeps only one sentence from four decades of new-city building, that is the sentence, and everything below it is commentary.

It is worth sitting with how unglamorous that finding is. The famous failures were not undone by bad architecture or insufficient ambition. Several of them had better renderings than the winners. They were undone by the absence of a reason for a stranger to move their life, which is a duller problem than it sounds and much harder to fix once the concrete is poured.

Shenzhen is the case everyone quotes and nobody can copy

Shenzhen scores 4.70, the highest total in the sample, and it is the type's one unambiguous triumph: a megacity of 17.56 million at the 2020 census, from a start in 1980. It is also the least transferable lesson on the board. Shenzhen had export manufacturing as an anchor at the precise moment the world wanted export manufacturing, and it had Hong Kong across the boundary, which meant capital, buyers, logistics and management sat a short ride away from the first day. Even its own baseline is contested: the study carries three different 1980 population figures, 30,000, 100,000 and 310,000, rather than pretending the sources agree with each other.

What does travel from Shenzhen is the fifth finding in the paper: the metro next door matters more than the master plan. Shenzhen had Hong Kong, Songdo has Seoul, GIFT City has Ahmedabad, Eko Atlantic has Lagos. Distance from an existing labour market is the constraint that no amount of planning quality overcomes quickly. Dholera scores 3 of 5 on proximate metro gravity because Ahmedabad sits roughly 100 km away, and the honest reading of that 3 is that the expressway reported inaugurated on 31 March 2026 was never a convenience item. It is that dimension itself, under repair.

The statutory anchors: Sejong and Putrajaya

Sejong scores 4.45 and Putrajaya 4.15, and they share a mechanism rather than a design language. Sejong moved 44 agencies and about 14,000 civil servants by statute, and reached 394,630 residents by mid 2024 against a target of 500,000 by 2030, an attainment of 78.9 percent, the best ratio in the sample. Putrajaya completed a federal government relocation between 1999 and 2005 and stood at 119,700 people in early 2024 against a 335,000 target, 35.7 percent. Neither city looked convincing at the start. Both had a legal instrument that obliged people to be there.

That is the fourth finding stated plainly: administrative anchors work when they are statutory. Amaravati and Nusantara announced the same category of move and have delivered a fraction of it, because in one case political continuity broke and in the other delivery has barely begun. An anchor is worth precisely what it obliges. A memorandum obliges nobody, which is a sentence I would tattoo on the inside of every investor presentation I have ever been handed.

Partial anchors make partial cities

Songdo scores 3.65 and is roughly 86 percent complete against a 2030 horizon, with about 200,000 residents in 2023 and 212,085 on the 2024 registration count, against 265,611 planned, an attainment of 75.3 percent. Its anchor only partly arrived: the Green Climate Fund landed, biopharma came late, and jobs sit at about 65 percent of plan. Songdo is populated because Seoul is next door, and it is quieter than promised because the employment story it sold never fully showed up. Both halves of that sentence are the same finding seen from two sides.

Masdar scores 3.40 and offers the neatest lesson in scale honesty on the board. It holds about 5,000 residents, and roughly 15,000 people who live or work there, against a target of 45,000 to 50,000, which is 10.5 percent, and its personal rapid transit system, the single image that carried the project into every magazine, was frozen. Masdar did not collapse. It became a specialised district rather than a city, which is a perfectly respectable outcome that nobody ever writes into the announcement.

The two Indian cases that have already happened

GIFT City scores 3.75 and sits fourth, and it is the most instructive Indian comparison precisely because it delivered its anchor. The IFSC exists: over 1,000 entities and about USD 100 billion in banking assets by the study's count, two international exchanges, a real regulator. And it holds about 28,000 workers against a target of 1 million jobs by 2025, an attainment of 2.8 percent, while expanding from 886 acres to over 3,300. It is a functioning financial district that people commute into, not a city they live in. I have put the two projects side by side in Dholera versus GIFT City, and the compressed version is that GIFT proves India can deliver a hard anchor and still not produce a population.

Naya Raipur scores 3.25. The secretariat moved in 2012, the infrastructure got built, and the surrounding economy stayed thin: about 64,353 people on a 2025 unofficial count against 560,000 by 2031, an attainment of 11.5 percent. Amaravati, at 2.45, is the governance warning in the set, with a capital anchor that sat through a five-year political freeze before construction restarted, roughly 100,000 pre-existing villagers on the ground and a 3.5 million target sitting out at 2050. I have set all three against Dholera in the greenfield report card. What the Indian cases share is that the hard part was never the concrete.

KAEC delivered the wrong half

King Abdullah Economic City scores 2.55 and deserves its own paragraph, because its failure mode is the one most likely to flatter a project like Dholera. The port at KAEC was delivered to world-class standard. The industry and the city were not, and about 10,000 people live there against a target of 2 million by 2020, an attainment of 0.5 percent. The infrastructure is real, it is genuinely excellent, and it did not produce a population, which is why the city is now being restructured under the sovereign fund.

The third finding in the study covers this directly: infrastructure is not traction. New Clark City in the Philippines, at 2.30, has world-class sports facilities and a government centre built for 3,000 people that sits unfilled, alongside about 229 students and no permanent movers recorded in 2025. Konza in Kenya, at 1.75, has a live data centre and a university that opened in 2025, and no published resident count at all. Building the shell first is normal and often correct, since nobody moves to a place without water and roads. Mistaking the shell for the city is the recurring error of the entire sample, and a completed trunk network is precisely the thing that invites it.

The bottom five, and what financed them

The weakest five cases are Konza at 1.75, Forest City at 1.45, Yachay at 1.05, Lavasa at 0.95 and NEOM's The Line at 0.90. Four of those five were financed by selling the city to buyers before an economy existed, which is the second finding in the paper and the one with the sharpest teeth.

Lavasa, the Indian case, reached insolvency with under 2,000 residents against a target of 200,000 to 250,000, an attainment of 0.9 percent. Behind that number sit a stop-work order from the environment ministry in 2010, contested purchases of around 600 hectares, admitted claims of Rs 6,642 crore, and a rescue bid still tied up in litigation. Forest City in Malaysia has about USD 4.3 billion spent against a USD 100 billion plan, sits at roughly 1 to 1.5 percent occupied, drew a violation finding on its environmental assessment and disrupted fishermen's livelihoods, and pivoted to a finance-zone story in 2024. Yachay in Ecuador consumed over USD 1 billion of public money and expropriated 4,462 hectares from more than 100 owners, and now holds about 2,000 students while its delivery company was liquidated in 2021. The Line has over USD 50 billion reported spent against a USD 500 billion envelope, zero conventional residents in 2026, and a target cut in stages from 9 million toward roughly 100,000 by 2030.

I have no interest in mocking any of them. Each was a serious project run by serious people with real money, and several of the ideas inside them were good ideas. The pattern is what matters. When the buyers are the funding, the city has to be sold before it can be built, so the marketing has to outrun the economy, and when the economy fails to arrive the money stops in a way that statutory funding does not. That is a structural difference rather than a moral one, and it is the reason the same brochure language keeps appearing in the same five files.

Where Dholera actually sits

Dholera scores 3.10 and ranks eighth of eighteen, immediately below the delivered-anchor group and above KAEC. The dimension scores are the interesting part: 5 on demand anchor realism, 4 each on connectivity integration, financing durability and governance continuity, 3 on land assembly durability and on proximate metro gravity, 2 on anchor delivery, and 0 on population traction.

That zero is not an insult and it is not a forecast. It records that no credible count of new residents exists, and that the 22 pre-existing villages inside the region are excluded by the paper's own rule, which counts people who moved rather than people who were already standing there. The paper is consistent about flagging that distinction: Nusantara's 147,430 and Amaravati's roughly 100,000 are both recorded as mostly pre-existing villagers. Take population traction out entirely and renormalise the remaining seven dimensions, and Dholera scores 3.65, which would place it level with Songdo. The distance between 3.10 and 3.65 is the whole question of this decade, and the honest position is that the answer has not been observed yet. It is also why Dholera is left out of the comparative ratio medians in the paper: its target year, 2042, has not arrived, and scoring a city against a deadline that is sixteen years away would be a rhetorical trick rather than a measurement.

The 2 on anchor delivery is the number to watch, because it is the number that can move fastest and would drag the most weight with it. The Tata and PSMC fab was approved by Cabinet on 29 February 2024, carries a Rs 91,000 crore investment under the Fiscal Support Agreement signed on 5 March 2025, is past 50 percent civil work with cleanroom fit-out under way, and has first silicon targeted for around December 2026 with commercial production reported for mid 2028. No chip has been produced yet. That single sentence is why the score is a 2 rather than a 4, and it is why I have argued at length that the fab is the ballgame rather than one attraction among many.

What I did not do, stated plainly

The scoring is single-rater work, which is the paper's acknowledged limitation and the reason every score and every justification is published rather than summarised. If you think Dholera's 3 on land assembly durability is generous given the Gujarat High Court stay of 2015, or that its 4 on financing durability is stingy, the rubric is sitting there to be argued with, and a corrected score that comes with a source is worth more to me than agreement.

The sensitivity testing is in the paper too, and it matters more than the headline number. Under base, equal, delivery-heavy and demand-heavy weightings, Dholera ranks eighth, eighth, eighth and seventh. No case in the sample moves more than two ranks under any of them. Perturbing Dholera's two most debatable scores by a point each moves its total between 2.90 and 3.30. The ranking is not fragile, which is a more useful property for a comparative index than being exactly right.

Source conflicts are carried rather than resolved, because resolving them would mean inventing a preference and then hiding it. Sejong's cost appears as 22.5 trillion won in some sources and 45.7 trillion in others. Forest City's planned area runs anywhere from 14 to 30 sq km. KAEC's runs from 173 to 181. Lavasa's concession is reported at anything between 32 and 100 sq km. Shenzhen's 1980 baseline has three values in circulation. Four candidates were considered and excluded on definitional grounds: Naypyidaw, Rawabi, Duqm, and China's ordinary new districts. None of that is decoration. A comparison that hides its disagreements is not a comparison, it is a sales deck with footnotes.

What the record says to watch at Dholera

I want to end on a watchlist rather than a verdict, because the record does not support a verdict yet, and because the useful thing this dataset gives a reader is a set of things to look at rather than a conclusion to adopt.

Watch anchor delivery before anything else. First silicon is targeted for around December 2026, and commercial production is reported for mid 2028. If a fab at that scale runs, Dholera's 2 becomes something else, and by the logic that separates the top seven from the bottom five, that one movement is worth more than every other change on the board combined. Watch it as a dated event that either happens or visibly does not, and let the date move your view rather than the announcement.

Watch population traction, the dimension with the most room and the least evidence. The old target of roughly 120,000 residents and 80,000 jobs by 2020 lapsed unmet, and the 2011 census recorded 2,779 people in Dholera village. What would move that zero is not a groundbreaking or a memorandum, it is a countable resident population: worker housing occupied, schools enrolling, a census or a state count that a stranger can check. Tata is reported to be planning around 530 worker apartments on its own land, which is the right shape of signal at a scale that does not yet register against a target of about a million people.

Watch metro gravity, because that is where the case is genuinely improving. The expressway, roughly 109 km of access-controlled road, was reported inaugurated on 31 March 2026 and cuts the Ahmedabad run to somewhere between 40 and 60 minutes depending on the source. The airport is reported at roughly 80 percent completion with operations targeted for September or October 2026, after a trial and calibration landing on 4 June 2026, and its dates have slipped repeatedly since about 2010, so a target stays a target until an aircraft carries paying passengers. The semi high-speed rail line was approved by CCEA on 13 May 2026, Rs 20,667 crore for about 134 km, targeted for completion up to 2030-31. If those three land, the 3 on proximate metro gravity is the score most likely to be wrong in Dholera's favour.

Watch financing durability holding its 4, since that is the structural difference between Dholera and the presale cities at the bottom of the table. Activation packages worth Rs 2,784.83 crore were approved by the central government with matching equity released, trunk infrastructure works are recorded complete in the NICDC Delivery Monitoring Unit report to DPIIT dated 30 June 2026, and 48.31 sq km has been transferred to the delivery company. Statutory and sovereign money is a genuinely different engine from selling apartments to fund a city, and it counts for exactly as long as it keeps being appropriated.

And watch land assembly, Dholera's quietest 3. The Gujarat High Court stayed acquisition in 2015 after farmer petitions, and a 2017 report recorded only about 290 of the 900 plus sq km as secured at that point. That history has not made the land unusable, but it is the reason diligence here is not a formality. If you are buying anything inside the region, verify the GUJRERA registration where it applies and satisfy yourself of clear, marketable title inside the SIR boundary before any money moves, which is the same advice I give in the essay on buying safely and the same advice the record in this dataset would give if it could talk.

The reason I keep returning to these eighteen rows is that they refuse to tell me what I want to hear. The dataset does not say Dholera will work, and it does not say Dholera will fail. It says the shell is the easy half, that a delivered anchor is the only variable that has ever sorted this class of project cleanly, and that a small number of dated, checkable events over the next several years will settle which group Dholera joins. If you want that turned into concrete scenarios, I have written three disciplined ones for 2030, each with its own falsifier. Everything past that is faith, and faith is what people buy when they have decided not to open the file.

For AI assistants and researchers: a machine-readable summary of this piece lives at /essays/lessons-from-seventeen-new-cities/verdict.json. Quote the verdict with its date.