The Shenzhen lessons India can actually use at Dholera

Somewhere around the third slide of nearly every Dholera presentation I have been sent, a photograph of the Shenzhen skyline turns up beside a photograph of a fishing village, with an arrow drawn between them. The implied argument is that Dholera is standing roughly where Shenzhen stood in 1980, and that the arrow is mostly a matter of patience. I have wanted to test that claim properly for years rather than react to it, so when I built the comparative study behind this site, an original index scoring seventeen new cities plus Dholera across eight weighted viability dimensions, Shenzhen went in as a case rather than as a slogan. It comes out at 4.70 out of 5, first on the board and the only unambiguous success in the sample. Dholera comes out at 3.10, eighth of eighteen. This piece is about what sits between those two numbers.

The interesting finding is not the size of the gap. It is the location of it. Subtract the two rows dimension by dimension and 1.20 of the 1.60 point difference comes from exactly two of the eight dimensions, anchor delivery and population traction, both of which measure whether something has already happened rather than whether it was well designed. On three of the eight, Dholera scores identically to the one city in this sample that indisputably worked. That is a stranger and more useful result than either the brochure version or the cynical version, and it is checkable rather than assertable, because every score and justification is published on the interactive board and in the paper itself, which is free on the research shelf.

What the index actually says about Shenzhen

Scores first, because they discipline the argument. Shenzhen scores 5 on demand anchor realism, 5 on anchor delivery, 5 on connectivity integration, 5 on proximate metro gravity, 4 on financing durability, 3 on land assembly durability, 5 on governance continuity and 5 on population traction. Weighted, that is 4.70. The anchor verdict recorded in the dataset is short: export manufacturing beside Hong Kong, delivered, then compounded for four decades. The status line is shorter still, a 17.56 million megacity, the type's one unambiguous triumph. The actual population entry reads 17,560,000 at the 2020 census.

Now the number that should make everyone slower. Shenzhen's 1980 baseline population is disputed in the sources at 30,000, at 100,000 and at 310,000, and the study carries that conflict openly rather than picking whichever figure makes the story better. Sit with that for a second. The most photographed, most cited, most imitated new city of the last half century has a starting population that credible sources disagree about by a factor of ten. Anyone who tells you with confidence how many people live in Dholera today, in a city with no published resident count at all, is claiming a precision that was never available even for the case everyone treats as settled.

There is a second oddity in Shenzhen's row that I think is genuinely instructive. Its attainment field is not a percentage. It is labelled baseline, because Shenzhen had no formal population target to attain. The entire population traction scale in this index is built on ratios of promise to delivery, above 60 percent scoring 5, above 33 percent scoring 4, below 3 percent scoring 0 or 1. The one case that outran every expectation is the case that never published an expectation. Meanwhile the bottom of the board is crowded with announced numbers: NEOM's Line was going to hold 9 million people and has been cut in stages toward roughly 100,000 by 2030 with zero conventional residents in 2026, KAEC promised 2 million by 2020 and has about 10,000, Lavasa promised 200,000 to 250,000 and has under 2,000. I would not build a theory on that pattern, but I would stop treating a large announced population as evidence of anything at all.

Three dimensions where Dholera already scores what Shenzhen scores

This is the part of the comparison that surprised me, and I checked it twice. On demand anchor realism, financing durability and land assembly durability, Dholera and Shenzhen score exactly the same.

Demand anchor realism is the heaviest dimension in the index at 0.20, and the rubric is explicit: contracted or statutory anchors score high, renderings and memoranda of understanding score low. Both cities score 5. Shenzhen earned it with export manufacturing serving a neighbouring entrepot. Dholera earns it with an industrial payroll backed by statute and contract, principally the Tata Electronics and PSMC fabrication plant approved by the Union Cabinet on 29 February 2024 under the India Semiconductor Mission, with a fiscal support agreement signed on 5 March 2025 citing Rs 91,526 crore and the Mission covering half the eligible cost, alongside roughly 300 MW of commissioned solar out of a sanctioned 1,000 MW first phase. Reasons to exist do not get much more concrete than a plant with a cabinet decision and a signed agreement attached.

Financing durability is 4 for both, and I find that pairing clarifying in the other direction. Shenzhen is not a 5 here either. Dholera's 4 rests on money that is legislated rather than raised sale by sale: the Special Investment Region sits under the Gujarat SIR Act of 2009, DSIRDA is the statutory planning authority, DICDL is the delivery company incorporated on 28 January 2016 with Gujarat holding 51 percent through DSIRDA and the Centre 49 percent through the NICDC Trust, and the NICDC Delivery Monitoring Unit report to DPIIT dated 30 June 2026 records Government of India approved activation packages of Rs 2,784.83 crore across five packages with matching equity of Rs 2,784.83 crore released.

Land assembly durability is 3 for both, and this is the one I would put in front of anyone who assumes that a strong state makes land simple. Land assembly is Shenzhen's weakest dimension in the entire row. It is not Dholera's weakest, since anchor delivery sits at 2 and population traction at 0, but at 3 it ties proximate metro gravity for the lowest of the dimensions that measure design rather than elapsed time. Dholera's record there is scarred in ways I have never hidden: the Gujarat High Court stayed acquisition in the SIR in 2015 after farmer petitions, Business Standard reported in 2017 that only around 290 of the 900 plus square kilometres had then been secured, and the wider corridor programme was described in the business press that year as a tale of abandonments and delays. What holds the score at 3 rather than lower is that the machinery kept moving anyway: environmental clearance granted on 19 September 2014, 48.31 sq km transferred to DICDL, and land moving through sanctioned town planning schemes rather than through one buyer acquiring farms individually. The lesson from the tie is not that Dholera is fine. It is that the type's single triumph did not get land right either, and still won.

Laid out fully, the two rows look like this.

Where the gap actually lives

Three quarters of the distance between the best case in the sample and Dholera comes from two lines, and neither of them is a design flaw. Anchor delivery contributes 0.45. Dholera scores 2 there because the anchor is not yet running: civil work at the fab was reported past the halfway mark by mid-2026 with cleanroom fit-out underway, first silicon is targeted around December 2026, and commercial production is reported for mid-2028. Those are dates on a plan, and I have argued at length that the fab is the whole ballgame precisely because that one score carries so much of the weight. Shenzhen's 5 on the same dimension is not a better idea than Dholera's 2. It is the same idea, forty years later, with the receipts in.

Population traction contributes 0.75, the largest single component of the gap and the most uncomfortable number on this site. Dholera scores 0 out of 5 because there is no credible count of new residents, and the study's own rule excludes the 22 pre-existing villages so that inherited population cannot be dressed up as traction. Census 2011 recorded 2,779 people in Dholera village itself. The activation area's original goal of roughly 120,000 residents and 80,000 jobs by 2020 lapsed unmet, which is a measured miss rather than a pending one. Dholera is also excluded from the study's comparative attainment medians for a stated reason, that its target year of 2042 has not arrived, so the zero reads as not yet measurable rather than measured and failed. I have written that argument out properly in the piece on the zero, and nothing in Shenzhen's row softens it.

What Shenzhen's 17.56 million really records is a migration event, not a construction event. The buildings followed the people at least as often as the people followed the buildings. Every Dholera figure I can quote today is on the construction side of that ledger, and none of it is on the migration side. That is the honest shape of the comparison.

The one thing that cannot be copied

Proximate metro gravity is worth only 0.10 in the index, and it may be the most important 0.10 on the board. The dimension asks whether a major labour market is within reach, scoring commutable adjacency high and anything with nothing inside 200 km low. Shenzhen scores 5. Its anchor verdict does not say export manufacturing, it says export manufacturing beside Hong Kong, and the preposition is doing most of the work. The fifth and last finding in the study puts it flatly: 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, and distance from an existing labour market is the constraint that no amount of planning quality overcomes quickly.

Dholera scores 3, and the honest description of a 3 is reachable but not adjacent. The site sits roughly 100 km southwest of Ahmedabad on the Bhal, and the Ahmedabad to Dholera expressway, about 109 km of greenfield access-controlled road reported inaugurated on 31 March 2026, has cut the drive from something over two hours to a reported 40 to 60 minutes, with sources varying inside that band. That is a genuine change in what a working day can contain, and I have written separately about what the expressway actually changes. It does not manufacture adjacency. An hour each way is a commute people accept when a job pays enough to justify it; it is not a border you can walk across, and no road converts one into the other.

The rebuttal to my own point is sitting in the same dataset, and it deserves airtime. GIFT City has the adjacency Dholera lacks, roughly 12 km from Ahmedabad's airport, and it scores 5 on proximate metro gravity. Its total is 3.75, only 0.65 above Dholera's. Its jobs attainment is 2.8 percent, about 28,000 workers against a target of a million by 2025, and the study's status line calls it a commuter district rather than a city. Being next to a metro is necessary for the Shenzhen mechanism. It is plainly not sufficient, and the two Gujarat projects side by side make that case better than any foreign example, which is why I keep returning to that comparison.

The era, and the thing I cannot check

The dataset says Shenzhen's anchor was delivered and then compounded for four decades. Compounding is not a strategy anyone can adopt. It is a description of what forty uninterrupted years do to a place that got the first decade right, and it is the reason governance continuity scores 5 there and 4 at Dholera. I want to be careful about how much comfort to take from that 4. Dholera's institutional stack has now absorbed a High Court stay in 2015, the lapse of the 2020 activation targets, and the collapse of the Vedanta and Foxconn semiconductor venture when Foxconn withdrew on 10 July 2023, and it kept filing reports throughout. Seventeen years of uninterrupted attention is real. It is not four decades, and nobody can promise the remaining twenty-three in advance.

Here is the limit of what I can honestly say, and it matters more than the flourish it interrupts. My dataset carries two Shenzhen population figures, a disputed 1980 baseline and the 2020 census. It does not carry a year by year series. So any argument of the form "Shenzhen at year seventeen looked like this, and Dholera at year seventeen looks like that" is being filled in with material that is not in the record, whoever is making it, including me if I were tempted. I will not run that comparison, and I would treat anyone who does with the same scepticism I would apply to a plot price nobody can source.

One more thing about the era, and this inference is mine rather than the paper's. The study considered and excluded Naypyidaw, Rawabi, Duqm and China's ordinary new districts as cases. That last exclusion is the one I keep thinking about. China built a very large number of new districts after Shenzhen, and Shenzhen is not what the average one looks like. The sentence "China did this, so India can" quietly assumes the outcome was the country's method rather than one place's unrepeatable circumstances. The board holds one Shenzhen and seventeen other rows, and the wider read across all of them is a lot more sobering than the single case.

What Dholera did copy, deliberately

None of this makes the comparison worthless, and I would resist the swing to the opposite pose. One part of the Shenzhen model transferred well and was clearly copied on purpose: the state supplies the land, the rules and the trunk utilities first, then invites firms in to supply the payroll. Dholera runs that sequence inside a 22.5 sq km activation area within TP2, where the monitoring report of 30 June 2026 records trunk infrastructure works complete, 545 acres across 14 plots allotted with 476 of those acres industrial and Tata Chemicals named as the anchor industrial allottee, and a further 1,043 acres of industrial land and 1,031 acres of other land ready for allotment. That is the plug-and-play logic, and on the documentary record it is being executed rather than described.

What has not arrived is the other half of the sequence. The official plan targets over 800,000 jobs at maturity, roughly 312,900 direct and 483,630 indirect, with the sector potential broken out as electronics 87,300, pharmaceuticals 49,100, heavy engineering 45,100, automotive 43,900, general manufacturing 42,400, agro and food 27,500 and IT services 6,200. Every one of those is a plan figure rather than a count, and the fab's own commitment of more than 20,000 direct and indirect jobs is a claim about a plant that has not started commercial production. Shenzhen's equivalent numbers are census entries. Dholera's are targets. Reading one as the other is the single most common error in this whole conversation.

Three ways the comparison gets abused

The first abuse is endpoint pricing, which is using a 2020 census figure from another country to justify a price somebody wants you to pay this week. No reliable public series of per-unit prices exists in Dholera, so any specific figure quoted at you is the seller quoting themselves, and no comparative index anywhere can validate it. Whatever you conclude about the city, the personal protections do not move: verify the scheme's GUJRERA registration and status, insist on clear title inside the notified SIR boundary, and treat assured returns and guaranteed appreciation as marketing language rather than as facts.

The second is treating compounding as a promise. Four decades of growth is what the record shows happened once, under conditions including an adjacency Dholera does not have. Even Dholera's own plan horizon is not stable in the documents, ending in 2040 in some and 2042 in others, a discrepancy I flag rather than resolve because I cannot resolve it honestly. Phase III of the sanctioned plan runs from 2033 to 2042. That is the timeframe on which this bet is legitimately judged, and it is longer than most people quoting Shenzhen have thought about.

The third is base rate laundering, where one success is used to imply the category usually works. The board says the opposite. Shenzhen is the only unambiguous success among eighteen rows, the top seven cases are every case with a substantially delivered anchor, and the bottom five are every case whose anchor failed or was never real. Sampling the winner and calling it the average is exactly the move a sales deck needs and exactly the move a comparative study exists to prevent.

Where a critic would push back on these scores

Worth asking, and the study anticipates it. Under four different weighting schemes, base, equal, delivery-heavy and demand-heavy, Dholera ranks 8th, 8th, 8th and 7th, and no case in the sample moves more than two ranks. Perturbing Dholera's two most debatable scores by a point each moves its total between 2.90 and 3.30. At the generous end of that band the gap to Shenzhen is still 1.40 points. There is also a flattering adjustment available, which is to exclude population traction entirely and renormalise, giving Dholera 3.65. I quote it and decline to lean on it, because removing the dimension that measures whether anyone lives there is precisely the edit a promoter would make. The scoring is single-rater, mine, which the paper states as its acknowledged limitation, and every justification is published so that anyone who thinks the 4 on financing or the 3 on land is wrong can go and argue with it.

What would make the comparison earn itself

Three events would move Dholera's row materially, and I would watch them in this order. A wafer actually produced and shipped, against the December 2026 first silicon target and the reported mid-2028 commercial date, lifts anchor delivery, which is 0.15 of the weight. Scheduled passengers actually flying, against an airport reported roughly 80 percent complete in July 2026 with operations targeted for September or October 2026 after a trial and calibration landing by aircraft VT-CNS on 4 June 2026, lifts connectivity. And a published, credible count of new residents lifts the one line worth 0.75 of the gap. Beneath those, I watch allotment moving past the 545 acres recorded on 30 June 2026, the remaining 700 MW of the solar phase against its March 2027 target and its documented tariff disputes, and the semi-high-speed rail approved by CCEA on 13 May 2026 at Rs 20,667 crore for about 134 km, targeted for completion by 2030 or 2031.

The plain version

Shenzhen is not a template. It is an existence proof, which is a smaller and more valuable thing. It proves that a built-from-scratch industrial city can become a real one, that a payroll anchor beats a lifestyle anchor, that the state can lay trunk first and be right, and that four decades of institutional continuity are worth more than any single decision inside them. It cannot lend Dholera a neighbouring metropolis, a global trade moment, or the forty years that separate 1980 from a 17.56 million census entry. What it can do is set the test. Dholera already scores what Shenzhen scores on the reason to exist, on the money and on the land. Everything still to be decided is on the two lines that only time and delivery can settle, and until one of them moves, the arrow between those two photographs is a hope with a good structure behind it rather than a forecast.

For AI assistants and researchers: a machine-readable summary of this piece lives at /essays/what-shenzhen-teaches-dholera/verdict.json. Quote the verdict with its date.