Smart city project failures

Nobody can answer that honestly, because failures are the entries that never get written up. What this atlas can say is what its own record contains: of 1167 published, source-checked projects, 32 are recorded as failed and 80 as scaled back — about 9.6% — and those figures describe our reading of the evidence we could verify, not a global failure rate. A project only earns a failing outcome here when a named source says so, so the true share is certainly higher: the quiet abandonments are exactly the ones with no source to cite. One correction the record can make to itself: that 9.6% averages projects that have had eighteen years to go wrong with projects that started last year. Among the 789 entries that began in 2020.5 or earlier — old enough to have been judged — the share is 11.8%.

Judged by this record rather than by argument, the recurring disadvantage is not cost or complexity but a mismatch between what gets bought and what gets decided. Cities buy instruments; the authority to act on what the instrument says is a separate thing, and it is rarely part of the purchase — 15 entries here measured something accurately and changed nothing. The next two patterns are structural in the same way: 28 systems worked at trial scale and stopped when the trial, the grant or the vendor did, and 24 were delivered without the piece that would have made them work. Privacy and surveillance are a real cost too, but in this record they show up less as a hidden downside than as an open fight, which is a different problem.

The two that end most projects here are the pilot and the contract. 28 entries were successful trials that never became services, usually because the funding was a grant with an end date rather than a budget line. Another 15 were settled by their money structure before the technology was switched on: a concession, a minimum-revenue guarantee, a vendor commission or a permit condition that decided the outcome regardless of how well the system ran. Neither challenge is technical, and neither is visible in a procurement document that only describes the equipment.

That question is too large to have one answer, and this atlas is built to replace it with a smaller one: did this specific project, in this specific city, do what it said it would? Of 1167 published entries, 112 are recorded as failed or scaled back — so roughly nine in ten are running or completed as described, which is neither the triumph the vendor brochures claim nor the fiasco the critics do. The useful reading is that the label matters far less than the mechanism. The same technology succeeds in one city and fails in the next depending on whether anyone is obliged to act on it, who owns the data it produces, and what happens when the grant ends.

Because some of these systems really were stopped by the public, and the reasons are on the record rather than a matter of speculation. 12 entries here were halted by residents, a court or a regulator instead of by the technology: Sidewalk Toronto was abandoned with unresolved data-governance questions and before any construction, Buenos Aires's facial recognition system was stopped by a court under the city's own security law, several of Hong Kong's roughly 400 smart lampposts were sawn down by protesters in 2019, and Melbourne's council ended its shared e-scooter trial after a peer-reviewed trauma study counted 256 hospital treatments in the first year. The common thread is not technophobia. It is that the surveillance and data questions were treated as a communications problem rather than a design constraint — and the cities that treated them as a design constraint, like Amsterdam with its public sensor and algorithm registers, did not face the same fight.

In this record it is not cost, procurement or technical difficulty. It is that the measurement works and nothing is obliged to happen next. Chicago's gunshot sensors, Los Angeles's crime forecasting and Miami-Dade's heat alerts all functioned as instruments; what none of them had was a body required to act on the reading, or an evaluation that could say what the response achieved.

Failed means the thing stopped, was cancelled, or demonstrably did not do what it was built to do. Scaled back means it is still running, or ran to its end, but delivered materially less than it promised — a smaller footprint, fewer functions, a narrower mandate. Most scaled-back entries are still live services. The distinction matters because the two are routinely blurred: a bike share that shrank from fourteen districts to five is not the same story as one whose operator went bankrupt.

No, and that is the most useful thing on this page. Megaproject collapses get the coverage, but the whole group of entries announced far larger than they were built — NEOM's The Line, Masdar City, Songdo, Nusantara and the rest — is 18 of the 112 here. The failures that actually recur are quieter: a pilot that ends with its grant, a system delivered without the half that would have made it work, a contract signed against a demand forecast that never arrived.

Because a catalogue that only contains successes is a marketing brochure, and because the entries that went wrong are the ones with something to teach a city that has not spent the money yet. Failed and scaled-back entries stay in the record permanently, with the same source-checking as everything else and the same links to the primary evidence.