Hardware Stops, Data Doesn't

Development money arrives with a ribbon, a press release and a photograph. What it leaves behind, five years later, is the interesting part. Across the atlas, forty-one verified entries name an external funder: a development bank, a bilateral agency, a foundation. Read them together and they sort into two piles that have almost nothing to do with which country, which donor or how much money. They divide on a plainer question: did the grant buy something that needs an operator, or something that needs nobody?

Two things a grant can buy

An asset needs an operator, a maintenance budget, spare parts and somebody whose job it is to care when it breaks. A dataset needs a licence and a copy. Donors are good at funding the first and structurally bad at funding the second half of it, because operating budgets are recurrent, domestic and unglamorous, while capital grants are finite, foreign and photographable.

This is not a new observation in development economics. What the atlas adds is a controlled comparison: the same donors, often the same cities, the same years, and two very different survival rates depending on what the money was spent on.

Cairo: a grant that bought bicycles

Cairo Bike is Egypt's first public bike share, owned by Cairo Governorate, paid for by the Swiss Drosos Foundation, delivered with UN-Habitat and technically supervised by ITDP. On paper it is exemplary: a philanthropic funder, a UN agency, and one of the most experienced transport NGOs in the world, all pointed at one city.

It collapsed twice in its first year. The first attempt in July 2022 lasted weeks: the governorate cancelled the maintenance contract that August after the contractor demanded more money, and the bicycles disappeared from the docks. It was inaugurated again on 20 October 2022, a few weeks before Egypt hosted COP27, with 250 GPS-tracked bicycles at around 25 solar-powered stations. It stalled a second time, the original consortium was dropped for a local operator, and the system reopened in September 2023 on a new app.

In June 2024 the newspaper Al-Masry Al-Youm watched the stations for seven days: the app refused bookings, the hotline printed on the signs was disconnected, and the bicycles were poorly maintained. Of roughly fifteen kilometres of protected lane planned, about two were built, behind 50-centimetre plastic barriers.

The grant did what grants do well. It bought bicycles, docks, an app and solar stations, all of which existed and worked on the day of the ribbon. What it could not buy was a maintenance contractor who would stay at the agreed price, or thirteen more kilometres of protected lane. Those are recurrent, political and domestic.

Nairobi and Dar es Salaam: grants that bought data

Now the other pile. In 2012 a Rockefeller Foundation grant paid for researchers from MIT, Columbia and the University of Nairobi to ride roughly 130 matatu routes with GPS-equipped phones and convert the traces into a modified GTFS feed. It became the first informal transit system added to Google Maps, in 2014. The grant ended long ago. The feed is still listed and verified, and it feeds several Nairobi routing apps.

Ramani Huria, Swahili for "open map", ran from 2015 to 2020 on World Bank and GFDRR money with UK development funding. It trained roughly 470 university students and hundreds of Red Cross volunteers to map flood-prone Dar es Salaam onto OpenStreetMap. By the time the programme closed it had covered about 49 of the city's 90 wards and around 450,000 buildings. The atlas entry records the outcome plainly: the data was absorbed by national statistics and disaster-management bodies and still guides response.

Neither project needed anyone to keep paying. A GTFS feed does not have a maintenance contractor who can walk away mid-contract. An OpenStreetMap import does not stop working when a foundation's programme cycle ends. That is not because the teams were smarter. It is because of what the money was spent on.

What was announced, and what was built

Cairo Bike protected cycle lane, ~2 of ~15 km press · seven-day field investigation 13 %

Algiers adaptive junctions, 22 of 500 contracted press · state news agency APS 4 %

Nairobi water dispensers, 4 built against a stated 12,000 peer-reviewed fieldwork + utility statement <0.1 %

One measure: how much of the announced build sources confirm was delivered. The three announcements are not equally binding, and that matters. Algiers was a signed contract for 500 intersections. Cairo's fifteen kilometres was a project plan. Nairobi's 12,000 was a utility's stated intention in 2018, not a tendered volume, and the four refers to the original Mathare pilot; a 2026 commentary puts the present total at 257, but that source returned an error to every attempt and the utility publishes no count of its own. Read the bars as orders of magnitude, not as a league table.

The Algiers row is in the chart deliberately, because it was not aid at all. A commercial joint venture with Spanish suppliers was contracted in 2017 to fit 500 intersections with adaptive signals. Only 22 of the 200 first-phase junctions were ever wired; fibre faults and a national-security refusal to switch on the cameras froze it, the Spanish partner left, and officials called it a failure by 2021. Whatever is going wrong here, it is not caused by the money being foreign or charitable.

The twist: open data survives because nobody owns it

Having praised the data pile, here is the honest damper, and it is sharper than the failures.

Map Kibera is the oldest project of this kind in the atlas. In November 2009 thirteen young residents were trained on GPS handsets and surveyed their own neighbourhood in about three weeks: water points, latrines, clinics, informal schools, pharmacies, refuse sites. The survey went into OpenStreetMap in January 2010. Seventeen years later it is still running, still on donor grants, and the data sits under the Open Database Licence, which places it above every African government catalogue this atlas has audited for reuse rights.

A map anyone may copy is a weaker instrument than a register a department is obliged to consult.

That is the finding, and it cuts against the easy conclusion. The project's own evaluation concedes it cannot tell who uses the maps. The recurring academic criticism is that the step from map to intervention is asserted rather than demonstrated. The very property that lets an open dataset outlive its funding, that nobody owns it and nobody has to maintain it, is also why nobody is accountable for acting on it.

Compare GMMA READY in Metro Manila, funded by Australian aid through UNDP from 2010. It mapped nine hazards across 24 local governments and handed each one an average of 87 maps in October 2013. The atlas entry is blunt about the shape of it: a handover, not a live service. The clearest proof anyone used the data is the metro-wide earthquake contingency plan built on it two years later. That is a real result, and it is also the only one anybody can point to.

This is not an Africa problem

Two cases make that unmistakable.

Dhaka's CASE project installed automated signals at 29 intersections between 2012 and 2023 with World Bank funding, for around 112 crore taka. The signals went largely unused because police kept directing traffic by hand. The hardware was delivered exactly as specified. The operating practice it assumed was never funded, trained or negotiated.

And then Miami-Dade County, in one of the wealthiest countries on earth. In 2021 an Adrienne Arsht-Rockefeller Foundation challenge grant seeded what the county and its funder both call the world's first Chief Heat Officer. The post produced a heat action plan, a formal heat season, and a National Weather Service pilot that lowered the county's heat-advisory threshold from a heat index of 108°F to 105°F. Then Florida barred counties from requiring heat protection for outdoor workers, and the county cut the officer post to help close a $400 million budget gap. The heat season still runs. Nobody holds the title.

Same lifecycle as Cairo Bike, different continent, different income bracket: external money creates a capability, the recurrent cost lands on a domestic budget, and the domestic budget has other problems.

Loans are not grants, and export credit is neither

A distinction the press routinely flattens, and which changes who carries the risk:

• Grants do not have to be repaid. Kampala's traffic control centre, opened in September 2025 with adaptive signals at 27 junctions, was Japanese grant aid.
• Concessional loans are debt on softer terms. Cairo's smart meter rollout runs on a Japanese ODA loan inside a package of about US$226 million. Karachi's Red Line BRT stacks four lenders, and its Lot 2 contractor was terminated in April 2026 with completion still uncertain.
• Export credit is not development finance at all. The Cairo Monorail sits behind a £1.7 billion UK Export Finance guarantee supporting a syndicated buyer credit. Its purpose is to sell the exporting country's goods. The recipient still owes the money.

There is also the case of donors returning to repair what donors built. Metro Manila's flood telemetry system was built with Japanese aid in 1993 and rehabilitated by JICA in 2014 to 2016. Two decades of instrumentation, one external rebuild.

The one design that anticipates its own exit

Since March 2026, Lagos State has held a parametric flood insurance policy. Instead of assessing damage after a flood, the contract pays when satellite-observed flood extent crosses an agreed threshold. Radar satellites supply the footprints, a risk modeller built the model, a Nigerian insurer fronts it, and global reinsurers carry it, with cover up to US$7.5 million per event across seven of the state's local government areas.

The interesting part is the financing shape. Germany's InsuResilience Solutions Fund paid 90 per cent of the first-year premium, with Lagos committed to raising its share in years two and three while the fund keeps co-financing. The subsidy is designed to taper. That is a different instrument from a capital grant: it buys a habit rather than an object, and habits can be handed over.

Three things are not public, and the atlas says so. The trigger thresholds are undisclosed, so no resident can check whether a flood should have paid. The premium is undisclosed. And as of late July 2026, in the middle of the rainy season, no payout had been reported. A premium somebody else pays is not yet a commitment. The tests are the first payout, and the first renewal Lagos funds itself.

The catch

• Survivorship runs both ways. Open datasets are easy to verify years later, which is partly why they look durable in a corpus built from public sources. A quietly maintained piece of hardware generates no news and may be undercounted here.
• Attribution is guesswork. Cairo Bike had a foundation, a UN agency and an expert NGO behind it and still failed. Nothing in this evidence isolates the funding model from the political and administrative context it landed in.
• "Succeeded" is a low bar for data projects. It usually means the dataset still exists and is still used by somebody, not that a measurable outcome improved. Map Kibera is the honest illustration.
• Most numbers here are self-reported. Donor consortia announce their own results; independent evaluations of these projects are rare. Where the atlas found peer-reviewed fieldwork, as in Nairobi's water dispensers, it contradicted the announcement.
• Aid is a minority of the corpus. Forty-one of 630 entries name an external funder. This is a pattern in a subset, not a law of urban technology.

The verdict

If you are designing a donor-funded urban technology project, the atlas offers one uncomfortable rule and one workable one.

The uncomfortable rule: the capital grant is the easy half, and it is the half everybody photographs. Whatever the money buys will need an operator, a maintenance line and a domestic budget owner, and if those are not named and funded in the same document, the ribbon is the high point. Cairo bought bicycles and could not buy a contractor who stayed. Dhaka bought signals and could not buy the police practice that would have used them. Miami bought a post and could not buy the state law that would have given it teeth.

The workable one: fund things that do not need you afterwards. A GPS survey converted into an open feed, a community mapping programme that lands in OpenStreetMap, a hazard model handed to the agencies that must plan around it. These survive because they have no running cost to abandon. The price of that durability is that nobody is obliged to act on them, which is a real cost and should be stated rather than glossed. The most promising instrument in the corpus, Lagos's tapering insurance premium, is interesting precisely because it tries to buy an obligation instead of an object, and it is too new to judge.

For the wider question of what a city keeps when somebody else owns the machinery, see our ownership essay; for the recurring shapes of failure across the whole atlas, why smart cities fail.