Urban technology and AI
The urban century meets the intelligence age
Two structural forces are converging. First, urbanization; 55 per cent of the world’s population lives in cities today, rising to a projected 68 per cent by 2050, a further 2.5 billion urban dwellers, with close to 90 per cent of that growth in Asia and Africa (UN DESA, 2018). Cities already generate more than 80 per cent of global GDP and accounted for 88 per cent of private-sector job creation between 2010 and 2020 (World Bank, 2026). Second, artificial intelligence; the McKinsey Global Institute estimates that AI and analytics could unlock US$11-17.7 trillion in annual economic value, with generative AI adding a further US$2.6-4.4 trillion each year, roughly the GDP of the United Kingdom, every year (McKinsey Global Institute, 2023).
Nowhere do these forces intersect more deliberately than in the Gulf. The UAE’s National AI Strategy 2031 targets an AED 335 billion (US$91 billion) AI contribution to GDP (UAE Government, n.d.); regional technology spending is forecast to reach US$169 billion in 2026 (Gartner, n.d.); and announced AI data-center investment across the GCC exceeds US$30 billion through 2030, with US$5-7 billion deploying in 2026 alone (Analysis Mason, 2025). Stargate UAE, a planned 5-gigawatt AI campus in Abu Dhabi, and Saudi Arabia’s PIF-owned HUMAIN, targeting 1.9 GW of AI factory capacity by 2030 and 6.6 GW by 2034 (Middle East Institute, 2026), are the compute substrate of something categorically different (the cognitive city).
From smart to cognitive, a step change, not a rebrand
A smart city senses and reports; a cognitive city learns, anticipates, and acts. Three shifts separate them:
- From dashboards to decisions, data stops being something officials look at and becomes something the city acts on: systems that reroute traffic, rebalance energy loads, and dispatch services autonomously, with humans governing by exception.
- From vertical silos to one operating model, mobility, utilities, safety, and health run on a shared data and AI layer rather than parallel stacks.
- From procurement to orchestration, no hyperscaler, telco, or integrator can deliver a cognitive city alone; value accrues to whoever orchestrates the ecosystem.
The prize is measurable, and the capture rate is the real contest
The McKinsey Global Institute, analyzing deployed applications across 50 cities, found smart technologies can improve key quality-of-life indicators by 10 to 30 per cent: commutes cut by 15-20 per cent, emergency response times by 20-35 per cent, crime incidents by 30-40 per cent, greenhouse-gas emissions by 10-15 per cent (McKinsey Global Institute, 2018). Yet even the most advanced cities have captured only about two-thirds of that potential (Remes, 2018), and most programmes stall in pilot purgatory before reaching city scale. The contest is the capture rate, not the technology. The capital is committed regardless (a US$1.96 trillion market in 2026, projected at US$4.06 trillion by 2031 (Mordor Intelligence, 2026)), with the AI-in-smart-cities segment growing at roughly 27 per cent a year towards US$345 billion by 2033 (Market.us, 2024).
The twist. The cognitive city’s most important customers do not live in it
The real customers of the cognitive city are not its local residents, but other countries facing the challenge of rapidly digitalizing their cities.
Close to 90 per cent of the world’s urban growth to 2050 will occur in Asia and Africa (UN DESA, 2018), in cities that will never fund gigawatt compute campuses of their own, and Africa is already the fastest-growing smart-cities region, at an 18 per cent CAGR through 2031 (Mordor Intelligence, 2026). The Gulf’s cognitive cities are therefore best understood not only as domestic infrastructure but as reference implementations of an exportable product, a packaged operating model of compute, platforms, and governance that rapidly urbanizing states can license, subscribe to, or franchise. The precedent already exists. Estonia, a nation of 1.3 million, turned its X-Road data-exchange layer into a de facto export, now adopted or adapted in more than 20 countries, from Finland and Japan to Brazil and Cambodia (e-Estonia & Nordic Institute for Interoperability Solutions, n.d.; GovInsider, n.d.). The Gulf can run the same play with capital Estonia never had. The trajectory is from exporting hydrocarbons to exporting urban intelligence, on one condition. The export must transfer capability, not dependency.
Five tests. The buyer’s due-diligence, run on yourself first
If the domestic contest is the capture rate and the international prize is the franchise, one set of questions decides both, because they are precisely the due-diligence questions a delegation from Jakarta, Nairobi, or Lagos will ask before licensing anyone’s cognitive-city stack. A program that fails them captures little at home and sells nothing abroad.
1. Who owns the data? This is where the capture-rate gap begins; programs stall on data readiness and unclear ownership long before they stall on models. A named executive with a budget, or a committee? If it is a committee, the program has already failed. Jurisdictions that codified accountability early, as Saudi Arabia did through its NDMO and PDPL frameworks, converted regulation from friction into a trust asset.
2. Where will the power come from? The gigawatts announced in Abu Dhabi and Riyadh make the constraint concrete; grid capacity, cooling, and clean-power procurement are now the first strategic decision, not a utility afterthought. A city that has not contracted its energy has a rendering, not a roadmap.
3. Can you fire your vendor? If switching costs make the honest answer no, the city has bought dependency, not sovereignty. Open interfaces and contractual exit ramps are worth more than any single product feature, for the Gulf’s own build, and doubly for the export thesis; a stack sold as a lock-in merely relocates the sovereignty problem to the buyer.
4. Would it survive a privacy backlash? Sidewalk Labs’ Quayside project in Toronto, backed by Alphabet’s balance sheet and world-class technology, was cancelled in May 2020 after two years of public resistance over data collection and opaque governance; the pandemic was the stated reason, trust was the operative one (Sidewalk Labs, 2020). Cognitive systems compound the risk; models trained on their own outputs degrade. And the franchise stakes are higher still; no government licenses a stack whose flagship implementation lost its own citizens.
5. Which pilot did you kill last quarter? If the answer is none, the city is funding activity, not outcomes; pilot purgatory is the capture-rate gap in daily practice. Every deployment needs a published KPI baseline (minutes of commute saved, response times cut, emissions avoided) and a sunset clause for whatever does not move the number. Those published baselines then become the export brochure; the reference implementation’s evidence base is the product.
Three predictions to test these theses against
Falsifiable beats fashionable.
First, by 2030, at least one GCC state will sign a government-to-government license for its cognitive-city stack with an Asian or African capital, following the Estonian playbook at sovereign-wealth scale. The deal will go to the state that can answer the five questions above — in writing.
Second, the first high-profile cognitive-city setback of this cycle will be caused by power procurement, not model performance.
Third, Toronto will not be the last. At least one flagship cognitive-city program will be cancelled because of backlash over algorithmic governance rather than economics.
For everyone else, the choice remains the same: orchestrate the cognitive city — or have it orchestrated for you.
Discover where we provide tailored solutions to enhance value for our clients.
- Data Centers
- Data Management, Applied AI & GenAI
- IoT & Smart-X
