What the KEZAD Floor Buys Noon

By the time the Dubai Roads and Transport Authority cleared Yango's autonomous delivery robots to operate on public walkways in Sobha Hartland, the machines had already logged more than 1,500 kilometres of live pilot mileage under Yango's AI-based navigation stack, per the partnership's own launch statement.

That is an operational detail, not a keynote-worthy one. Regulators do not sign off on a sidewalk-going robot because the pitch deck is convincing. They sign off because someone on the RTA side has walked through the failure modes — a pedestrian stepping in front of the machine, a curb cut in poor repair, a stalled robot blocking a fire lane — and the operator has produced logs the regulator finds credible. That is a form of qualification the Gulf's retail-tech announcements rarely showcase, and it matters, because the sidewalk pilot is not the story. It is the visible tip of a much larger industrial build-out that Middle Eastern retail has been quietly running through for two years. The question I keep circling is whether the payback shows up on the floor rather than in the press release.

What is being installed

Look at what has been poured into the ground over the last twenty-four months. In Abu Dhabi's Khalifa Economic Zone, Noon's 252,000-square-metre fulfilment centre has been built out as a build-to-suit facility, with automation systems for storage, material movement and sortation baked into the design rather than retrofitted on. In Jeddah, Agility opened a SAR 611 million logistics park in November 2025, 338,000 square metres of Grade-A warehouse space with smart conveyors and automated sortation on the floor from day one. Amazon's DXB5 in Dubai South added 2.1 million cubic feet of storage, a seventy percent bump to Amazon's UAE footprint, and sits under a wider AWS + HUMAIN infrastructure deal publicly costed at $5.3 billion that will stand up a new AI zone in-country by end of 2026.

If you strip the marketing language, what is being built here is a light-industrial site pretending to be a warehouse. The sortation lines look more like a semiconductor packaging fab than a Cash & Carry back room. The robots resemble automotive assembly kit, not retail. The maintenance regime is closer to a paper mill's than to a supermarket depot's, with condition-based monitoring on the drives, vibration sensors on the belts, and a controls layer that expects an OT engineer at the ready rather than a warehouse supervisor. And the labour attached to it, with Noon alone committing to 6,000 new hires at the KEZAD site, is not warehouse labour in any traditional sense. It is a mix of operations, controls, systems integration and technician work. That is the piece the ribbon-cutting speeches skate past. E-commerce fulfilment in the Gulf has become a manufacturing problem shaped like a retail one.

The productivity paradox that will not go away

Here is the part I do not want to soft-pedal. A 2026 study in Sustainability on digital transformation, employment and productivity in the GCC, analysing panel data across the six member states, reports that digitalisation has been negatively related to labour productivity in the region, even where employment moved the other way. That is not a comforting finding for anyone who has just poured half a billion dirhams into automation. It is also not an outlier. McKinsey's own warehouse automation retrospective surfaces a consumer-goods case where over $150 million disappeared into a fully automated consolidated facility that never produced the volumes the plan required, and identifies inadequate pre-project assessment as the single most reliable predictor of a failed deployment.

Two possibilities present themselves. Either the GCC's automation programme has genuinely been underperforming the capex, which the sector's operators will vigorously dispute, and which would carry serious implications for the next round of build. Or the productivity measurement itself is missing what is happening. The MDPI panel largely covers a phase when regional automation was still shallow: barcode systems, WMS deployments, some conveyor upgrades. What is being installed now at KEZAD, at Amazon.ae's DXB fleet, at the Jeddah Agility park is different in kind. Closed-loop robotic sortation, AI-orchestrated pick paths, computer vision on the return line, digital twins that simulate the day's expected orders against the current floor state before the shift even starts. Whether that changes the productivity signal is a matter of evidence, not confidence. I am reserving judgment for two more years.

Amazon and Noon are not doing the same thing

Grouping the Amazon investment and the Noon build-out under one Gulf-retail-AI headline flatters both and describes neither. Amazon's next-generation US fulfilment centres now integrate ten times the previous robotics density and are being coordinated by an internal generative model called DeepFleet, per Amazon's own newsroom. Treat that ten-times figure as vendor claim rather than independent evidence — Amazon has an interest in the headline, and no third party has confirmed the throughput yet. In the GCC, Amazon's play is different in shape anyway. It sits behind AWS infrastructure, cloud storefronts, marketplace SME onboarding. The conveyor system at DXB5 does not appear to run the DeepFleet-class stack on the public record.

Noon's KEZAD build is more industrially self-directed. The automation is on the physical floor, integrated by Noon's operations team and their construction partners at KEZAD Group, tied to Noon's catalogue and demand pattern rather than a marketplace of external sellers. They are different problem shapes. Amazon is building the AI substrate under a retail vertical. Noon is building the sortation and robotics inside its own four walls. If the productivity number moves, it will move for different reasons in each case, and the same policy lens should not be pointed at both. That is worth insisting on, because Gulf industrial policy is being written right now against a composite picture that flattens the two.

The workforce arithmetic is the second problem

The GCC's retail-industrial pivot has a labour-policy tension that Brussels-style analysis mostly misses. A 2025 Nature Humanities and Social Sciences Communications study of the GCC AI workforce audited six national strategies and 47 major initiatives and found the region investing hard in AI-native skills, with Saudi Arabia declaring 2026 its Year of Artificial Intelligence and standing up the SDAIA Academy as a dedicated upskilling arm. So far, so good.

The catch is on the receiving side. When Noon commits to 6,000 KEZAD hires while installing sortation that materially reduces the picker headcount per throughput unit, the difference has to go somewhere. In a Western market it would show up as displacement, redeployment, or an unemployment print. In the Gulf, under a labour-import regime, it shows up as fewer visas issued, or issued differently. That is a policy adjustment, not a payroll one, and it moves through channels a productivity model does not see. The Majid Al Futtaim / Advertima computer-vision rollout across Carrefour UAE is a smaller and different example of the same tension. Cameras identifying shopper archetypes for in-store retail media do not remove cashiers. Over a longer horizon they slowly move the mix, from customer-facing headcount to systems and analytics headcount, from a labour-visa question to a data-processing consent question. That the UAE has a self-assessment window under its emerging AI regulatory framework arriving in September 2026 will matter for how these deployments are documented. Whether it changes what is deployed is a different question, and I am not confident it will.

What I do not know

I do not know whether the KEZAD floor is now running at design throughput. I do not know whether Noon's sortation MFU-equivalent, the fraction of designed sort capacity actually used across a full week, is meaningfully above what a conventional depot achieves. I do not know how many autonomous last-mile trips it takes for the Yango robots to pay back the fleet capital, or how much of the Sobha Hartland pilot's cost sits on the operator's balance sheet versus Noon's. I do not know the mean time between manual interventions on the KEZAD sortation line, and I would not trust the number the operator reports for the first eighteen months anyway — that is a figure that only converges after the maintenance regime has been through a full summer. Those are the numbers a serious answer to the productivity paradox would need, and they are not in the press releases. If someone on either operator's side wants to publish them, even a redacted, week-averaged version, the conversation about Gulf retail industrialisation would get honest overnight.

Until they do, the ribbon-cutting is not the evidence. And the sidewalk robot in Sobha Hartland, the one the RTA signed off on last winter, is not the evidence either. What is the number, in kilos moved per labour-hour or dirhams of throughput per square metre of installed automation, that would show the productivity paradox reversing at these sites?

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.