Why the UAE Is Pouring Billions Into Desert Data Centers — and the Cooling Problem Nobody's Fully Solved

Why the UAE Is Pouring Billions Into Desert Data Centers — and the Cooling Problem Nobody's Fully Solved

The UAE wants to be one of the physical homes of the AI boom, not just a buyer of it. Over the past eighteen months, Abu Dhabi and Dubai have signed up for gigawatts of new AI data-centre capacity, backed by some of the most recognizable names in the industry. But building that capacity in one of the hottest, driest climates on Earth creates an engineering problem the rest of the industry doesn't have to solve quite as urgently: how do you cool a few hundred thousand GPUs when you can't count on cheap, abundant water or a mild climate to help you do it?

The Headline Project: Stargate UAE

The centerpiece is Stargate UAE, announced in May 2025 as a 1-gigawatt AI data-centre cluster built inside a much larger 5-gigawatt "UAE-US AI Campus" planned for Abu Dhabi. The project groups together G42 (the Emirati AI firm building it), OpenAI and Oracle (who are expected to operate the compute), and Nvidia, SoftBank and Cisco as partners, according to Khaleej Times' coverage of the announcement. Reported estimates put the investment at $8–10 billion, with the first 200 megawatts of capacity originally targeted to come online in 2026 and the full 1-gigawatt buildout expected to take around three years.

As of the most recent public tracking (an Epoch AI directory update in July 2026, based on satellite imagery and company disclosures), the site was still listed as under construction, with no IT power confirmed online yet — a reminder that "coming online in 2026" and "online" aren't the same milestone, and that these timelines tend to slip in both directions as hardware deliveries and grid connections catch up.

Stargate UAE isn't the only big number on the table. Microsoft announced in November 2025 that it would expand its AI and cloud infrastructure in the UAE through a partnership with Khazna Data Centers (itself a G42 unit), with reported investment of $7.9 billion through 2026–2029. Separately, telecom operator du has said it will build an AI data centre with Microsoft, backed by more than $500 million. Dubai-based Emirates NBD Research projected in an October 2025 report that the country's overall data-centre capacity would surge roughly 165% to around 950 megawatts by 2028. Whatever the exact final figures turn out to be, the direction is unambiguous: the UAE is trying to become a regional compute hub, not just a consumer of AI services built elsewhere.

Why the Desert Makes This Harder

A gigawatt of GPUs generates an enormous amount of waste heat, and the traditional way data centres get rid of it — evaporative cooling, which relies on large volumes of water — is a much harder sell in a country with essentially no natural fresh water and heavy reliance on energy-intensive desalination. Mohammed Soliman of the Middle East Institute has argued publicly that Gulf operators can't lean on the "cheap, abundant cold water" model that cools much of the data-centre capacity in cooler, wetter regions, and will instead need to lean harder on closed-loop and immersion cooling systems that reuse the same fluid rather than evaporating fresh water into the air. That approach saves water, but it's costlier to build and, as Soliman has noted, still imperfect.

You can already see operators hedging toward liquid in practice rather than air. HPE and Khazna launched what they billed as the UAE's first managed AI hosting service built around direct liquid cooling in 2024, with HPE claiming roughly 20.7% better performance per kilowatt than an air-cooled equivalent. Separately, du and AI Hosting Hub's Nvidia-based supercluster at Dubai Silicon Oasis uses direct-to-chip liquid cooling across a dedicated data hall with more than 5 megawatts of capacity. Khazna has also talked up "adiabatic-free" chillers it says cut cooling energy use by 70% compared with conventional systems — a meaningful claim given that cooling alone typically eats up around 40% of a data centre's total power draw.

There's also a quieter, AI-on-AI angle: the UAE's Ministry of Energy and Infrastructure, together with Khazna, Agility and the cooling-optimization company Phaidra, has been running a pilot since early 2026 that uses machine learning to trim energy use in data centres and connected district-cooling systems, with plans to scale it if the results hold up. It's a smaller story than the gigawatt announcements, but arguably more telling: a lot of the actual savings in Gulf data centres right now are coming from software squeezing efficiency out of existing hardware, not just newer chillers.

The Bigger Trade-Off

None of this is free, and the UAE's own officials and analysts have been fairly open about the tension. Soliman and others have floated pairing new AI campuses with the country's existing desalination plants, so the waste heat from thousands of GPUs could theoretically help offset the energy desalination already burns through — an idea that's more conceptually appealing than it is currently deployed at scale.

The honest summary is that the UAE's AI infrastructure push is real, well-funded, and backed by serious partners — but it's also still mid-construction, both literally (Stargate UAE's buildings) and figuratively (the cooling and water strategy that will actually make gigawatt-scale AI campuses sustainable in a desert climate). The headline numbers — gigawatts, billions of dollars, "online in 2026" — are the easy part to announce. Getting thousands of liquid-cooling loops, grid connections and chiller systems to actually hold up under a Gulf summer, at a price and water footprint the country is willing to live with long-term, is the part that will take the next few years to prove out.

What to Watch

Two things will tell you whether this is working: whether Stargate UAE's first 200-megawatt phase actually shows measurable IT load (not just finished buildings) in the independent trackers that monitor these sites, and whether operators like Khazna start publishing real operational water-use figures rather than efficiency percentages. Until both of those show up, the UAE's AI data-centre boom is still mostly a very well-funded construction project.