I am following AI investment in Saudi Arabia and the UAE to see how announced commitments turn into computing capacity that companies and researchers can use. The combination of state-backed capital, energy and partnerships with major U.S. technology companies is substantial. My question is how far individual projects have progressed beyond those agreements.
For this note, I used Madar Strategy's audit of seven major Saudi and UAE projects or programs announced or materially updated between 2024 and July 22, 2026. Their headline capacities add up to about 4,000 megawatts. In the public record the audit assembled, none of the seven had a dated document establishing that power had been turned on or that the facility was operating.
I read that as a reason to look for more specific evidence of delivery, not as proof that the projects were inactive. It also says nothing about whether either country already had other working data centers. To see what the announcements do establish, I have kept national targets, company plans and individual facility records separate.
National targets and company plans
The first difficulty in comparing the announcements is that their megawatt figures describe different things. Saudi Arabia's national data-center strategy, published in June 2025 by the Ministry of Communications and Information Technology and SDAIA, targets about 1.5 gigawatts by 2030, and the U.S. International Trade Administration repeats that figure. It works as a statement of national ambition. It does not tell me the status of any particular AI campus.
Nor can I use the ministry's operating-capacity series to fill that gap. As reported by vision2030.ai, the series includes colocation, cloud regions, and government and research halls as well. It also leaves unclear whether capacity means the power used by computing equipment or by the whole facility, including its supporting systems.
HUMAIN's plans complicate the comparison further. Its interim target of 1.9 gigawatts by 2030 is about 27% larger than the Saudi national target, and I would need to reconcile the two definitions before treating HUMAIN's figure as evidence of how much capacity Saudi Arabia is building. Its AMD partnership spans Saudi Arabia and the United States without a country split, and the announcements do not establish how its AMD and NVIDIA projects relate to one another. So I have kept those amounts separate rather than adding them into a Saudi total.
For the UAE, I have not found a single national figure equivalent to the Saudi strategy. What I found instead is a campus announcement: the Abu Dhabi AI Campus proposes up to five gigawatts, with a first phase of one gigawatt. For my comparison, keeping that gap visible is more useful than substituting the campus plan.
The seven projects
For evidence of delivery, I rely on the individual project records in Madar's audit, whose sample includes projects reporting at least 50 megawatts. The table pairs each headline capacity with the highest status the audit could support, so that planned size does not read as completion.
| Project or program | Headline capacity | Status supported by the audit | Public milestone or target |
|---|---|---|---|
| Stargate UAE | 1,000 MW | Announced | May 2025 announcement; first 200 MW expected in 2026 |
| DataVolt Oxagon | 1,500 MW | Site identified | February 10, 2025 NEOM agreement; 2028 target |
| HUMAIN-NVIDIA | 500 MW | Announced | May 13, 2025; 18,000 GPUs in the first phase; 2030 target |
| HUMAIN-AMD | 500 MW | Announced | May 13, 2025; Saudi-U.S. network; 2030 target |
| Microsoft-G42 UAE expansion | 200 MW | Announced | November 2025 announcement through Khazna; target before the end of 2026 |
| Khazna Dammam | 200 MW | Site identified | December 9, 2025; 225,000-square-meter site; no public completion date |
| Khazna QAJ01, Ajman | 100 MW | Under construction | Steel completed; design certification dated February 24, 2026; December 2026 target |
Even the roughly 4,000-megawatt sum needs care. It combines computing-equipment load with campus and AI-factory figures, some of whose definitions remain unclear. With those qualifications, it records the scale of the announcements. I would not use it as an inventory of comparable usable power.
QAJ01 is the project I can follow furthest in this sample. It has a named facility, a stated 100 megawatts of IT load and completed steelwork. It is also the only project with a dated construction start and a named utility, Etihad Water and Electricity. Those records support a specific account of construction progress, but they still do not establish energization.
Madar's separate review of 13 power arrangements hits a similar limit. The only project-specific evidence it found was QAJ01's named utility and grid-asset procurement for Naqaa in Riyadh. That gives me a possible supplier and a procurement step, but neither record establishes a firm connection or that power was turned on.
Financing and the next delivery records
Alongside construction, I want to know which financial commitments have become binding. The public record I used includes Microsoft's $1.5 billion equity investment in G42, dated April 16, 2024, and the MIS-Naqaa expansion contract, signed September 7, 2025, for SAR 227.8 million including VAT.
I treat those as financial events in their own right. Neither lets me mark a particular facility as supplying computing services. The investment and the contract show that something has been committed; a delivery record would have to show what that commitment produced.
For QAJ01, I know more precisely what to look for next. Its construction progress already has public support. What is missing is a dated account of energization, commissioning or entry into operation. Following those milestones would carry the project beyond the announcement and toward the usable capacity I am trying to understand.
This article reports the source record through July 22, 2026. The roughly 4,000-megawatt total and project classifications come from Madar Strategy's audit, not an independent recount of all Gulf capacity.