Aleria Said Oct. 21 It Plans 60MW UAE Data Center, 14,000 Rubin GPUs

Aleria, an Abu Dhabi-based AI company, said in a press release dated Oct. 21 that it is developing a dedicated 60MW data center in the UAE built around about 14,000 NVIDIA Rubin GPUs.

The release lists three NVIDIA compute components: roughly 14,000 Rubin GPUs, 6,000 Groq 3 LPX accelerators and 8,000 Vera CPUs; the release does not say whether the LPX figure counts individual LPUs or LPX racks. NVIDIA Vera Rubin NVL72 systems are the foundation, though the release gives no count of systems. Storage combines NVIDIA CMX Context Memory Storage, integrated with the Dynamo software, and multiple petabytes of a DDN data layer that Aleria rates at more than 17 TB/s of aggregate read throughput. All figures are the company's own.

Chief executive Eric Leandri said in the release that the UAE has made its AI ambition clear and that the question now is how quickly capacity can be added. He described the model as already running real workloads in the country.

What 60MW means for the UAE grid

The release does not say whether 60MW is critical IT load, total campus load or utility interconnect capacity, and those are different numbers for a grid operator. It also names no city, no energisation date and no power source. For a site built on Rubin-generation racks, those three details decide when the capacity can actually run, so they are the first things to watch.

What 60MW Buys in Rubin Racks

NVIDIA's Vera Rubin NVL72 is a liquid-cooled rack that holds 72 Rubin GPUs and 36 Vera CPUs, according to NVIDIA. Its Oct. 14 datasheet lists rack power at 227 kW at maximum performance (MaxP) and 187 kW in its MaxQ mode. Groq 3 LPX is a separate rack, which NVIDIA describes as 256 Groq 3 LPUs working as one domain beside the GPU racks.

That is why a site's megawatt figure and its chip counts cannot be turned into each other without a rack count, which the release does not give. Cooling, storage and the LPX racks all draw power on top of the GPU racks, so the load a grid operator plans for depends on figures Aleria has not stated.

A smaller Abu Dhabi site and an earlier plan

A second Aleria release, also dated Oct. 21 with a Berlin dateline of Oct. 20, describes a separate 10MW modular AI data center in Abu Dhabi with Hyperfusion. That project is under construction, targets service entry in the first quarter of 2027, and, according to the Hyperfusion release, is designed for rack densities above 200kW with direct-to-chip liquid cooling. The 60MW release does not mention it.

Aleria had already announced a Rubin deployment in the UAE. In an earlier report, DatacenterDynamics said the company planned 28 racks of NVIDIA DGX Vera Rubin NVL72 systems there, alongside Grace Blackwell Ultra GPUs in the United States. The sources Voltgrid read do not explain how the 14,000-GPU plan relates to the 28-rack plan: whether it replaces it, absorbs it or comes on top.

For scale, NVIDIA's DSX POD reference design packs 504 Rubin GPUs into one container, and the Firebird licence for 45,000 NVIDIA GPUs in Kazakhstan shows how large sovereign GPU plans outside the US and China are becoming.

What is not confirmed yet

Aleria says the facility is being developed. The release gives no completion date, no customers and no financing, and it carries no independent verification of the GPU, accelerator or CPU counts. A search on Oct. 28 found no newer Aleria statement on the 60MW site. Voltgrid will update this report if Aleria names the site or its power supply.