Aleria Technology announced plans on 21 October for a 60-megawatt UAE data centre with approximately 14,000 Nvidia Rubin GPUs, following a March plan for 28 DGX Vera Rubin NVL72 racks in the country, Konsulteer reported. The Abu Dhabi-based company intends the facility to expand computing capacity for AI training and inference.
Key points
- The proposed UAE configuration includes 6,000 Nvidia Groq 3 LPX accelerators and 8,000 Nvidia Vera CPUs.
- Aleria specifies DDN storage with aggregate read throughput exceeding 17 terabytes per second.
- The company separately announced plans to deploy 3,000 Nvidia Blackwell Ultra GPUs in Switzerland.
Aleria sets out its 60 MW configuration
The proposed UAE system combines Rubin GPUs with approximately 6,000 Nvidia Groq 3 LPX accelerators and 8,000 Nvidia Vera CPUs. Rubin is intended for model training and inference, while the Groq accelerators are positioned for inference work that requires rapid responses and token generation, according to Konsulteer.
Aleria also specifies multiple petabytes of DDN storage, with aggregate read throughput of more than 17 terabytes per second. The configuration includes Nvidia CMX Context Memory Storage alongside Nvidia Dynamo, software used to serve and manage inference workloads. These are components of the proposed facility rather than measurements from an operating system.
Aleria says it plans to connect the computing and storage equipment to its own software for workload orchestration, data management and enterprise applications. The company says the additional capacity will serve large-scale training, inference and AI-agent workloads, including deployments in private environments where an organisation controls its infrastructure and data.
Aleria describes its existing UAE operations as already running AI workloads. The new 60 MW figure refers to a planned data centre, not capacity reported as operating at that site.
The announcement supplied no expected utilisation rate, complete-system performance benchmark or inference-latency measurement, Konsulteer reported. Aleria’s specified storage throughput is a component target. It is distinct from a measured result for customers running training or inference across the proposed configuration.
March rack plan preceded the UAE facility
In March 2026, Aleria disclosed plans to deploy 28 racks of Nvidia DGX Vera Rubin NVL72 systems in the UAE. That earlier disclosure concerned racks of systems. The October proposal describes a data centre by its planned power capacity and the number of individual processors in its configuration.
Aleria also has a separate October 2026 collaboration with Hyperfusion on a 10 MW modular AI data centre in Abu Dhabi. The project aims to begin operations in the first quarter of 2027. The 60 MW proposal is a separate announcement.
The announcements do not establish whether the proposed 60 MW facility and the earlier UAE rack deployment share a site, a delivery schedule or customer commitments. The public disclosures do not establish whether hardware for the 60 MW project has been ordered, delivered or installed.
Aleria plans 3,000 GPUs for Switzerland
In a separate announcement on 21 October, Aleria said it would deploy 3,000 Nvidia Blackwell Ultra GPUs in Switzerland, supported by DDN storage. The company described the deployment as an expansion of its European computing infrastructure. It is an Nvidia Cloud Partner that supplies AI infrastructure and software, according to its announcement.
Aleria said the Swiss capacity would be offered through GPU as a Service, allowing customers to access computing hardware without purchasing and operating dedicated GPUs. The company identified training, fine-tuning and inference as workloads for the deployment. That customer-access model was stated for Switzerland in the company’s announcement.
Eric Leandri, Aleria’s chief executive, said the Swiss deployment would bring computing and storage resources together for organisations running more demanding AI workloads. Aleria said DDN storage would support the data-access requirements of those workloads, including training on large datasets.
