Al-Farabium AI Computing Platform with 400 NVIDIA H200 GPUs

Create locally hosted computing capacity for AI model training and inference in Kazakhstan. Give businesses, public-sector organisations, researchers and developers access to GPU resources without building their own clusters. Support domestic data storage and processing, reduce reliance on foreign cloud services and provide a platform for internal automation and new digital services.

2025

The system achieved 17.93 PFLOP/s in the HPL Linpack benchmark (Rmax), with a theoretical peak of 26.84 PFLOP/s (Rpeak). It ranked 103rd in the November 2025 TOP500 list and 122nd in June 2026.

Operational results include the industrial deployment of the procurement chatbot in 2025 and dozens of hosted projects reported in May 2026, including contact-centre assistance. The platform has also been adopted for power-system modelling, extraction-process optimisation and AI service development.

50 servers with 400 NVIDIA H200 GPUs; 6 PB storage; upgraded data centre engineering systems; expanded network capacity; resource access management and billing.

TOP500 configuration: Supermicro SYS-821GE-TNHR servers, Intel Xeon Platinum 8558 48-core processors at 2.1 GHz, H200 SXM5 141 GB accelerators and ConnectX-7 400G InfiniBand NDR. The published benchmark software stack is Ubuntu 22.04.4 LTS, HPL-NVIDIA 23.0.0, cuBLAS and HPC-X.

The project does not use solutions from the Global CIO catalog

The project combined data centre engineering upgrades and network expansion with deployment of computing, storage, access management and billing. Coordinating these interdependent components formed the implementation scope for an operational AI service.

Al-Farabium is Kazakhtelecom's domestic AI computing service. It was publicly launched on 3 October 2025 and entered industrial operation in 2025. Customers can rent virtual or bare-metal GPU resources rather than purchase and operate their own cluster.

The platform is already supporting applied work. A procurement-portal chatbot for Samruk-Kazyna entered industrial operation on the supercomputer in 2025, using the Alem-LLM language model in a closed environment. The fund also reports that SKAI is deployed on Al-Farabium and identifies use by KEGOC for power-system modelling, Kazatomprom for extraction-process optimisation, Kazakhtelecom for AI Mektep development, and domestic technology companies.

An official report dated 5 May 2026 describes dozens of projects deployed on Al-Farabium, including a contact-centre AI assistant that processes customer enquiries automatically. These applications establish operational use of the infrastructure; their individual financial effects are not claimed as measured results of the cluster.

Almaty, Kazakhstan: the H200 cluster is deployed in a modular data centre. The project serves users in Kazakhstan.