NVIDIA said Kazakhtelecom has deployed 400 H200 and 480 B300 GPUs in Astana and Almaty, with more than 1,000 additional GPUs planned at each of two further sites next year. The figures appeared in an October 22, 2026 post by NVIDIA Vice President Rev Lebaredian, published as part of the GTC Berlin news sequence. The disclosure matters because it signals a shift from planning to operating sovereign AI capacity available to enterprises and developers at scale.
Kazakhtelecom GPU deployments and next-year sites
Kazakhtelecom, described as the largest telecommunications company in Kazakhstan, has placed the current 880 GPUs across Astana and Almaty. Next year the operator plans more than 1,000 GPUs in the Data Center Valley and more than 1,000 Vera Rubin GPUs in the Ekibastuz region. According to NVIDIA, that infrastructure is intended to serve diverse workloads for enterprises, developers and institutions. The post framed the buildout as covering infrastructure, models and applications, with more local capacity allowing teams to train, fine-tune and serve increasingly capable models.
The deployment rests on a cloud delivery model paired with NVIDIA software and safety platforms. The May 2025 sovereign AI factory was specified at 56.4 petaflops in FP16 and was to include full access to the NVIDIA AI Enterprise software suite. The robotaxi assessment relies on the Hyperion sensor and compute platform with the Halos safety system. Pony. ai links that stack to its Level 4 domain controller work, including single-chip and multi-chip configurations that can combine two DRIVE Thor chips for up to 4,000 FP4 TFLOPS through NVLink.
The October post follows a longer sequence that began with the May 29, 2025 launch of what Kazakhtelecom called Central Asia's first commercial sovereign AI factory, scheduled to go live in summer 2025. On June 15, 2026, Prime Minister Olzhas Bektenov met Lebaredian and Firebird co-founders Razmig Hovaghimian and Alexander Yesayan, after which AI and digital infrastructure agreements totaling 10 billion dollars were signed. Those included a Strategic Cooperation Agreement with the Ministry of Artificial Intelligence and Digital Development and a Binding Term Sheet between Kazakhtelecom and Firebird for the Data Center Valley project.
What the capacity and robotaxi test mean for business
For companies operating in Kazakhstan and Central Asia, the operating 880-GPU base plus cloud access changes where training and inference can run. Government institutions, private businesses, startups and universities were named as customers of the sovereign factory, which points to local procurement of model development, fine-tuning and hosted inference. Small teams gain rental access without building clusters, while large organizations can keep data and workloads inside national infrastructure and plan multi-site capacity across Astana, Almaty, Data Center Valley and Ekibastuz.
The limits center on timelines, power and unconfirmed commercial terms. The Data Center Valley is described as a flagship project in Ekibastuz with 300 MW available and phased expansion to 1 GW, while officials discuss 100,000 GPU chips including GB300 and Vera Rubin and at least 3 billion dollars in annual export revenue. The inDrive and Pony. ai work is framed as a feasibility assessment for seventh-generation robotaxis in Astana via the inDrive app, not a launched service. Buyers should ask about live availability, pricing, software entitlements, support terms and safety validation before committing roadmaps.
The marker to watch is a 2027 launch described by Firebird CEO Hovaghimian, which would place Kazakhstan among the top ten countries for AI infrastructure. If Data Center Valley and Ekibastuz additions arrive on schedule and the Astana robotaxi assessment converts into operations, the buildout will have moved from agreements to revenue-generating services. If those dates slip, the project remains an infrastructure plan rather than a production platform.
