Firebird Secures U.S. License for 45,000 Blackwell GPUs in Kazakhstan

Kazakhstan has moved from the stage of theoretical plans to the deployment of accelerated computing infrastructure on the ground, after Firebird obtained an official export license from the United States to ship 45,000 Nvidia Grace Blackwell GPUs. The step was capped by the signing of an official investment agreement between Firebird and the Kazakh government at the Akorda presidential palace in the capital Astana, in the presence of co-founder and CEO Razmig Hofagimian, alongside a visit by Nvidia founder and CEO Jensen Huang and his meeting with President Kassym-Jomart Tokayev. The first phase, now under construction in partnership with Kazakhtelecom in the Data Center Valley, targets commercial launch in the third quarter of 2027.

The bet goes beyond the first batch toward an operational capacity of roughly half a gigawatt and the conversion of power into exported intelligence.The announced plans indicate that the second phase will raise the potential total capacity of Firebird’s infrastructure in Kazakhstan to about 500 megawatts, continuing the company’s expansions that began with its first artificial-intelligence factory in Armenia 180 days before the new project was shaped. In addition to the forthcoming hardware, Kazakhtelecom, one of the country’s largest telecom operators, has begun operating 400 H200 and 480 B300 Nvidia processors in Astana and Almaty to provide computing for enterprises and developers on a broad commercial scale.

Next year’s roadmap adds a notable hardware dimension to the provisioning race; the Kazakh telecom operator plans to deploy more than a thousand additional GPUs in the Data Center Valley, as well as over a thousand units of the next-generation Nvidia Vera Robin processors in the Ekibastuz region. This staged hardware flow links abundant power in industrial zones with the localisation of super-computing centres, a point Huang summarised in his public dialogue with Deputy Prime Minister and Minister of Artificial Intelligence and Digital Development Jaslan Madiyev at the AI World centre, saying that energy-exporting countries are now able to turn excess power into artificial intelligence and export it abroad.

At the level of the local ecosystem of startups and sovereign models, Firebird is establishing Firebird Kazakhstan Labs to provide funding, compute access and research partnerships for local talent, while Nvidia announced during the visit its direct investment in the Kazakh-based startup Hedgesfield AI, which is expanding into generative-media applications using cloud-based Vera Robin processors. The government side also outlined plans to build language models that master Kazakh and its cultural context, and to deploy physical artificial intelligence in mining, metallurgy, agriculture and logistics sectors.

The equation taking shape in Central Asia sends direct signals to decision-makers in the Arab region.For telecom and data-center sectors in the Gulf, Egypt and the Levant, this expansion shows that U.S. export licences for the latest super-computing chips are becoming available to emerging markets outside the traditional allies once strict regulatory requirements are met. This obliges cloud-infrastructure operators in the Gulf to revisit pricing strategies and capacity competition; the emergence of regional computing centres with capacities up to 500 megawatts equipped with Blackwell and Robin processors creates geographically close alternatives for hosting heavy training and inference models that do not require ultra-low latency.

At the level of engineering teams and tech startups from Riyadh and Dubai to Cairo and Amman, the Kazakh model reflects a shift in the role of national telecom operators toward becoming direct operators of sovereign compute layers, and Nvidia’s investment in Hedgesfield AI demonstrates that local startups can attract chip-giant capital when they build global applications atop advanced compute infrastructure. Leveraging these shifts requires regional teams to combine secure energy supplies and advanced hardware with the development of personnel capable of managing distributed workloads and creating models that understand local languages and industrial sectors.