The facility at Hrazdan marks the operational debut of Firebird’s international AI infrastructure platform. The company plans to expand the Armenia operation to 300 megawatts of computing capacity through a three-phase programme, potentially establishing the country as one of the larger AI computing centres serving Europe and neighbouring markets.
Built around NVIDIA’s DSX AI Factory reference design, the facility uses Spectrum-X Ethernet networking and Dell Technologies infrastructure. Schneider Electric has supplied power systems, while Vertiv is providing cooling technology designed for the high-density computing requirements of advanced AI processors.
Firebird says its infrastructure configuration can accommodate as much as 40% more GPU capacity within the same physical footprint. Such density is becoming increasingly important as operators confront rising electricity, cooling and property requirements from increasingly powerful AI hardware.
The first phase represents an investment of about $500 million. Earlier plans envisaged more than 6,000 NVIDIA Blackwell GPUs running at roughly 18 megawatts, but Firebird has substantially increased the scale of its planned deployment. The current roadmap calls for more than 70,000 Blackwell and Rubin GPUs and 300 megawatts of infrastructure by the end of next year.
The company broke ground on the site only about six months before operational readiness, underscoring the pace at which specialised AI data centres are being constructed as global demand for computing capacity increases.
NVIDIA also intends to invest in Firebird, adding another major technology partner to the project. The size and terms of the proposed investment have not been disclosed. AI cloud operator CoreWeave invested in Firebird earlier this year.
Perplexity is among the first customers of the Armenia facility. The AI company is expected to use Firebird’s computing infrastructure for its answer engine and AI-powered workplace products, giving the Hrazdan operation an international workload as it begins scaling capacity.
Firebird co-founder and chief executive Razmig Hovaghimian has positioned the project as part of a broader effort to bring advanced computing infrastructure into markets that historically lacked direct access to the largest AI clusters. The strategy centres on providing countries, research institutions and companies with locally accessible computing resources rather than relying entirely on data centres concentrated in established technology hubs.
Armenia has increasingly sought to use its engineering workforce and technology sector as the foundation for a larger role in artificial intelligence. Government planning around the Hrazdan development has included scientific research, robotics, aerospace, biomedical technologies, engineering simulations and enterprise AI among the workloads that could use the infrastructure.
The centre occupies a site of about 200,000 square metres. Its communications infrastructure includes a next-generation fibre network capable of bandwidth of up to one terabit per second. The cooling system uses a closed-loop water configuration intended to reduce water replacement requirements.
NVIDIA chief executive Jensen Huang said AI factories were becoming infrastructure that countries could use to develop technology domestically, linking access to computing power with economic competitiveness. The Armenia project illustrates a broader shift in AI investment towards so-called sovereign computing, where governments and domestic technology ecosystems seek control over strategically important processing capacity.
Firebird is also extending the model beyond Armenia. Kazakhstan has been identified as its second major market, where the company has secured 125 megawatts of planned AI infrastructure capacity at Data Center Valley. The project has received government approvals and the necessary US export authorisation for controlled advanced computing technology.
The company is targeting approximately two gigawatts of AI infrastructure capacity worldwide by the end of 2028, with Armenia and Kazakhstan forming the initial anchors of a network intended to extend into additional emerging markets.
The scale-up will coincide with NVIDIA’s transition from its Blackwell architecture towards Rubin systems. Blackwell has become central to the current generation of high-performance AI training and inference installations, while Rubin is designed as its successor for increasingly demanding model development and deployment.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
