In August 2026, the Firebird AI factory officially opened in the Armenian city of Hrazdan, which NVIDIA positions as the largest data center in the CIS. The project, implemented on the reference NVIDIA DSX AI Factory platform, has become a symbol of the rapid growth of regional artificial intelligence infrastructure. The construction of the first phase took only six months, made possible thanks to modular architecture and close cooperation with leading technology partners.
Technological breakthrough and infrastructure
The first phase of the DC-1 complex includes 6,144 NVIDIA B200 accelerators with a total capacity of 15 MW. To ensure uninterrupted operation, Schneider Electric provided the power infrastructure, including medium and low voltage switchgear, three-phase UPS systems, and rack solutions. Meanwhile, Vertiv developed a cooling system combining chilled water cooling technology, advanced management systems, and Vertiv TrimCooler technology for efficient heat dissipation. The Vertiv iCOM CWM chilled water management system centrally coordinates cooling resources based on workloads.
Scaling plans and energy challenges
By the end of 2027, Firebird plans to deploy more than 70,000 NVIDIA Rubin and Blackwell GPUs — totaling 300 MW of infrastructure capacity. The planned DC-2 cluster is designed for 75,000 NVIDIA VR200 GPUs with a total capacity of 125 MW. The company also plans to construct a third phase of similar scale. However, scaling from 15 to 300 MW will require solving the power supply issue, noted the TECHi resource. This will be one of the key challenges for the project, given the limited energy resources in the region.
Investments and international cooperation
In November 2025, Firebird received a US export license to supply advanced NVIDIA chips to Armenia. The cost of the first stage of the AI factory construction is estimated at $500 million. In February 2026, during a visit to Armenia by the US Vice President, it was announced that the project cost would increase to $4 billion, with the deployment of about 50,000 GPUs and the licensing of an additional 41,000 NVIDIA GB300 chips. One of the first clients of the data center in Armenia was Perplexity, which will gain access to Firebird's AI infrastructure to support its answer processing engine and AI agent platform.
Contradictory data
There are discrepancies in the estimates of the project cost and the volume of equipment being deployed. According to Firebird, the cost of the first stage is $500 million, however, in February 2026 during the visit of the US Vice President, a figure of $4 billion was announced. Furthermore, the number of planned GPUs varies from 50,000 to 70,000, creating uncertainty about the scale of the project. These contradictions may be related to different planning stages or different sources of financing.
Global ambitions and NVIDIA strategy
With the support of NVIDIA, the company is implementing a plan to develop AI infrastructure with a capacity of about 2 GW, covering Armenia, Kazakhstan, and other markets. Following a cyclical scheme, NVIDIA acquires a stake in the company, which then purchases chips from it on a large scale. NVIDIA has already invested more than $40 billion in the stocks of companies engaged in AI technologies. This confirms the strategic importance of the project for the global artificial intelligence market.
