Firebird Launches the CIS Region's Largest AI Factory
U.S. cloud provider Firebird has opened the largest AI data center in the CIS region in Armenia, marking a major expansion of high-performance computing into frontier markets.
On August 8, 2026, Firebird officially opened its flagship DC-1 facility in Hrazdan, Armenia, establishing the largest AI factory in the CIS region. The opening ceremony was attended by Armenian Prime Minister Nikol Pashinyan, Kazakhstan’s Deputy Prime Minister Zhaslan Madiyev, and U.S. chargé d’affaires David Allen. Built on NVIDIA's reference architecture with Dell PowerEdge servers, the liquid-cooled DC-1 site currently operates 6,144 NVIDIA B200 GPUs across 15 megawatts of capacity. This marks the first operational phase of a massive regional buildout, which began with a strategic collaboration in June 2025 and secured a crucial U.S. export license in November 2025.
The project is scaling rapidly under a $4 billion Phase 2 expansion announced in February 2026 by U.S. Vice President JD Vance, which built upon a $500 million Phase One. Backed by an export license for an additional 41,000 GB300 units, Firebird plans to deploy more than 70,000 NVIDIA Rubin and Blackwell GPUs and 300 megawatts of capacity in Armenia by the end of 2027. This includes a planned DC-2 facility featuring 75,000 VR200 GPUs across 125 megawatts, alongside a third site of equal scale. Ultimately, Firebird aims to establish a 2-gigawatt infrastructure footprint across Armenia, Kazakhstan, and other frontier markets, supported by CoreWeave and an upcoming investment from NVIDIA.
For AI practitioners, this facility offers high-performance bare-metal and multi-tenant cloud capacity interconnected via InfiniBand east-west and Ethernet north-south networks, with a BlueField-3 DPU on every node. Early customers like Perplexity are already leveraging this infrastructure to power answer engines and AI agents. The data center relies on the NVIDIA DSX platform, which integrates compute, networking, power, and cooling. By utilizing MaxLPS software to dynamically manage power and Vertiv's iCOM CWM Chilled Water Manager, the system recovers stranded power to run up to 40% more GPUs on the same footprint. This coordination directly improves tokens-per-dollar efficiency, allowing developers to train and deploy massive models with lower latency and optimized operational costs.
This is our own summary of reporting by Unite.AI
