Firebird, a U.S.-based AI cloud firm, opened the CIS area’s largest AI manufacturing facility in Armenia on August 8, 2026: a facility in Hrazdan constructed on NVIDIA accelerated computing and Dell PowerEdge servers, with Armenian Prime Minister Nikol Pashinyan, Kazakhstan’s Deputy Prime Minister Zhaslan Madiyev, and U.S. chargé d’affaires David Allen attending the opening ceremony.
The buildout plan is giant by any regional commonplace: Firebird plans to deploy greater than 70,000 NVIDIA Rubin and Blackwell GPUs and 300 megawatts of AI infrastructure capability in Armenia by the tip of 2027. The corporate says the positioning went from plan to working capability in simply over six months — a quick supply for a challenge of this class, and one which leaned on a provide chain of established distributors. Schneider Electrical delivered the ability infrastructure, together with medium- and low-voltage switchgear, three-phase uninterruptible energy provides, and rack enclosures, whereas Vertiv provided a chilled-water cooling structure managed by means of its iCOM CWM Chilled Water Supervisor, which coordinates cooling sources as AI workloads shift.
The power is constructed on the NVIDIA DSX platform, which codesigns compute, networking, energy, and cooling as one system. NVIDIA says DSX lets an operator run as much as 40% extra GPUs on the identical footprint by recovering stranded energy — the direct lever on tokens per greenback in a power-constrained construct.
Firebird additionally stated NVIDIA intends to put money into the corporate, following an earlier funding by CoreWeave in 2026. Early demand is coming from AI-native clients: Perplexity is working with Firebird to entry high-performance infrastructure for its reply engine and AI agent platform.
What the AI manufacturing facility runs on
Firebird’s personal information heart web page specifies the opened flagship, DC-1, at 6,144 NVIDIA B200 GPUs throughout 15 MW, liquid-cooled and constructed on NVIDIA’s reference structure. A Section 2 enlargement, DC-2, is listed at 75,000 VR200 GPUs throughout 125 MW, with a 3rd deliberate website on the similar scale. The fleet interconnects over InfiniBand east-west and Ethernet north-south, with a BlueField-3 DPU on each node, and exposes capability each as naked metallic and thru a multi-tenant cloud.
The DSX layer issues due to how the Armenia construct is constrained. Fairly than treating energy, cooling, and compute as separate procurements, DSX coordinates them as one system: its MaxLPS software program dynamically manages energy on the GPU, rack, and workload ranges to maintain the positioning inside its megawatt funds, whereas DSX OS handles fleet operations and DSX Flex can adapt workloads to grid indicators. For a facility scaling from 15 MW towards a 300 MW goal in roughly 18 months, that coordination is the place the capability math will get made.
The story to this point
The Armenia challenge has moved in quick, well-documented levels. Firebird introduced a strategic collaboration with the Armenian authorities and NVIDIA at GTC Paris in June 2025, then secured a U.S. export license in November 2025 — the regulatory key that allowed superior NVIDIA {hardware} into the nation in any respect.
In February 2026, Firebird and the U.S. authorities introduced Section 2, scaling the megaproject to $4 billion and roughly 50,000 GPUs with an export license for an extra 41,000 GB300 models (on high of the $500 million Section One). U.S. Vice President JD Vance disclosed the enlargement at a Yerevan press briefing alongside Pashinyan, NVIDIA’s Rev Lebaredian, and Firebird co-founders Razmig Hovaghimian and Alexander Yesayan. The August 8 opening turns the primary section of that paper capability right into a working facility.
What occurs subsequent
Firebird’s acknowledged goal is greater than 70,000 GPUs and 300 MW in Armenia by the tip of 2027. Past Armenia, the corporate is pursuing an roughly 2-gigawatt AI infrastructure roadmap spanning Armenia, Kazakhstan, and extra markets, with NVIDIA’s help. The corporate describes its enlargement as an effort to construct the biggest and most superior compute clusters throughout frontier markets, with the meant NVIDIA and earlier CoreWeave investments funding the broader footprint.
