Firebird has opened what NVIDIA describes as the CIS region’s largest AI factory in Armenia.
NVIDIA says the site is powered by NVIDIA accelerated computing and Dell Technologies high-performance AI infrastructure. The opening ceremony included Armenia’s prime minister, Kazakhstan’s deputy prime minister, and the U.S. charge d’affaires in Armenia.
The bigger number is the expansion plan. Firebird plans to deploy more than 70,000 NVIDIA Rubin and Blackwell GPUs and 300 megawatts of AI infrastructure capacity in Armenia by the end of 2027.
Firebird’s own site presents Armenia as the first AI factory in a wider emerging-markets platform. It lists three purpose-built data centers, 150,000 GPUs, 400 MW total capacity, and more than 5 TiB/s interconnect bandwidth as platform-level claims. Its current Armenia DC-1 is described as 6,144 NVIDIA B200 GPUs across 15 MW, with later Armenia phases planned around NVIDIA VR200 capacity.
AI capacity is becoming regional infrastructure
The project is not only another GPU cloud announcement. It is part of a shift toward regional AI capacity as a strategic asset.
NVIDIA’s post frames the Armenia factory as local compute for national languages, local industries, universities, public institutions, developers, and startups. That is the sovereign-AI argument in infrastructure form: countries and regions do not only consume global AI services; they want enough local capacity to build, fine-tune, and serve models around their own priorities.
Firebird is also pitching the facility as a commercial platform. Its site lists bare-metal GPU capacity, managed inference endpoints, Kubernetes, Slurm, programmable network fabric, observability, and high-performance storage. That makes the factory both a national-capacity story and a customer-acquisition story.
Power density is the constraint hiding inside the GPU count
NVIDIA says Firebird’s AI factory is built on the NVIDIA DSX platform, which codesigns accelerated compute, networking, power, and cooling. NVIDIA also says DSX can run up to 40% more GPUs on the same footprint.
That is the operational hinge. Large GPU counts are only useful if the site can feed, cool, connect, schedule, and monetize them. Firebird’s plan depends on power infrastructure from Schneider Electric, high-density liquid cooling, resilient facilities, and network links that make Armenia capacity usable from broader AI hubs.
