The launch pairs Firebird’s cloud ambitions with NVIDIA accelerated computing, Dell Technologies servers and NVIDIA’s DSX reference platform. NVIDIA says the project is intended to turn a combination of compute, networking, power and cooling into a repeatable AI-factory design rather than a conventional data-centre build.
A new compute hub in Armenia
The opening ceremony included Armenian and Kazakh government representatives alongside partners from NVIDIA, Dell Technologies and CoreWeave. Firebird positions the site as infrastructure for local-language AI work, research and commercial applications, rather than capacity used only by overseas customers.
That regional framing is important. Access to large-scale AI infrastructure is still concentrated in a small number of markets, while many organisations must send workloads and data elsewhere to obtain sufficient compute. A local facility does not remove those constraints on its own, but it can give regional teams another option for model development and inference.
What is operating now — and what is planned
The announcement confirms the factory has launched, but the largest capacity figures remain future targets. Firebird says it plans to deploy more than 70,000 NVIDIA Rubin and Blackwell GPUs and 300 megawatts of AI infrastructure in Armenia by the end of 2027. Those figures should therefore be read as a build-out plan, not as an installed-capacity claim for the day of launch.
NVIDIA says the facility uses DSX to co-design accelerated computing, networking, power and cooling. The company claims that approach can allow more GPUs within a given footprint, but it has not published independent operating metrics for the Hrazdan site. Prospective customers will need to assess actual available capacity, service terms, model support and network connectivity as the project expands.
Partners line up behind a broader expansion
Firebird says it is pursuing an approximately two-gigawatt infrastructure roadmap across Armenia, Kazakhstan and other markets. NVIDIA also says it intends to invest in Firebird, following an earlier CoreWeave investment. Dell PowerEdge systems provide the server layer, while Schneider Electric and Vertiv are supplying the power and cooling components described in the announcement.
The partner list shows how AI infrastructure projects now combine chip suppliers, server makers, cloud operators and facilities specialists. It also makes delivery dependent on more than GPU supply: power availability, construction, cooling, network links and local approvals will determine how quickly the planned capacity arrives.
Early demand and the questions ahead
NVIDIA identifies Perplexity as an early customer seeking access to Firebird’s infrastructure for its AI agent platform and answer engine. The companies have not disclosed capacity commitments, pricing or a timetable for customer access beyond the launched facility.
For teams evaluating regional AI compute, the announcement is a significant infrastructure signal rather than a finished capacity map. The useful next indicators will be deployed GPU counts, commercial availability, energy performance and the extent to which the site supports workloads for local organisations as well as international AI companies.