Summary
- Firebird has opened its first NVIDIA-based AI factory in Hrazdan, Armenia, after bringing the initial 15MW facility into operation in 2026.
- Its Armenian roadmap calls for more than 70,000 NVIDIA Blackwell and Vera Rubin GPUs and 300MW of AI infrastructure by end-2027.
- The company is also pursuing Kazakhstan and other emerging markets as part of a global infrastructure pipeline targeted at roughly 2GW by end-2028.
Firebird has opened its first AI factory in Hrazdan, Armenia, and set out plans to expand the Armenian platform to 300MW and more than 70,000 NVIDIA GPUs by the end of 2027.
The opening moves the project from an announced AI-infrastructure programme into operation. Firebird says its Hrazdan flagship currently comprises 15MW and 6,144 NVIDIA B200 GPUs, giving the company a functioning base from which to attempt a much larger expansion.
NVIDIA says Firebird’s Armenian roadmap calls for more than 70,000 Blackwell and Vera Rubin GPUs and about 300MW of AI infrastructure by the end of 2027. NVIDIA also says it intends to invest in Firebird, following an earlier investment by CoreWeave, although financial terms for the planned investment have not been disclosed.
The company is also expanding beyond Armenia. Firebird has secured 125MW at Data Center Valley in Kazakhstan and is targeting a wider infrastructure pipeline of roughly 2GW by the end of 2028 across Armenia, Kazakhstan, and additional emerging markets.
NVIDIA says Schneider Electric is supplying power infrastructure for the Armenian deployment and Vertiv is providing cooling systems. Firebird has also identified Perplexity as an early customer of its AI infrastructure.
Compute growth becomes an infrastructure programme
The 300MW target puts the Armenian project into a different engineering category from the operating 15MW phase. Scaling from tens of megawatts to hundreds turns the project into a power, cooling, network, construction, and supply-chain programme as much as a GPU deployment.
Dense AI clusters require electrical distribution that can support rapidly rising rack loads, cooling systems capable of removing concentrated heat, resilient backup architecture, high-bandwidth networking, and building systems that can be deployed in step with hardware deliveries. A delay in any one of those layers can strand expensive compute or leave power capacity unused.
The location is also notable. Much European AI-capacity discussion has centred on established western markets where grid queues, constrained land, and permitting can slow new capacity. Firebird is instead building in Armenia and pursuing Kazakhstan, putting infrastructure investment into markets with smaller existing hyperscale footprints but potentially greater room to assemble land and power.
That does not remove infrastructure risk. Large GPU deployments still depend on reliable high-voltage connections, resilient generation and transmission, specialist cooling equipment, suitable logistics, and a supply chain capable of installing and maintaining dense computing systems. Expansion speed depends on those systems arriving together rather than simply securing processors.
There is also a sovereignty dimension. Countries seeking domestic AI capability increasingly need physical compute infrastructure inside their borders or region rather than relying entirely on capacity concentrated in established cloud hubs. Firebird is explicitly positioning its Armenian and Kazakh projects around that demand.
Commercial utilisation will be as important as construction as Firebird moves beyond the opening phase. A 300MW build-out requires sustained demand as well as access to processors and energy, and the company will need to convert its infrastructure pipeline into contracted, operating capacity.
The immediate milestone is concrete: Firebird now has a functioning Armenian AI factory. The harder part is the next scale jump. Reaching more than 70,000 GPUs and 300MW by the end of 2027 would require a rapid sequence of power, cooling, building, networking, and hardware deployments, with the Kazakhstan programme developing alongside it.
