Armenia has recently launched the largest AI factory in the region, made possible by a license signed by Washington.

      Firebird, a US-based AI cloud company, inaugurated what NVIDIA labels as the largest AI factory in the CIS region, located in Hrazdan, Armenia, on Saturday. Armenian Prime Minister Nikol Pashinyan was present, along with Kazakhstan's deputy prime minister Zhaslan Madiyev and David Allen, the US chargé d’affaires in Armenia.

      The significance lies in this guest list; a US company has established a facility within Russia's traditional sphere of influence, observed by an American diplomat and a Kazakh minister.

      Key to this development is the export approval. Firebird received a US export license in November 2025, which was essential for bringing advanced NVIDIA hardware into Armenia.

      The political context was clarified in February when Firebird and the US government jointly unveiled Phase 2 at a press conference in Yerevan during Vice President JD Vance’s visit, expanding the project to $4 billion and approximately 50,000 GPUs, with permits for an additional 41,000 NVIDIA GB300 units. Phase One had an initial investment of $500 million. The announcement suggested that Armenia would host one of the five largest AI GPU clusters globally.

      Currently operational is the flagship facility, DC-1, equipped with 6,144 NVIDIA B200 GPUs operating at 15 megawatts, utilizing liquid cooling based on NVIDIA’s reference architecture. By the end of 2027, the target is to exceed 70,000 Rubin and Blackwell GPUs with a total capacity of 300 megawatts. This represents a twentyfold increase in power capacity in approximately 17 months, with a planned DC-2 featuring 75,000 VR200 GPUs at 125 megawatts, along with a third site of a similar scale.

      The speed of the project’s delivery is particularly impressive; NVIDIA reports that the site transitioned from planning to operational status in just over six months, with Schneider Electric supplying essential switchgear and rack systems, while Vertiv provided cooling solutions.

      The facility operates on NVIDIA’s DSX platform, which integrates computation, networking, power, and cooling into a unified system instead of treating them as separate components. NVIDIA asserts that this method allows for up to 40% more GPUs within the same footprint by optimizing power usage. For a site that is scaling from 15 to 300 megawatts, effective coordination is crucial for maximizing capacity and, according to NVIDIA, this results in a profitable arrangement rather than merely focusing on performance.

      Firebird announced that NVIDIA has plans to invest in the company, following a previous investment in CoreWeave earlier this year. This approach is not new; NVIDIA frequently invests in companies that subsequently purchase NVIDIA hardware in large quantities. The company has committed over $40 billion to AI equity positions for 2026, adopting increasingly creative financing structures. For example, it took a $2.1 billion warrant in IREN as part of a five-gigawatt data center agreement.

      This pattern extends to NVIDIA's sovereign deals, wherein it invested $1 billion in Naver while entering a $500 billion contract with SK Group, an approach that some critics have described as circular financing.

      Firebird represents a category that emerged within the last three years: companies that acquire GPUs at scale to lease them out to model builders. CoreWeave, an early investor in Firebird, has arranged a multi-year contract with Anthropic to operate Claude at full production capacity. Europe has a similar case with Nscale, which pledged €695 million to Portugal alongside Microsoft, achieving a valuation of $14.6 billion in two years, based on a similar premise regarding sovereign capacity.

      Firebird's first identified client is Perplexity, which plans to utilize the facility for its answer engine and AI agent platform.

      Looking ahead, Firebird is working towards a two-gigawatt roadmap that includes Armenia, Kazakhstan, and additional markets, which accounts for the presence of a Kazakh deputy prime minister at the inauguration in Armenia. The strategic reasoning is clear: nations are increasingly aligning around computing power, and the United States has found a method to extend its influence into a region where it was previously minimal.

      The critical unresolved issue is demand. Constructing 300 megawatts in the Caucasus assumes that customers will prefer to direct their workloads there instead of utilizing more cost-effective or geographically closer options, and merely one answer engine does not definitively address that concern.