Until recently, Armenia would hardly have been called a notable player in the global computing infrastructure market. Now, the Firebird AI factory is being built in Hrazdan, and by the end of 2027 it is expected to receive more than 70,000 Nvidia accelerators and about 300 MW of infrastructure capacity. Nvidia calls the project the largest AI factory in the CIS, Bloomberg reports.

But the Firebird story is not only about a data center and scarce GPUs. As it emerged from a The Wall Street Journal report, Armenia’s access to advanced Nvidia chips became one of the economic incentives American negotiators used in the process that led to a preliminary Armenian-Azerbaijani peace understanding in Washington in August 2025. This is an unusual example of so-called “chip diplomacy,” when access to computing technologies becomes a foreign policy tool.

For Armenia itself, this means much more than the appearance of another large data center. The country is trying to leverage its longstanding expertise in electronics and chip design, its diaspora, and its ties with American technology companies to secure a place in the emerging infrastructure of global AI.

From Soviet electronics to an AI factory

Armenia has an unusual technological history for a small country. Back in Soviet times, the republic was one of the USSR’s centers of electronics and computer engineering. After the collapse of the Soviet Union, that industrial base changed significantly, but the country retained its talent pool in engineering and software development.

One of the key areas became chip design. Synopsys operates an Armenian division in the country, and the technological ecosystem gradually became linked with major American companies. It was against this backdrop that Prime Minister Nikol Pashinyan visited Nvidia’s headquarters in Santa Clara in 2019. At that time, generative AI had not yet become a mass phenomenon, and ChatGPT did not exist.

In 2022, Nvidia opened an office in Armenia. After the start of Russia’s full-scale invasion of Ukraine, the company relocated some engineers from its Moscow office to Yerevan. A year later, Nvidia CEO Jensen Huang visited Armenia. According to Firebird founder Razmig Hovaghimian, those contacts later helped form the basis for the project.

However, the path to large-scale AI infrastructure proved difficult. In 2024, after more than six months of negotiations, an Armenian state university project managed to obtain U.S. export licenses for 64 Nvidia H100 accelerators. At the time, it was the largest such computing project in the region.

This experience proved important not only because of the chips themselves. Armenia had to learn how to build infrastructure that meets the requirements of modern computing systems, while also navigating U.S. export rules for high-performance accelerators. The expertise gained later proved useful for projects on a much larger scale.

Why Nvidia chips became part of diplomacy

This is where the most unusual part of the story begins.

U.S. restrictions on the export of advanced AI accelerators were introduced under the Joe Biden administration. For a long time, Armenia was among the countries for which deliveries of the most powerful AI chips required special U.S. authorization. Washington feared that American technologies could reach Russia through third countries.

Those concerns did not arise out of nowhere: Armenia remains a member of the Moscow-led Eurasian Economic Union and formally maintains allied relations with Russia. At the same time, starting in 2023, Yerevan began tightening restrictions on electronics exports to Russia while also more actively developing ties with the U.S. and Europe.

In 2025, Firebird requested permission for deliveries of Nvidia GPUs for the future data center in Hrazdan. According to the WSJ, the initial request concerned about 6,000 accelerators. In June of that year, the team of U.S. special envoy Steve Witkoff prepared a document in which licenses for Firebird chip deliveries appeared as one element of a broader economic package for Armenia within the framework of peace talks with Azerbaijan.

After the August meeting in Washington between Donald Trump, Nikol Pashinyan, and Ilham Aliyev, the U.S. authorized deliveries of a significantly larger number of advanced accelerators. According to the WSJ, this involved tens of thousands of Nvidia chips, including GB300. In February 2026, U.S. Vice President JD Vance publicly announced that approvals had been granted for deliveries of high-performance Nvidia chips to Armenia.

Importantly, the link between the chips and the diplomatic negotiations had not been publicly disclosed in detail before the WSJ report. The Azerbaijani side, for its part, said it did not need economic incentives to conclude the agreement. Therefore, it is more accurate to describe the chips as one of the tools in the American negotiating strategy rather than the sole reason an understanding was reached.

What exactly is being built in Hrazdan

In August 2026, Firebird officially launched its AI factory in Hrazdan. According to the company’s plan, by the end of 2027 the infrastructure is expected to scale to more than 70,000 Nvidia Blackwell and Rubin accelerators and roughly 300 MW of capacity. Nvidia calls the facility the largest AI factory in the CIS region.

This is no longer an ordinary data center where you can simply place servers. Modern AI infrastructure simultaneously requires a huge number of computing accelerators, a high-speed network, data storage systems, power supply, and cooling.

Firebird uses the Nvidia DSX platform, which combines computing, network infrastructure, power supply, and cooling into a single system. For the Hrazdan site, Nvidia also lists Dell Technologies equipment, Schneider Electric power systems, and Vertiv cooling infrastructure.

And it is not only about Blackwell. Firebird also plans to use the new generation of Nvidia Rubin. The company expects to turn the Armenian site into a cloud infrastructure that developers and companies both inside the country and abroad will be able to use.

One of the first international customers is already Perplexity, which uses Firebird’s infrastructure for its AI services.

At the same time, about one-fifth of the computing capacity, according to the project’s original plans, is intended for use within Armenia — by universities, researchers, government agencies, and local startups. The remaining capacity is intended for international customers.

This is what distinguishes the project from simply building another commercial data center. Armenia is gaining the ability to use part of the computing infrastructure for its own needs without sending every major AI project outside the country.

Why does a small country need so much computing power?

For a state with a population of about three million people, 70,000 modern accelerators look like a scale that far exceeds the needs of an ordinary domestic market.

But the point of the project is precisely not to be limited to that market.

Nvidia describes AI factories as infrastructure needed to train, fine-tune, and run AI models. In Armenia’s case, this means the ability to create models and services for local needs, work with the Armenian language, conduct scientific research, and provide computing resources to startups. At the same time, the same cluster can serve foreign companies.

What emerges is a kind of bet on exporting computing power. Armenia can sell not only software or developer services, but also access itself to high-performance infrastructure.

This is especially important for small countries. Building your own global cloud platform on the level of the largest American companies is practically impossible, but building a regional computing hub and attracting international customers to it is a much more realistic strategy.

And Nvidia is already viewing Firebird in exactly that context. In August 2026, the company described the Armenian project as one example of a new model of regional AI clouds, in which local companies provide land, electricity, and operational capabilities, while Nvidia supplies the technological platform.

Armenia is trying to turn technology into an economic and geopolitical asset

The project also has a broader context.

After relations with Russia deteriorated in 2023, Yerevan began to more actively seek economic and political ties with the West. At the same time, Armenia found itself at the center of plans to reshape the transport architecture of the South Caucasus.

In August 2025, in Washington, Armenia and Azerbaijan, with U.S. participation, signed a joint declaration and preliminarily agreed on the text of a peace agreement. At the same time, a transport route through Armenian territory was discussed, which later became known as the Trump Route for International Peace and Prosperity, or TRIPP. In parallel, the U.S. and Armenia signed memorandums on cooperation in artificial intelligence and semiconductors.

Thus, several processes converged at one point at once: the normalization of relations between Armenia and Azerbaijan, the expansion of the American presence in the region, the development of transport links, and the construction of AI infrastructure.

For Washington, this is an opportunity to simultaneously promote American technologies and strengthen economic ties with a state that is gradually distancing itself from its former dependence on Russia.

For Armenia, it is a chance to gain access to computing resources that until recently were difficult to import, attract international companies, and use the technology sector as one element of its economic strategy.

For Nvidia, it is a new market and a major regional project, moreover in a country where the company has already built an engineering base.

From one data center to a regional AI network

Firebird does not intend to limit itself to Armenia. The company has announced plans to build about 2 GW of AI infrastructure in different countries, including Kazakhstan. Nvidia views the project as part of a broader network of regional AI clouds.

This is especially interesting in the context of the South Caucasus and Central Asia. The region is located between Europe and China and is simultaneously trying to expand transport and economic ties without relying exclusively on Russian infrastructure.

If Firebird’s plans are implemented, Hrazdan will turn out to be not just a place where tens of thousands of Nvidia GPUs are installed. It could become one of the nodes of a new regional computing infrastructure connecting local companies with the global AI market.

At the same time, a huge cluster by itself still does not guarantee a technological leap. A полноценной ecosystem requires specialists, developers, research groups, startups, software, stable power supply, and customers willing to pay consistently for computing.

That is why the main question for Armenia now is not whether it will be possible to install tens of thousands of accelerators. The first stage has already been launched. The question is whether the country will be able to turn access to this computing resource into a sustainable ecosystem — from university research and local startups to international AI companies.

So far, Firebird and Nvidia are betting on exactly this scenario. In August, Nvidia described the Armenian AI factory both as a source of computing resources for local developers and as a way to attract international companies to the region.

And in this sense, the story of Hrazdan goes far beyond a single data center. In a world where access to advanced AI accelerators is becoming a strategic resource, several tens of thousands of chips may prove to be for a small country not just a technological asset, but part of its new economic and foreign policy role.