What Does Technological Sovereignty Mean for a Small State Like Armenia?

Listen to the AI generated audio article. 

For a long time, sovereignty was understood in familiar terms: borders, security, territory, the ability of a state to make decisions for itself. In the age of artificial intelligence, that idea has started to stretch. Now the term AI sovereignty is appearing more frequently in countries’ innovation and technology policies—the idea that states should be able to control the technological systems they depend on, including models, infrastructure, data and the rules that shape their use.

Armenia is also moving toward this form of sovereignty. Expanded computing capacity at Yerevan State University (YSU), the National Supercomputing Center in Engineering City, and the announcement of a much larger Firebird-linked supercomputing project are all signs of a country trying to build more technological capacity at home. 

These investments were at the center of the AI conference held at Yerevan State University in April 2026. Much of the event reflected Armenia’s growing AI ambitions, from larger infrastructure to the hope of building more of its own AI capacity. But among all these conversations, one speech stood out. Armen Aghajanyan, CEO of Perceptron AI and a former researcher at Meta, presented a doctrine that pushed this idea of sovereignty in a very different direction. What made it striking was that it began with refusal: a refusal of the idea that every country must build the most advanced AI models, showcase massive computing power, or control every layer of the technology stack. 

The core claim of the doctrine is that small states often make a mistake in confusing total technological independence with sovereignty. In a private conversation, Aghajanyan put it even more simply: technological dependence, by itself, is not the problem. For Armenia, some reliance on outside models, compute, and providers is unavoidable. The question is how it is structured, and whether the country keeps enough freedom to adjust when outside conditions change.

The Setting

Before saying what Armenia should do in AI, Aghajanyan starts with a simpler question: what are the real conditions Armenia has to work with? 

In his view, Armenia faces three structural constraints from the outset: a limited pool of highly skilled technical talent, limited government funding, and finite electricity generation. Aghajanyan’s argument is straightforward: Armenia lacks the scale to think about AI the way much larger powers do. It does not have enough engineers to build every layer of the AI stack, enough public funding to compete in a global race for frontier models, or enough energy to treat compute as if it were an unlimited resource.

The AI side of the setting follows from that. Frontier models (the most advanced large-scale AI systems like GPT, Claude, or Gemini) are becoming more expensive to train, while the cost of using them keeps falling. Armenia is not going to build its own ChatGPT or Gemini. The gap in money, compute, and talent is too large. That is why the doctrine relies on a mix of access to frontier models and open-weight models (AI models that can be run and adapted locally). The question, then, is not how Armenia can own the whole stack, but how it can use the strongest systems when needed and still keep enough capacity at home when technology or geopolitics shift. 

The Claim

From there, Aghajanyan arrives at the doctrine’s main idea: use frontier models for the hardest tasks, run open-weight models locally where more control is needed, and build more of Armenia’s own capacity where AI enters the physical world.

In practice, the process is fairly simple. First, when a complex decision needs to be made, Armenia can use frontier models, because they are better at solving difficult problems quickly and accurately. But what those models produce cannot be taken as final. Their output still has to pass through Armenian law, administrative rules, clinical protocols, local terminology, and institutional context. That is where local adaptation enters, and where open-weight models become useful, because they make it possible to reshape that output into something that fits Armenian reality. Only after that does it become something real: a law, a prescription, an assessment, or another decision with consequences inside the country, carried through the country’s own processes. 

Coherence

A doctrine like this has to remain viable under multiple future scenarios. Otherwise, it becomes little more than a gamble on how the global AI landscape evolves. If open-weight models improve, Armenia can do more at home. If frontier models remain superior for the hardest tasks, that still fits the doctrine, since Armenia was never expected to develop those systems itself. The real problem begins if the gap between open and frontier models grows too wide, or if access to outside systems becomes unstable.

This is why the doctrine gives such importance to the continuity reserve. It is meant to keep Armenia from starting over every time access to outside systems changes. If that happens, an important workflow, for example in law, healthcare or public administration, should be able to move onto a new model within around 90 days. For that to happen, Armenia needs people who know how to run those workflows, records of how they have actually worked in practice, and enough compute to run them again on a different model. The continuity reserve, in other words, is the part of the doctrine meant to keep disruption from turning into collapse. 

This also explains why the doctrine is not against compute itself. It is against compute without a purpose. A GPU cluster, by itself, proves very little. It matters when it is tied to real work: adapting open-weight models to Armenian needs, testing whether frontier tools are worth their cost, running local systems in daily use, or keeping enough reserve capacity to move important workflows to a new model if conditions change.

Aghajanyan was directly involved in drafting the original proposal for Yerevan State University’s GPU-based supercomputing project, an initiative designed to strengthen Armenia’s local AI computing capacity, though he was not part of the local implementation itself. In his view, that compute helps Armenia retain talent, and makes serious work possible locally. But it also reveals the cost of delay. The longer it takes for an idea to move from proposal to implementation, the more of its value is lost. If decisions were made faster, the gains from that compute would arrive sooner and go further.

What Needs to Be Done

What the doctrine means to do is answer a practical question: if Armenia cannot suddenly become bigger, richer, or less geopolitically vulnerable, how can it still increase its capacity to function effectively?

The doctrine assumes that AI can increase the productivity of Armenia’s relatively small workforce, allowing the same number of people to accomplish far more — roughly 1.25 to 1.6 times more. Aghajanyan argues that the impact could be especially important inside the state apparatus. “One of Armenia’s deepest weaknesses is how slowly the state reacts, and how much the speed of decision-making can cost the country,” he said, adding that in some areas, especially where decisions are being made, the multiplier can feel “closer to 10 than 1.25.” 

The Commitment

That leads to the final question: who could actually lead this, and how would Armenia know it is working?

This is where the doctrine becomes less forgiving. Armenia already has many of the necessary pieces — policy bodies, research institutions, language data, and technical talent. The problem is that too much exists side by side without one center of gravity. Separate institutions, separate budgets, and separate priorities can easily produce a series of pilots without ever becoming a real platform. In Aghajanyan’s view, that leading body should not be just another ministry office. It would have to be something closer to a semi-public, semi-private structure, close enough to industry to understand what needs to be built, but public enough to carry authority, budget and responsibility. Without that kind of ownership, the doctrine can sound convincing in theory and still go nowhere when it meets reality. 

The doctrine is also clear about how it can fail, for example, by building infrastructure that is rarely used or by renting frontier tools without building enough capacity at home. But when asked which failure worried him most, Aghajanyan didn’t point first to technology. “The real challenge is organizational,” he said. “To understand what this doctrine demands, ask yourself: what is the biggest project Armenia has ever actually managed to organize and carry through? In the case of the United States, people think of the Manhattan Project or the building of railroads across the country. In Armenia, we should ask what the equivalent even is, and then add another ten percent of complexity on top of it.” 

The Test

Five years is roughly the amount of time this doctrine would need to prove it can stand on its own feet. Five years from now, could Armenia still function better in Armenian, with more throughput per worker, and more real-world autonomy under stress, if a major frontier vendor disappeared, if no new open-weight release arrived for a year, and if the geopolitical setting changed? If the answer is yes, the doctrine worked. If the answer is no, the doctrine also points to the main places where it may have failed: dependence, local capacity, continuity, or implementation. 

The harder point is that failure here would not always announce itself dramatically. It could look modern, active, even impressive from the outside while remaining hollow underneath. New hardware can arrive, new partnerships can be announced, pilots can launch, and the deeper weakness can remain. 

With this doctrine, Aghajanyan is not promising Armenia total technological independence. He is proposing a framework for what a small state can realistically build, keep, and protect in a future it will never fully control. What happens to this doctrine now depends on whether Armenia has the speed, discipline, and institutional capacity to carry out the kind of organizational change it demands.

Creative Tech