Rev Lebaredian describes the first phase of Firebird's AI facility in Hrazdan, Armenia, as operating with 6,144 B200 GPUs. The discussion frames an AI factory as infrastructure that converts data and energy into token output, with reliability and energy efficiencyCompute performance per watt measures how much useful computing work a system completes for each watt of electrical power it consumes. shaping its economics. Rev Lebaredian says designing chips, networking, cooling and grid connectionsAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models. together can improve capacity within fixed space and power constraints.
Rev Lebaredian argues that countries should develop their own AI capabilities rather than depend entirely on imported intelligence. Sovereignty, in this account, includes infrastructure as well as agency over national data, language and cultureData sovereignty is the principle that data is governed by the laws and authority of the jurisdiction where it is stored or processed.. The explanation follows five connected layers: energy, chips, infrastructure, models and applications, with local applications providing the eventual practical value.
Rev Lebaredian says the initial Armenian capacity primarily serves overseas customers, while a portion is reserved for local universities, researchers and startups through a government arrangement. The proposed economic benefits include infrastructure revenue, engineering experience, research access and attracting outside talent. These are presented as expectations for the project's development, rather than demonstrated national economic outcomes.
Watch on YouTube
