Armenia’s $4B, 41,000-GPU AI Bet Isn’t About Building Chips—It’s About Sovereignty

Let’s kill the most viral misconception right up front: NVIDIA is not building a chip fabrication plant in Armenia. You’re not going to see a TSMC-style foundry popping up in Yerevan. The Blackwell GPUs at the heart of this story are still made on TSMC’s 4NP process. What is actually happening is, in some ways, more interesting for a country of Armenia’s size.

This is a story about a small nation trying to muscle its way into AI sovereignty by becoming a compute hub. The centerpiece is a U.S.-based company called Firebird, operating between San Francisco and Yerevan. The Armenian government says phase one, targeted for 2026 near Hrazdan, involves a $500 million investment, over 6,000 NVIDIA Blackwell GPUs, and 18 MW of capacity delivering up to 110.6 exaflops of FP4 compute. A wildly ambitious second phase reportedly balloons to $4 billion and more than 41,000 additional GPUs. If delivered, that’s an absurd amount of compute for a country this size.

The confusion stems from a legitimate web of associations. U.S. export approval was required for the chips, Dell is providing PowerEdge servers, and the U.S. and Armenia signed an AI and Semiconductor Innovation Partnership MOU in August 2025. That agreement covers secure supply chains and IC development, which sounds close to “chip production,” but the reality is AI infrastructure and R&D. As Eurasianet and OC Media have reported, the political milestone of getting U.S. regulators to greenlight advanced chip transfers to Armenia, a country previously caught in export-control crossfire, is a story in itself.

Armenia’s logic isn’t crazy. It already has a genuine tech base with over 1,000 people at Synopsys Armenia working on electronic design automation and semiconductor IP. Pair that diaspora-connected engineering talent with imported, high-density systems like NVIDIA’s GB300 NVL72 racks—which NVIDIA now calls “AI factories”—and you get a pitch that’s less about manufacturing physical chips and more about manufacturing intelligence. Whether the full $4 billion vision materializes is a different question entirely, but the first approved GPU shipments make clear this is no longer just a pitch deck.