Recent reports of Kazakhstan signing a massive $10 billion AI deal with Firebird using NVIDIA chips have generated significant buzz, but the reality appears to be more measured.
As of now, there is no official confirmation of such a deal. What we do know is that Kazakhstan's government announced a concrete plan on June 9, 2026, to develop a sovereign exascale AI infrastructure. This confirmed project is in partnership with Freedom Holding Corp., with an anticipated budget of $2 billion and a 100 MW data center site. The $10 billion figure seems to be an exaggeration, likely stemming from a mix of separate events.
So, where did this rumor come from? The most likely source is a combination of two key events. First, in late April 2026, Kazakh President Kassym-Jomart Tokayev did meet with leaders from NVIDIA and Firebird AI to discuss potential cooperation. These were exploratory talks, not a contract signing. Second, Firebird AI is already building a large-scale AI project in nearby Armenia, which has secured U.S. export licenses. It's plausible that news from these separate threads became conflated into one large, unverified story about Kazakhstan.
This situation highlights a critical geopolitical trend: the rise of 'sovereign AI.' Nations, particularly in regions like Central Asia, want to build their own AI capabilities to ensure digital independence. However, they must navigate a complex landscape. The U.S. is tightening export controls on advanced chips to prevent them from being diverted to sanctioned countries like Russia. This creates a dual effect: it increases the risk and compliance burden for projects in the region, but it also encourages the development of U.S.-aligned AI hubs in countries willing to comply, like Kazakhstan.
The economics of such a project are also revealing. An NVIDIA GB300 NVL72 rack, a powerhouse for AI, costs an estimated $4 million. A $2 billion budget would purchase around 500-667 racks, consistent with the power of a 100 MW site. A hypothetical $10 billion project, however, would scale this up to 2,500-3,333 racks, requiring multiple large data centers and a far more complex, multi-year build-out. While even a $2 billion deal provides a solid order for NVIDIA, it represents less than 1% of its annual revenue, underscoring the scale of the global AI market.
- Sovereign AI: An initiative by a nation to develop its own artificial intelligence capabilities and infrastructure, independent of foreign countries or corporations, to protect its national interests and data.
- Exascale: Refers to computing systems capable of at least one exaFLOPS, which is a quintillion (10^18) floating-point operations per second. It represents an exceptionally high level of computational performance.
- GB300 NVL72: A high-performance, rack-scale computing platform developed by NVIDIA, combining 72 Blackwell GPUs and 36 Grace CPUs. It is specifically designed for training and running extremely large AI models.