NVIDIA and Firebird Bring 45,000 GPUs to Kazakhstan: Inside the Country's AI Factory Bet

 

Kazakhstan is not waiting to be a customer of the AI revolution. During NVIDIA founder and CEO Jensen Huang's first visit to the country on October 23, 2026, US AI infrastructure company Firebird signed an investment agreement with the Kazakh government — and NVIDIA confirmed that Firebird has secured a US export licence for 45,000 NVIDIA Grace Blackwell GPUs for Kazakhstan. The first phase is already under construction with Kazakhtelecom at Data Center Valley in Ekibastuz, with commercial operations targeted for the third quarter of 2027. That date is a target, not a launch that has already happened — but the scale of the licence alone makes this one of the most ambitious AI infrastructure bets outside the US, China and the Gulf.

What was actually announced

At the heart of the announcement is the export licence: 45,000 Grace Blackwell GPUs authorised for Kazakhstan. Firebird's first phase, built with Kazakhtelecom, will turn that authorised hardware into working computing capacity at Ekibastuz. A planned second phase would take Firebird's total potential AI infrastructure capacity in Kazakhstan to about 500 megawatts, according to NVIDIA. Firebird also plans to establish Firebird Labs Kazakhstan to support local startups, researchers and talent development through funding, access to computing resources and research partnerships. This is not Firebird's first AI factory. NVIDIA notes that the Kazakhstan expansion builds on Firebird's first AI factory in Armenia, and the signing ceremony took place at the Akorda presidential palace during Huang's visit to Astana.

Why Kazakhstan, and why now

Huang's message in Astana was blunt: Kazakhstan could become "an AI factory hub for all of Europe." Instead of only exporting energy, he argued, the country could transform energy into intelligence — and export intelligence. That pitch fits Kazakhstan's own declaration of 2026 as its Year of Digitalization and Artificial Intelligence. The country has power, land and a stated ambition to move from consuming computing resources to providing them. Kazakhtelecom CEO Bagdat Mussin put it the same way: competitiveness in AI is now determined not only by algorithms, but by whether a country has its own computing infrastructure. The same logic is driving other large GPU bets in the region. Saudi Arabia's HUMAIN, for example, recently expanded its NVIDIA capacity in an 84 MW deal covered in our article on HUMAIN and NVIDIA's 84 MW AI expansion, while governments elsewhere are treating AI computing as national infrastructure — as the UK did when it put £600 million behind AI security in our report on the UK's AI security package.

What is already running

Kazakhstan is not starting from zero. According to NVIDIA, Kazakhtelecom has already deployed 400 NVIDIA H200 GPUs and 480 NVIDIA B300 GPUs in Astana and Almaty — 880 GPUs in total. Next year, Kazakhtelecom plans to deploy more than 1,000 additional GPUs at Data Center Valley, along with more than 1,000 NVIDIA Vera Rubin accelerators in the Ekibastuz region. Vera Rubin is NVIDIA's next-generation platform, and its arrival in 2027 would move the project beyond Grace Blackwell hardware alone. Put in perspective, the 45,000-GPU licence is about 51 times the 880 GPUs Kazakhtelecom has already deployed. That comparison is about chip count, not equivalent computing performance — newer GPUs are far more powerful per chip — but it still shows how large the jump is from pilot deployments to industrial-scale capacity.

The $10 billion context — and what is still unproven

Reports around the visit describe a planned $10 billion AI infrastructure package involving Firebird and Kazakhtelecom, with computing systems based on NVIDIA technology, and Kazakh officials have talked about compute exports eventually bringing in at least $3 billion a year. Those figures should be read carefully. They are reported plans and ambitions, not revenues already earned or hardware already installed. Power supply, cooling, networking and commissioning still have to accompany processor deliveries, and NVIDIA's own account gives a service target — the third quarter of 2027 — without a shipment-by-shipment installation schedule. The same caution applies to the wider GPU race. NVIDIA's record market run, examined in our piece on what NVIDIA nearing $6 trillion really means, reflects investor belief in AI infrastructure demand — not a guarantee that every announced cluster will be built on time. Open-weight rivals such as Moonshot AI's Kimi K3.1 Pro, covered in our report on Kimi K3.1 Pro, also show why demand for computing is spreading: more capable models, open and closed, all need more GPUs to train and run.

What to watch next

Three milestones will tell us whether Kazakhstan's AI factory bet is real. First, whether the first phase at Ekibastuz hits its third-quarter 2027 commercial target. Second, whether the Vera Rubin deployment in 2027 actually lands on schedule. And third, whether Firebird Labs Kazakhstan turns computing access into local startups, research and jobs — or whether the capacity mainly serves customers elsewhere. If those milestones land, Kazakhstan will have done something few countries outside the biggest AI powers have managed: turning an export licence and a presidential-palace signing into a working AI factory. If they slip, it will be a reminder that in AI infrastructure, announcements are cheap and megawatts are hard. Would you bet on Kazakhstan becoming Europe's AI factory hub — or is this another infrastructure announcement that will take years longer than promised? Tell us in the comments below.