Freedom builds AI supercluster: Who will access its capacity?
Renat Tukanov, CTO of Freedom Holding Corp. and CEO of Freedom AI, said the first phase of the Freedom AI supercluster will be completed in the second quarter of 2027. During a panel discussion titled «Freedom AI: Inside Infrastructure» on Oct. 29, he discussed the capacity of the first phase, why Kazakhstan needs to build new data centers for GPU servers from scratch and who will be the primary users of the cluster’s capacity.
Project timeline
The cluster is being constructed in phases. The first phase was initially planned at 2 megawatts, but the scope has since expanded to about 4 megawatts. NVIDIA B300 servers will power it, with a launch scheduled for the second quarter of 2027.
By the end of 2027, the holding company plans to bring total capacity to approximately 50 megawatts, including 25 megawatts in Astana.
«When we were planning for a 2-megawatt cluster, Freedom was already consuming half of that capacity. Since then, consumption has risen even higher,» Tukanov said.
Asked whether foreign corporations could occupy the cluster’s capacity for years to come, Tukanov clarified that the cluster is being built as sovereign infrastructure and is intended primarily to meet domestic demand.
«Of course, we wouldn’t be happy to find ourselves in a situation where everything ends up in the hands of foreign companies. We are thinking about that,» he noted.
Freedom Cloud CEO Temirlan Zinalabdin also confirmed in an interview with Kursiv that the timeline for the data center in Alatau — a project developed in collaboration with NVIDIA — remains unchanged. The facility is slated to open in the second quarter of 2027.
Costs and funding
Freedom Holding expects to use approximately 2 megawatts of the cluster’s total capacity. According to Tukanov, the equipment required for this capacity would cost approximately $120 million. Graphics cards account for 65% to 75% of a server’s total cost, with the remainder consisting of CPUs and supporting infrastructure — components essential for services built around AI models.
«If we are consuming these resources, it implies that a corresponding value must be generated for the end clients,» Tukanov explained.
The company will sell access to the cluster by the GPU-hour. Under this model, clients will pay for the computing time required for tasks such as model training rather than renting a server for an entire year.
The holding company sees large enterprises, small and medium-sized businesses (SMEs), and individual users as potential clients. Many of these clients face legal restrictions on transferring personal or banking data to foreign cloud services.
Government agencies, for example, are seeking ready-to-use services, such as video analytics or speech recognition, rather than just hardware, Zinalabdin said.
«When government or quasi-government entities approach us, they aren’t saying, ‘Buy us some chips.’ They are asking us to ‘offer a solution for a specific problem,’» the Freedom Cloud CEO said.
He also pointed out that Freedom’s B2B cloud platform is «significantly» cheaper than those offered by international hyperscalers, although he did not disclose specific pricing.
Why buying servers isn’t enough
Existing data centers in Kazakhstan cannot accommodate GPU-based servers. Tukanov explained that these facilities were originally designed to handle 5 to 15 kilowatts per rack.
A single air-cooled B300 server consumes about 15 kilowatts, effectively occupying an entire rack. A liquid-cooled GB300 rack requires 130 kilowatts, while the next generation of NVIDIA hardware will demand 230 kilowatts.
«You cannot simply install servers into an existing data center. Therefore, alongside purchasing the servers themselves, it is necessary to build new data centers — data centers of a new class,» Tukanov said.
To illustrate the scale of the shortage, he noted that seven data centers are currently operating in Kazakhstan, with two more under construction. Together, these facilities house approximately 1,900 racks.
Meanwhile, the cluster’s construction faces several challenges. For instance, lead times for cooling chillers have increased from four months to 15 months.
Furthermore, only three companies worldwide produce high-bandwidth memory (HBM) modules for NVIDIA graphics cards. One of them, SK Hynix, reported last year that its production capacity had been fully booked through the end of 2026.
Importing these chips also requires an export license from the U.S. Bureau of Industry and Security. On top of that, constructing a data center takes approximately a year.
Meanwhile, demand within the holding company continues to rise. Freedom’s current AI-related power demand is approximately 1 megawatt, and Tukanov predicts it will reach roughly 10 megawatts over the next five years.
What do we know about Freedom’s data centers?
In April 2025, Freedom Holding Corp. began construction of the Freedom Cloud data center at the Innovative Technologies Park «Alatau.» The 2.5-hectare facility is designed to accommodate 480 racks, each with up to 15 kilowatts of capacity. It will provide 7.2 megawatts of usable IT power, with total power consumption of 13 megawatts. The two construction phases were valued at about $165.6 million, with an additional $220.8 million allocated for equipment, bringing the total investment to $386.4 million.
In November 2025, Timur Turlov’s holding company announced plans to build a $2 billion AI center in partnership with NVIDIA. At the time, it was also reported that the Akashi data center, currently under construction in Astana, was already 61% booked.
In February 2026, Turlov revealed plans to raise an additional $200 million to $300 million for telecommunications infrastructure and data centers, noting that procuring chips for an AI hub could cost up to $100 million. In June, he explained that Freedom was financing construction through its profits and that offtake agreements had already been signed for the new facilities, which were projected to generate tens of millions of dollars in revenue during their first year of operation.
In July 2026, Freedom Holding Corp. signed a memorandum with the Ministry of AI and Digital Development and Citi to establish the Freedom AI Supercluster. The signing ceremony took place on the sidelines of a meeting of the Foreign Investors’ Council under the president of Kazakhstan. The first phase was scheduled to launch in the first quarter of 2027, with the second and third phases slated for 2027 and 2028, respectively. The first phase is now expected to be completed in the second quarter of 2027. Citi has committed to evaluating financing options involving international capital.
In September, Tukanov said the supercluster would allow the country to store banking secrets and Kazakhstani citizens’ personal data domestically.
