Search
IDCNOVA

Kazakhstan’s $10 Billion AI Data Center Project in Ekibastuz Moves to Deployment Phase

By: IDCNOVARegion: Central Asia
Kazakhstan has announced that its ambitious $10 billion AI data center project in the northern city of Ekibastuz is moving from planning into active deployment. Prime Minister Olzhas Bektenov informed President Kassym-Jomart Tokayev on July 13 that partners have begun rolling out 250 megawatts of infrastructure, with the project having already attracted more than $10 billion in foreign investment. The announcement gives firmer shape to a plan first unveiled only six months ago, marking a significant step in the country’s broader digital transformation strategy.

The project is part of the government’s “Data Center Valley” initiative, which aims to repurpose Ekibastuz—a city historically built around coal mines and power stations that suffered economic decline after the Soviet collapse—into a hub for computing power. The city lost more than 1,000 residents during 2025, but officials now hope to leverage its industrial grid to attract global AI companies. The first 125 MW data center is expected to come online in the first half of 2027, with a second facility of the same size planned for 2028. These initial centers are only the beginning of a much larger development: Pavlodar officials have allocated 177 hectares, including 124.4 hectares for the opening phase, and up to ten facilities could be built across the cluster between 2029 and 2033. The site could eventually reach 1 GW, with an earlier government estimate putting total investment near $30 billion.

Construction activity has already moved beyond paper. By May 25, crews had completed geodetic work and started engineering and geological surveys, with workers excavating pits for modular blocks and equipment and personnel arriving on site. The first center will draw power from an existing 215 MW substation, and across the wider development, 300 MW is already available, with capacity eventually rising to 1 GW. The Satpayev Canal will supply water for staff and site operations, with daily use estimated at 2,300 cubic meters, while separate reserves will be kept for fire protection. The project’s reliance on coal remains a notable factor: thermal plants generated 74.4% of Kazakhstan’s electricity in 2025, and the country still imports nearly 1.5 billion kilowatt-hours from Russia annually. A 250 MW data center complex running continuously would consume about 2.19 billion kilowatt-hours per year, roughly 1.8% of Kazakhstan’s total electricity use in 2025.

NVIDIA Vice President Rev Lebaredian underscored the importance of energy for AI infrastructure, stating in June: “Everything begins with energy. If you do not have energy, you cannot build the rest.” He added that Kazakhstan had energy “in abundance.” In Ekibastuz, the harsh winters that once made industrial life difficult are now being seen as a commercial advantage, as cold temperatures can help cool server halls that generate vast amounts of heat. The planned Trans-Caspian fiber cable would then link the city’s abundant power to faster international data routes.

The funding structure remains somewhat opaque. In June, officials put the first phase at $5 billion, including $1 billion from Kazakhtelecom, with another $5 billion planned for a larger computing cluster. By July 13, however, the government was describing more than $10 billion as foreign investment, leaving questions about how the two versions align. At a June 15 ceremony, the AI ministry and Firebird signed a strategic cooperation agreement, while KT-Telecom, a wholly owned Kazakhtelecom subsidiary, signed a binding term sheet with Firebird covering technical and organizational work. NVIDIA participated in discussions and presented technology but was not named as a signatory. Firebird described the agreements as a “foundation for future collaboration” but gave no specific investment figure. Kazakhtelecom chairman Bagdat Mussin said Firebird had already secured service agreements with major global players, but he named none and disclosed no contract values. There is no public timetable showing when the money will be committed or released.

The proposed hardware order would be enormous. AI and Digital Development Minister Zhaslan Madiyev said the cluster would use 100,000 GPUs, including NVIDIA GB300 and Vera Rubin technology, and forecast at least $3 billion in annual export revenue. Firebird co-founder Razmig Hovaghimian said the 2027 launch would put Kazakhstan “among the world’s top ten countries in AI infrastructure.” Kazakhstan already has two systems on the June 2026 TOP500 list—Alem.Cloud ranked 104th and AI-Farabium ranked 122nd, both using NVIDIA H200 accelerators—but the Ekibastuz project would serve a much larger commercial market. The agreements also call for Firebird Labs Kazakhstan at Alem.ai to support research, startups, and engineering talent, reinforcing the government’s case that the development will build domestic skills alongside export capacity.

Firebird points to its project in Armenia as evidence it can build quickly. The first phase of its AI factory near Hrazdan is designed for 18 MW of power capacity and more than 6,000 NVIDIA Blackwell GPUs, and Firebird said it moved from initial construction to deployment readiness in just over six months. The first Ekibastuz center would be almost seven times larger by power capacity. Officials expect more than 1,500 temporary construction jobs and over 250 permanent posts once the facilities enter service, though data centers typically spend heavily on chips, electrical equipment, and cooling while running with relatively small teams. The larger economic promise depends on export customers using the computing capacity.

President Tokayev has tied the Data Center Valley to the Digital Qazaqstan strategy and promised personal oversight, calling it part of Kazakhstan’s “digital sovereignty” and a “high-tech hub along the New Silk Road.” The timetable is tight: by mid-2027, Ekibastuz must turn an excavation site into a working 125 MW data center, with power, cooling, fiber links, and computing hardware in place.