AIdataNewsGlobal General: Data Center Valley Construction in Ekibastuz Moves Kazakhstan AI Infrastructure Plan From Licensing to Deployment
By David Nakamura | AI & Data News | AIdataNewsGlobal General
Why the Kazakhstan compute story is a sequencing story
For readers following the intersection of AI, industrial infrastructure and policy, the key question about Kazakhstan is not whether a single large target has been achieved, but how several distinct commitments are unfolding. NVIDIA and Firebird are advancing tens of thousands of GPUs, additional data centre capacity and Vera Rubin systems, while a secured US export licence for 45,000 NVIDIA Grace Blackwell GPUs gives the first construction phase at Data Center Valley in Ekibastuz a clearer near-term operational target in the third quarter of 2027.
The announcement therefore needs to be read in layers. One layer is hardware already deployed. Another is a facility under construction. A third is a future platform and cluster plan. A fourth is a longer-term expansion ambition that should not be collapsed into installed capacity. The Astana meeting on 23 October 2026 connects those layers to national policy conversations involving Kassym-Jomart Tokayev and Zhaslan Madiyev.
Key facts at a glance
- Firebird has secured a US export licence for 45,000 NVIDIA Grace Blackwell GPUs.
- The first phase is being built at Data Center Valley in Ekibastuz with Kazakhtelecom.
- Commercial operations are scheduled to begin in the third quarter of 2027.
- A second phase may lift total AI infrastructure capacity to approximately 500 MW.
- The wider Data Center Valley project has a longer-term expansion target of 1 GW.
- Kazakhtelecom has separately deployed 400 NVIDIA H200 GPUs and 480 NVIDIA B300 GPUs across Astana and Almaty.
- During 2027, the operator plans to install more than 1,000 additional GPUs at Data Center Valley.
- Planned deployments also include exceeding 1,000 Vera Rubin GPUs in the Ekibastuz region.
- NVIDIA announced an investment in Higgsfield, a Kazakhstan-founded company developing generative AI tools for video and image production.
- Firebird plans to establish Firebird Labs Kazakhstan for local startups and researchers.
What has already been deployed and what is still under construction
The cleanest distinction is between deployed systems and announced construction. Kazakhtelecom has already placed 400 NVIDIA H200 GPUs and 480 NVIDIA B300 GPUs across Astana and Almaty, meaning H200 and B300 capacity is present now. By contrast, the Firebird installation at Data Center Valley in Ekibastuz is in construction, with the first commercial start date set for the third quarter of 2027.
Government updates published in August confirmed work on an initial 50 MW facility. Those updates also noted acquisition of a 215 MW electrical substation, with the first facility expected to enter operation in June 2027. That gives readers two near-term dates that can easily be confused: the initial facility operation date reported by the government, and the commercial operations date in the broader NVIDIA and Firebird announcement.
How to read the capacity and investment figures
| Figure | What the announcement associates it with | Reader takeaway |
|---|---|---|
| 45,000 NVIDIA Grace Blackwell GPUs | US export licence secured by Firebird for the first phase at Data Center Valley | A hardware licensing milestone, not a measure of installed capacity |
| 50 MW | Initial facility construction update published by Kazakhstan’s government in August | The smaller construction scope is the closest operational signal |
| 215 MW | Electrical substation acquisition noted alongside the August construction update | Power infrastructure is part of the Ekibastuz development narrative |
| approximately 500 MW | Second phase that may lift total AI infrastructure capacity in Kazakhstan | A possible phase-level capacity target, not a current build-out figure |
| $10 billion | Earlier agreements announced by Kazakhstan’s government with Firebird and NVIDIA | An investment scope tied to a broader plan rather than the first commercial phase alone |
| up to 100,000 GPUs | Plans for a computing cluster included in earlier agreements | A longer-range cluster ambition that differs from the first phase |
| 1 GW | Longer-term expansion target for the wider Data Center Valley project | A broad project target that should not be read as installed capacity |
Vera Rubin, Grace Blackwell and the platform stack
The platform dimension of the update is as important as the capacity dimension. Grace Blackwell GPUs appear in the US export licence and first construction phase, while Vera Rubin represents the next-generation architecture referenced for future deployment. The Vera Rubin architecture combines next-generation GPUs with the Vera CPU and high-bandwidth memory in rack-scale computing systems.
A reporting reference noted as eeNews Europe states that NVIDIA began supplying customer samples of Vera Rubin hardware earlier this year, ahead of broader deployments. For Kazakhstan, the relevant implication is that the announcement is not only about current H200 and B300 installations, but also about planned Vera Rubin capacity in the Ekibastuz region.
Applications, research access and the local ecosystem
The programme is not limited to data centre construction. Firebird Labs Kazakhstan is described as a future access point for local startups and researchers, providing computing resources, funding and research partnerships. That element suggests the infrastructure plan is intended to serve domestic development as well as regional compute deployment.
NVIDIA also announced an investment in Higgsfield, which develops generative AI tools for video and image production. The investment amount was not disclosed. Higgsfield intends to expand its computing capacity using Vera Rubin infrastructure supplied through cloud service providers.
Policy conversations and energy constraints
The timing of the announcement aligned with the Kazakhstan visit of NVIDIA founder and CEO Jensen Huang on 23 October 2026. Huang met President Kassym-Jomart Tokayev in Astana and joined a discussion with Deputy Prime Minister Zhaslan Madiyev, the country’s minister of artificial intelligence and digital development.
During the visit, Huang discussed the possibility of using Kazakhstan’s energy resources to support AI computing services for international customers. Madiyev also outlined government interest in developing Kazakh-language models and applying AI to manufacturing, mining, agriculture and logistics.
Energy availability remains central to the economics of the project. Kazakhstan’s government has identified competitive electricity costs and access to existing power infrastructure as advantages for the Ekibastuz development. At the same time, the proposed expansion will require substantial additional generating and distribution capacity.
Checklist for AIdataNewsGlobal readers
- Separate deployed hardware from planned hardware: the 400 NVIDIA H200 GPUs and 480 NVIDIA B300 GPUs are already in place, while future Vera Rubin and Grace Blackwell scale-up remains staged.
- Track first commercial operations in the third quarter of 2027 for the Firebird first phase, and distinguish that date from the June 2027 initial facility operation expectation reported in August.
- Watch power readiness around the 215 MW electrical substation and the 50 MW initial facility, because the larger approximately 500 MW and 1 GW targets depend on further generation and distribution capacity.
- Follow the planned 2027 build-out involving more than 1,000 additional GPUs at Data Center Valley and exceeding 1,000 Vera Rubin GPUs in the Ekibastuz region.
- Observe whether Firebird Labs Kazakhstan, Higgsfield and Kazakh-language model initiatives convert compute investment into application-layer activity.
Frequently asked questions
Is Kazakhstan already operating the full announced capacity?
No. The announced facts distinguish already deployed Kazakhtelecom systems from the first Firebird construction phase at Data Center Valley. The deployed items include 400 NVIDIA H200 GPUs and 480 NVIDIA B300 GPUs. The larger GPU, power and cluster figures relate to construction, future deployment, or longer-term expansion targets.
Why is the US export licence important?
It gives the first phase a licensing milestone: Firebird has secured the licence for 45,000 NVIDIA Grace Blackwell GPUs. That supports the planned construction and deployment path at Data Center Valley in Ekibastuz, with commercial operations scheduled to begin in the third quarter of 2027.
Are $10 billion and up to 100,000 GPUs installed now?
No. Those figures refer to earlier agreements announced by Kazakhstan’s government with Firebird and NVIDIA, including cluster plans. They describe different stages and scopes, not installed capacity.
What is Vera Rubin?
Vera Rubin is the next-generation architecture referenced in the announcement. It combines next-generation GPUs with the Vera CPU and high-bandwidth memory in rack-scale computing systems. The reporting note referencing eeNews Europe says NVIDIA began supplying customer samples of Vera Rubin hardware earlier this year.
What should readers watch next?
The practical next indicators are construction progress, whether the initial facility expectation reported in August holds, whether commercial operations begin in the third quarter of 2027, and whether the planned 2027 additions of more than 1,000 additional GPUs and exceeding 1,000 Vera Rubin GPUs move from scheduled deployment to operational capacity.
Bottom line for the General column
For AIdataNewsGlobal, the story is a layered infrastructure programme rather than a single completed deployment. NVIDIA, Firebird, Kazakhtelecom, Higgsfield and Kazakhstan’s government appear across the announcement as participants in a sequence that moves from licensed GPU access and current H200 and B300 deployments to construction, platform transition and application partnerships. The first commercial start in the third quarter of 2027 will be the initial checkpoint for understanding how much of the tens of thousands of GPUs, approximately 500 MW and 1 GW narrative can become available computing capacity.