Kazakhstan’s AI Ambition Depends on Turning Power into Export Revenue

Kazakhstan is trying to build an export industry around AI computing. Data Center Valley in Ekibastuz brings together electricity infrastructure, large development ambitions and international technology partners. The commercial question is whether that combination can produce computing services that customers will buy consistently at a price that covers the full cost of delivery.

My reading is that the recent announcements make the first stages of this strategy more credible. They also move the central question towards execution and customer economics. Kazakhstan can attract a large AI installation without first reproducing the entire ecosystem of an established cloud market. Sustaining expansion will require a repeatable business case beyond the first anchor project.

The new evidence strengthens the development case

In its account of Jensen Huang’s October visit, NVIDIA reports that Firebird has obtained a US export licence for 45,000 Grace Blackwell GPUs. Its first phase is under construction with Kazakhtelecom at Data Center Valley, with commercial launch targeted for Q3 2027. A subsequent phase could bring Firebird’s potential AI infrastructure capacity in Kazakhstan to approximately 500 MW.

An export licence removes a specific obstacle to procurement. Construction activity adds evidence that the partnership is moving into implementation. Neither establishes completed delivery, but together they provide a stronger basis for assessing execution than an expansion target alone.

The Kazakh ministry’s update on 23 October adds a commercial signal: it describes the first-phase envelope of 215 MW as fully reserved, within a development programme with energy potential of up to 1 GW. Reservation terms are not disclosed. The announcement nevertheless supports a more focused question about who has committed, on what conditions and against which delivery milestones.

The scale references need only a short clarification. The government’s August construction update identifies a 50 MW first facility targeted for June 2027 and an acquired 215 MW substation. Firebird’s approximately 500 MW ambition covers its planned AI infrastructure across phases, while 1 GW describes the wider energy potential. These figures overlap and cannot be added. June 2027 and Q3 2027 also refer to different described milestones, rather than proving that the same delivery date has changed.

The development case is becoming more concrete. Its financial strength will depend on how those milestones connect.

A route into global compute demand

PowerlandMap’s Kazakhstan inventory, checked on 26 October, contains 14 published supply records, including nine classified as live and five in development or construction. It includes existing facilities in several cities alongside larger development programmes. This is a coverage snapshot, not a census of the national market, and the records include different scopes such as facilities, expansions and programmes.

The distinction matters for the strategy. An established domestic hosting market and a large export-oriented AI cluster serve different customer needs. The existing inventory provides context, but counting local facilities cannot establish how much international AI demand a new cluster could attract.

A plausible initial route is to secure a limited number of substantial compute customers, then expand as delivery and utilisation become demonstrable. That would let a new location build around identifiable contracts while developing the wider supplier, engineering and service ecosystem over time.

There is some public support for this reading. In the government’s account of the cooperation agreements, Kazakhtelecom says Firebird has secured service agreements with major global players. The account does not disclose customer names, committed volumes, prices or termination provisions. It supports the existence of a commercial strategy, while leaving the quality and duration of its revenue base open to assessment.

Concentration is the trade-off. A few large customers can make an initial phase possible, but their delivery requirements and contract decisions can also determine the utilisation of much of the asset.

Electricity savings need to survive the full cost of service

The economic attraction of locating compute near competitively priced power is straightforward. Electricity contributes to the cost of every hour of operation. At sufficient scale, a modest difference in electricity price has a material annual effect.

Consider a hypothetical site with 100 MW of electrical import capacity and an average draw equal to 80% of that capacity. Its annual consumption would be 700,800 MWh. The following sensitivity isolates electricity cost:

The calculation is 100 MW multiplied by 80%, 8,760 hours and the assumed price difference. It uses total site electricity, so no additional PUE multiplier is applied. These are illustrative assumptions, not disclosed tariffs, consumption or savings for Data Center Valley. Network charges, financing, equipment, staffing, taxes and other costs are excluded.

The commercial implication is that an electricity advantage can create room for competitive pricing. It cannot establish the final cost per useful GPU-hour on its own. Purchased hardware still incurs financing and depreciation costs during periods when it is available but not generating revenue.

A project that secures cheaper electricity but struggles to fill its machines may therefore lose much of its advantage. Conversely, dependable customer utilisation can make the location more attractive even before later expansion phases are delivered. That is why contract quality and service reliability deserve as much attention as the energy price.

Expansion will test the whole delivery system

The first operating phase will establish whether the development can coordinate power, buildings, cooling, connectivity, equipment installation and customer acceptance. Each later phase will test whether that delivery can be repeated economically.

For equipment suppliers, early construction creates an opportunity to qualify project requirements and delivery packages. A long-term campus target alone does not establish the size or timing of a purchase order. Local service capability may become commercially valuable where it shortens repair times and helps customers maintain availability.

For customers, the decisive evidence will be the service they can actually contract: technical configuration, acceptance tests, availability commitments, network performance and remedies if delivery slips. For investors, it will be the relationship between committed capital, completed assets and enforceable revenue.

Power sourcing also matters to the customer mix. The government’s cooperation announcement explicitly links the project’s export strategy to Ekibastuz’s coal-generated electricity. That makes the emissions profile of the contracted supply a substantive commercial consideration for buyers with carbon requirements. Any future improvement in the energy mix should be assessed against documented supply arrangements.

A reasonable counterargument is that strong initial commitments could justify expansion before the first phase has accumulated a long operating history. That can work when customers accept the delivery risk and the financing structure reflects it. The evidence needed is the substance of those commitments, rather than the reservation headline alone.

The first phase will set the value of the expansion

Data Center Valley could give Kazakhstan a meaningful position in internationally traded AI computing. The combination of power infrastructure, access to advanced chips and an identified operator partnership creates a credible route towards that outcome.

The next developments that would strengthen this assessment are equipment deliveries, completed power and facility milestones, customer acceptance and evidence of sustained commercial use. They would show whether the initial energy advantage translates into a service that can compete and support further investment.

The strategic opportunity is to build an export business whose economics improve with reliable delivery and repeat customers. The first phase will determine how much confidence the market should place in the larger expansion.

Analysis by Matthieu Gallego. Sources and PowerlandMap inventory checked on 26 October 2026. Interpretations and the electricity sensitivity are analytical assessments, not disclosed project forecasts.