Crazy Rich Kazakhstan | Supercomputers, Data Centres, AI Compute and Digital Mining begins with a useful correction: artificial intelligence does not live “in the cloud”.
Did you know Kazakhstan’s National AI Platform currently lists two sovereign supercomputing systems in the global TOP500? Alem.Cloud is shown at rank #104 with an Rmax of 20.48 PFLOP/s, while Al-Farabium is listed at #122 with an Rmax of 17.93 PFLOP/s. Search terms such as “Kazakhstan supercomputer”, “Alem.Cloud”, “Kazakhstan data center”, “Kazakhstan AI”, “Kazakhstan crypto mining” and “Kazakhstan digital infrastructure” all lead to the same physical truth: computation consumes machines, buildings and electricity.
The government also reported that two new data centres with a combined capacity of 7.4 MW entered service in Almaty and Astana during 2025, with three more facilities totalling 12.9 MW planned for launch in 2026.
Crazy Rich Kazakhstan becomes silicon, cooling and electricity here.
Did You Know? AI Has a Physical Address
A language model may feel like software floating on the internet.
Somewhere, processors are actually running the calculations.
They sit in racks.
The racks sit in data centres.
The data centres require electricity, cooling, fibre, security and maintenance.
The “cloud” is somebody else’s building full of computers.
Alem.Cloud Is Sovereign Compute Infrastructure
Kazakhstan’s National AI Platform describes Alem.Cloud as one of Central Asia’s largest computing complexes for artificial intelligence, machine learning and high-performance computing.
Its current 2026 listing gives an Rmax of 20.48 petaflops on the TOP500 benchmark.
The system uses NVIDIA H200 accelerators in a high-density compute architecture.
That makes the project a national infrastructure asset rather than an ordinary office IT system.
What Does 20.48 PFLOP/s Mean?
PFLOP/s means quadrillions of floating-point operations per second under the relevant benchmark measurement.
It is a way of expressing how quickly a machine can perform certain numerical calculations.
The important word is benchmark.
Real AI performance depends on model architecture, memory, interconnects, software and workload as well as the headline number.
One score is a useful indicator, not the whole machine.
Al-Farabium Adds a Second National Compute Layer
The National AI Platform lists Al-Farabium at #122 on the 2026 TOP500 with 17.93 PFLOP/s Rmax.
It is intended for scientific computing, AI development and industrial research.
Two systems matter because national compute capacity benefits from redundancy, specialisation and broader access.
Compute is easier to scale when one machine is not the entire national bottleneck.
Supercomputers Are Knowledge Multipliers
A supercomputer does not invent useful questions.
Researchers and engineers do.
The machine multiplies how many calculations they can test.
That is why compute and education are complements.
More compute without skilled people creates underused capacity.
Skilled people without compute hit another ceiling.
Capability emerges when both grow together.
This Is the Education Ouroboros in Hardware Form
Education trains people to use advanced systems.
Advanced systems let those people solve harder problems.
The solutions create new industries and better tools.
Those tools feed the next generation of education.
Read Crazy Rich Kazakhstan | Nazarbayev University, Research, Engineering and the Talent Economy.
Compute becomes valuable when learning can reach it.
Data Centres Are Factories for Computation
A data centre converts electricity and hardware into digital services.
Its production output may be:
- AI inference;
- model training;
- cloud computing;
- databases;
- government services;
- web applications;
- scientific simulation;
- digital-asset computation.
The factory floor is a server hall.
The raw input is electricity and data.
Kazakhstan Added 7.4 MW of Data-Centre Capacity in 2025
Government reporting says two new data centres began operation in Almaty and Astana in 2025 with combined power capacity of 7.4 MW.
Three more centres totalling 12.9 MW were planned for 2026.
Megawatts matter because data centres are increasingly constrained by power availability.
The digital economy eventually meets the electricity grid.
Why AI Data Centres Consume So Much Power
Modern AI accelerators perform enormous numbers of calculations.
That generates heat.
Servers consume power.
Networking consumes power.
Cooling consumes power.
Backup systems consume capacity.
The building therefore behaves more like an industrial facility than an ordinary office.
Cooling Is Compute Infrastructure
Computer chips become less reliable if they operate outside safe temperature ranges.
High-density AI racks can concentrate substantial heat in small spaces.
Cooling systems remove that heat using air, liquid loops or combinations of both.
The invisible product of a data centre is computation.
One of its largest physical engineering tasks is simply moving heat somewhere else.
Electricity Reliability Is as Important as Electricity Price
A short power interruption can disrupt services, corrupt workloads or force systems onto backup infrastructure.
Data centres therefore care about:
- grid reliability;
- redundant supply;
- backup generation;
- uninterruptible power systems;
- power quality.
Cheap electricity that disappears unpredictably can be more expensive than reliable power.
Kazakhstan Generated 123.1 Billion kWh in 2025
The Energy Ministry reported national electricity generation of 123.1 billion kWh in 2025 and targeted 126.5 billion kWh for 2026.
Digital infrastructure is entering that larger electricity system at the same time as industry, households and transport continue to demand power.
Compute therefore competes for a scarce real-world resource.
Renewables Are Growing but the Grid Still Has to Balance
Kazakhstan reported 162 renewable-energy facilities with 3.5 GW of installed capacity and a 7% share of national generation in 2025.
For 2026, renewable generation was targeted at 8.8 billion kWh.
Wind and solar can support new digital infrastructure, but their variable output has to be balanced within the wider grid.
AI cannot run on a slogan.
It needs electrons at the right second.
The 70/30 Model Links New Power Capacity to Data Centres and Digital Miners
In 2026, Kazakhstan introduced a proposed mechanism allowing up to 70% of newly added capacity from certain modernised power plants to be sold directly to major electricity consumers such as data centres and digital miners, with the remaining 30% supplied into the national grid.
The stated objective is to attract private capital into generation while strengthening the wider power system.
This is infrastructure coupling:
large digital users help justify new power investment; new power investment helps enable large digital users.
Digital Mining Is Not the Same Thing as AI Compute
The machines may both consume electricity and occupy data-centre-like facilities.
The workloads are different.
AI compute trains or runs models and performs general high-performance workloads.
Digital mining performs specialised cryptographic computation associated with blockchain networks.
Conflating them creates poor policy analysis.
Same substrate.
Different service.
Kazakhstan Regulates Digital Mining
Kazakhstan’s digital-assets framework requires licensing and reporting for regulated digital-mining activity.
Government resources published updated lists of licensed digital miners and accredited mining pools during 2026.
Tax authorities also carried out compliance checks on electricity consumption used for mining.
This is the institutionalisation stage of an industry that previously expanded faster than regulation.
Why Digital Mining Became an Electricity Question
Mining profitability depends heavily on the cost of electricity.
When large numbers of miners cluster around low-cost power, aggregate demand can become significant for the grid.
That is why regulation increasingly connects digital-mining policy directly to energy policy.
Software economics becomes power-system economics.
The 2026 Digital-Assets Decree Adds a Resource-Efficiency Idea
A 2026 government decree on digital assets included a mechanism allowing associated petroleum gas or natural gas at oilfields to support autonomous electricity generation for digital mining where those resources are not needed for state priorities.
The logic is to use energy that might otherwise be underutilised or difficult to transport.
The important evaluation question is lifecycle efficiency:
does the arrangement create useful economic output while meeting environmental and energy-security requirements?
Compute Location Is a Geography Decision
Data centres care about more than taxes.
They care about:
- electricity;
- fibre;
- cooling conditions;
- land;
- latency;
- regulation;
- skilled labour;
- physical security.
Kazakhstan has abundant land and substantial energy resources.
Its challenge is turning those advantages into reliably connected compute capacity.
Cold Climate Can Help — but Does Not Eliminate Cooling
Kazakhstan’s cold winters can reduce some cooling loads under appropriate facility design.
But modern AI systems create so much concentrated heat that data centres still require engineered thermal-management systems year-round.
Nature can assist infrastructure.
It rarely replaces infrastructure.
Fibre Makes Compute Economically Useful
A supercomputer that cannot move data efficiently becomes isolated.
High-bandwidth connectivity is therefore the other half of the compute economy.
Models need data.
Users need low-latency access.
Cloud services need networks.
A server without fibre is capability trapped in a room.
Astana Hub Creates Demand for the Compute Layer
Kazakhstan’s startup ecosystem increasingly includes AI products.
Read Crazy Rich Kazakhstan | Astana Hub, AI Startups and the Digital Economy.
Startups create demand for GPUs and cloud services.
Compute infrastructure lowers the cost of experimenting.
Better experiments create stronger startups.
Another Ouroboros appears.
The Singapore Comparison: Scarce Land, Dense Compute
Singapore built one of Asia’s major data-centre markets despite severe land and energy constraints.
Read Crazy Rich Singapore | Data Centres, Cloud and AI Infrastructure.
Kazakhstan has abundant land but a different challenge: network density, power-system development and demand concentration.
Same machines.
Opposite geography.
Digital Infrastructure Is a Civilisation Substrate
Government services, banking, logistics, healthcare and education increasingly depend on computation.
That means data centres are becoming comparable to roads and electricity grids.
Most citizens will never enter one.
Many will notice immediately if one stops working.
Invisible infrastructure becomes visible through failure.
What Students Can Learn from Kazakhstan Compute
Computing
Supercomputers introduce parallel processing, GPUs and high-performance workloads.
Physics
Electricity and heat transfer determine how dense computing can operate.
Economics
Data centres show how infrastructure, electricity prices and capital costs shape digital services.
Geography
Location affects latency, power access, fibre and cooling.
English
Useful terms include supercomputer, accelerator, petaflop, latency, workload, redundancy and data centre.
Ten Vocabulary Words for Kazakhstan’s Compute Economy
1. Supercomputer
A computer system designed to perform extremely large numbers of calculations very quickly.
2. Accelerator
Specialised hardware, such as a GPU, designed to speed up particular types of computation.
3. Petaflop
A unit representing one quadrillion floating-point operations per second.
4. Data centre
A facility housing servers, storage, networking and supporting power and cooling systems.
5. Latency
The delay between sending a request and receiving a response.
6. Workload
A computing task or collection of tasks processed by a system.
7. Redundancy
Duplicate capacity or components designed to keep a system operating after a failure.
8. Digital mining
Specialised computation used to validate or support certain blockchain networks under applicable rules.
9. Sovereign compute
Computing infrastructure operated under national jurisdiction and control for strategic or public purposes.
10. Cooling load
The amount of heat that cooling systems must remove to maintain safe operating conditions.
Frequently Asked Questions
What is Alem.Cloud?
Alem.Cloud is Kazakhstan’s national supercomputing infrastructure for AI, machine learning, high-performance computing and digital public services.
How powerful is Alem.Cloud?
Kazakhstan’s National AI Platform currently lists it at 20.48 PFLOP/s Rmax and #104 in the 2026 TOP500 ranking.
What is Al-Farabium?
Al-Farabium is another national high-performance computing system, listed by Kazakhstan’s AI platform at 17.93 PFLOP/s Rmax and #122 on the 2026 TOP500.
Is Kazakhstan building more data centres?
Yes. Government reporting says two centres totalling 7.4 MW entered service in 2025 and three more totalling 12.9 MW were planned for 2026.
Is crypto mining legal in Kazakhstan?
Digital mining operates under a regulated licensing and tax framework. Travellers and investors should use current official regulations rather than older information from the industry’s early boom period.
Why does AI need so much electricity?
High-performance processors perform huge numbers of calculations and generate substantial heat, while data-centre cooling and networking add further electricity demand.
Helpful Reading Across the Kazakhstan and Singapore Graph
- Crazy Rich Kazakhstan | Astana Hub, AI Startups and the Digital Economy
- Crazy Rich Kazakhstan | Nazarbayev University, Research, Engineering and the Talent Economy
- Crazy Rich Singapore | Data Centres, Cloud and AI Infrastructure
- Crazy Rich Singapore | Batteries, Energy Storage and the Grid Flexibility Economy
References and Current Sources
- Kazakhstan National AI Platform, Sovereign computing infrastructure and TOP500 systems, 2026.
- Kazakhstan Ministry of Artificial Intelligence and Digital Development, 2025 digitalisation results and 2026 data-centre plans.
- Kazakhstan Ministry of Energy, 2025 electricity results and 2026 production target.
- Kazakhstan Ministry of Energy, Renewable generation plans for 2026.
- Kazakhstan Ministry of Artificial Intelligence and Digital Development, 70/30 energy-infrastructure model, 2026.
- Government of Kazakhstan, Measures to develop the digital-assets industry, 2026.
Crazy Rich Kazakhstan Is Building the Physical Side of Intelligence
The model is software.
The GPU is hardware.
The rack is infrastructure.
The data centre is real estate.
The grid supplies the electrons.
Did you know? Artificial intelligence only feels weightless when somebody else has already built the machines heavy enough to run it.