The question “where do we put our servers in Kazakhstan” almost always starts with the price per rack. In practice price turns out to be the last thing that matters: if a site fails on reliability class, kilowatts or timelines, there is nothing left to compare on cost.
Below is the framework corporate teams actually use. Not a directory of every operator in the country — the criteria apply to any facility, ours included.
Context: why the market is tight
Three things happened at once. Data-localization rules removed “a cloud region somewhere nearby” from the shortlist. Regional expansion by Chinese and Middle Eastern IT companies created demand for owned hardware in Central Asia. AI workloads pushed rack density from a comfortable 5 kW to 20 kW and beyond — older halls are not physically built for that.
The result: fewer commercial racks of the right class than the demand being voiced, and those racks move into long-term contracts before the buildings are commissioned.
Step 1. Reliability class
This is the first filter, because it cannot be changed after the building is up.
| Class | What it means for the workload |
|---|---|
| Tier II–III | engineering can be maintained without a full stop, but a failure of individual elements can still reach IT |
| Tier IV | the failure of any single element does not stop the IT load; designed availability 99.995% |
For mission-critical systems, financial services and regulated workloads the benchmark is Tier IV. Distinguish design certification (TCDD) from certification of the constructed facility (TCCF): the first validates documentation, the second validates what was actually built.
Step 2. Kilowatts, not square meters
A rack is sold by power, not by floor area. Check:
- how many kW per rack are guaranteed, and how that is written into the contract;
- whether your density is supported without “dilution” — that is, without leaving neighbouring racks empty;
- whether liquid cooling or hot-aisle containment is available if density will grow;
- the power redundancy scheme: number of independent feeds, UPS redundancy class, generator autonomy.
Step 3. Connectivity
A single carrier is a single point of failure, no matter how the hall is engineered. Look at the number of physically independent optical entries, how many carriers are present on site, the cost and lead time of cross-connects, and the routes to your users and partners.
Step 4. The legal contour
Personal data of Kazakhstan citizens is processed and stored in-country. Colocation gives what a foreign cloud cannot: a specific physical address for the hardware, a defined access perimeter, and the ability to audit who approached your rack, when, and on what grounds.
Check the equipment import regime separately — customs clearance times and certification requirements feed straight into your go-live date.
Step 5. Horizon and reservation
The corporate cycle is 12–18 months. On that horizon you should compare not only what is free today, but what can be locked in advance: power, density, room-readiness dates, expansion terms.
Questions to ask the operator
- Which reliability class, and which certifications are already obtained versus planned?
- How many kW per rack are contractually guaranteed, and under which cooling scheme?
- How many independent optical entries, and which carriers are present on site?
- What is the power redundancy scheme and generator autonomy at full load?
- How is a capacity reservation formalised before commissioning, and what happens if dates slip?
- What are the SLAs, the compensation mechanics, and the 24/7 remote hands procedure?
- How are access control, video surveillance and visit logging organised?
How this looks at Akashi
Akashi Data Center is being built in Astana as Central Asia’s first commercial Tier IV facility: 4,224 racks across four buildings on 11 hectares, 100 MW of IT capacity, designed availability of 99.995%.
The engineering contour: two independent 110 kV feeds in an active/active scheme, UPS with 5/4N redundancy, 12 hours of generator autonomy at full load, and four independent optical entries onto the East–West digital corridor. Density runs up to 20 kW per rack, with liquid cooling for high-density configurations. Free cooling and hybrid modes deliver up to 40% energy savings; the facility is designed to ISO 14001.
The first phase launches in 2027, with the TCCF certification audit scheduled for April 2027. Pre-launch commitments for the first phase already exceed its capacity — which is why placement is discussed as a reservation rather than as free racks.
Frequently asked questions
What is colocation and how is it different from renting a server? Colocation means placing your own hardware in someone else’s data center. You own the servers and control the configuration; the facility provides power, cooling, security and network paths. With rented hardware the provider owns the machine and your configuration is limited to their catalogue.
Does a foreign company have to keep its data inside Kazakhstan? Personal data of Kazakhstan citizens must be processed and stored in-country. A foreign cloud does not close that requirement; colocation gives the hardware a physical address inside the jurisdiction and an auditable access path.
How much power per rack do CPU servers need? Classic CPU racks fit within 5–10 kW; dense and mixed CPU/GPU configurations need 15–20 kW and above. Akashi is designed for densities up to 20 kW per rack.
Why reserve capacity in a facility still under construction? The deployment cycle runs 12–18 months, and free racks of the right class are contracted before hardware arrives. A reservation locks power, density and dates.
What does Tier IV mean in practice? The failure of any single element of the engineering infrastructure does not stop the IT load, and maintenance happens without taking client equipment offline. Designed availability is 99.995% — about 26 minutes per year.
Planning to place your own servers in Kazakhstan? Contact sales — we will discuss power, reservation timing, and connectivity requirements.