Moving servers into colocation in Kazakhstan without downtime means the old site stays in service until the new one has passed its checks. The outage usually comes from a dependency found inside the window: shared DNS, a licence tied to the old hardware, a database still writing outside the country, or one link carrying both users and replication.
This is a working order for a team that brings its own equipment. On the Akashi Data Center campus in Astana the first phase is not accepting servers yet. The first-phase launch is scheduled for May 2027, so the path starts with an audit, a format and a capacity reservation. The route from the request to the rack is in Where to place servers in Kazakhstan.
Audit the infrastructure and its dependencies
The audit answers what has to move together and what can wait. A server list without a map of links will not.
Inventory. Physical servers, virtual machines, storage, network gear, out-of-band management. For each: owner, current location, measured draw, and the date the current site contract ends.
Dependencies. Which services call each other, and which share a disk, a cache or a queue. Mark flows that contain personal data of Kazakhstan citizens and flows that end in a foreign cloud. The legal frame is in the localisation section and in the data localisation guide.
Entry points. DNS names, certificates, IP allow-lists, VPNs, hardware-bound licences, and the people who need physical access. Include maintenance windows already promised to users.
Outage limits. How much data each estate can lose, and how long it can stay down. Those targets are yours. They set the move order.
Some estates switch by copy and a DNS change. Some move only as a set: application, database and queue. Some wait on a dependency. Stateless ones can be stood up early. The output is a move order and a list of what stays powered at the old site until acceptance. Check the map on a typical estate — web, API, database, queue, object storage — and on calls that leave the rack: payments, mail, identity, backup.
Placement format: cabinet, cage or dedicated server room
Choose the format by the access perimeter and by how custom the engineering must be. The campus publishes three.
A cabinet is one lockable rack of up to 20 kW. Server cabinet rental specifies a width of 1200 mm, coded access and dual PDUs. You can start with one cabinet; the commercial range runs past 1,000. That range is commercial. The campus is planned for 4,224 racks across four data-center buildings. For a migration, a cabinet is the smallest placement: one perimeter, a standard size, a stated power limit.
A cage is a section of the server room with its own perimeter, open only to authorised staff, around the clock. Standard and non-standard cabinets of any size fit inside, and power is guaranteed for the cage load. See the private cage. Before shipping you need the access list and the layout, not only the kilowatts.
A dedicated server room is the room under the client’s control. Cooling, access, monitoring and fire suppression follow the client’s requirements, with on-site workspaces and 24/7 on-site support. See the dedicated server room page. Custom engineering is agreed before anyone sets the date the old site goes dark.
Standard servers in standard cabinets — a cabinet. Your own perimeter on a group of racks — a cage. The whole room under your control — a dedicated server room. How reliability class shows up in a contract is covered in Tier III or Tier IV and on the Tier IV page.
Power, cooling and connectivity
Size from measured watts. The label on the power supply is only its ceiling. A request needs sustained draw, peak, and headroom to run on one feed: the second PDU carries the load when the first is down. The published cabinet limit is 20 kW. If the sum reaches that, say so in the request.
Campus power figures are different metrics. The four buildings are stated with 100 MW of dedicated power. That figure is not a confirmed IT load, and it is not capacity already in operation. The planned IT load of phase one is 5.28 MW. The launch is scheduled for May 2027. Expansion toward the full 4,224 racks is marked as 2027 and later. See construction progress.
The engineering the placement relies on: two active/active 110 kV feeds, dual electrical redundancy per building, UPS at 5/4N with up to 10 minutes of battery, generators with 12 hours of autonomy at full load, and two fuel suppliers with a 4-hour delivery SLA.
Cooling follows the same watts. Astana’s climate supports free cooling for much of the year. Published material describes free-cooling and hybrid modes as designed to cut energy use by up to 40% versus a conventional scheme, with a target PUE below 1.4. Those are design targets. The design includes readiness for liquid cooling of high-density racks. The invoice side is in colocation cost structure and green engineering.
Check mass and size against a floor load of 2 tons per m², a 1.6 m raised floor, and two loading terminals with freight elevators, while the document in play is still a specification.
One carrier remains a single point of failure whatever the facility’s tier. The campus has four independent optical entries. Carriers and cross-connect terms are discussed per workload: state where traffic must go and what path independence you need. A team arriving with its own fleet can use the checklist. Site-comparison criteria are in Colocation in Kazakhstan.
Phase-one design documents carry Uptime TCDD Tier IV (Certification of Design Documents). TCCF, the audit of the built facility, is planned for April 2027. TCDD is not TCCF. The calculated availability of the Tier IV topology is 99.995%, by design. It is a topology figure, not a promise for a hall already in operation.
Migration plan, testing and rollback
The window is the short part. Before it, the source stays authoritative. After it, the source stays reachable, so rollback is a return.
- Write the target down. Addresses, networks, access rules, DNS, the people allowed in. Until handover this is a specification and a reservation. See reserving colocation before launch.
- Copy without moving users. Replication, storage sync, a parallel application path. The new copy takes test traffic only.
- Test the work. Login, payment or identity, the overnight job, a restore from backup, one uplink down. A restore you have never run is not a rollback.
- Lower DNS TTL and sign the go/no-go sheet before the window: who may stop the cutover, and on which numbers.
- Switch briefly. Freeze writes, catch the replica up, move the names, watch errors. If a written criterion trips — error rate, lag, a silent dependency — return to the source.
- Leave the source up through acceptance. The team sets that period. Old hardware and the old contract stay in place until it closes.
- Then retire the source and update the paperwork, including the registered address of a personal-data database.
Rehearse on a staging pair now. The first phase is not taking equipment yet, and power-on night in Astana should not be the first time the checklist appears. The site-selection guide puts a corporate deployment cycle at 12–18 months: requirements, hardware lead time, network, compliance. That is a planning horizon, not an Akashi service interval. Move monitoring and a proved restore before application traffic, or the new rack is blind at the switch.
Data localisation requirements in Kazakhstan
If the servers hold personal data, the move is also about where that data is stored.
Law of the Republic of Kazakhstan No. 94-V of 21 May 2013 “On Personal Data and Their Protection” (as amended) separates the place of storage from protection duties. Paragraph 2 of Article 12 states that the owner and/or operator, and a third party, store personal data in a database and/or a digital object located on the territory of the Republic of Kazakhstan. Article 22 sets duties to protect personal data. It does not set the place of storage.
Cross-border transfer is Article 16. It is a transfer to the territory of a foreign state. It is allowed when that state protects personal data. A transfer to a state that does not protect personal data is allowed only in the exceptions the article lists; one of them is the consent of the data subject or their legal representative. The article does not make prior notice to an authority a condition of the transfer.
Inventory of the databases, registration of their location and keeping the facility contract are covered in Data localisation in Kazakhstan. Colocation gives the equipment a physical address in the country. What the data contains, and who the owner or operator is, is for the company to determine.
A consultation and an Akashi capacity reservation
Until May 2027 a conversation with Akashi is a reservation of configuration. The pre-launch pipeline is 107% of DC-1 capacity, with 5.54 MW reserved. The 5.28 MW figure is the planned IT load of phase one. The two describe different things.
Use the contacts page. State the format (cabinet, cage or dedicated server room), rack count, power per rack, any size limit, whether you need a separate perimeter, and the date the equipment must be under power. Attach which estates move as a set and where personal data of Kazakhstan citizens sits. Sales replies within one business day, checks available capacity and sends pricing, timelines and technical detail. An NDA is available on request. Direct line: sales.department@akashi.cloud, +7 700 810 3010.
Common questions
How do you move servers into colocation in Kazakhstan without downtime? The source stays on until the copy has passed its checks. Audit, replication, test, a window with a rollback written in advance, and the source powered down only after acceptance. The first-phase launch is scheduled for May 2027, so the procedure is rehearsed ahead of time and capacity is reserved now.
How do a cabinet, a cage and a dedicated server room differ for a migration? A cabinet is one rack of up to 20 kW. A cage is several racks behind their own perimeter. A dedicated server room is the whole room under client control. The wider the perimeter, the earlier power, access and dimensions are fixed.
How do you size power, cooling and connectivity before the move? From measured draw, with headroom to run on one feed, inside the cabinet limit of 20 kW. Check mass against 2 tons per m² and a 1.6 m raised floor. For links, look at four independent optical entries and the real traffic paths.
What does Kazakhstan’s personal data law require when servers move? Paragraph 2 of Article 12 of Law No. 94-V “On Personal Data and Their Protection” requires personal data to be stored in a database and/or a digital object located in Kazakhstan. Article 22 sets protection duties, not the place of storage. Cross-border transfer is Article 16: to countries that protect personal data, and otherwise only under the exceptions, including the subject’s consent. The law does not make prior notice to an authority a condition.
When can equipment be installed at Akashi, and how is capacity reserved? The first-phase launch is scheduled for May 2027. Until then sales locks a reservation. The DC-1 pipeline is 107% of capacity, with 5.54 MW reserved. The request is on the contacts page.
Need a plan for your own estate? Send a request to sales — we will look at format, power and a reservation ahead of the first phase.