Move your own GPUs straight into a high-density data center
Owning the hardware but finding no data center that can host it is the most common obstacle to deploying high-density GPUs. Servers placed in a KONST data center get a rack, power, cooling, and network. Equipment ownership does not change, KONST collects and pays the data center and telecom fees on the owner's behalf, and the owner receives one combined bill each month.
Specs do not match,
so high-density equipment cannot find a data center
Power per rack, air intake conditions, rack weight capacity, and power distribution type must each be checked against the equipment specs. A mismatch on any one of them usually means reconfiguring entire rows of racks or the site itself, and adjusting settings will not solve it.
Power per Rack
At 20kW or more per rack, matching power distribution capacity and cold aisle containment are required.
Cooling requirements
Intake air temperature and airflow must match; liquid-cooled models also need water piping and CDU space.
Rack requirements
Rack depth and load capacity must be sufficient to support the equipment.
Power distribution type
The number of power phases and the outlet type must match the equipment.
Six colocation items in one contract
Racks, power, cooling, network, cross-connects, and electrical and mechanical maintenance are covered by a single contract. Data center and telecom fees are collected and paid together, so the owner reviews one contract and receives one bill.
Rack space
DetailsCapacity reserved under the contract, including racks, cabling, and installation space
Power
DetailsUtility power, UPS backup, and generators, with distribution paths configured by redundancy level
Cooling
DetailsAir cooling or liquid cooling, selected by power density per rack
Network
DetailsExternal bandwidth over dedicated overseas high-speed telecom network lines
Cross-connects
DetailsPhysical connections within the site to other tenants, telecom carriers, or cloud services
Electrical and mechanical maintenance
DetailsRegular inspection of power and cooling equipment
Overseas lines for external links,
with capacity and power billed separately
External connectivity uses dedicated overseas high-speed telecom lines, and within the site you can cross-connect directly to other tenants or cloud services. Billing has two parts: space is priced by the capacity reserved in the contract, and power is passed through at actual consumption.
Network and cross-connects
External links use dedicated overseas high-speed telecom lines, which suit cross-border transfer. In-site cross-connects let equipment connect directly to other tenants and to telecom or cloud service access points without going over the public internet.
Billing method
Space is billed by the capacity reserved in the contract, regardless of actual power use. Power is passed through at actual consumption.
A three-way split of responsibility among the data center, colocation, and operations
The data center operator is responsible for the site and for electrical and mechanical systems. KONST is responsible for securing capacity, coordinating move-in, and settling fees. Equipment monitoring, troubleshooting, and manufacturer RMA belong to compute operations, which is charged separately but can be handled by the same team.
Handled by the data center operator
Building the data center itself is within the scope of the AI compute infrastructure build service.
Handled by KONST
KONST handles securing capacity, move-in coordination, incident tracking, and fee settlement, so the owner does not need to track progress or accounts.
Compute operations, charged separately
Equipment monitoring, troubleshooting, and manufacturer RMA coordination are not included in colocation. The same team can take them on together when needed.
Four-stage move-in
A site assessment comes first: the equipment list is checked against data center conditions, and rack count and power are estimated. After confirmation, a contract is signed that sets capacity, term, and billing basis. Next come racking, power distribution, and cabling. Finally, the equipment is powered on and tested for connectivity, and it goes live once it passes.
- 01Site assessment
Check the equipment list against data center conditions, and estimate how many racks fit and how much power is available.
- 02Contract signing
Sign the colocation contract after agreeing on capacity, term, and billing basis.
- 03Move-in work
Equipment arrives, is racked, and is connected to power and network.
- 04Power-on testing
Power on and test connectivity, then go live once everything is confirmed.
Our self-operated data center can host racks from 20 kW
The self-built Taichung data center has 24 racks in four rows at 20 kW per rack, cooled by in-row air conditioning with a chilled water system, and is a working example in this density range. New Taipei has a further 1.5 MW of incoming power, which can accommodate new-generation high-density models.
FAQ
What is KONST's advantage in AIDC?
How do colocation requirements differ between GPU racks and standard server racks?
How is the electricity charge calculated for colocation?
Can I arrange the data center before the equipment is purchased?
What certifications and safeguards do KONST's data centers have?
Where does KONST have sites?
Services often evaluated together
Bought your own GPUs but have not found a data center?
Send us your equipment specs, and we will help check site conditions and the move-in schedule.