Rent dedicated compute
at a lower cost than cloud services
For customers whose needs are settled and who do not build their own data center. You can rent GPU bare metal in the data center, delivered remotely over a dedicated VPN, or move equipment you bought yourself into racks for colocation. Contracts run by the year, capacity is reserved exclusively, and unit cost is lower than cloud services.
The workload is clear, but data center and staffing remain gaps
The workload size can already be estimated. What is missing is a site that can carry it: power distribution and cooling for high-power racks, 24-hour on-site staffing, and site checks before equipment moves in. The fixed investment in these three is out of proportion to the workload size.
Two ways to rent,
delivering compute and data center space separately
One route rents GPU servers in the data center, with compute accessed remotely over a dedicated VPN. The other leases racks and power, the lessee's equipment moves in, and KONST collects and pays the fees on the lessee's behalf.
- Problem
Building a whole data center
for a defined workload is a disproportionate investment The fixed investment in a whole data center does not shrink with workload size, so unit cost runs high.
SolutionDeliver usage rights to a full batch of GPU servers
Rent GPU equipment in the data center. The lessee holds the usage rights to the whole batch of servers and can put them to work on compute right away.
- Problem
High-density models
need a high-power data center and on-site staff Once equipment arrives, you still need a data center with matching power capacity and 24-hour on-site staff.
SolutionSelf-purchased equipment moves in,
and KONST provides racks and powerThe lessee moves its own purchased equipment into the data center, and KONST provides racks, power, and network, collecting and paying all fees on the lessee's behalf.
- 01Enterprise workload
Confirm the GPU model, capacity, and lease term.
- 02Dedicated access
Use compute remotely over a dedicated VPN.
- 03GPU bare metal
Get usage rights to dedicated servers.
Bare-metal rental and colocation,
with different asset ownership
with different asset ownership
The two modes differ in who owns the equipment. Equipment that KONST buys goes through bare-metal rental, and equipment the owner already has goes through colocation, so responsibilities and billing differ accordingly.
Bare-metal rental: the lessee gets usage rights only
KONST buys and operates the servers, high-speed networking, and storage, and the lessee gets usage rights. When the contract ends, the equipment stays with KONST, and the lessee does not bear equipment replacement or asset disposal.
Colocation: assets registered in the owner's name
The owner buys the equipment, and the assets are registered in the owner's name. KONST handles racks, power, and network, plus liaison and fee settlement with the data center operator. The owner decides on equipment warranty, replacement, and expansion.
The longer the term, the lower the unit cost
Long contracts for continuous loads cost less than usage-based billing
For the same GPU model, the annual-term unit price is lower than usage-based billing, and the gap grows with longer terms.
Data stays in a known location,
with an auditable path
Physical machines are not shared with other tenants, data moves over a dedicated line and a dedicated VPN, and the path can be written into audit documents.
Equipment already purchased,
site environment and operations staff still needed
The equipment has been bought and is in place, and what is missing is a hosting environment and operations staff. In this case colocation fits better than renting equipment.
Bare metal per GPU per hour,
colocation itemized
Bare metal is priced per GPU per hour, settled monthly, and uses take-or-pay. Colocation bills space, power, and network as separate items, with space charged on reserved capacity and not varying with power use. Unit prices are calculated by GPU model, quantity, and term.
Settlement Monthly, and most contracts include take-or-pay
Power Per the contract, with the actual unit price calculated by GPU model
Network Delivered remotely over a dedicated VPN
Idle periods Charged at the agreed capacity
Settlement Space is billed on reserved capacity and does not vary with power use
Power Passed through at actual consumption
Network Cross-connects and dedicated lines charged separately
Idle periods Space and minimum power are still charged
FAQ
How should I choose between rental and cloud services?
If the GPUs are already purchased and only a site to house the equipment is needed, is that possible?
Under take-or-pay terms, is unused capacity still charged?
How is colocation electricity calculated? Is there still a charge while equipment sits idle?
When bare metal is leased, who owns the equipment after the contract ends?
Services often evaluated together
Looking for stable long-term compute capacity?
Describe your workload and scale, and we will reply with the GPU models available to rent and the timeline.