The data centre.
Colocation means you own the servers and we own everything around them: the building, the power, the cooling, the network connection and the security. You rent capacity, measured in kilowatts, and you pay for the electricity your equipment actually uses. It is the standard way organisations run their own hardware without running their own facility.
What we provide
Space
Rack positions in a secured, access-controlled enclosure. Standard 19-inch racks; full or partial racks depending on your load.
Power
A metered electrical feed to your rack at the capacity you contract, supplied from a renewable grid. Distribution, protection and monitoring are ours.
Cooling
Liquid cooling to the rack. Your servers are cooled by water, not air, which is what allows their heat to be recovered. Servers that are not liquid-cooled can be fitted with rear-door heat exchangers.
Connectivity
Fibre connection to the site with carrier options; cross-connects to your own provider on request. Latency to Europe is the same as for any Icelandic facility.
Operations
Monitoring around the clock, remote hands for physical tasks, and a named contact. Hardware installation and replacement on your instruction.
Removal attribution
The heat from your equipment is measured, and the carbon removal it supports is calculated from your energy use and documented to you. No other colocation offers this.
What you bring
Your servers, storage and network equipment; your software, licences and data; remote management of your own systems. You decide what runs, we keep it running.
How power is measured.
Two figures matter. IT load is the electricity your equipment draws, measured at the rack in kilowatts. Facility power is IT load plus the power the building needs for cooling, distribution and losses. The ratio between them is the facility's PUE, power usage effectiveness. A PUE of 1.2 means that for every kilowatt your servers use, the site draws 1.2 kilowatts in total. Liquid-cooled facilities in cold climates typically achieve a PUE between 1.1 and 1.2; air-cooled facilities in warmer places are often above 1.5. Your bill is based on IT load; the difference is our cost to keep low.
Density.
A rack of general-purpose servers typically draws 5 to 15 kW. A rack of GPU servers for AI training draws 30 to 100 kW or more, which is beyond what air cooling can handle and is the reason modern AI facilities are moving to liquid cooling. Nordvar racks are liquid-cooled by default and designed for high density.
Partial rack
- Standard 19-inch racks; full or partial racks depending on your load.
- A metered electrical feed at the capacity you contract.
Full rack, general-purpose
- Liquid cooling to the rack; heat recovered as warm water.
High-density GPU racks
- Beyond what air cooling can handle.
- Liquid-cooled by default, designed for high density.
A worked example.
Take one rack with an IT load of 10 kW running continuously. Over a year it uses 10 kW × 8,760 hours, about 87,600 kWh of electricity, and produces the same amount of heat. At a PUE of 1.2, the facility draws about 105,000 kWh to support it. About 85% of the rack's heat, roughly 75,000 kWh, leaves it as warm water and is available to the capture process; the rest escapes to the room air and is rejected by the facility cooling.
- IT load, continuous
- 10 kW
- Hours in a year
- 8,760
- Electricity used
- about 87,600 kWh
- Facility draw at a PUE of 1.2
- about 105,000 kWh
- Heat available as warm water
- roughly 75,000 kWh (about 85%)
What colocation costs.
Colocation is priced in three parts. A capacity fee per kilowatt per month for the space, cooling, connectivity and operations. Electricity at the metered rate, passed through from the supplier with no margin or a small handling fee. A one-time setup fee for installation. Industry ranges in 2026: capacity fees of roughly USD 100 to 200 per kW per month for standard-density retail colocation in Europe, and roughly USD 150 to 250 per kW per month for liquid-cooled high-density racks. Electricity prices vary widely by country; Iceland's industrial rates have historically been among the lowest in Europe and are not exposed to gas prices. Contracts usually run 12 to 36 months with a committed minimum load.
Nordvar's rates will be published when first-phase capacity is quoted. Expect the capacity fee to sit within the industry ranges above, with electricity passed through at cost and the removal attribution included, not charged separately. Tenants who can offer flexible operation, pausing or shifting workloads on request, are offered a lower capacity fee in return.
If you do not have hardware.
Not every organisation wants to own servers. For tenants who prefer to rent computing rather than space, Nordvar can supply and operate the hardware and charge for compute instead of kilowatts. The heat recovery and the removal attribution work the same way. Ask which model fits when you write to us.
Helpful to have ready when you write
None of this is required to get in touch - it simply lets us give you a real answer sooner.
- Your expected IT load in kW, now and in two years
- Whether your hardware is liquid-cooled, air-cooled, or not yet bought
- Latency and connectivity requirements
- Preferred contract length
- Whether your workloads can pause or shift in time
How we build.
Nordvar is built in phases, and each phase passes through gates. A gate is a written decision, made on documents, with criteria fixed in advance. Passing a gate releases the spending of the next stage. Failing one stops the work.
The first build is small
The first phase is a pilot liquid-cooled module on a leased plot - the smallest unit that is truly a data centre - then a building fitted out in stages as tenants sign. The pilot is small enough to move if a site falls through, and real enough to produce operating data: power, heat, removal. Each later stage repeats what the previous one proved.
The five gates.
- Gate 01
Concept confirmed
Written answers exist that power, site and demand are plausible. Nothing is spent beyond enquiries until they do.
- Gate 02
Site and tenant
A specific plot, a written indication from the grid, and a tenant's written interest in defined capacity.
- Gate 03
Financing committed
The money for the first phase exists in writing. Until it does: no lease, no orders.
- Gate 04
Ready to build
Permits granted, the connection date and delivery dates confirmed in writing, the tenant signed. Only then does building start.
- Gate 05
In operation
The site energised, the tenant live, metering and heat data flowing. The next phase is planned on what this one measured.
What a gate cannot do
A gate cannot be argued past. If the criteria are not met, the work pauses, and the plan is revised on what was learned. A paused project with letters and data is an asset; a half-built one is a liability.
Build the smallest thing that proves the next thing.
Planned · decided site by site, not yet confirmed
Built into the land.
The intent is that the building borrows its surfaces from the landscape it stands in instead of standing against it. This is planning, not commitment - it is fixed in the landscape plan for the final site, and nothing here is confirmed until that plan is.
A roof of native moss
The roof planted with moss raised from native stock, so from above the building reads as ground and weathers the way the land does.
Concrete that behaves like basalt
Facades in textured concrete, coated black - taking the colour and grain of the rock around the site rather than standing against it.
Paths of black sand
Roads and paths surfaced in black sand, keeping the ground plane the colour it already was.
How much of this a site carries depends on the environment it is placed in. Direct air capture wants open, moving air around it, so a wide, open plot suits the capture unit and may keep its surroundings sparse; a more sheltered site may take more of the planting.
Dates are not published: the dates that bind belong to counterparties and do not exist until they are in writing. Progress is shared directly with the people involved rather than announced here.