Need to Talk to a Real Person? Call us at 877.378.3769.
Unit 03 · Colocation
You own the hardware.
We own the building problem.
A general-purpose data hall is engineered for racks drawing single- digit kilowatts. A populated GPU rack draws several times that, and a current-generation liquid-cooled node several times more again.
The Constraint
Most Facilities Physically Cannot Host a Modern GPU Cluster
This is not a service-level problem or a willingness problem. It's electrical and thermal. A hall built around a few kilowatts per rack cannot be persuaded into carrying ten times that, and the retrofit is measured in millions and quarters.
The density gap is an order of magnitude
Traditional halls are designed around a handful of kilowatts per rack. AI deployments start around fifteen to twenty-five and extreme- density liquid-cooled configurations run past a hundred. Nothing about that gap closes with better airflow management.
01
Air cooling runs out before the silicon does
Beyond a certain heat density, moving enough air becomes physically impractical and acoustically absurd. Direct-to- chip liquid removes heat at the source, which is why current reference architectures assume it rather than offer it.
02
Power availability is the real scarcity
In constrained markets the binding constraint isn't floor space or even cooling — it's whether the utility feed and switchgear can deliver contracted megawatts at all. Capacity has to be secured ahead of need. 03
How we do it
From Assessment to a Powered,
Cabled Cage.
Deployment planning starts with power and thermal reality and works outward. Everything else — cabinets, containment, cabling, cross-connects — follows from those numbers.
Power, thermal and space assessment
We work from your actual node specifications: draw per chassis, heat rejected, inlet requirements, weight per rack and the growth curve you expect over the term. This produces a real power and cooling envelope rather than a rack count.
OUTPUT POWER AND THERMAL ENVELOPE
Facility and site selection
Requirements are matched against available capacity — feed headroom, cooling type, redundancy topology, carrier density and proximity to where your data and your people already are. Latency to your users and your existing estate is part
OUTPUT SITE RECOMMENDATION WITH CAPACITY CONFIRMED
Cooling design
Direct-to-chip loops, rear-door heat exchangers or hybrid liquid-air are selected against your node design and density, then engineered with the coolant distribution, containment and redundancy the configuration requires.
OUTPUT COOLING DESIGN AND THERMAL MODEL
Cage, cabinet and containment build
Physical build-out to the design: cabinets and rails matched to your chassis, structured cabling and fibre pathways, power distribution to the correct redundancy level, containment, and physical security boundaries.
OUTPUT BUILT CAGE READY TO RECEIVE
Receive, rack and interconnect
We take delivery, handle the rigging that heavy GPU nodes genuinely require, rack to plan, terminate and label, then establish cross-connects and transit. Cabling is documented as built, not as drawn.
OUTPUT RACKED, CABLED, DOCUMENTED
Commissioning and handover
Power-on under load, thermal validation at design density, connectivity verification, then handover with as-built documentation, access procedures and escalation paths. Ongoing hands coverage starts here.
OUTPUT COMMISSIONING REPORT AND ACCESS LIVE
What's included
What Sits Inside a
Colocation Contract.
Contracted Power.
Committed kW per rack with defined redundancy, not best-effort capacity.
Liquid Cooling.
Direct-to-chip and hybrid liquid-air where density requires it.
Space and containment.
Cabinets, cages, private suites and dedicated halls.
Physical Security.
Multi-layer access control, monitoring and audited entry.
Compliance Posture.
Facilities operated to recognized audit and reporting standards.
Carrier Density.
Multiple transit options plus cross-connects to clouds and peers.
Remote and Smart Hands.
On-site engineering that can diagnose, not only power cycle.
Spares Storage.
Secured on-site space for your spares kit, adjacent to the deployment.
Receiving and Rigging.
Delivery acceptance and handling for heavy, awkward GPU nodes.
As-built documentation.
Cabling maps, power schedules and asset placement kept current.
Density tiers
What the Facility
has to do Changes
with Density.
These tiers describe engineering regimes, not price bands. The jump
between them is where most deployments discover their existing facility
can't follow them.
Standard IT
AI baseline
High
Density
Extreme
Density
Questions
Before You
Commit
No — and the distinction matters commercially. In colocation you own the accelerators outright; they sit on your balance sheet and you carry the depreciation and the residual value. We supply the facility, the power and cooling, the connectivity and the hands. If you'd rather not own hardware at all, rental is a different product with different economics, and we'd point you there instead of blurring the two.
Yes. We regularly accept shipments directly from OEMs and distributors, inventory and tag the hardware, and complete the installation and cabling according to your specifications.
Our on-site Smart Hands team handles the physical swap and RMA dispatch with your vendor, while you maintain direct software and management control.
Yes. We can reserve adjacent contiguous cabinet footprint and power headroom to ensure seamless scaling as your cluster expands.
You can renew your contract, upgrade to newer cages/power densities, or have our team securely decommission, pack, and ship your hardware back to your designated location.
Send us your node specs.
We'll tell you what the facility has to do.
A power and thermal envelope, a site recommendation, and confirmed available capacity before you commit to anything.
Talk to an engineer
The Rest of the Stack
Six Units. One
Assembled Stack.
GPU Procurement
GPU Financing
GPU Colocation
IB & RoCE Networking
High Performance Storage
GPU Management
Note on figures. Density tiers and per-rack capacities describe engineering regimes and vary by facility, cooling design and contracted redundancy. Available capacity is site-specific and confirmed per deployment.