Technology
The infrastructure behind large-scale AI
Every deployment we manage is built around liquid cooling, high-density power, and GPU configurations purpose-designed for AI factory and sovereign cloud workloads. Here is how it works.
Cooling
Liquid cooling, matched to your site
We specify the right cooling approach for each deployment based on facility readiness, power density, and budget. Not every site needs full immersion — we match the solution to the workload.
Direct Liquid Cooling (DLC)
Best for: GPU clusters, AI factory workloads
Coolant is routed directly to cold plates on GPUs and CPUs. Highest thermal efficiency, compatible with GB300 and H100/H200 configurations.
Rear-Door Heat Exchangers
Best for: Mixed workloads, retrofit deployments
A liquid heat exchanger replaces the standard rear door, capturing heat before it enters the room. Works alongside existing air-cooled infrastructure.
Full Immersion Cooling
Best for: Ultra-high density, maximum PUE
Servers submerged in dielectric fluid. Best PUE of any cooling method. Requires custom facility infrastructure and specific hardware compatibility.
Air-Assisted Hybrid
Best for: Entry deployments, standard colocation
Precision air cooling supplemented with in-row coolers. Suitable for facilities not yet liquid-ready, or for staging before full DLC conversion.
Compute
GPU configurations we deploy
We support the full NVIDIA Blackwell stack from GB200 through GB300, plus H100 and H200 where price or site constraints apply.
GB300 NVL72
72× Blackwell Ultra GPUsHighest compute density available. Full NVLink all-to-all communication across 72 GPUs in a single rack.
Power
~130kW per rack
Cooling
DLC required
Fabric
NVLink 5th gen
GB200 NVL72
72× Blackwell GPUsProduction-proven configuration. Strong availability across our APAC and Middle East partner sites.
Power
~120kW per rack
Cooling
DLC required
Fabric
NVLink 4th gen
GB200 NVL36
36× Blackwell GPUsHalf-rack format. Suitable for sites with existing rear-door heat exchangers or partial DLC infrastructure.
Power
~60kW per rack
Cooling
DLC or RDHx
Fabric
NVLink 4th gen
H100 / H200 SXM
8× GPUs per serverPrevious generation, widely available. Strong option for price-sensitive deployments or air-cooled colocation sites.
Power
10–14kW per server
Cooling
Air or DLC
Fabric
InfiniBand NDR
Process
From brief to first GPU online
A typical GB300 deployment takes 30–90 days from site selection to a validated, running cluster. Here is what happens at each stage.
Site qualification
We assess power capacity, grid connection SLAs, structural loading, and cooling infrastructure against your workload requirements.
Infrastructure design
Our engineers specify cooling topology, power architecture (PDU, busbar, UPS, generator), and network fabric design for your GPU configuration.
Procurement and staging
Hardware is sourced, staged, and tested off-site before shipment. GPU cluster firmware and baseline configuration is validated at this stage.
Installation and commissioning
On-site installation of cooling infrastructure, power systems, and GPU racks. Commissioning includes thermal validation, burn-in, and fabric verification.
Handover and operations
We hand over a running cluster with monitoring, remote hands, and optional ongoing operations support through our local partner teams.
Get started
Ready for a partner who stays in the deal?
Submit a requirement and we build a priced, structured proposal — or register your facility and we bring it to qualified demand. Either way, we're in it with you.
Or browse anonymized listings without an account.