01
MDC500 · AI INFERENCE · 300–500 KW IT LOAD
Build the infrastructure for your inference servers.
MDC500 brings power, cooling and IT space together for NVIDIA H200 and RTX PRO 6000 Blackwell inference systems. Select the configuration around your servers and site.
- 300–500 kW
- Target IT-load range
- 10 × 42U
- Air-cooled reference rack positions
- Up to 50 kW
- Per rack in the air-cooled reference
- 3
- Technical modules in the issued air-cooled design
- H200 / RTX PRO 6000
- Target inference equipment
Match the facility to the inference workload.
The selected servers determine rack power, airflow, network and service access. MDC500 coordinates those requirements in a modular facility.
The issued air-cooled design supports ten racks and up to 500 kW IT load. An immersion configuration needs a separate review of the server and cooling interfaces.
Run H200 and RTX PRO 6000 Blackwell inference systems.
Start with the server model and deployment plan. Rack geometry, power and cooling limits are checked against the chosen equipment.
Applications can include:
02
Engineering and scientific HPC
03
Rendering and visualisation
04
Dense private cloud and virtualised infrastructure
05
Storage and data-processing clusters
For larger rack-scale AI clusters, compare MDC2000.
Choose the cooling around the server configuration.
Air cooling
Air-cooled servers with rear-door heat exchange in the issued reference design.
Immersion by project
Project-specific option, subject to server compatibility and a dedicated thermal and service design.
- IT intake air
- servers
- hot rack exhaust
- active rear-door heat exchanger
- closed chilled-water loop
- external heat rejection
01
The servers remain air cooled.
IT equipment draws conditioned air through the rack.
02
The rear door captures exhaust heat.
Active fan modules move hot server exhaust across a chilled-water coil.
03
The closed loop carries the heat away.
Pumps, buffer capacity, isolation valves, filtration, instrumentation and controls circulate chilled water.
04
The external plant rejects the heat.
Three 200 kW modular air-cooled chillers stage with load and use free cooling when ambient conditions allow.
The issued design keeps servers air cooled and uses rear-door heat exchangers to move exhaust heat into a closed cooling loop. Immersion is a separate project option; it requires compatible equipment and its own cooling and service design.
The air-cooled reference: three coordinated zones.
The issued design separates the power equipment, ten-rack IT area and cooling services. Its component ratings and geometry apply to that configuration.
- 01
Electrical & energy zone
Switchgear, scalable modular UPS, batteries and AC distribution. Multiple utility or generator inputs and independent A/B rack paths are coordinated to the project power design.
- 02
IT zone
Ten 42U rack positions, each 800 × 1,200 mm, with active rear-door heat exchangers and independent rack distribution.
- 03
Thermal & services zone
Hydronic equipment, filtration, controls, fire-protection interfaces, security and technical access.
The project defines redundancy, fire protection, access and the acceptance tests for the selected configuration.
Add inference capacity as demand grows.
Repeated 500 kW reference units provide a route to staged expansion. Plan shared power, cooling and connectivity with the first deployment.
| Capacity plan | MDC500 units | Rack positions |
|---|---|---|
| 500 kW | 1 | 10 |
| 1 MW | 2 | 20 |
| 2 MW | 4 | 40 |
| Multi-MW | Repeat standard units | +10 per unit |
The table shows repeated units of the issued air-cooled design. Other rack populations and immersion configurations follow their own engineering and capacity plan.
A/B power at the rack. Resilience designed end to end.
MDC500 supports multiple utility or generator inputs, online double-conversion UPS, scalable power modules and independent A/B rack delivery.
The project decides how those elements work together. A/B rack feeds alone do not establish 2N, concurrent maintainability or Tier certification. The design has to trace every path from source to rack.
Use the climate to reduce cooling energy.
The air-cooled reference uses modular chillers with free-cooling capability. When the ambient conditions allow, free cooling reduces compressor operation.
We model seasonal performance for the site climate, IT load, cooling temperatures and controls.
Integrate before the site gets busy.
- 01
Define the rack population.
Hardware, dimensions, peak loads, airflow and growth.
- 02
Freeze the interfaces.
Power, chilled water, heat rejection, controls, fire protection, network, civil works and site access.
- 03
Assemble the selected configuration.
Integrate the agreed electrical, cooling and control systems in the factory.
- 04
Complete factory acceptance testing (FAT).
Execute the agreed inspection and functional test scope.
- 05
Install at site.
Position and interconnect the modules and external plant.
- 06
Complete SAT and commissioning.
Retest transported connections and the assembled facility before handover.
Factory work and site preparation can run in parallel. The programme still depends on equipment availability, site readiness, transport, permits and final configuration.
Issued air-cooled reference configuration
- Reference IT-load capacity
- 500 kW
- Rack configuration
- 10 × 42U positions, 800 × 1,200 mm
- Maximum rack load
- Up to 50 kW
- Architecture
- 3 transportable technical modules operating as one system
- Rack cooling
- Active rear-door heat exchangers; IT remains air cooled
- Water system
- Closed-loop chilled-water distribution
- Heat rejection
- 3 × 200 kW modular air-cooled chillers with free-cooling capability
- Power
- Multiple utility/generator inputs; online double-conversion modular UPS; independent A/B rack paths
- Batteries
- Lithium-ion option
- Redundancy
- Configurable 2N or N+1 for critical systems, subject to project design
- Fire resistance
- E60 / I60
- Expansion
- Repeatable 500 kW increments
- Acceptance
- Factory pre-integration and testing, followed by site acceptance and commissioning
What the site must provide
- Confirmed power sources and upstream protection
- Foundations for the selected modules and external heat-rejection equipment
- Delivery, crane and maintenance access
- Fibre and communications
- Hydraulic and electrical interconnection routes
- Ambient and environmental design conditions
- Drainage, water-quality and freeze-protection strategy where applicable
- Local fire, security and permitting coordination
Frequently asked questions
Which product fits my AI servers?
MDC60 is a two-rack, 60 kW air-cooled option for B200-class servers. MDC500 targets 300–500 kW inference deployments using H200 or RTX PRO 6000 Blackwell systems. MDC2000 targets larger inference and training clusters, with a GB300 NVL72 reference configuration. The exact server model, rack power and cooling interface determine the fit.
Is the equipment cooled by air or liquid?
Conventional CPU and storage systems and MDC60 use air cooling. MDC500 has air-cooled and proposed immersion configurations, selected by project. MDC2000 uses an approximately 85% direct liquid / 15% air cooling split. MDC4000 is planned for 100% direct liquid cooling. Compatibility follows the exact server configuration.
Do you supply the infrastructure or the servers as well?
Our core offer is the data-center infrastructure. We can plan integration around customer-supplied equipment or discuss an equipment package in the project scope. The proposal states who procures and supplies the servers, GPUs, networking and storage; they are not automatically included.
Does liquid cooling need a permanent water connection?
Liquid circulates in closed cooling loops. Water consumption depends on the outdoor heat-rejection system. MDC2000's dry-first approach can use stored, treated water for adiabatic assistance in hot conditions, without a permanent utility-water connection. Storage, annual use and refill arrangements follow the site climate.
Can I start small and add capacity?
Yes. Capacity can be phased around equipment deployment and demand. We plan the first module together with the power, cooling, fibre and access needed for later stages. Expansion work and any effect on the operating facility are defined in that plan.
How long does delivery take?
We agree a programme for design, procurement, factory assembly and testing, transport, installation and commissioning. Factory work and site preparation can run in parallel. The quotation separates those stages and identifies the equipment and site dependencies that affect the date.
What does a modular data center cost?
Price follows the equipment, IT load, cooling, resilience and delivery scope. Site works and computing hardware can change the budget substantially. Send the location and intended capacity so we can define the scope before quoting.
Tell us what your inference deployment needs.
Send the server configuration, intended IT load and location. We will assess the product fit and cooling option.