XRT-AI Rack Nodes, Halls & Campuses | UPS Solutions
20+ years • 100,000+ UPS units supported • 24/7 support
AI Data Centres, Delivered Complete
You Install the GPU Nodes. Nothing Else.
An AI cluster is not a rack of servers with a bigger power bill. It is a thermal and electrical problem that most existing rooms cannot solve, because the racks arrive needing 40 kW to well past 100 kW each and the building was designed for 5. The usual answer is a construction project measured in quarters.
XRT-AI removes that. The enclosure, liquid cooling, UPS, batteries, distribution, racks, monitoring, access control and fire suppression are installed and load tested in the factory, and the unit ships as a finished facility. Site works reduce to footings, a supply connection, a data path and commissioning. Your team racks the GPU nodes and nothing else.
Four deployment forms cover the whole range: a single rack node for a pilot or an edge site, containerised training halls at 400 to 800 kW, and multi-pod prefabricated campuses where compute, power, network and operations each get their own module.
Scoping an AI deployment?
- Free heat-load assessment
- Cooling method selection
- Factory acceptance testing
- Install, commission & support
XRT-AI-R90-DLC rack node
XRT-AI-C800-IMX immersion hall
AI Has Outgrown Air Cooling
Liquid carries roughly four thousand times more heat per unit volume than air. Beyond a certain density you stop cooling servers and start moving air at velocities that cost more than the compute. Drag the density, or select a technology, to see where each one stops working.
Cooling Technology Coverage by Rack Density
Every XRT-AI configuration is built on one of these. The scale is logarithmic, because the useful range spans two orders of magnitude. Colour shows what actually carries the heat away.
kW per rack
Four Deployment Forms, One Platform
From a single rack for an AI pilot to a multi-pod prefabricated campus. All are factory assembled and acceptance tested before dispatch. Click a plan or a card to isolate one.
The Three Footprints, Drawn to the Same Scale
A rack node next to a containerised hall next to a campus. This is the comparison a floor plan never shows you.
Click any card, or any plan above, to isolate that form
Rack node
Liquid cooling or high density air cooling integrated into an all-in-one rack platform. Supports flexible, scalable computing for AI experimentation, development and prototyping.
- 25 kW, 50 kW and 90 kW configurations
- Integrated 60 to 120 kVA modular UPS
- Single rack footprint, Tier I
Containerised hall
AI training demands peak computational power and generates extreme heat. Direct-to-chip, rear door and immersion configurations are engineered for that load, with rapid deployment and staged scalability.
- 400 to 800 kW configurations
- 10 racks at 50 to 80 kW, or 4 to 8 tanks at 100 kW
- 2N power and cooling, Tier III
AI farm
Modular design allows rapid deployment of AI farm infrastructure in almost any location. Dedicated pods for compute, power, network and operations combine into a single site.
- Mixed cooling architectures on one site
- Power pods separated from compute
- Expands by adding modules
XRT-AI-R Rack Nodes
AI pilots, prototypes and edge sites
A complete AI micro data centre inside a single rack footprint. Cooling, UPS, distribution, monitoring, access control and suppression all ship inside the one platform, so an AI pilot does not need a room built around it first.
Three configurations cover the range. The R25-AIR is high density air cooled at 25 kW using a 30 kW in-row DX unit and an air cooled condenser, which suits a prototype or an edge node where there is no water and no appetite for a loop. The R50-RDX adds a 65 kW rear door heat exchanger and reaches 50 kW on a free cooling chiller. The R90-DLC goes to 90 kW with a 70 kW direct-to-chip loop plus a 30 kW in-row unit for the air side, rejecting through a cooling tower.
Every node carries a modular UPS sized to the load, from 60 kVA on the R25 to 120 kVA on the R90, with five minutes of battery autonomy as standard and basic or smart PDUs. A 10-inch colour touchscreen runs the node locally and reports over Modbus or SNMP.
Delivered assembled and factory tested. Site works reduce to a supply connection, a data path and commissioning, and the unit is craned into position rather than built in place.
What is included
- Single IT rack with integrated cooling
- Modular UPS, 60 to 120 kVA
- Five minutes of battery autonomy
- Basic or smart rack PDUs
- 10-inch colour touchscreen and full sensor set
- Three-in-one and remote access control
- Novec 1230 fire protection, optional
XRT-AI-R Technical Data
Every published figure for the three rack node configurations.
XRT-AI-C Containerised Halls
Training clusters in a 12.8 metre enclosure
A complete liquid cooled data hall in a single prefabricated enclosure, 12.8 by 7.3 metres, holding ten racks or eight immersion tanks. Tier III with 2N power and cooling, built and load tested before it leaves the factory.
The liquid cooled halls come in two forms. The C800-DLC runs ten racks at 80 kW each on direct-to-chip cooling, served by two 1520 kW coolant distribution units with four 65 kW in-row units handling the air side, and rejects through a cooling tower. The C500-RDX runs ten racks at 50 kW on rear door heat exchangers, ten of them at 65 kW, on a free cooling chiller.
The immersion halls trade racks for tanks. The C800-IMX carries eight tanks at 100 kW each and the C400-IMX four, with 2N coolant distribution per tank. There are no server fans at all, which cuts IT power draw and makes the unit deployable where noise limits apply, and the fluid runs warm enough that mechanical refrigeration is not needed in most Australian climates.
Enclosures are rated for the conditions Australian infrastructure actually faces: sealed against dust and moisture, thermally insulated, and available with high corrosion protection for coastal and remote sites. Cooling runs independent internal and external loops, so nothing from outside reaches the IT space.
What is included
- 10 racks, or 4 to 8 immersion tanks
- 2N UPS power, 600 or 1000 kVA modules
- Direct-to-chip, rear door or immersion cooling
- Cooling tower, free cooling chiller or dry cooler
- 21-inch colour touchscreen and full sensor set
- Three-in-one and remote access control
- Novec 1230 fire protection
- Optional C5M corrosion protection for coastal sites
XRT-AI-C Technical Data
All four containerised hall configurations, liquid cooled and immersion.
XRT-AI-F: From Container to Campus
Prefabricated AI farms, built by adding modules
For AI farms and factories, a prefabricated campus gives you rapid, scalable deployment without a construction programme. Dedicated pods for compute, power, network and operations combine into a single site, with flexible liquid cooling that optimises energy efficiency and reduces operating cost at scale.
A four-pod XRT-AI-F campus. Every module arrives finished, is craned into position and connects to the next one.
A Campus Is Assembled From Three Module Types
These are the pods drawn in the footprint comparison further up this page, in the same colours: compute in blue, power in green, network and operations in grey. A site is specified by choosing how many of each you need, and grows later by adding more of the same.
Compute pods
Mixed cooling architectures
Direct-to-chip pods, immersion pods and air cooled pods coexist on one site, each matched to the density of the hardware it holds. You do not have to pick one architecture for the whole campus.
Power pods
Separated from compute
UPS, batteries and distribution sit in their own modules with dedicated fire compartments, so battery energy never shares air with the compute hall.
Support pods
Network and operations
Dedicated network pods, monitoring rooms, lobbies and personnel space complete the campus without borrowing floor area from the data halls.
And it goes where the power is, not where a building is
Because every pod is prefabricated, compute can be sited where land, grid capacity and fibre are actually available rather than where a data hall already exists. Pods ship on standard transport, are craned into position, and the site expands by adding modules instead of rebuilding the originals.
What Is Already Inside Before It Ships
Every XRT-AI configuration arrives with all of this installed, wired, charged and load tested. Your team racks the GPU nodes.
Liquid cooling
CRX4-RDX rear door, CRX4-DLC direct-to-chip or CRX4-IMX immersion, with the coolant distribution, manifolds and fluid supplied and commissioned as one system.
UPS and batteries
Modular UPS from 60 kVA on a rack node to 2N 1000 kVA on a hall, with five minutes of autonomy as standard and battery in its own fire compartment on campus builds.
Racks and distribution
IT racks, containment where the architecture needs it, basic or smart rack PDUs and the full distribution path from incomer to rack.
Monitoring and DCIM
A 10-inch touchscreen on rack nodes, 21-inch on halls, with a full sensor set and Modbus or SNMP reporting into the platform your team already runs.
Access control
Three-in-one access control by PIN, card or fingerprint, with remote authorisation and surveillance options.
Fire protection
VESDA class detection with Novec 1230 clean agent suppression, and dedicated fire compartments separating battery energy from compute.
A Full Solution Provider, Not Just the Box
Behind the prefabricated range sits a complete precision cooling and critical power platform. Whatever the density, the architecture and the site, the equipment comes from one supplier with one point of accountability.
Platform Capability
The ranges behind every XRT-AI configuration.
Why Choose UPS Solutions for AI Infrastructure
Australian owned critical power and cooling specialists. We engineer the whole facility, prove it under load before it ships, and maintain it for the life of the asset.
Australian field team
Local technicians across Sydney, Melbourne, Brisbane, Perth, Adelaide and beyond. The people who commission your cluster are the people who come back to service it.
Load tested before dispatch
Built, wired, charged and proven under load at the factory, with witness testing available, so site works reduce to footings, a supply connection, a data path and commissioning.
Power and cooling, one scope
UPS, batteries, distribution, liquid cooling, heat rejection, monitoring, access control and suppression from one engineering team with one point of accountability.
24/7 support after handover
Preventive maintenance, battery replacement, fluid and filtration service and emergency response across Australia, New Zealand and the Pacific Islands.
How an XRT-AI Project Runs
Load and site review
Rack count, kW per rack, the GPU platform you are deploying, grid capacity, fibre and what the site can physically take.
Form and architecture
Rack node, containerised hall or campus, with the cooling architecture and redundancy chosen against your real numbers rather than a catalogue range.
Factory build and test
Assembled, wired, charged and load tested before dispatch, with factory acceptance testing you can attend.
Install, commission, maintain
Craned into position, connected, commissioned under load, then maintained on the same 24/7 response as your UPS.
Send us the load, the platform and the site
Rack count, kW per rack, the GPU platform you are deploying, and whether the site has grid capacity and fibre. That is enough for us to recommend a deployment form and put a configured price in front of you.
Prefabricated AI Data Centre FAQs
The questions we are asked most often when scoping AI infrastructure.
What is a prefabricated AI data centre?
A prefabricated AI data centre is a complete facility for accelerated computing that is built and tested in a factory rather than constructed on site. The enclosure, liquid cooling, UPS, batteries, distribution, racks, monitoring, access control and fire suppression are installed and proven under load before dispatch, so the unit arrives ready to energise. Site works reduce to footings, a supply connection, a data path and commissioning.
How much IT load can an XRT-AI system carry?
From 25 kW in a single rack node up to 800 kW in one containerised hall, and into the multi-megawatt range as a prefabricated campus. The rack nodes are 25, 50 and 90 kW. The containerised halls are 400, 500 and 800 kW. Above that, XRT-AI-F campuses add compute, power, network and operations pods to one site rather than rebuilding the original units.
What is the difference between XRT-AI and XRT-CDC?
XRT-CDC is the containerised data centre platform, configured to whatever a site needs across defence, mining, oil and gas or general IT. XRT-AI is a set of pre-engineered configurations for accelerated computing, six of which are built on that platform. The exception is the XRT-AI-R rack nodes, which are a single rack footprint rather than a container, so the AI range is not simply a subset of the container range.
Which liquid cooling architecture should I choose?
It is decided by rack density and by what the platform vendor supports. Rear door heat exchangers cover roughly 20 to 66 kW a rack and change nothing inside the server. Direct-to-chip covers 50 to 130 kW and is the route rack-scale GPU platforms are designed around. Immersion starts around 100 kW and goes beyond, and removes the air path entirely. All three are available in XRT-AI and all three are commissioned by us.
Do I need liquid cooling, or will air do?
Air cooling holds a rack to roughly 30 to 40 kW before the volume of air required stops being practical. Below that, the XRT-AI-R25-AIR configuration uses a 30 kW in-row DX unit and an air cooled condenser, and it is the cheaper answer. Every current AI training platform sits above that line, which is why the rest of the range is liquid cooled.
How long does an XRT-AI deployment take?
Because the unit is prefabricated and factory acceptance tested, the build happens in parallel with your site works rather than after them. Lead time depends on the configuration, the cooling architecture and the UPS sizing, so we quote it against your brief. What is predictable is the site end: footings, a supply connection, a data path and commissioning, rather than a construction programme.
What redundancy and Tier level is available?
Rack nodes are Tier I with five minutes of battery autonomy as standard. Containerised halls are Tier III with 2N power and cooling, using 600 or 1000 kVA UPS modules. Immersion halls add 2N coolant distribution per tank. Campus builds separate power pods from compute pods so battery energy never shares air with the compute hall.
Can XRT-AI be deployed on a remote or harsh site?
Yes, and that is much of the point. Enclosures are sealed against dust and moisture, thermally insulated, and available with high corrosion protection for coastal and remote deployments. Cooling runs independent internal and external loops so nothing from outside reaches the IT space. Immersion in particular isolates hardware from dust, pollutants and humidity, which extends service life on mine sites and offshore support bases.
How much does a prefabricated AI data centre cost?
It is driven by IT load, cooling architecture, redundancy target and site conditions rather than by a list price. A 25 kW air cooled rack node sits at one end of the range and an 800 kW liquid cooled hall with 2N power at the other. Send us the rack count, the kW per rack, the platform you are deploying and the site location and we will configure and price it rather than quote a range.
What is the PUE of an XRT-AI system?
It depends on the cooling architecture and the climate. Immersion adds as little as 0.03 to overall PUE, because the fluid runs warm enough that mechanical refrigeration is not required in most Australian climates. Direct-to-chip and rear door systems benefit from the same warm water principle through free cooling chillers and dry coolers. Treat any single figure as a cooling contribution rather than a whole-facility number, since PUE also carries UPS losses, lighting and any remaining air handling.
Do you install and maintain the system after delivery?
Yes. UPS Solutions designs, supplies, installs, commissions and maintains the complete system across Australia, New Zealand and the Pacific Islands, including preventive maintenance, battery replacement, coolant and filtration service, cooling service and 24/7 emergency response.