Guides & Blog · 2026-09-15 · eWEB TEC

Installing GPU and AI Servers in a Sydney Data Centre

A GPU or AI server install is a normal rack-and-stack job with less room for error: heavier chassis, much higher power per rack, dense high-speed cabling and, more and more often, liquid cooling. Most failed installs trace back to power, weight or cabling that nobody checked before the truck arrived.

Sydney is adding AI-ready capacity fast. NextDC markets S6 as a purpose-built AI facility, and Macquarie Data Centres is opening IC3 Super West with direct-to-chip liquid cooling. Whichever hall you land in, the floor work looks the same. This guide is the checklist eWEB TEC uses for smart hands GPU installs in Sydney.

What is different about a GPU server install?

Before the truck arrives: the pre-install checklist

1. Power budget confirmed with the facility: kW for the rack, circuit count, A and B feeds, and PDU socket type.
2. Weight and rack depth checked against the chassis spec and rail kit. Heaviest units go at the bottom.
3. Rack elevation drawn up: which U each server takes, where switches sit, and space for cable management.
4. Port map for every high-speed and management cable, with cable lengths and labels agreed.
5. Delivery booked with the loading dock, including dock hours, pallet size and where boxes are unpacked.
6. Remote contact who will be on the bridge for BMC access and first boot.

If you have not written one before, start with our MOP and ticket guide.

On the floor: how the install runs

1. Unbox away from the rack, check for shipping damage, record serial numbers and photograph labels.
2. Fit rails to the rack elevation, then lift and seat each chassis with two people.
3. Connect A and B power to the correct PDUs and confirm load is balanced.
4. Run high-speed and management cabling to the port map, label both ends and dress it so airflow is not blocked.
5. Fit blanking panels in empty U. Missing panels on a GPU rack cause hot-air recirculation.
6. Power on, confirm fans, status LEDs and link lights, and confirm your team can reach each BMC.
7. Send photos of the front and rear of the rack, serial numbers and the as-built port map.

What about liquid-cooled GPU servers?

Direct-to-chip liquid cooling is becoming common for dense AI racks. The split of work matters: the facility or hardware vendor owns the cooling loop — manifolds, quick-disconnects, coolant and leak testing. eWEB TEC racks and cables the servers and follows the vendor and facility procedure for connecting to the loop. If that procedure is not written down, the install should wait.

Common mistakes that cost a second visit

• Power cords that do not match the PDU sockets
• Rail kits for the wrong rack depth or rack type
• Optics or cables that are the wrong length, speed or connector type
• No agreed port map, so cables are moved after first boot
• Assuming the rack has enough power because it has free U space

Which Sydney data centres does eWEB TEC attend?

We attend Equinix SY1–SY5 (Mascot and Alexandria), NextDC S1/S2 (Macquarie Park), Global Switch Ultimo, and CDC Eastern Creek and Marsden Park. Other halls are booked once access is confirmed. See Sydney data centres and, for western sites, the Western Sydney data centre guide.

FAQ

Related service: view this service page · contact eWEB TEC.