Rising component costs may adjust prices on selected items without prior notice. Supply constraints may add shipping time — we’ll notify you if affected. OEM costs and supply tightening — contact us to secure stock and pricing. For projects or bulk orders, choose “Request for Quotation” on any item.
Call our specialists(65) 6978 3502 Email our sales teamsales@sourceit.com.sg

Servers & Power

Can Your Singapore Colocation Hall Take a 140 kW Rack? A Pre-Quote Checklist

Most rack-scale AI projects in Singapore do not stall on budget. They stall on a phone call to the data centre, usually made too late. A rack like the NVIDIA GB300 NVL72 draws 132 to 140 kW according to HPE. That single figure changes the power feed, the cooling, the floor and the people you need.

This checklist is meant to be used before you request a quote. It will not make the answers easier. It will make them arrive earlier, when they are cheap.

One rack, by the numbers

What HPE's published figure Why it matters
⚡ Rack power 132 kW (GB200), 132 to 140 kW (GB300) The feed to one rack is the size of a small data hall's
🔌 Power delivery Eight 33 kW power shelves on a 50 V DC busbar; 400 V 60 A power whips Not a pair of standard rack PDUs
💧 Heat to liquid 115 kW (GB200); about 90% (GB300)
Needs a coolant distribution unit and a facility loop
💨 Heat to air 17 kW (GB200); about 10% (GB300)
The "leftover" still needs real air cooling
🧊 Cooling hardware In-row or in-rack CDU; rear-door heat exchanger options Extra equipment, extra floor space, extra power

Sources: HPE GB200 NVL72, HPE GB300 NVL72, HPE NVL72 QuickSpecs

A point worth dwelling on: the air-cooled remainder of a GB200 rack is 17 kW. Many enterprise racks never reach that figure in total. "Liquid-cooled" does not mean the hall's air system gets a holiday.

The seven questions

# Ask your data centre 🟢 Good answer 🔴 Stop sign
1 ⚡ What is the maximum power you can deliver to a single rack position, and at what voltage? A figure above the rack's draw, with headroom for peaks "We average X kW per rack across the hall"
2 💧 Is there a facility water loop to the white space today? Yes, with stated temperature and flow "It is on the roadmap"
3 🧊 Who supplies, installs and owns the CDU? A named party with a service contract Nobody is sure
4 💨 Can the hall absorb the air-cooled share of the rack? Yes, confirmed for that position "It is liquid-cooled, so it should be fine"
5 🏋️ What is the floor loading limit, and what is the delivery path from loading bay to cage? Written limits checked against the vendor's rack weight "We have never had a problem"
6 🧰 Who responds to a leak alarm at 3 a.m., and what is the procedure? A documented runbook and trained staff "The vendor handles that"
7 📜 Does your liquid cooling design follow SS 726:2026? Yes, or a clear explanation of the equivalent "What is SS 726?"

Why each question earns its place

Power: averages hide the problem

A hall's average density tells you nothing about one position. HPE's QuickSpecs list eight 33 kW power shelves feeding a 50 V DC busbar, supplied by 400 V 60 A whips. Ask for the per-position figure in writing, and ask about peak draw, not only nominal. If you are already watching PDU loads on ordinary racks, our piece on overloaded server racks covers the basics.

Liquid: the loop is the long-lead item

Most data centres in Singapore still rely on air cooling, as the legal commentary on the new national standard notes. A rack can be delivered in months. A facility water loop is a construction project. HPE validates both in-row and in-rack CDUs for the NVL72, which gives some flexibility, but neither works without somewhere to send the heat.

Source: Rajah & Tann Asia on SS 726

Weight and access: boring, and decisive

A fully populated liquid-cooled rack is far heavier than a conventional one, and it usually arrives pre-integrated. Get the weight and dimensions from the vendor's site preparation guide, then check floor loading, lift capacity, door heights and ramp gradients along the whole route. We are not quoting a weight here on purpose: use the figure for your exact configuration.

Operations: liquid is a commitment

HPE's services list for these racks includes site readiness assessment, coolant fill, CDU commissioning, leak detection and thermal monitoring. That list is a fair summary of the skills someone on site needs to have. Decide who that someone is before the rack ships.

Standards: there is now a local reference

Singapore published SS 726:2026 in August 2026 to guide liquid cooling in tropical data centres. An operator who knows it well is a good sign.

If the answers are mostly red

That is useful information, not a failure. There are three honest routes:

  • 🏢 Move the workload to a facility built for it. Purpose-built liquid-cooled capacity exists in Singapore and across the Causeway.
  • 💨 Drop a tier. An air-cooled 8-GPU server such as the HPE Compute XD690 gives current-generation GPUs without a liquid loop.
  • ☁️ Rent the peak. Use cloud GPU capacity for the largest jobs and own a smaller baseline.

The controversial part is simple: for many Singapore organisations, the correct outcome of this checklist is not buying the rack. Finding that out with seven questions is a great deal cheaper than finding it out with a purchase order.

Related on SourceIT

SourceIT Approved

SourceIT Approved

Researched and verified by the SourceIT team for Singapore buyers.

Share This Article
Facebook

More from our blog


← Older Post Newer Post →