Five years ago, grey space was the unglamorous half of the data centre – the power, cooling, and cabling infrastructure that supported the heartbeat of what was happening in the white space beyond it. According to Brendan Dessent, founder and CEO of Computer Room Solutions and head of Legrand Data Centre Solutions, that hierarchy no longer holds.
“Grey space is definitely a big play,” Dessent tells W.Media, then immediately corrects a common industry shorthand. Most people describe grey space as two disciplines, he says, but he counts three: electrical, mechanical, and telecommunications. The third is easy to overlook. “A lot of people probably don’t know what an MMR room means,” he says. “I heard it 25 years ago… it’s called Meet Me Room.”
Telco pits outside the building feed the MMR from diverse paths for redundancy in case of a
communications outage. North and south, or east and west, so a single street excavation can’t take out the network. “So, three paths: you’ve got power, cooling, and your network,” Dessent says. “Your telco network is part of your grey space as well.”
Rack level not rooms
What has genuinely shifted, in Dessent’s view, is the unit that infrastructure is now designed around. “Everything’s looking towards rack level now with the advent of AI,” he says, illustrating it with his own numbers: “Once upon a time you’d be having, say, four megawatts in a room – 100 racks at four kilowatts each. Now you’re looking at 100-kilowatt racks,” which is the same footprint carrying much higher loads.
That means denser power delivery through smart, intelligent power rails, and liquid-to-chip manifolds running vertically through the rack “very similar to a power rail,” feeding coolant to each piece of compute. It also means far more network density at the rack itself – a jump, he notes, from the 12-port copper patch panels of the 1990s to 48-port equivalents in the same 1RU form factor, on top of a steep rise in fibre density.
It’s a shift Dessent doesn’t dispute when it’s put back to him in those terms: that the rack, not the room, is now the unit of design, “forcing data centres to be engineered with the same level of detail and responsiveness as the servers that are in the racks.”
Compression is real
That compression is not abstract. Dessent recently measured it himself, tape measure in hand, at a new-build data hall. Traditionally, he says, a hot aisle containment pod might run around 15 metres long, with two or three metres of incoming grey-space corridor feeding into it. “That’s your heartbeat; that’s where your server cabinets sit,” he said.
At this site the rack PODs themselves ran nine metres, but the electrical infrastructure serving them measured 9.6 metres, wider than the racks, with a further six to seven metres of mechanical infrastructure, including the primary liquid-to-chip loop, sitting in what is still technically grey space. “That’s a first-hand account of one of the biggest changes I have noticed, and I
have been in this game well before we even talked about hot and cold aisles,” he said.
New security issues
The physical boundary between white and grey space is dissolving for another reason: more infrastructure now needs hands-on maintenance inside the data hall itself. In-row cooling, rear-door heat exchangers, secondary water loops running above the racks – all of it requires technicians working further into the IT space than before, which in Australia carries a security dimension.
“Federal government clearances for technicians are fast becoming a requirement to enter white spaces. You’ve got to be an Australian citizen, but you’ve got to put people through what’s called a baseline or an NV1 clearance,” Dessent says, referring to the federal accreditation levels. “There are a lot more moving parts…technicians need access to those data halls rather than just being outside the building.”
Liquid cooling itself has become a baseline design consideration for new facilities, whether implemented from day one or built in as future capacity. “Liquid cooling is definitely a basis of design for all new data centres,” Dessent says, noting that colocation operators in particular “are always going to allow for liquid to chip as a standard, or as a future medium.”
He draws a separate line around existing, air-cooled facilities: retrofitting them for liquid-to-chip, he says, “will be another industry angle – industry discipline” in its own right, one that will still require the same technician access and clearances into the data hall. Among colocation operators – the bulk of the current Australian market, as distinct from hyperscaler self-builds – he’s blunt about where things stand: “Colocation providers need to look to the future and be future-proofed.”
Power problems
On what is now driving site selection, Dessent doesn’t hesitate: “Certainly power…you’ve got to identify where the power is first. It’s no good going to get permit approvals if you don’t have the power.” But two examples he raises suggest power access alone doesn’t decide a site. The first is Gippsland, in regional Victoria, where he once put the case to a hyperscaler: with the Latrobe Valley’s coal-fired power stations nearby, “you’ve got all the transmission towers… it’s not always about the power, it’s how do you get the power from A to B.”
The response, he says, came down to connectivity rather than power and the region’s fibre wasn’t up to it. “Someone has to put their hand up and say I’m going to spend the capital to get the dual-feed fibre optic cabling network down there,” he says. The second is a site he’s watched go up close to major transmission infrastructure elsewhere, where he uses his own shorthand for the towers themselves: “coat hangers…that’s something that can’t be overlooked.”
Rack density
Rising rack densities are also raising the bar on what operators are expected to know about their own infrastructure in real time. “With the increased rack densities comes increased dependency on the reliability of critical supporting infrastructure,” Dessent says. “Dependency on monitoring, early detection, predictive analysis of equipment performance. Transparency of infrastructure performance data is now expected by AI operators.”
SLAs between operators and their customers aren’t new, he adds, but the depth of visibility expected under them is: with more moving parts inside the data hall, predictive monitoring and utility power-quality tracking, including the ability to ride through brownouts and blackouts, have become baseline expectations rather than a differentiator.
Standardisation is emerging as the industry’s other lever against the pace of AI-driven build-out, particularly as more capacity moves to regional sites where skilled trades are scarce. Dessent is pretty direct about the constraint driving that shift: “Where do you find the required amount of
tradespeople such as electrical, mechanical and plumbing, essentially MEP?” he asks, describing city builds that can require “1,500 staff on site per day, a workforce that simply doesn’t exist 300 kilometres outside a capital city.
The rise of modular
Modular, prefabricated power and cooling blocks bring three concrete advantages in response, in Dessent’s framing: consistency of product, improved procurement leverage through repeat purchasing, and simpler maintenance, since “if they’re going to site, they’re working off the same maintenance manuals…whether that be CBD, regional, wherever it might be.”
Prefabricated switch rooms, pushed into position as complete containerised units, were a novelty when Dessent first saw them on a pre-COVID build; by his account they are now simply how the industry will build.
Asked where the greatest infrastructure innovation is happening today, Dessent points not to the room or even the rack, but to the chip itself. “Creating high GPU heat transfer tolerance will be key,” he says. “The system’s design and dependency – GPU tolerance to voltage and frequency changes – will influence the power distribution.”
He sees GPU designs evolving towards higher ambient operating temperatures, “thereby reducing or eliminating central cooling plants where heat rejection occurs within typical ambient conditions” – chips engineered to run hotter, reducing how hard the mechanical plant around them has to work.
It’s basically a reminder that, in Dessent’s view, the causality in data centre design now runs in one direction: “Everything starts there…that’s where all the heat and power requirements are, which obviously reverse engineers into how the data centre is built.”
That same forward-shift is happening at the design stage. Where colocation fit-out once waited until a customer signed, Dessent has watched integrated fit-out, such as rack layouts and hot aisle containment included, move earlier and earlier, now sometimes specified before planning permission is even granted.
“It’s getting even further back in the process,” he says. It’s a small detail, but for Dessent it captures the broader shift AI has forced on the industry: infrastructure that used to follow demand is now being built to anticipate it.
The Sydney Cloud & Datacenter Convention takes place on 16–17 September 2026 at the ICC Sydney, bringing together data centre operators, cloud providers, technology vendors and enterprise leaders to discuss the future of digital infrastructure in Australia and the Asia-Pacific region.