Data centre operators face a pivotal moment as sustainability targets and AI-driven rack densities force a complete rethink of power and cooling infrastructure.
Data centres have hit an inflection point in 2025. The need to meet sustainability targets and the rapid escalation of AI workloads – from large language models to unprecedented rack densities – has meant operators must rethink every aspect of their facilities.
And as data centres evolve from static power consumers to dynamic participants in the energy ecosystem, the industry is shifting towards modular, hybrid, and collaborative solutions that balance performance with resilience and low-carbon objectives.
Here’s a closer look at the challenges that lie ahead for data centre operators, and the technology and infrastructure developments designed to overcome them.
Securing robust grid connections
The very first consideration for any data centre project is access to high-voltage power lines and substation connection points, said Mark Deguara, General Manager Data Centres at Schneider Electric (SE).
“Is it from a single network within the grid or is it available from multiple sources within the grid?” he asked. “Because the cost of having to get high-voltage transmission lines to your facility is significant – and you can’t just run them wherever you like.”
Even when lines are nearby, there may be additional infrastructure required to support that capacity – which energy providers may not have on-tap.
But beyond siting, data centres must compete with growing grid demand from electrified transport and industrial decarbonisation – meaning even relatively flat 24/7 loads put stress on the grid during evening peaks.
While many operators source renewables to offset usage, Deguara predicts a shift toward prosumer models – with on-site battery storage and even gas turbines enabling data centres to both draw from and feed into the grid.
Maximising compute density
In today’s data-driven world, the push to pack ever more compute power into ever smaller footprints is at an all-time high.
“Higher-density installations deliver higher efficiencies on cooling and better utilisation of your electrical infrastructure and physical building infrastructure,” Deguara said.
Fitting the same compute capacity into a smaller building envelope translates directly into both operational savings and lower capital outlays. “You don’t have the capital cost of constructing that building,” Deguara said.
Yet it’s the relentless speed of semiconductor innovation that has truly accelerated densification.
“Chip manufacturers are driving a lot more compute in a much smaller footprint,” he said.
Today’s fabrication processes push transistor geometries down to “one micron,” enabling far more processing cores per square millimeter than ever before.
“Earlier this year NVIDIA announced over the next two to three years we’re going to move from 130 kilowatts to a 1 megawatt rack,” Deguara said.
Achieving 1 MW of power delivery within a standard 600 mm-wide, 1200 mm-deep rack creates significant engineering challenges – but this is the scale of power density that AI workloads demand.
“AI has been driving high-performance compute which is driving the density and the applications to support that,” he said.
Embracing hybrid-liquid cooling
AI workloads are pushing compute and power densities far beyond the reach of traditional air cooling, making a hybrid approach not just advantageous but fundamental.
“If you look at the mix in a data centre at the moment you’ve got roughly between 80 per cent traditional and 20 per cent liquid ,” Deguara said. “Moving into 2030, this will probably eventuate in a 40/60 split.”
Even purpose-built liquid-cooled servers cannot eliminate air handling altogether.
“If you have a liquid-cooled server, for example, they still have a certain amount of air cooling that’s required,” he said. “So by default, even a liquid cooled AI data centre is hybrid in terms of its cooling methodology.”
Traditional air-cooled servers draw in 25° Celsius air, pass it across heat sinks, and expel it roughly 10°C hotter – adequate for chips dissipating up to 40 kW, but nowhere near the needs of modern AI accelerators.
“We’ve hit the thermal limitations of air to be able to reject the amount of heat that is coming off these new servers,” Deguara said.
SE’s latest liquid-cooling release addresses that ceiling by replacing airflow over heat sinks via direct heat transfer.
“Heat coming off the chipset is transferred directly into a cold plate,” he said. “That cold plate is then cooled via water and a glycol propylene mixture.”
Balancing low carbon-power and high AI energy demands
Modern AI-driven data centres must both ensure resilient, low-carbon utility power and meet the soaring energy demands of high-performance workloads.
And energy procurement is just one piece of a much larger sustainability puzzle. “How was the building itself built? Is it using low-carbon steel and concrete?” Deguara asked.
Every stage – from manufacturing to onsite assembly – must be optimised. To that end, SE just released a new uninterruptible power supply (UPS) that’s one-third the size of the previous iteration.
“That means two-thirds less materials such as steel and copper as well as a reduction in packaging, transportation and onsite labour,” he said.
But sustainability in data centres is about more than hardware and procurement.
“Energy in isolation is a combination of how you build your data centre, the efficiency of the products and the whole supply chain,” Deguara said.
As governments introduce mandatory sustainability reporting, every operator will need this holistic approach to remain compliant and competitive.
“Is the software and algorithms that run on these servers optimised? If not, there’s a lot of potential energy being wasted.”
Digital tools for design, monitoring and optimisation
By leveraging digital twins for electrical infrastructure, real-time monitoring dashboards and customisable analytics – operators can seamlessly guide the development of AI-ready data centers.
SE offers a comprehensive suite of digital solutions specifically engineered to support every stage of a data center’s lifecycle.
“We’ve just recently released an Electrical Transient and Analysis Program (ETAP), which is a digital twin designed to support electrical infrastructure design through the construction phase and then into the operational phase,” Deguara said.
By generating a digital twin of the entire electrical system, ETAP lets operators add additional equipment virtually first to see what impact it may or may not have as a data center is developed. This model-first approach drastically reduces risk when scaling to meet ever-growing AI power requirements.
And beyond design, SE’s EcoStruxure platform provides end-to-end monitoring and analytics. But these tools aren’t one-size-fits-all, with success significantly hinging on collaboration.
“It’s very much around working with the customer and working on a platform that supports their requirements and outcomes as opposed to just trying to push or promote a particular piece of software,” he said.
Scalability strategies for the future
Between 2025 and 2030, soaring AI workloads will push data centres to their limits in power capacity, build speed, space efficiency and regulatory compliance. But SE is already tackling these challenges head-on.
“The next-generation UPS is modular, in that its ability to start with a smaller capacity and grow lets operators scale power exactly as demand rises,” Deguara said.
At the same time, SE is embracing off-site prefabrication.
“We build the prefabricated module and ensure that the product is tested before it comes to site,” he said.
This accelerates timelines, enables parallel civil and technical work and guarantees quality. With rack densities climbing, every square metre counts.
“The white space is becoming smaller because the physical infrastructure itself isn’t changing, so we’re engineering quieter, more compact UPS and switchgear to free up critical room.”
Anticipating tighter environmental regulations, as is occurring in Europe, SE is also moving away from sulfur hexafluoride (SF₆) gas.
“We’re moving to what we call ‘pure air’, which has the same insulation performance in a nearly identical footprint but without the greenhouse-gas impact,” Deguara said.
By combining modular growth, factory-built pods, a reduced footprint and SF₆ free designs, SE is empowering data centre operators to expand AI infrastructure quickly, flexibly and sustainably.
