By Laurent Segneri, R&D executive director at Apx Data Centre Solutions
TL;DR
- The AI “Thermal Wall”: As AI-driven demands push rack densities to 100kW, traditional air-cooling methods have reached their physical limits, rendering standard, transactional cooling procurement obsolete.
- From Vendor to Collaborative Partner: Because every data centre now faces unique physical constraints and staffing shortages, operators must shift from buying cooling equipment as a commodity to treating providers as integrated engineering partners.
- Precision Engineering for Sustainability: Collaborative cooling strategies prevent wasteful “over-cooling” by tailoring compressors and closed-loop systems to specific thermal loads, helping operators optimize critical green metrics like pPUE and WUE.
# # #
As we move through 2026, we aren’t just seeing a slight uptick in data demand; we’re in the middle of a full-scale industrial revolution, powered by AI. While this is good news for the sector, we’re now in the era of the 100kW rack, with power consumption pushed to its limit by the new generation of chips and their successors.
This isn’t just an anecdotal shift. A January 2024 International Energy Agency (IEA) report forecasted that the global data centre industry’s energy consumption would hit a record high 1,000 TWh annually by 2026 —roughly equivalent to the electricity consumption of Japan. With these updated racks, it’s a prediction set to become reality.
While this innovation drives the industry forward, it’s also created a massive hurdle for those of us on the ground. Traditional air-cooling methods, which have served us well for decades, have officially hit a physical ceiling. It is pushing us towards a thermal wall, where blowing more cold air at a server isn’t just inefficient, it’s physically impossible. Operators need reliable and efficient solutions that are tailored to real-world challenges. In this new era, success in data centre cooling goes beyond simply providing equipment.
The commodity trap
As we’ve reached a physical limit, we can no longer treat cooling as a commodity. For too long, procuring the equipment has been a transactional, box-ticking exercise. In the past, operators asked for a set amount of kilowatts for cooling their data centres and the supplier sent them over the correct amount. No questions asked.
In 2026 though, cooling is taking a front seat, and this approach might just be a recipe for operational disaster. At Apx, we’re seeing first-hand that ‘standard’ data centres no longer exist. Every site, whether a hyperscale, enterprise or colocation has its own challenges.
Some operators find that their sites have structural load limits that can’t handle the weight of new liquid-cooling infrastructure. Others are facing a staffing crisis. According to recent Uptime Institute data, 53% of operators are reporting difficulties with finding qualified staff to manage increasingly complex systems.
A true collaborative partner fills the technical void that a mere equipment provider cannot. Being able to trust the person that you are working with and knowing they can rely on you is really important. It goes beyond providing the product– it’s being focused on customer needs and delivering them as best you can, right until the end.
We can’t solve today’s problems with an outdated approach. It’s time to treat cooling providers as partners rather than vendors. It doesn’t matter if the equipment used is modern, it’ll underperform if it hasn’t been designed with the people who have to maintain and install it.
I’ve worked in the HVAC industry for years and understand its intricacies as well as the importance of mechanical engineering. In both sectors, I’ve seen first-hand why working together is critical for infrastructure like this. Focusing on an ‘engineering in partnership’ mindset and collaborating to share our experience creates far better systems; not just built for the now, but the future. And, in doing so, the future’s problems.
Engineering for the boots on the ground
Mechanical engineering backgrounds mean we understand the fundamentals of what it takes to design products that actually work for specific environments. This means we’re not just building hardware; we’re designing around real-world challenges. As part of the LFB Group, we have over 60 years of HVAC experience with state-of-the-art testing facilities that enable us to put the solutions through rigorous, real-life tests first.
Integrating engineering expertise at the start of the process means that logistical issues can be solved or ‘designed out’ before they become expensive problems further down the line.
The industry is littered with cautionary tales of high specification cooling units that arrived on site, but didn’t fit through the door. Focusing on the ‘boots on the ground’ approach prevents logistical problems like this and makes installing a cooling unit a simpler process, reducing the risk of projects from running over on time and budget.
We also have to think about longevity. Although these highly specialised units are projected to have a lifespan of 15 years, they won’t last unless we future-proof them. We need to swap parts for these systems without them shutting down, and they must be easily maintained and have critical components that can be reached by maintenance engineers.
Without collaboration with specialist engineers, operators risk deploying units that are technically brilliant but logistically catastrophic.
Metrics with meaning
Crucially, collaboration doesn’t just mean making life easier for engineers; it also means hitting important sustainability targets. Over 100 of Europe’s data centres are now part of the Climate Neutral Data Centre Pact, striving to be carbon neutral by 2030.
Cooling, and the industry as a whole, often gets a really bad rep, primarily because there’s still a lack of understanding around the world about what data centres do, and where they draw their energy from. However, data centres actually use no more water than a typical golf course.
For us, it’s about flipping the script and working as closely with operators and suppliers as we can to ensure that we are delivering solutions tailored to operators’ specific needs and priorities so that we are providing effective solutions. We are part of an effort to change the narrative through our precision engineering mindset.
The success of cooling units in data centres isn’t just measured by its design, it’s also validated by green metrics such as pPUE (partial Power Usage Effectiveness) and WUE (Water Usage Effectiveness).
Tailoring fans and compressors to the specific thermal load of an AI cluster prevents the ‘over-cooling’ waste that’s less cost-effective. Collaboration with cooling specialists will allow them to implement closed-loop systems that keep units at an optimum temperature without draining the local water supply.
Precision engineering is far more important than updating systems at an unsustainable speed. It’s about being right, not just being fast.
A change in mindset
Three to five years ago, cooling sat in the shadows. But now, thanks to the demand for AI, we’ve gone from the proverbial ‘Fred Flintstone’ era, where they’re powering their car with their legs, to a Formula One car. And because of that, cooling can no longer sit quietly in the background or be seen as a ‘box-ticking’ exercise. Instead, it’s a vital and collaborative engineering discipline which determines whether a data centre thrives or fails.
Future success relies on listening to customers, understanding their physical constraints and addressing operational realities—crucially, ahead of time. Viewing cooling providers as partners rather than vendors enables the construction of solutions that not only fit the space, but operate effectively in real-world scenarios, ensuring that the 100kW racks of today keep running efficiently in 2040.
# # #
About the Author
Laurent Segneri is an experienced leader with a strong focus in R&D Management and transformation, portfolio management and end-to-end project execution in cross-cultural matrix environment while keeping a lean and pragmatic approach to all things.
