17 June 2026
|
Oliver Sanders, Data Centre Commercial Director UK&I at Carrier, outlines why demand for data centres is rising and why it’s vital to have the right products and support in place before, during and after deployment.
We’ve all spent the last year talking about AI. Every campaign seems to have AI stamped across it, urging us to get excited about the next era. We hear plenty about the magic: generative models and the speed of their output. But I’m less interested in the magic and more interested in the cost of it.
Because strip away the software and the hype, AI is really just heat. It’s electricity converted into calculations, with heat as the byproduct. And as we barrel along into this new era, the rules for managing that heat are changing fast.
The physics of densification
First, let’s ask: why are we densifying? Why are we building data centres? Why are we packing 100 or 200 kilowatts into a rack? We’re not just doing it to save on floor space and rent. We’re doing it to maximise the speed of calculations. That means we need to move the data faster. The further apart the processors are, the higher the latency, so to achieve the performance AI demands, we must shrink the physical distance between the chips. But when you shrink the physical distance, when you shrink the volume, you concentrate a huge amount of power into a tiny box. And physics dictates that all that heat and energy must go somewhere.
Therein lies the challenge. The data centre cooling industry tends to ‘slice’ this problem into pieces. IT buys the server. Contractors and data centre builders buy the chillers. Mechanical engineers and consultants size the pipes in cooling loads. But heat couldn’t care less about job titles. No matter how we choose to slice it, heat f lows in a chain. It flows from the chip to the rack, to the room, to the loop, and finally, to the outside world. But every time that heat is handed off, a penalty is introduced called a delta T (the difference between two measured temperatures). And if a contractor or consultant only looks at their slice, inefficiencies compound across the whole system.
If you take away anything from reading this, remember the balance. To move heat, you need flow and a temperature difference. But as density goes up, and you try to pump more water and blow more air, you’re going to hit a wall. Higher flow means massive pressure drops, which means huge pumps and enormous pipes. The result is that CapEx explodes, and no one can afford it. To overcome this imbalance, we have to stop designing from a ‘chip forward’ mentality in the hope that legacy buildings can cope. Instead, we need to start designing from the destination – the chip – backwards.
There are a hundred brands out there selling a thousand products. But if you look at physics, there are broadly four destinations where heat can go. There’s no magical ‘fifth dimension’, no matter how hard we look. Heat is either expelled into the air, through evaporation, into heat recovery or into a large body of water.
Let’s explore those options.
Four thermal destinations
- Dry cooling: Globally, air can act as a virtually infinite sink. But locally, it’s finite. And air is a relatively poor conductor of heat. To cool a 500-megawatt AI factory, you can’t just blow air at 200 miles an hour and think you’re going to get the efficiencies and Power Usage Effectiveness (PUE) you need in a data centre. You need to make the building bigger. Much bigger. Stick to dry cooling, and the facility size might eventually outpace available land, and you might eventually hit a wall.
- Evaporative cooling: Water usage is a sensitive topic. Nobody wants to consume water if they don’t have to. But honestly, evaporation is thermodynamic magic. It’s an order of magnitude that can be more efficient at ejecting heat than dry air. As densities double or triple, when you look at the statistics of what’s being built in the DC landscape, however we feel about it, the physics may force us to use more water. Because without the phase change of water, heat might not be siphoned off fast enough.
- District heating: My favourite option. Where feasible, we reuse. We take whatever heat we can remove from the rack, and we heat homes and sanitary water. Technologically, it’s easy. We have the heat exchanges and highly efficient products on the market. The problem is geography. We need a population to accept and own the heat. While it can’t be done in the middle of the desert, it can in a condensed city area or a new build site. If built near a data centre, that waste heat can be turned into a charitable donation or a revenue stream.
- Direct water: Rivers, lakes, the sea. These are the ultimate thermal mass, absorbing heat almost instantly. But while river water doesn’t care about a 100-kilowatt data rack, the local ecosystem certainly does. Discharging hot water creates thermal pollution, depleting oxygen in the environment and drawing strict regulatory scrutiny. Also, tapping into natural waterways is a logistical headache requiring highly specialised infrastructure to combat corrosion and constant biofouling from aquatic life. Even taking the logical extreme and sinking the data centre itself, as seen with Microsoft’s underwater Project Natick1, reveals a huge compromise: thermodynamic brilliance comes at the cost of prohibitively difficult maintenance for rapidly scaling AI workloads.
Ultimately, the fundamental point is this: before you go out and buy those GPUs, you need to know which of these four destinations you’re going to use, because that decision dictates everything else inside the building. The issue is that opting for just one of those to cross the thermal wall is insufficient – and, as with most things, that comes with added cost.
The myth of pure liquid cooling
How then do we scale the thermal wall without going bankrupt? How do we span the gap between legacy cooling and an AI liquid-cooled future? Bridge technology.
We know where the heat goes and where to capture it. The ‘hottest’ heat capture in the industry right now is direct-to-chip liquid cooling. Not only is it thermodynamically superior, but it also funnels heat directly away from the heat source (the chip). But here’s a reality check: pure (non-hybrid) liquid cooling is a myth. Because even in a rack full of liquid-cooled GPUs, we’re still burdened with power supplies, memory and networking gear, all of which – no matter how high the loads and temperatures – still need air cooling. And we know the demand is going up.
While we might capture 80% of that heat in liquid, the remaining 20% is still hot air. We can’t just erase air handling and mechanical cooling elements from the equation. That has challenges of its own. To illustrate, let’s consider the hybrid car.
The hybrid car dilemma
Think of the modern-day data centre as a plug-in hybrid car. On a physics level, a hybrid is technically not the best at being a car. It’s heavier than others in its size class, because it drags around both a petrol engine and a massive battery. That added bulk means it’s neither optimised for electric efficiency nor petrol mileage.
Yet we in the data centre industry are doing the same thing. We’re building hybrid halls. We’re buying massive chillers for the air side and immense CDUs for the liquid side, thereby expanding our infrastructure footprint. Why do we build these expensive, heavy hybrids? For the same reason people choose to buy the car: security.
They can drive in the city and on the motorway using electricity and petrol. In a hybrid-cooled data centre, legacy workloads and high-density AI can operate in the same room. It’s an expensive bridge, granted, and you’re likely paying for 100% air capacity and 80% liquid capacity at the same time. But right now, it’s the main vehicle available that’s transporting us from the past into the future.
Abandoning legacy thinking
So, we’re facing a thermal wall. To cope, we’re building complex, heavy hybrid bridges. And we know that physics dictates there are four ways to get the heat out. The temptation now is to go and buy a bigger chiller, a stronger pump. But that’s legacy thinking. That’s trying to smash through the wall with brute force. To handle the gigawatt era, we don’t need incremental steps or a plaster. We need a fundamental shift in how we connect the chip to the atmosphere and science.
To illustrate my point, take Carrier QuantumLeap™. It isn’t just a product brochure. It’s Carrier’s framework for integration. It’s our approach to stopping the ‘slicing’ problem; a methodology designed to take the heat from the AI chip, manage the hybrid cooling loop, and deliver it efficiently to the four destinations.
Here’s the takeaway: Density is difficult to avoid. Hybrid is expensive. But it’s necessary, because the laws of physics aren’t going to change. The biggest risk to a data centre project isn’t buying the wrong chiller or CDU. It’s the risk generated by the gap between them, of failing to connect the physics of the chip to the physics of the destination. We’re entering an era where the data centre is no longer just ‘a big building with air conditioning’. It is a complex thermal power plant with a promise of untold future opportunities.
So, the challenge is this: before buying any components, stop looking at the chiller in isolation from the rack. Instead, start designing for the entire thermal lifecycle. It’s a proven method to provide what data centres need right now: a quantum leap.
Learn how the Carrier QuantumLeap™ team can help you embrace the future of data centre cooling at quantumleap.carrier.uk.
Source
1. https://news.microsoft.com/source/features/sustainability/project-natick-underwater-datacenter/
