How Ecolab Tames the ‘Wild West’ of AI Data Centre Cooling
The AI revolution is reshaping the data centre industry at breakneck speed, forcing operators to abandon decades-old cooling paradigms and embrace technologies that bring unprecedented complexity and risk – but also transformative opportunities.
As power densities soar from 10 kilowatts to a projected one megawatt per rack, the industry finds itself navigating what insiders describe as chaotic territory – a ‘Wild West’ where established best practices struggle to keep pace with innovation.
Rob Lowe, Director RD&E – Global High Tech at Ecolab, has spent nearly 20 years in water innovation, with the past six years specifically focused on data centres. His perspective offers a unique vantage point on an industry in the midst of dramatic transformation.
“Everybody says it's the ‘Wild West’ in AI right now, because there’s not only the speed of change but everybody is trying to figure out the best practice at the same time,” he explains.
Against this backdrop of uncertainty, Ecolab is establishing order and empowering its customers through collaborative standards development, advanced monitoring technologies and the application of more than a century’s worth of water expertise to cutting-edge challenges.
The scale and speed challenge
The transition from air cooling to direct-to-chip liquid cooling represents more than just a technological shift – it marks a fundamental reimagining of data centre infrastructure.
“AI is changing everything. It's reshaping the way business is done,” Rob observes. “But it’s also reshaping the way that data centres are being built. They’re moving from CPUs to GPUs, and those are getting faster, hotter and they're using more electricity.”
The evolution from CPUs to increasingly powerful GPUs has created heat densities that air cooling simply cannot address – but there is an alternative.
Direct-to-chip liquid cooling itself is hardly a new technology. Supercomputers have employed these systems for years. What distinguishes the current moment, according to Rob, is “really the scale of it and the speed of implementation”.
Data centre operators are racing against the clock, driven by what the industry calls speed-to-first compute – the crucial timeline between breaking ground and training AI models.
“Getting those data centres set up so they can train the models and then do the work of the AI algorithm to produce the results is critical to their business success,” Rob says. “It’s big implementations, and they’re unfolding very, very fast right now.”
This urgency creates a perfect storm of risk. Traditional air cooling posed minimal threat to sensitive computing equipment.
Liquid cooling, by contrast, introduces coolant directly into server racks, down to individual chips mounted on cold plates.
“Just think about the complexity and the risk that comes along with that. Blowing air across chips to keep them cool is one thing,” Rob notes. “Now we’re pumping water or some coolant into the system to do that, and the risk and the complexity are just amazing.”
Unknown operational realities
The rapid deployment of liquid cooling infrastructure has revealed troubling knowledge gaps. Unlike mature data centre operations built on decades of institutional learning, AI-focused facilities are being constructed and commissioned without established operational playbooks.
“This entire operation, the AI business, is coming into being as we speak,” says Rob.
“It’s not like these companies have 30 years of experience, and there’s a leader in the industry. Everybody is trying to figure this out at the same time, and we don't know how the operational conditions are going to impact the systems.”
One particular concern centres on glycol-based cooling systems. Traditionally viewed as a “fill it and forget it” solution, according to Rob, glycol has earned a reputation as a commodity product requiring minimal oversight.
This perception, while based on experience with conventional utility loops, may prove dangerously inadequate for AI data centre applications.
“There are so many unknowns in the AI systems right now,” warns Rob. “How many touchpoints are we going to have in these systems that could potentially introduce contamination or dilution events that could change the coolant health?”
The stakes extend beyond operational efficiency. Glycol degradation, material compatibility issues and unexpected corrosion byproducts can threaten assets worth billions of dollars. “I’ve seen several examples where some unknown materials are showing up, and suddenly there’s a corrosion byproduct that they weren't expecting,” recalls Rob.
When investigating these incidents, basic questions arise: “Did you have the right stainless steel? Were your tubes cured in the right way? Because all those things can impact what's going on in the system,” says Rob.
Applying legacy expertise to novel challenges
Ecolab’s response to this uncertainty draws on its century-long history in industrial water treatment and cooling systems.
The company’s 3D TRASAR technology, which has monitored and maintained water systems for decades, is now being adapted for direct-to-chip applications.
“We’ve been using 3D TRASAR in cooling systems for a long time, more than 30 years,” Rob explains. “We are experts at monitoring fluid and understanding what's going on, and then taking that information and saying, ‘We can help you prevent something from happening because of this change. We see it coming.’”
This expertise enables what Rob describes as “the speed of insight to action”: the ability not merely to collect data, but to interpret it and respond proactively. For glycol-based liquid cooling systems, this means continuous monitoring capable of detecting subtle changes in fluid composition, identifying contamination events and predicting degradation before it impacts operations or damages equipment.
The company’s approach acknowledges that while the technologies may be new to data centres, the fundamental challenges are not.
“The problems that we see will be the same. They’re just going to manifest in different ways,” Rob suggests. “There are going to be different ways we need to troubleshoot because of materials of construction, operational conditions, or other operational dynamics. But the fundamentals are still the same.”
Collaborative standards development
Recognising that technical solutions alone cannot address industry-wide uncertainty, Ecolab has positioned itself at the centre of collaborative standards development.
The company is working with the Open Compute Project (OCP) and partnering directly with equipment manufacturers to establish best practices.
A recently published white paper co-authored with Nvidia focuses on pre-commissioning procedures – a starting point Rob identifies as offering “the most value and the most need around how you set these things up”.
Additional white papers addressing commissioning practices, operational best practices, and monitoring protocols are in development. “We’re working with our customers – customers and the OEMs – to really drive some certainty and get rid of some of the crazy Wild West aspects of the AI data centre space right now,” says Rob.
This collaborative approach has accelerated Ecolab’s own innovation cycle dramatically.
The company’s 3D TRASAR programme for direct-to-chip cooling moved from whiteboard concept to commercial deployment in less than 12 months – a timeline enabled by intensive customer engagement.
“We sketched out the scenarios with a simple proposition: ‘Here’s everything that we think we might need to have to help us monitor what's going on in the glycol loop.’ In less than 12 months, we had a commercial installation of that device,” recounts Rob.
The rapid iteration process involved customers at every design stage, with weekly cycles of feedback and refinement.
“We didn't wait for a complete prototype to be built, and then take it out to a customer site and try it,” explains Rob. “We had the iterative approach where we said, ‘Here’s the design, let's take a look at it together.’ That was the key. It's increasing the expediency, and this comes directly from the collaboration with our customers and the OEMs.”
Legacy infrastructure and future technologies
The industry’s evolution must also account for existing facilities. Not all data centres can be purpose-built for liquid cooling, creating demand for hybrid approaches that leverage existing infrastructure.
Rob points to Amazon Web Services’ recent announcement of in-row liquid-to-air cooling as an example of this practical reality.
“There is a lot of data centre space in operation today that is not set up for liquid cooling. We can't just ignore that,” he notes. “Legacy technologies are going to still be very important. It's not like we’re going to transition to entirely new things.”
Looking forward, Ecolab is tracking several parallel development paths. The company is investigating how long current liquid cooling technologies can support ever-hotter chips before more advanced systems – such as two-phase cold plates or immersion cooling – become necessary. “If we can prolong that transition, that's really valuable for the industry because you don't have to build an entirely new cooling system to cool the new chips,” Rob suggests.
Water reuse represents another emerging priority. “There’s going to be more and more non-potable, reclaimed water that can be used in cooling, and figuring out how to do that in the safe way,” says Rob. The challenge extends beyond basic water treatment to understanding how various water sources interact with complex cooling systems and what implementation considerations arise.
Fundamentally, Rob advocates for a holistic view that recognises the interdependence of water, energy and thermal management.
"We're heading towards this convergence,” summarises Rob. “I think that we have to view these as cooling systems versus cooling assets. You can't optimise to provide the best solution if you're not thinking holistically about the cooling process."
The transformation of data centre cooling infrastructure shows no signs of slowing. As liquid cooling becomes mainstream – potentially outnumbering traditional cooling towers within the near future – the ability to adapt quickly whilst maintaining operational reliability will separate industry leaders from laggards.
For Rob, the path forward is clear: "Liquid cooling is growing so fast. The thing is, it's only going to evolve even more quickly. The winners will be the ones that can adapt and scale the fastest. This collaboration in the water space is going to be something that we need to continue – this will define who the winners will be. But this depends on collaboration, speed and the ability to adapt to these new technologies as fast as possible."
