Andrew Schaap is CEO and Board Member of Aligned Data Centers.
getty
A team of scientists uses a supercomputer to zero in on the most effective Covid vaccine. A busy executive saves time by prompting ChatGPT to draft an email. Taking stock of its contents, your smart fridge tells you what to pick up at the grocery store.
What do those three things have in common? They’re all powered by data centers, the backbone of the digital world.
They also generate heat and lots of it.
As a CEO in the rapidly growing data center industry, I’m excited by the massive demand that AI, machine learning and supercomputing are creating for our business. Besides making people’s lives easier and helping companies run better, there are many examples of how these advanced technologies are doing good for humanity.
But for data centers, they also pose a big challenge: how to keep all that hardware cool.
And what those unfamiliar with this space don’t know is that traditional cooling methods have started to outlive their usefulness. That’s putting data center providers in a race to create a cooling technology for the modern world—one that not only turns down the heat for high-performance computing but also uses energy more efficiently.
The sooner we can do that, the better. Demand for data centers will only keep growing, with their power consumption in the U.S. projected to double between 2022 and 2030.
Data Center Cooling 101
For data center providers, keeping things cool for customers is a priority. Doing that means responding to the laws of physics.
When electricity travels through a semiconductor, it creates heat. The more powerful the chip, the hotter it runs. Nvidia’s latest graphics processing unit (GPU)—which underpins many cutting-edge AI applications—contains a staggering 80 billion transistors that need cooling. It also uses as much energy as the typical resident of a U.S. household.
Pulling the resulting heat from a circuit board and its wiring is crucial to stop it from malfunctioning or even breaking down.
So far, the data center industry has mostly relied on air cooling to get the job done. This technology uses two sets of air ducts, bringing in air from outside and pushing it across the front of the equipment, then sending the warmed air to the back so it can be expelled.
The downsides? Besides consuming energy and water—needed to chill the machinery—the typical air-cooling system is bulky, taking up precious real estate. With its many moving parts, it’s also prone to breakdowns.
For data center providers, those sustainability and reliability concerns have prompted a race to develop alternatives to air cooling. The rise of GPUs and other hot-running hardware has only upped the ante over the past couple of years.
Enter Liquid Cooling
Increasingly, data centers are turning to liquid cooling to meet their needs.
Before I continue, let’s be clear: In the data center world, there is no flux capacitor that will solve everyone’s cooling problems overnight. Until innovation in electricity takes a quantum leap with superconductors that let computers operate at room temperature, we’ll have to make do with incremental improvements to existing technologies.
My money is on liquid cooling for one simple reason: Water is a better heat rejector than air, making it ideal for GPUs and other high-density computing hardware. Historically, the argument against liquid cooling has been that water shouldn’t be anywhere near electronic equipment. But the huge demand for Nvidia and other AI chips has brushed those concerns aside.
Liquid cooling systems also use less power—and, perhaps surprisingly, less water—than their air-driven rivals. They have fewer moving parts, too. From a maintenance point of view, it’s like the difference between an electric vehicle and one with an internal combustion engine.
But no matter what happens, some air must flow at data centers. Most facilities will keep using air to cool equipment running at lower computing densities, and not all customers are keen to make dramatic changes.
Why Liquid Is The Cooling Tech Of Tomorrow
Although the jury is still out, I believe liquid cooling is the future for data centers and will ultimately become the status quo.
Why? The business case is compelling. For customers and investors, liquid cooling means less capital expenditure for the same output. The payoff: Companies that depend on data centers will be able to crush the competition by making higher margins.
Switching to liquid cooling also means better water and power usage effectiveness (WUE and PUE), two key metrics in our industry. Compared to air cooling alone, it can shrink facility power by almost 20% and total data center power by 10%, a recent study found.
The leap to liquid cooling will happen without anyone really noticing. As innovation sweeps the data center industry, legacy cooling equipment will inevitably be replaced. To me, it’s like the gradual shift from the old MD-80 passenger plane to the Dreamliner that is now an airline industry standard.
Helping drive that change is an enormous amount of infrastructure investment. From $250 billion in 2022, annual capital spending on data centers is forecast to surge to more than $500 billion by 2027. With investment in high-performance servers for AI set to grow at five times the outlay on their general-purpose counterparts, expect liquid cooling to take center stage.
Of course, for some, the transition to liquid cooling will mean overcoming a few hurdles. It can be capital-intensive, and there’s still a lack of standardization for this relatively new technology, along with a learning curve to make the shift.
But looking ahead, one thing’s for sure: Because AI and so many other industries rely on data centers, innovations that serve customers better will only keep gathering pace.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
