The Thermal Tipping Point: Why Air Cooling Can’t Handle AI’s Future

Future-proofing your data center starts with how you cool it. The method chosen is no longer a technical consideration, but a strategic one. The reason for this is simple: as data centers scale to support AI and increase their compute density, air cooling is hitting its physical limits, beyond which it is no longer useful. 

The IT industry has traditionally relied on air cooling for data centers, but as GPU-based AI systems and high-performance computing (HPC) consume more power and generate more heat, the technology simply cannot keep up. Data center operators are facing difficult decisions. How will they cater for user demand for more power while dealing with increasing temperatures, all without breaking their budgets? 

The answer lies in a new approach, which will be a gamechanger for the data centers of tomorrow. 

Air Cooling Has Reached Its Limit

Air-cooled systems can support up to approximately 70 kilowatts per rack in the data center, but hit a physical limit when it comes to removing heat known as ‘specific heat capacity’. Beyond this, operators need to look at alternative cooling methods. Adding to the complexity is the rise in demand for AI workloads across every industry, requiring five times the amount of power. The industry is now seeing a shift from chips operating at 120 watts to 600 watts or more. 

This change is pushing air cooling beyond the threshold at which it is viable. It mirrors a previous shift in the automotive industry, where manufacturers had to adopt liquid cooling for engines as consumers demanded more power. Citroen developed the last mass-produced air-cooled engine in the 1970s, with a few holdouts such as Porsche still using air cooling into the last years of the 20th century. Air-cooled cars are now largely limited to vintage or specialist equipment, whereas modern cars use liquid cooling.

Similar lessons can be applied to the data center industry. Water is the perfect choice because it has the highest capacity to retain and move heat out of any common liquid or gas. It is more than 3,000 times more efficient at moving heat than air. To put that in context, if walking was 3,000 times more efficient, you could stroll from London to Rome in the time it takes to brew a cup of coffee. 

Water makes an enormous difference to the efficiency of data centers, which use as much as half of their total energy on cooling systems. Liquid cooling techniques, such as direct-to-node cooling, can remove at least 98% of heat from servers, which can then be reused to heat facilities. Modern cooling systems are designed to be serviceable and safe, making them practical in a wide variety of environments. 

Liquid Cooling is Scalable, Practical and Proven

Liquid cooling is not an exclusive technology reserved for the biggest hyperscalers like Amazon, Google and Microsoft. As demand for AI grows, along with concerns over sustainable computing, we are seeing universities and Tier 2 and Tier 3 data centers (rated for reliability) adopting liquid cooling to keep pace. For those switching to liquid cooling, there is no need for a complete overhaul: the technology integrates neatly with existing infrastructure. The latest water-cooled systems work across multiple configurations, including hybrid systems using both air and liquid, which can make adoption easier. 

With the energy used by data centers predicted to rise 150% in Europe in the coming decade, according to think tank Ember, measuring the efficiency of data centers in power use (or Power Usage Effectiveness) will come under scrutiny. Liquid cooling is the easiest way to stay compliant while boosting performance, and warm-water cooling, which does not waste water to keep it cool, also avoids putting pressure on local water supplies. Energy is used to compute, not cool, and can play a major part in cutting power consumption. 

A Cleaner Future

Liquid cooling might seem like a daunting prospect, but the reality is that as demand for AI grows, the option is either to adopt liquid cooling or be left behind. With air cooling hitting its thermal barrier, CIOs and IT leaders who hope to drive efficiency, hit ESG targets and embrace a data-driven future have to look seriously at liquid cooling. Making the shift brings immediate benefits in terms of running costs, sets organizations up for a longer-horizon future.