The rise of AI, hyperscale clouds, and the general need for performance have significantly transformed the design and cooling of data centers in 2025. However, the rapid evolution of AI and adjacent technologies will change the scale of data centers and drive the adoption of even more sophisticated cooling technologies over the next decade, so let's examine what the future holds.
Just 10 years ago, in 2015, data center liquid cooling was primarily limited to specialized applications such as supercomputers and mining farms; its global adoption rate at the time was at best around 5%. Air cooling dominated due to lower upfront costs and simpler infrastructure. But by 2020, the adoption of liquid cooling began to rise and reached around 10% as cloud hyperscalers started to strive for efficiency . However, with power densities of approximately 5 –10kW per rack on average, air cooling was sufficient for most.
Fast forward a few years, and power density per rack increased to 15kW, 20kW, 30kW, or even 40kW (in the case of Nvidia H100-based racks). These soaring power densities led to the steady adoption of liquid, as the technology moved from specialized setups to enterprise and cloud data centers.
By 2024, liquid-based cooling had conquered 46% of the entire data center cooling market, but air cooling retained 54% of the market, according to Mordor Intelligence. Air cooling still accounts for the majority of legacy and mid-tier facilities, but liquid cooling is projected to dominate new builds, especially in AI and hyperscale cloud data centers, as power-per-rack continues to increase. For example, one Blackwell Ultra rack consumes up to 140kW, and this is not the end of the soaring power densities.
As cooling systems get more sophisticated, their costs increase, resulting in more expensive solutions. The global data center cooling market is valued at $10.80 billion in 2025 and is projected to reach $25.12 billion by 2031, growing at a compound annual growth rate of 15.11%, according to Mordor Intelligence. This strong expansion is driven by the ongoing shift from air-based to liquid-based thermal systems in AI and hyperscale facilities, owing to factors such as rising AI accelerator heat densities that make air cooling unsuitable.
Air cooling
Traditionally, data centers have used air cooling to keep operations running, and for many operators, that's not going to change any time soon. However, as thermal densities increase, the technology's limitations become more apparent. So, let's examine how it all works in practice.
(Image credit: Research Gate)
Air cooling in data centers works by circulating conditioned air to absorb and remove the heat produced by servers and networking equipment. Typically, air cooling keeps intake air around 21°C to 24°C — the range recommended for safe operation — by continuously pushing cool air toward the servers and drawing warm exhaust air back for reconditioning.
Most air-cooled facilities are organized using a hot aisle/cold aisle layout to prevent mixing of hot air and cold air, and to ultimately lower energy bills. Racks are arranged so that the fronts (intakes for cool air) of servers face each other across a cold aisle, while the backs (exhausts for heated air) face each other across a hot aisle. Some data centers go further by enclosing either the hot or cold aisles with physical barriers to completely separate air streams and minimize energy waste.
... continue reading