Posted on by

Air cooling remains prevalent, but liquid cooling is gaining momentum

Source: .shock/Storage/Adobe Stock.

Datacenter infrastructure, encompassing power, cooling and rack systems that support the operation of core IT hardware, remains a topic of interest and concern in our survey data. The rise of accelerated computing, particularly in AI, is rapidly increasing rack density, posing challenges for traditional air cooling and accelerating the shift toward liquid cooling solutions. A survey conducted by S&P Global Market Intelligence 451 Research focused on liquid cooling technology seeks to understand the key dynamics driving the datacenter liquid cooling market.

The Take

Datacenter infrastructure, which includes power, cooling and rack systems, is essential for supporting and protecting core IT functions such as computing, storage and networking. It ensures high availability and reliability, ideally in a sustainable manner. However, as rack density increases, both power and cooling systems face challenges, particularly in transitioning from air cooling to liquid cooling. Datacenter operators are working to choose the right technology and architecture to handle changing workloads, improve efficiency, reliability and sustainability, and lower costs.

The hybrid nature of IT infrastructure will persist, with air cooling still the most common method. However, the rise in server power and rack density driven by AI is pushing for quicker adoption of liquid cooling. Various types of liquid cooling deployments exist, with direct-to-chip technology leading the way. Improving standardized connections between coolant distribution units (CDUs), manifolds and servers could help speed the use of liquid cooling technology.

Summary of findings

The hybrid nature of IT infrastructure remains consistent. Cloud services broadly — including infrastructure as a service, software as a service, platform as a service, and hosted private cloud — are the predominant choice for workloads, with 79% of participants (versus 73% in 2024) indicating that their organization uses these services. Additionally, a significant number of organizations maintain their own server rooms or server closets (51%, versus 47% in 2024), as well as datacenters (38%, versus 39% in 2024). A smaller portion of organizations own facilities but rely on a facilities management provider for operations (20%, versus 16% in 2024), or use colocation services (18%, versus 17% in 2024).

Air cooling remains the predominant method, but liquid cooling is gaining ground. When asked about the type of cooling system used in their primary data centers, 45% of respondents (versus 48% in 2024) report having a fully air-cooling system, while 42% (up from 38% in 2024) indicate a combination of air and liquid cooling systems. Additionally, 12% (versus 13% in 2024) say they use a purely liquid cooling system.

Significantly, 59% of respondents (versus 56% in 2024) say their organization plans to implement liquid cooling in the next five years, with 21% (up from 13% in 2024) intending to do so in the next 12 months, 26% (versus 31% in 2024) in two to four years, and 12% (identical to 2024) in five years. Just over half of respondents (51%) anticipate that the proportion of their datacenter loads supported by liquid cooling will rise, while 40% expect it to stay the same.

The types of liquid cooling deployments remain varied, with direct-to-chip maintaining its leading position. When asked about the liquid cooling technologies their organizations use or are considering, 45% of respondents (versus 46% in 2024) cite liquid-to-air direct-to-chip, while 41% (versus 43% in 2024) choose liquid-to-liquid direct-to-chip. Additionally, 41% (versus 32% in 2024) opt for rear-door heat exchange (RDHX), 37% (versus 33% in 2024) select full immersion, and 23% (versus 16% in 2024) indicate chassis immersion.

Addressing the rise in server power is recognized as the top benefit of liquid cooling, surpassing the advantage of higher rack density. More than half of respondents (55%, up from 45% in 2024) identify accommodating increased server power as a key benefit of liquid cooling. This is followed by 51% (versus 54% in 2024) citing higher rack density related to GPU deployments. Additional benefits include improved facility power usage effectiveness at 49% (up from 39% in 2024), optimized total cost of ownership at 46% (up from 38% in 2024), increased chip thermal design power at 38% (up from 27% in 2024), and quieter operation at 38% (up from 33% in 2024). Notably, most benefits are cited at substantially higher rates in 2025 versus 2024.

High costs and lack of perceived necessity remain the primary barriers to widespread adoption of liquid cooling. Despite being available for decades, liquid cooling has not yet gained significant traction. When asked about the reasons for this, 56% of respondent (compared to 55% in 2024) cite the high costs of installation and maintenance, while 53% (versus 49% in 2024) indicate that air cooling remains adequate for most cooling needs. Additionally, 43% (versus 28% in 2024) point to operational habits and resistance to change, 29% (versus 30% in 2024) mention the lack of standardization for connecting components, and 29% (versus 28% in 2024) highlight a shortage of skilled personnel.

Rack density in liquid cooling deployments is anticipated to exceed 50 kW. When asked to specify the typical power densities of their liquid-cooled racks (current or planned), most respondents report densities from 50 kW to 200 kW. When allowed to select multiple options, the most commonly cited ranges are 50 kW-80 kW (37%, versus 31% in 2024), 80 kW-120 kW (39%, versus 44% in 2024), and 120 kW-200 kW (23%, up significantly from 6% in 2024).

There is a broad consensus that improving standardized interconnections between CDUs, manifolds and servers could speed the adoption of liquid cooling. More than three-fourths of respondents (79%, versus 76% in 2024) agree that the absence of standardized interconnections between components has hindered or delayed their organization’s implementation of liquid cooling technology. Specifically, 26% (versus 22% in 2024) strongly agree, while 53% (versus 54% in 2024) somewhat agree.

What is an AI Datacenter?

Read Report

Want insights on datacenter trends delivered to your inbox? Join the 451 Alliance.

Published by Perkins Liu

Perkins Liu is a senior research analyst on the Datacenter Services & Infrastructure team of 451 Research, a technology research group within S&P Global Market Intelligence. He covers datacenter activities in China and datacenter technology more broadly. He joined S&P Global Market Intelligence in April 2021. Before that, Perkins had been in the datacenter industry in different roles for more than 20 years, most recently as a leader of strategy and business development for the cloud and service provider segment at Schneider Electric. Perkins’s research areas include supply and demand activities, market share analysis, industry growth projections, government incentives and regulation, and pricing dynamics of leased datacenters, as well as the technology innovation trend in datacenter industry. His recent research topics include alternative energy sources of datacenter and liquid cooling in datacenters. Perkins holds a bachelor’s degree in electrical engineering from Beihang University and an MBA degree from the University of Rhode Island, where he was honored to join the Beta Gamma Sigma Business Honor Society, the highest recognition in a business program accredited by AACSB International. He is fluent in Chinese.