Data center liquid cooling in 2025: Evidence, economics and the road to mainstream adoption

Download Listen

Liquid cooling is quickly becoming a cornerstone of data centre design as AI workloads drive up rack density and thermal demand. With direct-to-chip cooling emerging as the leading solution for hyperscalers, the technology is now seeing broader adoption across high-performance and edge environments. This article explores the technologies behind liquid cooling, compares their architectures, and assesses the business case for legacy and greenfield data centres. As efficiency and sustainability take centre stage, liquid cooling is poised for rapid growth and long-term relevance.

Why is the liquid cooling market accelerating?

The densification of data centres is firmly entrenched as a key determinant of cooling vendor product development investment. With the recent boom in spending on ‘AI factories’ showing no signs of abating, the weight of the market, and opportunity of collecting a slice of this AI pie, is pulling vendors firmly in the direction of enabling dense (>100kW) AI racks. Such racks require a majority liquid cooling solution, and direct to chip appears to be emerging as today’s preferred technology for the hyperscalers, whose preferences play a significant role in determining industry direction of travel. For example, Microsoft’s recent Sustainability Report 2025 outlined their lean towards closed-loop direct-to-chip cooling and its sustainability benefits.

Direct-to-chip (D2C) cooling architectures involve two main components:

• Chip-adjacent cold plates – small heat exchangers positioned in direct contact with CPU/GPUs, and responsible for the transfer of heat away from these most energy-intensive components, which are in turn responsible for the bulk of waste heat. This waste heat is transferred via conduction to the liquid coolant within the cold plate, and subsequently pumped away from the rack by the CDU.

• Coolant Distribution Unit (CDU) – responsible for pumping rack-side coolant through a loop of piping in a rack (IT-side), and heat transfer between this and the wider building loop, the chemistry of which is isolated from the rack-level loop and chemistry. CDUs can either distribute coolant to either a single rack or a row of racks.

The key competing technology for D2C under the liquid cooling umbrella is immersion cooling. The key difference in immersion cooling is that the IT hardware is submerged in a dielectric coolant, as opposed to being in direct contact with a cold plate containing the liquid coolant. Immersion cooling requires a complete redesign of racks as watertight ‘pods’, as well as cutting edge coolant chemistry and a CDU specifically designed to circulate coolant in an immersion cooling pod.

Both D2C and immersion architectures are most commonly found using a ‘single phase’ approach, where the coolant remains in liquid form throughout the process. However, a two-phase approach, pioneered by companies such as LiquidStack, where the coolant is boiled and condensed elsewhere can allow for a greater intensity of heat transfer, and as an enabler of rack densities towards the far end of Nvidia’s roadmap, which charts an accelerated increase in densities to 600kW racks by the end of 2027.

Is cooling innovation limiting AI innovation?

In short, no. Between these two technologies, cooling technology is not the limiting factor for AI scalability — power generation and distribution are. While D2C is likely to dominate in the next 12-24 months, if rack density forward guidance deadlines are met, and the weight of greenfield building demands immediate scaled implementation of such cutting edge densities, then we expect to see a rise in immersion cooling barring technological breakthroughs in the D2C cooling field.

However, ‘legacy’, non-AI cloud computing workloads are not going anywhere, and indeed are projected to grow significantly with the long tail of global digital transformation and migration to off-premise hosting, as well as AI-adjacent workload growth (e.g. generative AI for code generation lowers the marginal cost to code generation, in turn growing the demand for physical infrastructure underpinned this generated code).

This shift is placing greater pressure on the infrastructure that keeps data centres running efficiently. Cooling systems, in particular, are struggling to cope with the increased thermal load generated by high-density, compute-intensive workloads. Traditional air-based cooling, long the standard in most facilities, is increasingly unable to manage the heat generated by modern, high-density hardware without significant trade-offs in energy use, space, or performance.

To address these limitations, operators are turning to liquid cooling, a more efficient and scalable approach that removes heat directly from components using fluid-based systems. Liquid cooling not only improves thermal performance but also enables greater compute density and supports long-term sustainability goals. As a result, industry demand is accelerating, with the global liquid cooling market expected to grow from US $4.1 billion in 2024 to US $19.4 billion by 2031.

What is liquid cooling?

Liquid cooling refers to the use of liquids, usually water or special non-conductive fluids to dissipate the heat generated by IT equipment such as servers, CPUs, and GPUs. It’s a more thermally efficient alternative to traditional air cooling systems, which rely on fans and chilled airflow to regulate temperatures.

The key principle is simple: liquid absorbs and transfers heat more efficiently than air. This allows data centres to cool components more effectively, enabling greater computing density and reduced energy consumption.

There are several forms of liquid cooling:

• Direct-to-chip (D2C) cooling: Coolant is delivered directly to metal plates (cold plates) mounted on high-heat components like processors and accelerators. This allows heat to be absorbed right at the source and transferred away rapidly.

• Immersion cooling: Entire server units are submerged in a non-conductive (dielectric) liquid. Heat is drawn away from all components simultaneously, and the warm liquid is circulated and cooled externally.

• Rear door heat exchangers: Liquid-cooled panels are fitted to the back of server racks. These absorb and remove heat from the exhaust air before it enters the data hall, reducing the ambient temperature.

How does liquid cooling work in practice?

The process may vary by cooling type, but the general steps are as follows:

1. Heat absorption

In direct-to-chip cooling, a closed-loop system pumps cold liquid through pipes and onto cold plates attached to the processor and memory. In immersion cooling, the entire server is bathed in coolant, which immediately begins absorbing heat from every surface.

2. Heat transfer

The now-warmed liquid exits the server and travels through a piping network to an external heat exchanger, a system that removes the heat from the liquid and transfers it into another medium (often water or air). This prevents overheating and keeps the cycle running smoothly.

3. Heat rejection and recirculation

Once cooled again, the liquid re-enters the loop and returns to the servers to repeat the process. Heat may be rejected outside the building using cooling towers, dry coolers, or even reused for district heating or other sustainable purposes.

What are the key advantages of liquid cooling?

• Improved thermal performance: Liquids conduct heat far more efficiently than air, enabling effective removal of heat directly from high-power components like CPUs and GPUs. This allows systems to maintain optimal performance under intense workloads.

• Greater compute density: With better thermal control, servers can be packed more closely together without overheating. This supports higher-performance configurations in the same physical footprint, ideal for AI and HPC environments.

• Lower energy use and environmental impact: Liquid cooling reduces reliance on large-scale air conditioning systems, leading to lower energy consumption — by as much as 40% compared to traditional air cooling. This not only lowers operating costs but also supports sustainability goals, especially when combined with heat reuse strategies.

• Reduced noise and maintenance requirements: Liquid cooling systems generate less noise and involve fewer mechanical components, such as fans, resulting in quieter operation and lower ongoing maintenance needs compared to traditional air-based systems.

What challenges does liquid cooling face?

Despite its clear advantages, liquid cooling presents several challenges that data centre operators must consider:

• Higher upfront investment

The initial cost of deploying liquid cooling infrastructure, such as specialised racks, cooling loops, and facility retrofits, can be significantly higher than traditional air-cooling systems. This may pose a barrier for smaller operators or facilities with limited capital budgets.

• System design and operational complexity

Implementing liquid cooling requires careful planning and specialised engineering expertise. Designing an efficient system that integrates with existing data centre infrastructure can be complex, and while many solutions are becoming more standardised, ongoing management may still demand specific technical knowledge.

• Fluid management and leakage risk

Although modern systems are designed with safeguards, there remains a small risk of coolant leaks, which could damage sensitive IT equipment if not properly contained. This necessitates rigorous monitoring, high-quality seals, and robust failure protocols.

What types of providers use liquid cooling?

• Hyperscalers and cloud operators

Hyperscalers such as Amazon Web Services (AWS), Microsoft Azure, and Google Cloud are investing heavily in liquid cooling to support large-scale AI and machine learning workloads, which demand vast amounts of power and produce high thermal loads.

• High-performance computing (HPC) facilities

National laboratories, weather modelling centres, and academic institutions operating supercomputers often adopt immersion or direct-to-chip cooling due to the extreme processing demands of their workloads.

• Edge and colocation providers

Edge data centres, smaller facilities located closer to users, often face space and cooling constraints. Liquid cooling helps enable high-performance compute at the edge while maintaining efficiency and minimising noise.

Colocation providers that offer rack space to multiple customers are also turning to liquid cooling as a premium service, particularly for customers running GPU-intensive AI, analytics, or financial modelling applications.

Is liquid cooling ready for mainstream adoption?

Liquid cooling represents a significant evolution in data centre design. While not yet mainstream across all facilities, it’s rapidly gaining traction among those prioritising high-density computing, energy efficiency, and sustainability. As AI, edge computing, and digital services continue to grow, so too will the need for smarter, greener, and more powerful cooling methods

Frequently Asked Questions (FAQ)

What is the difference between direct-to-chip and immersion cooling?

Direct-to-chip cooling places cold plates directly on heat-generating components, while immersion cooling submerges the entire server in a dielectric liquid. Both improve thermal performance but differ in complexity and hardware requirements.

Why is liquid cooling important for AI data centres?

AI workloads generate intense heat due to high-power GPU configurations. Liquid cooling enables higher performance and greater rack density without thermal bottlenecks.

Can liquid cooling reduce energy costs?

Yes. By reducing the need for large air conditioning systems, liquid cooling can lower cooling-related energy use by up to 40 percent.

What are the risks of using liquid cooling?

Modern systems are safe, but there is still a small risk of leaks. These are mitigated through high-quality seals, monitoring, and engineering controls.

Is liquid cooling only for hyperscalers?

No. While early adoption was driven by hyperscalers, edge and colocation providers are now deploying liquid cooling to meet performance and efficiency demands.