By Brooks Vaughan
Sr. Product Marketing Manager
Liquid cooling was initially proposed as a thermal solution to a thermal problem; some computing operations needed a better cooling medium because they simply ran too hot for air to do the job. That’s still true but the story is now bigger. Liquid cooling’s importance has grown to encompass energy efficiency compliance, financial forecast reliability, asset longevity, and regulatory permitting. Facing increasing rack densities and tightening regulatory imperatives, operators are driving liquid cooling adoption rates ever higher—particularly for new construction.
Hot Air vs Cool Water
It comes down to physics.
Air cooling is expensive. Traditional data centers rely on room air conditioners, chillers, cooling towers, and massive arrays of fans, all of which consume a lot of power. Air has a low thermal capacity, making it a poor medium for moving heat; facilities have to put a lot of power behind the effort just to make it work.
Air cooling is inefficient. Water's thermal conductivity is roughly 25x that of air, so liquid cooling can remove the same amount of heat with far less energy input. Direct liquid cooling (cold plates, immersion) targets heat at the source—the chip—rather than cooling an entire room.
The engineering required for liquid-cooling systems is considerably more complex but, at scale in GPU-dense AI data centers, the operational result is much more efficient and cost-effective, reducing total facility overhead and boosting power usage effectiveness (PUE).
Air Cooling Is Near Its Practical Limit for AI Data Centers
For those AI data centers, which generate more heat than many air-cooling systems can handle at all, liquid cooling is less a luxury and more a necessity.
Air cooling is viable up to only about 50 kW per rack with strict containment, high-efficiency systems, and rear-door heat exchangers. In contrast, current AI training workloads demand 100–200 kW per rack.
The market data reflects this reality. In 2024, analysts estimated that liquid-based cooling captured 46% of the data center cooling market, with direct-to-chip cooling responsible for nearly half of that. Liquid cooling is even stronger in the AI training server submarket, where Goldman Sachs projects liquid cooling will have a 72% penetration in 2026..
The PUE Gap Is an Energy, Cost, and Compliance Gap
Average data center PUE was around 1.56 in 2024, according to Uptime Institute's survey, meaning more than half of energy consumed doesn't contribute to compute. PUE isn't just an efficiency metric: it's increasingly regulatory exposure.
The EU requires PUE below 1.3 by 2030. Singapore permits new data centers only with PUE under 1.2. China's policies require new large data centers to achieve PUE better than 1.3 by 2025, with national hub nodes held to 1.25 and 1.2.
Liquid cooling consistently achieves PUE below 1.2, and often as low as 1.05 in advanced implementations. For larger and hyperscale data centers, switching to liquid cooling could save millions in electricity costs every year.
Thermal Predictability as Financial Governance
A more nuanced argument in favor of liquid cooling suggests that it makes energy cost projections, a key financial governance variable, more accurate by enabling tightly controlled and measurable thermal management. Look at it this way: If thermal behavior is unstable, energy consumption is unpredictable. And if energy consumption is unpredictable, financial forecasts are unreliable.
Amazon Web Services says it developed a custom liquid cooling solution that reduced mechanical energy consumption during peak cooling periods by 46%, lowering Its global PUE to 1.15 in 2024—well ahead of regulatory thresholds. Microsoft research, published in “Nature” in 2025, found that liquid cooling technologies reduce greenhouse gas emissions by 15–21% compared to air cooling across their full lifecycles.
Liquid Cooling Is the Future—And the Present
The data center liquid cooling market was $4.8 billion in 2025 and is projected to grow 18% annually, reaching $27.1 billion by 2035 (Global Market Insights, 2026). Many industry analysts predict that every new AI data center will be fully liquid cooled within two to three years.
Take the Next Step: Find Your Ideal Liquid-Cooling Solution at Delta
The liquid-cooling decision is the efficiency decision, the compliance decision, and the financial reliability decision, all wrapped up in one. Delta's liquid-cooled energy storage systems maintain battery cell temperature differentials within 3° C, extending battery lifespan and eliminating thermal runaway risk during the high-current charge/discharge cycles common to AI compute environments.
Visit Delta’s Liquid and Air Cooling web page
The “Rise of the AI Data Center” white paper lays out a microgrid strategic framework as "a new normal for power." Read the white paper
Want to see how much your data center could save with only minor improvements in efficiency? Check out our Power Efficiency Savings Calculator
To start this blog series at the beginning, go to Load Volatility: The Invisible Killer in AI Data Centers
Follow us on LinkedIn
Q&A
Is liquid cooling relevant only for large hyperscale data centers?
No. The key driver is rack density, not scale, so any facility running GPU-intensive AI workloads—whether for training, inference, or both—will encounter heat loads that make air cooling inefficient or unworkable. And regulatory pressure on PUE in places like the EU, Singapore, and China apply regardless of facility size.
Does liquid cooling require replacing an entire facility’s infrastructure?
Not necessarily. Many operators begin with a hybrid approach, deploying direct-to-chip or rear-door heat exchanger solutions in the highest-density zones while retaining air cooling elsewhere. This phased transition manages capital expenditure and avoids disrupting existing workloads. Full immersion cooling is typically adopted in new construction or major retrofits.
How does liquid cooling affect hardware maintenance and reliability?
Generally positively. Lower and more consistent operating temperatures reduce thermal stress on components, and tighter temperature control also means hardware performs more predictably over time. But liquid systems introduce fluid management issues such as leak detection and coolant maintenance, which require different operational expertise.
If liquid cooling uses electricity too, how does it actually improve energy efficiency?
The efficiency gain comes from the physics of heat transfer. Water conducts heat roughly 25 times more effectively than air, so liquid systems remove the same amount of heat while consuming far less energy than the fans, chillers, and air conditioning units they replace. Even accounting for pump energy, the net power draw of a liquid cooling system is substantially lower than an equivalent air-based system.