If you're planning a micro-module data center today, the cooling decision is no longer just about CRAC units and raised floors. Liquid cooling data center designs are becoming a practical option for mainstream deployments, not just for research supercomputers. This shift matters because rack densities are climbing past 20 kW per rack, and air alone struggles to keep up.
Why Air Cooling Hits a Wall
Air cooling works by moving heat from components to air, then pushing that hot air out of the room. But air has poor thermal capacity. For a typical 10 kW rack, you need around 3,000 CFM of airflow to keep inlet temperatures at 20°C. That's a lot of fans and a lot of energy.
When you push racks to 30 kW or more, the required airflow becomes impractical. You'd need massive ducting, high static pressure, and even then you risk hotspots. Thermal throttling kicks in as processors hit their temperature limits, which means performance drops just when you need it most.
In practice, most air-cooled facilities top out around 15-20 kW per rack. Beyond that, you're fighting physics. Liquid cooling changes the game by using water or other coolants that have far higher heat capacity. Water can absorb about 3,500 times more heat than the same volume of air, so you can move a lot more heat with much less infrastructure.
The Rise of High-Density Racks
Modern CPUs and GPUs are getting more powerful, but they're also getting hotter. AMD's latest EPYC processors, for example, can draw over 400 W each. When you pack 8 or 16 of those into a single server, you're looking at 3-6 kW per server. A rack with 10 such servers easily hits 30-60 kW.
That kind of density is becoming common in AI training, high-performance computing, and even some enterprise workloads. But traditional air cooling can't handle it efficiently. You'd need to space racks out, reduce density, and still risk overheating. That's a poor use of floor space and capital.
Liquid cooling lets you pack those racks tight. You can achieve 100 kW per rack or more with direct-to-chip cooling. That's a huge advantage if you're building a micro-module data center where space is at a premium.
Comparing Cooling Methods: Air vs. Liquid
Let's break down the main options. The table below gives a quick comparison of typical characteristics. Note that numbers vary by design and climate, but the trends hold.
| Cooling Method | Max Rack Density | PUE Range | Typical Cooling Cost |
|---|---|---|---|
| Air (CRAC/CRAH) | 15-20 kW | 1.3-1.6 | Medium |
| Air-Assisted Liquid | 30-50 kW | 1.2-1.4 | Medium-High |
| Direct-to-Chip Liquid | 50-100+ kW | 1.1-1.2 | High |
| Immersion Cooling | 100+ kW | 1.05-1.15 | High |
Direct-to-chip liquid cooling is the sweet spot for many micro-module deployments. It uses cold plates attached to the CPU/GPU, with a coolant loop carrying heat away. You still need some air cooling for other components like power supplies and drives, but the bulk of the heat is handled by liquid.
How Liquid Cooling Works in a Micro-Module
A typical liquid cooling data center setup has a few key components. First, there's the coolant distribution unit (CDU) that manages the flow of coolant to racks. Then, within each rack, you have manifolds that distribute coolant to individual servers. Finally, the heat is rejected to the outside via a dry cooler or cooling tower.
For micro-module data centers, the CDU can be integrated into the module itself. That means you get a self-contained cooling loop that doesn't require major facility modifications. VERHI offers modular solutions that can be configured with liquid cooling options, but you should always check specific specs with our engineers.
One of the big questions is whether to use water or a dielectric fluid. Water is cheap and has great thermal properties, but it conducts electricity. So you need careful leak detection and robust plumbing. Dielectric fluids, like those used in immersion cooling, are safer but more expensive and harder to manage.
Practical Steps for Adopting Liquid Cooling
If you're considering liquid cooling for your next micro-module data center, here's a rough roadmap. It's not exhaustive, but it'll get you thinking in the right direction.
- Assess your actual rack density. Don't assume you need liquid cooling if you're running 8 kW racks. But if you're planning for AI or HPC, plan for 30 kW+.
- Talk to your server vendor early. Not all servers support direct-to-chip cooling. Check for compatibility and warranty implications.
- Design for redundancy. In liquid cooling, a pump failure can stop cooling quickly. Plan for N+1 pumps and backup cooling modes.
- Consider the facility side. Where will the heat go? Dry coolers are common, but they need space and make noise. Check local regulations.
- Work with an experienced integrator. Liquid cooling is not just about swapping out AC units. It's a different approach to thermal management.
That last point is worth repeating. Liquid cooling data center projects fail when people treat them as a drop-in replacement for air cooling. You need to think about the entire heat path, from chip to outside air.
The Future Is Hybrid
I don't see liquid cooling completely replacing air cooling anytime soon. Most data centers will run a mix: air for low-density racks, liquid for high-density ones. That's already happening in colocation facilities and large cloud providers.
For micro-module data centers, the flexibility to mix cooling types is a big plus. You can start with air cooling and add liquid cooling later if your needs change. That's why modular designs are so popular – they adapt.
Another trend is using liquid cooling to improve PUE. By eliminating most of the fan energy and using higher temperature coolants, you can get PUE down to 1.1 or even lower. That's a big deal for sustainability goals and operating costs.
What About Retrofits?
If you already have an air-cooled micro-module data center, you might be wondering if you can retrofit liquid cooling. It's possible, but not always straightforward. You need to check your racks, servers, and piping. Often, it's easier to deploy a new liquid-cooled module alongside the existing one.
VERHI's modular approach means you can add a liquid-cooled module without disrupting your current operations. That's a practical path for many organizations.
Key Takeaways
- Liquid cooling is no longer just for supercomputers; it's becoming mainstream in micro-module data centers.
- Air cooling struggles above 20 kW per rack; liquid cooling handles 50-100 kW easily.
- Direct-to-chip cooling is a good middle ground, balancing cost and performance.
- Plan for redundancy and facility integration from the start.
- Hybrid cooling (air + liquid) is the realistic future for most deployments.
Frequently Asked Questions
What is the typical PUE for a liquid cooling data center?
Liquid cooling can achieve PUE values between 1.1 and 1.2, depending on the design and climate. Direct-to-chip systems often report PUE around 1.15, while immersion cooling can get closer to 1.05.
Is liquid cooling safe for electronics?
Yes, when implemented correctly. Direct-to-chip systems use sealed cold plates that don't expose electronics to coolant. Leak detection and proper maintenance are essential, but the technology is mature and widely used in HPC environments.
Can I retrofit liquid cooling into an existing air-cooled data center?
It's possible but often challenging. You need to check server compatibility, rack design, and facility plumbing. In many cases, it's more practical to deploy a new liquid-cooled module rather than retrofit the entire facility.
What is the cost difference between air and liquid cooling?
Liquid cooling has higher upfront costs due to CDUs, piping, and specialized racks. However, operational costs are lower because cooling is more efficient and you can run higher densities, which reduces the cost per kilowatt of IT capacity.
Thinking about liquid cooling for your next micro-module data center? VERHI's engineers can help you evaluate your options and design a solution that fits your needs.
