跳到正文
Data Center Knowledge· Christopher Tozzi·· 5 天前AI 評分39

浸沒式冷卻為何仍未成為主流

Why Immersion Cooling Still Hasn't Gone Mainstream

AI 導讀

浸沒式冷卻雖具能源效率及可持續優勢,在數據中心仍屬小眾,前期成本高昂及行業操作模式難以轉變是主要障礙。估算顯示,每 MW IT 負載的浸沒式冷卻系統約需 $1 million,直接芯片冷卻則需 $650,000。若 AI 訓練和推理令運算環境變得更密集、熱負載更高,浸沒式冷卻的吸引力或會增加。

正文

Technician Inspects Server in Immersion Cooling Tank

Immersion cooling offers strong efficiency and sustainability advantages, but it remains a niche approach in data centers.Getty Images

There is a great deal for data center operators to love about immersion cooling, a technique that keeps IT equipment cool by immersing it in liquid. It is highly efficient from an energy perspective. It requires no water. And because it eliminates many of the moving parts and airflow dependencies associated with traditional cooling, it can reduce certain operational risks.

And yet, very few data centers are actually using immersion cooling or appear to be planning a large-scale shift to it. Other liquid cooling methods – such as direct-to-chip cooling, which circulates coolant through pipes placed adjacent to processors rather than immersing entire servers – are projected to grow rapidly. Immersion cooling, by contrast, remains a niche.

That raises an obvious question: If immersion cooling is so effective and sustainable, why aren’t more data centers using it? And what technical, financial, or operational changes would the industry need to make for immersion cooling to take off?

Related:Liquid Cooling Retrofits: Hard, but Sometimes Worth It

To answer that, it is worth taking a closer look at the current state of immersion cooling, the barriers holding it back, and the odds that it becomes more mainstream in the years ahead.

Why Immersion Is Still a Niche

Immersion cooling technology has been around for decades. As of 2026, however, it still has not become a mainstream production cooling strategy among major data center operators. Most of the headlines on immersion cooling in recent years have focused on pilots, demonstrations, and proofs of concept, such as Microsoft’s well-known experiment earlier this decade.

Some of the large AI data centers under construction may be planning to deploy immersion cooling. In some respects, the technology is a natural fit for facilities designed for extremely high power densities and significant heat loads. But if large operators were preparing to embrace immersion cooling at scale, you might expect them to say so loudly. So far, few have.

What’s Holding Immersion Back

Why aren’t more data centers taking advantage of immersion cooling, given that it is a proven and reliable way to dissipate heat? Two factors likely explain much of the hesitation.

1. The Economics Are Hard to Ignore

First, immersion cooling systems are significantly more expensive to install than most alternatives. Estimates vary, but immersion generally carries materially higher upfront costs than direct-to-chip cooling and vastly higher capital costs than conventional evaporative or air-based approaches. According to one study, it costs about $1 million per MW of IT load to implement an immersion cooling system, compared to $650,000 using direct-to-chip. From a CapEx standpoint, it is hard to make immersion look attractive, especially for operators under pressure to bring new capacity online quickly and cost-effectively.

Related:RDHx: A Simple, Affordable Path to Lower Data Center Energy Use

To be sure, immersion cooling can deliver savings over time through lower operating costs. It can reduce energy required for cooling, improve thermal efficiency, and shrink the physical footprint needed to support a given amount of compute. Adam Silva Consulting estimates that “cooling a 64-rack AI cluster over 10 years costs $42M with advanced air, $31M with direct-to-chip liquid, and $28M with single-phase immersion.”

But those long-term benefits do not automatically outweigh the short-term economics. Data center operators may still be too focused on near-term financial targets to invest in systems whose payoff could take years to materialize. It may also be that, when compared to direct-to-chip cooling in particular, immersion’s operational savings are not compelling enough to justify the extra upfront expense.

2. The Industry Doesn’t Change Easily

Cost, however, is probably not the only factor. Part of the issue may also be cultural.

The data center industry is accustomed to designing and operating facilities in ways that have remained fundamentally consistent for decades. Those norms include placing servers in conventional racks, arranging hardware in familiar floor layouts, and relying on established approaches to maintenance, monitoring, and serviceability.

Related:Can Evaporative Cooling in Data Centers Still Make Sense?

Immersion cooling upends much of that model. With immersion systems, airflow is no longer a primary design constraint, which means data center architecture and white-space layouts can and arguably should change in more fundamental ways. Maintenance procedures change, too, because servers are no longer housed in traditional cabinets and are not accessed in the same way. Monitoring practices, service workflows, and technician training all have to evolve.

Direct-to-chip cooling also requires changes to data center design and operations, but overall, those changes are less disruptive than those immersion cooling demands. Direct-to-chip allows operators to preserve more of the conventional server-and-rack paradigm, making incremental adoption easier.

The cultural barriers to immersion cooling can be overcome. But doing so would require architects, operators, and technicians to rethink not just how they cool IT equipment, but what a data center should look like and how it should function.

Adoption Will Depend on Future Pressures

Whether that rethinking happens, and whether operators decide to make the financial commitments that immersion cooling requires, will largely depend on how data center cooling demands evolve in the coming years.

If power-hungry, heat-intensive workloads such as AI training and inference continue to expand, immersion cooling could become increasingly attractive. The denser and hotter the compute environment becomes, the stronger the case for cooling technologies that can efficiently handle extreme thermal loads. But if the AI boom plateaus before immersion gains real momentum, the technology may remain largely experimental in the data center market.

Another factor is the level of pressure operators face to reduce water and energy use. Public and regulatory concern about data center water consumption does not seem quite as intense as it did a year or two ago, which may reduce the urgency around solutions such as immersion cooling that use no water at all.

At the same time, concerns over data centers’ impact on power grids and electricity prices have helped drive moratoriums and tighter scrutiny in some markets. That could encourage more operators to seriously consider immersion cooling to improve power usage effectiveness (PUE) and reduce cooling overhead. That said, cooling is only one part of the energy equation. The main driver of data center power consumption is the electricity required to run IT equipment itself, with cooling systems playing a secondary role. For that reason, immersion cooling alone is unlikely to become a silver bullet for the industry’s broader power challenges.

Immersion cooling may yet find its moment. For now, even its technical advantages have not been enough to overcome the combined weight of higher upfront costs, operational disruption, and industry inertia.

About the Author

Christopher Tozzi

Technology Analyst

Christopher Tozzi is a technology analyst with subject matter expertise in cloud computing, application development, open source software, virtualization, containers and more. He also lectures at a major university in the Albany, New York, area. His book, “For Fun and Profit: A History of the Free and Open Source Software Revolution,” was published by MIT Press.

來源:Data Center Knowledge · datacenterknowledge.com