New Cooling Tech Tackles AI’s Soaring Energy Demands

A new cooling technology has been developed by engineers at the University of California, San Diego that can drastically improve the energy efficiency of data centers, which are crucial for AI development and deployment.

Data centers are facilities that house the infrastructure required to train and deploy AI models. This includes advanced compute, network, and storage architectures as well as energy and cooling capabilities to handle the massive data processing demands of AI workloads.

While traditional data centers have many of the same components as an AI data center, including servers, storage systems, and networking equipment, their computing power varies greatly. This is due to the extraordinary demands of high-intensity AI workload, which requires high-performance graphics processing units (GPUs).

Not only does the very large number of GPUs necessary for AI use cases require far more square footage, but they also require advanced energy and cooling capabilities.

In fact, the explosive usage of AI and its continuing expansion have the demand for data processing skyrocketing.

Data centers are actually projected to be among the largest consumers of energy globally. Estimates suggest that data center energy consumption could grow by as much as 160% to over 1,000 terawatt-h (TWh), accounting for 3% to 4% of the world’s electricity consumption, by 2030.

This surge is driven by the increasing computational requirements of advanced AI applications, leading to significant heat generation that necessitates effective cooling solutions. 

As a matter of fact, 40% of a data center’s total energy use goes toward just cooling the high-performance computing hardware, specifically GPUs and accelerators. 

These components generate a lot more heat than traditional CPUs, especially during AI training and large language model processing. Then there’s the increased density of computing power in modern AI servers, which also leads to higher heat concentration, requiring more efficient cooling solutions. 

“Cooling costs are a significant overhead cost operators must face,” said  Rithika Thomas, a senior analyst for sustainable technologies at ABI Research, whose ‘Chilling Out: Cooling Systems for Data Centres’ report calls for operators to adopt effective strategies to maintain performance, stability, and equipment lifespan.

If this trend continues, the global energy use for cooling could more than double by the end of this decade, making it crucial to find a solution to this problem.

“Effective cooling strategies demand a holistic, technology-agnostic approach to optimise Power Usage Effectiveness (PUE), Water Usage Effectiveness (WUE), and thermal management and reduce operating costs.”

– Thomas

AI’s Insatiable Heat Needs Smarter Cooling Solutions

The issue is that traditional air-based cooling methods are becoming increasingly inadequate to manage the heat loads of modern servers. These methods already consume up to 45% of a data center’s total energy.

When it comes to the maximum heat flux from advanced CPU and GPU chips, it is currently well above 50 W/cm². Heat flux or thermal flux is simply the rate of energy transferred from one place to another in the form of heat.

For instance, NVIDIA’s Hopper, an AI darling and a GPU microarchitecture specifically designed for data centers to accelerate AI and high-performance computing (HPC) workloads, has a thermal design power (TDP) of 700 W on an 814 mm² chip size (86 W/cm² heat flux) for AI applications. 

Hopper is NVIDIA’s successor to the Ampere architecture and is built with over 80 billion transistors using the TSMC 4N process. 

Then there’s the miniaturization of the transistors, which can further lead to about ten times higher transistor density in the 1 nm node by 2030. This can involve over 200 W cm−2 heat flux, a level that cannot be removed efficiently and cost-effectively with air cooling. As such, it creates a need for liquid cooling.

Liquid cooling technologies aren’t anything new. The industry has already adopted them for their next-generation systems to improve thermal management and energy efficiency.

This cooling method actually transfers the heat away from servers more effectively than air-based systems, which in turn, reduces cooling costs and energy waste.

Advanced liquid cooling techniques include microchannel cold plates and immersion cooling with dielectric fluids. Traditional cooling techniques, such as single-phase or boiling heat transfer, however, are usually limited to heat fluxes in the range of 100 W/cm².

So, what we need is high-performance heat dissipation strategies. They need to be passive ones, especially as that would require no additional power. While desirable, these strategies remain elusive.

Then there’s the fact that, cooling is important not only for CPU and GPU devices that are used for computing and AI but also for power electronics including light-emitting diode (LED) chips, high-power radio frequency (RF), gallium nitride (GaN) high electron mobility transistors (HEMTs), and pump lasers.

Many of these electronics have power dissipation over 100 W/cm².Here, evaporation heat transfer offers a promising technology for cooling high-power electronics, with several significant advantages over traditional cooling methods.

Compared to single-phase cooling systems, the large latent heat of liquid-vapor phase change allows for efficient heat removal while offering better stability and no hysteresis (the dependence of a system’s state on its history). Also, there are reduced pumping power requirements when utilizing passive capillary-driven flow. 

Evaporation-based systems actually provide an efficient and more controlled heat transfer mechanism, especially for high-power applications, unlike boiling.

In the face of energy-efficient thermal management being crucial for data centers, capillary-driven thin-film evaporation in membranes with extremely small pores presents a promising approach for dissipating high heat fluxes.

Click here to learn how AI can help keep a check on rising energy demand among AI data centers.

New Cooling Tech to Curb Rising Energy Demands

With the need for advanced cooling strategies critical to curbing the trend of rapidly rising energy consumption in data center cooling, reducing operational costs, and supporting carbon reduction goals, engineers have developed a new evaporative cooling technology for data centers and high-powered electronics.

Supported by the National Science Foundation, this new technique offers a promising alternative to traditional cooling systems, such as fans, liquid pumps, and heat sinks, as detailed¹ in the journal Joule. It could also cut down the water usage of many current cooling systems.

The new tech features a specially designed fiber membrane that passively removes heat via evaporation. The low-filter membrane used here involves a network of tiny, interconnected pores that use capillary action to pull in cooling liquid across its surface.

As the liquid evaporates, the heat is efficiently removed from the electronics underneath without requiring any extra energy. The membrane is placed on top of microchannels above the electronics, drawing liquid that flows through the channels and efficiently dissipates heat.

According to the study’s co-leader, Renkun Chen, professor in the Department of Mechanical and Aerospace Engineering at the UC San Diego Jacobs School of Engineering:

“Compared to traditional air or liquid cooling, evaporation can dissipate higher heat flux while using less energy.” 

Evaporation for cooling isn’t anything new, as many applications such as evaporators in air conditioners and heat pipes in laptops currently use it. But applying the method to high-power electronics has been a challenge.

Attempts at using porous membranes have been unsuccessful. Porous membranes have high surface areas that are ideal for evaporation, but previous efforts either had too small pores that would clog, or too large that triggered unwanted boiling, leading to failure.

“Here, we use porous fiber membranes with interconnected pores with the right size,” said Chen. This way, the engineers were able to achieve efficient evaporation without those downsides.

Upon testing the technology across various heat fluxes, the researchers found that their membrane was working and achieving record-breaking performance.

The membrane handled heat fluxes exceeding 800 watts per square centimeter (W/cm²). This is one of the most elevated levels recorded for this type of cooling system. Not just this, but it also proved stable over multiple hours of operation, highlighting a scalable and energy-efficient approach to next-generation electronic cooling.

“This success showcases the potential of reimagining materials for entirely new applications.”

– Chen

He explained that originally, the fiber members were designed for filtration. “No one had previously explored their use in evaporation,” he added. However, the team recognized the distinct structural features, which were interconnected pores of the right size, that made them perfect for efficient evaporative cooling. Chen stated:

“What surprised us was that, with the right mechanical reinforcement, they not only withstood the high heat flux. They performed extremely well under it.”

Besides their thermal performance, the fiber membranes are cost-effective, scalable for manufacturing, and exhibit both mechanical flexibility and strength, highlighting their potential for thin-film evaporation and indicating their usefulness for incorporation into high-flux electronic cooling systems, thereby offering a robust solution for next-generation thermal management.

The results are certainly promising, but the technology is still operating significantly below its theoretical limit. The team is now working on refining the membrane and optimizing its performance.

In the next phase, the team will integrate their technology into prototypes of cold plates, the flat components that attach to GPUs and CPUs to dissipate heat. As for commercializing the technology, the team will be launching a startup to take it to the market. The Regents of the University of California have also filed a patent related to this work.

Click here to learn about the new chip that will shrink LLM power usage by 50%.

Next-Gen Cooling Innovations

Given the wide usage of AI, which is projected to contribute trillions of dollars to the world economy, there’s now a growing focus on reducing its energy usage through various means.

In May 2025, tech giant Microsoft released a paper that quantifies the energy and water consumption, as well as the greenhouse gas (GHG) emissions produced by data center cooling techniques across their entire lifespan.

This life cycle assessment assesses not only the resources consumed during data center operations but also takes a deep dive into the resources required to produce all the virtual machines, servers, chips, cooling, and other equipment. This data, according to Microsoft, can help companies design data centers to use less water, energy, and carbon.

According to the study lead Husam Alissa, who is a director of systems technology in Cloud Operations and Innovation at Microsoft:

“We’re advocating in this paper for the use of life cycle assessment tools to guide engineering decisions early on and also sharing the tool with the industry to make adoption easier.”

Talking about the purpose of the study, Alissa said:

“What we’re trying to do here is tell the industry, ‘Here’s how you build an end-to-end life cycle assessment that takes cooling into account. And here is a tool for you that you can customize to your specific needs and then make a decision.”

Their study actually took two years, during which they evaluated four cooling technologies, viz. cold plates, air cooling, one-phase immersion, and two-phase immersion for servers.

The team expects the liquid approaches to perform better than air cooling, which is the industry standard, on carbon emissions, as well as energy and water consumption.

While Microsoft has installed cold plates in its data centers, it is also exploring other cooling techniques, like the rack-scale cold plate cooling technique that utilizes heat exchanger units. The tech giant is also developing a new technology, which it says can reduce water consumption by 30% to 50% and bring down energy demand and GHG emissions by about 15% across the entire life spans of data centers. 

Most recently, MIT Lincoln Laboratory also developed a special chip to assess cooling options for packaged chip stacks. This unique chip dissipates very high power to produce heat via the silicon layer and in specific hot spots. Next, cooling technologies are applied to the stack, upon which the chip measures changes in temperature. 

When put into a stack, it will allow researchers to examine just how heat travels through the stack layers and then measure the progress to keep them cool. 

“If you have just a single chip, you can cool it from above or below. But if you start stacking several chips on top of each other, the heat has nowhere to escape. No cooling methods exist today that allow industry to stack multiples of these really high-performance chips.”

– The study lead, Chenson Chen of the laboratory’s Advanced Materials and Microsystems Group

Investing in the AI Sector

A global leader in AI-focused GPUs, Nvidia (NVDA +2.05%) is a full-stack computing infrastructure company that operates through Compute & Networking and Graphics segments.

Nvidia has partnered with key players in semiconductors, server manufacturing, data storage, and enterprise software to accelerate the deployment of AI applications. The chipmaker is also heavily involved in funding startups, especially AI firms such as OpenAI, xAI, Inflection, Mistral AI, Perplexity, Lambda, Scale AI, and others to help advance the sector. 

NVIDIA Corporation (NVDA +2.05%)

When it comes to Nvidia’s market performance, it is one of the best-performing stocks, recording an eye-watering 1,350% gain in the past five years. As of writing, the shares of the world’s 2nd largest company by market cap of $3.5 trillion, are trading above $145, up 8.33% YTD and a mere 3% of its peak of almost $150 that was hit in Nov. 2024.

It has an EPS (TTM) of 3.10, a P/E (TTM) of 46.86, and an ROE (TTM) of 115.46% while offering a dividend yield of 0.03%.

In May 2025, Nvidia reported financial results for the first quarter of fiscal 2026, during which its revenue was $44.1 billion.

In the data center area, the company’s revenue was $39.1 billion. For this sector, Nvidia announced building factories in the US, introduced Blackwell Ultra and Dynamo for scaling AI reasoning models, and plans to speed up the IT infrastructure transition to enterprise AI factories with RTX PRO™ Servers. Among several other initiatives, NVIDIA AI Data Platform was also introduced as a customizable reference design for AI inference workloads.

Another big development involves the use of 100,000 Nvidia chips to build a new AI data center in the UAE, which will come online next year.

In another exciting news, Nvidia is reportedly working with Taiwan’s Foxconn to deploy humanoid robots at a new Foxconn factory that will produce Nvidia AI servers. Expected to be finalized in the coming months, the deployment would mark a milestone in the adoption of human-like robots to transform manufacturing processes.

Latest NVIDIA Corporation (NVDA) Stock News and Developments

Final Thoughts on AI Cooling Innovation

As AI continues to reshape industries, its massive computing needs are driving unprecedented energy consumption and heat generation. Here, the promising research into capillary-driven thin-film evaporation represents a crucial step toward building more energy-efficient, sustainable, and scalable AI infrastructure for the future.

With its high theoretical critical heat flux (CHF), capillary-driven thin-film evaporation in nanoporous membranes offers a promising thermal management strategy for high-power electronic devices. The latest study found that fiber membranes open pores that are interconnected, as efficient evaporators, facilitating rapid and uniform liquid transport through multiple routes while successfully reducing clogging and ensuring even wetting across the surface. 

By showing long-term stability, the 3D fiber membrane evaporators offer a highly promising solution for advanced thermal management, providing efficient cooling solutions to meet the demands of modern electronic systems.

Click here to learn all about investing in artificial intelligence.

Studies Referenced:

1. Feng, T.; Pei, Y.; Zhang, H.; Asai, B.; Dong, G.; Joshi, A.; Saha, A.; Cai, S.; Chen, R. High-Flux and Stable Thin-Film Evaporation from Fiber Membranes with Interconnected Pores. Joule 2025, 9 (6), 101975. https://doi.org/10.1016/j.joule.2025.101975