AI Supercharges Hunt for Next-Gen Sustainable Cooling Materials

In the world of materials, thermal nanophotonics is critical for enabling fundamental breakthroughs across technological applications.

Thermal nanophotonics combines nanophotonics and thermal science to manipulate and control heat transfer at the nanoscale. It utilizes nanostructures and materials to tailor thermal radiation and heat flow, resulting in advancements in various applications, including energy harvesting, thermal management, and sensing.

Nanophotonics deals with the behavior of light on the nanometer scale. Nanophotonic materials, meanwhile, offer spectral and directional control over thermal emission.

The traditional method of finding such materials is hindered by trial-and-error approaches, but the advent of machine learning (ML) and artificial intelligence (AI) has revolutionized the field of materials science by significantly accelerating the processes of material discovery, design, and optimization.

While the technology has shown its powerful abilities in the design of nanophotonic and metamaterials, it’s a challenge to develop a general design methodology to customize high-performance nanophotonic emitters with ultrabroadband control and precise band selectivity. 

This is due to being restricted by traditional algorithms, local optimization traps, and predefined geometries and materials. 

However, this is now being addressed by scientists from the University of Texas at Austin, who collaborated with researchers from Umeå University in Sweden, the National University of Singapore, and Shanghai Jiao Tong University.

Together, they have designed an ML technique¹ to design complex, 3D thermal meta-emitters. 

Meta-emitters are engineered materials designed to control and manipulate electromagnetic radiation, offering applications in energy efficiency and thermal management. 

“Our machine learning framework represents a significant leap forward in the design of thermal meta-emitters. By automating the process and expanding the design space, we can create materials with superior performance that were previously unimaginable.”

– Study co-lead Yuebing Zheng, a professor in the Cockrell School of Engineering’s Walker Department of Mechanical Engineering

AI-Powered Materials for Climate-Resilient Urban Design

Published in Nature, the study details the new ML-based framework that helped design materials that can bring down temperatures indoors and, in turn, energy costs.

Using their framework, the scientists have actually been able to generate more than 1,500 new materials that can selectively emit heat in a controlled manner. They can also offer higher accuracy in heating and cooling to achieve improved energy efficiency.

Their framework can design ultrabroadband and band-selective thermal meta-emitters by optimizing multiple parameters with limited data that covers material diversity and 3D structural complexity.

According to the study, their architecture enables dual design capabilities. First, it automates the inverse design of a multitude of possible metastructures as well as material combinations for spectral tailoring. Second, it has an “unprecedented ability” to design different 3D meta-emitters by applying a three-plane modelling method that overcomes the limitations of traditional, flat, 2D structures. 

In their study, the team presents seven proof-of-concept meta-emitters that demonstrate superior optical and radiative cooling performance, exceeding the current advanced designs. The seven classes of meta-emitters are tailored to specific functions.

The generalizable framework developed is for fabricating 3D nanophotonic materials, which the researchers noted, “facilitates global optimization through expanded geometric freedom and dimensionality and a comprehensive materials database.”

Now, to assess the viability of their design system, the researchers produced four sample materials and then tested their performance. 

One of the meta-emitter materials was applied to the roof of a model house. To analyse its cooling ability, the material was compared with standard commercial white and gray paints. What the researchers observed after the roof had four hours of direct midday sunlight exposure was that the newly created material was, on average, 5 to 20 degrees Celsius cooler than traditional paints.

Based on this, the team estimates that their material will save approximately 15,800 kilowatt-hours (kWh) per year in cooling costs for an apartment building in a hot city like Bangkok. A standard AC unit typically consumes around 1,500 kWh annually.

So, the materials created by the team can be utilized for residential and commercial energy savings. In cities, they can help bring down temperatures by reflecting sunlight and releasing heat at targeted wavelengths. This way, the material can potentially reduce the urban heat island effect caused by limited greenery and dense concrete structures.

But that’s not the extent of their usage. The material can also be used in space applications, where it efficiently manages incoming solar radiation and emitted heat, helping to regulate spacecraft temperatures.

The use cases of thermal meta-emitters extend far beyond just this. For instance, by integrating them into fabrics and textiles, we can improve the cooling technology in clothing and outdoor equipment. 

Automobiles are yet another one. By wrapping cars with thermal meta-emitters and embedding them into interior materials, the heat that builds up when cars sit in the sun can also be reduced.

Despite their many advantages, these materials haven’t been able to gain mainstream adoption due to their laborious design process, and even automated options have faced difficulty in dealing with their complex 3D hierarchical structure. But all that can finally change with the latest AI framework.

“Traditionally, designing these materials has been slow and labor-intensive, relying on trial-and-error methods. This approach often leads to suboptimal designs and limits the ability to create materials with the necessary properties to be effective.”

– Zheng

Therefore, researchers will continue to work on their technology, refining it and applying it to additional aspects of nanophotonics.

“Machine learning may not be the solution to everything, but the unique spectral requirements of thermal management make it particularly suitable for designing high-performance thermal emitters.”

– Co-author Kan Yao

How AI Accelerates Discovery of New Materials

Focused on the structure, properties, processing, and performance of materials, material science forms the foundation of everything from aerospace, electronics, and energy to medicine and many other fields. 

In fact, the discovery and development of new materials have been crucial in shaping human history for centuries, advancing technology.

For decades, we have relied on the trial-and-error approach to find new inorganic materials with favorable properties. This approach is extremely resource-intensive, requiring hundreds of thousands of hours of experiments to first identify and then synthesize just a handful of potential new materials.

The complexity of materials at their molecular and atomic levels is what makes the discovery of new materials a lengthy and expensive process. So, the increased availability of supercomputers changed materials science by allowing for the simulation of the behavior of materials.

And now, the arrival of AI is leading to a revolution in the field by accelerating the computational approach to material science. By providing the desired properties for a material as well as constraints, generative AI systems can now create entirely novel materials.

After all, today’s advanced models, trained on massive datasets, when combined with high-throughput computing, are able to rapidly screen candidate materials against desired parameters, thereby predicting properties for numerous substances in a very short time.

Not only can AI save significant development time and human and material resources, but it can also do so while precisely meeting complex and varied market requirements.

As Kristin Persson, a professor in materials science at the University of California, Berkeley, noted, we are currently in a paradigm where science is driven by big data and AI. Today, we have enough data to train machine learning algorithms, and “that brings a whole new level of speed in terms of innovation,” she said.

Interestingly, AI also benefits from the discovery of new materials. AI is data-hungry, and the field of materials science lacks data. By utilizing this technology, material properties can be simulated, and the resulting data can be used to train machine learning models.

Persson is currently leading the multi-institution, multi-national effort called  Materials Project, which leverages supercomputing and advanced stimulated methods to calculate the properties of all known inorganic materials and beyond. The data is made freely available to design novel materials.

Click here to learn about the adaptive roof tiles that can help reduce heating/cooling energy consumption.

Breakthroughs in AI-Driven Materials Discovery

Researchers from U of T Engineering introduced a new tool² recently that predicts just how a new material could be best used in order to reduce the lag between the discovery of a material and its deployment.

The multimodal AI tool uses early-stage data to predict the potential real-world use for a new material with a focus on a specific category of porous materials called metal-organic frameworks (MOFs). 

Just last year, researchers developed over 5,000 different types of these materials, which have adjustable properties, noted study lead Professor Seyed Mohamad Moosavi from the University of Toronto (ChemE). He added that the challenge is that an MOF created for one particular application often turns out to have suitable properties for an entirely different one.

“In materials discovery, the typical question is, ‘What is the best material for this application?’” said Moosavi. “We flipped the question and asked, ‘What’s the best application for this new material?’ With so many materials made every day, we want to shift the focus from ‘what material do we make next’ to ‘what evaluation should we do next.’”

So, ChemE PhD student Sartaaj Khan developed a multimodal machine learning system trained on various types of data. Multimodality was key here as it gave the model “a more complete picture” to make more accurate predictions without requiring post-synthesis.

Researchers from Argonne National Laboratory, meanwhile, used a generative AI diffusion model to generate over 120,000 MOF³ candidates in a matter of just over half an hour using a supercomputer. The modified neural network reduced the number of MOFs to 364, which were believed to be high-performing.

With a few more days and further computational analysis, the team found 102 stable MOFs in the dataset. 6 of these had a CO2 capacity ranked among the top 5% of materials in the popular hMOF database.

In another example, scientists used AI to design completely new nanomaterials⁴ that are as light as Styrofoam while having the strength of carbon steel. 

Strength and toughness tend to be at odds in many materials, including nano-architectured ones, which are composed of ultra-small building blocks. When repeated, these building blocks make the material strong, but can also cause stress concentrations that lead to sudden breakages. 

To find better ways to design nanomaterials, researchers simulated possible geometries and then ran them through an algorithm that learned from their designs to predict the best shapes for evenly distributing applied stresses while carrying a heavy load.

The researchers used a 3D printer to bring these shapes to life and found them to be able to withstand a stress of 2.03 megapascals (MPa) per cubic meter per kilogram, which is five times higher than that of titanium.

Researchers see their potential application as ultra-lightweight components in aerospace applications to reduce fuel demands and the high carbon footprint of flying.

According to first-author Peter Serles, an engineering researcher at Caltech:

“This is the first time machine learning has been applied to optimize nano-architected materials, and we were shocked by the improvements. It didn’t just replicate successful geometries from the training data; it learned from what changes to the shapes worked and what didn’t, enabling it to predict entirely new lattice geometries.”

AI-based material discovery is also being used extensively for urban planning. A collaborative research⁵ by Peking University and the University of Southern Denmark developed an advanced framework that integrates deep learning with remote sensing to identify building materials with unprecedented precision.

Besides energy efficiency, AI can elevate urban planning by helping with environmental monitoring and conservation, housing and infrastructure development, and public safety and disaster response.

Click here to learn about the bladeless wind turbines, often referred to as the future of clean energy.

Investing in AI-based Material Discovery

If we look into the investing potential of AI, it is massive, with the market projected to be worth trillions in the coming years. When it comes to companies leading this technological advancement, especially in material science, two names stand out: Microsoft (MSFT +0.23%) and Google (Alphabet Inc.) (GOOG +2.44%), which have launched their own models to elevate the scale and precision of materials research. However, for the purposes of this article, we will focus on Alphabet Inc.

Alphabet Inc. (GOOG +2.44%)

Late in 2023, Google’s DeepMind released an AI tool called Graph Networks for Materials Exploration (Gnome) to speed up the material discovery process. At the time, it reported⁶ finding 2.2 million new crystals with the help of the deep learning tool.

This, Google noted, is “equivalent to about 800 years’ worth of knowledge and demonstrates an unprecedented scale and level of accuracy in predictions.” The newly discovered crystals included 380,000 stable materials, making them promising candidates for experimental synthesis and capable of powering future technologies.

The Gnome model is a graph neural network (GNN) model, where the input data is represented as a graph. Gnome was trained on data from the Materials Project, including crystal structures and their stability, to generate novel candidate crystals and predict their stability.

Google assessed its predictive power by repeatedly checking its performance using Density Functional Theory (DFT). For the ‘training process,’ it used ‘active learning’ in which the resulting data was fed back into the model, significantly boosting Gnome’s performance. 

The model’s stability prediction accuracy, according to Google, soared from 50% to 80%. The efficiency of the model, meanwhile, was scaled from under 10% to over 80%.

Moreover, about 736 materials predicted by the Genome have been independently synthesized by external researchers. Google also collaborated with Lawrence Berkeley National Lab to synthesize 41 new materials, validating the tool’s predictive strength and the power of autonomous experimentation.

Now, let’s take a look at the $2.2 trillion market cap giant’s performance. As of writing, its shares are trading around $182, down 3.86% YTD. It has an EPS (TTM) of 8.97 and a P/E (TTM) of 20.29. The dividend yield paid is 0.46%.

As for company financials, Google’s parent company, Alphabet, reported revenue of $90.2 billion for Q1 ended March 31, 2025. Earnings per share were $2.81. These numbers, CEO Sundar Pichai said, “reflect healthy growth and momentum across the business. Underpinning this growth is our unique full-stack approach to AI.”

Latest Alphabet Inc. (GOOG) Stock News and Developments

Conclusion

AI is transforming every aspect of our lives, including how we design materials that shape our future. The technology’s integration into material sciences represents a truly paradigm shift, accelerating discoveries that once took years to come to fruition, now taking days or even hours.

To put it simply, AI is driving the near future of materials innovation by utilizing massive datasets, high-throughput computing, and generative models, allowing researchers to predict, design, and optimize novel materials with unprecedented efficiency and precision.

Click here to learn about the clean energy stocks with an environmental focus.

References:

1. Xiao, C.; Liu, M.; Yao, K.; et al. Ultrabroadband and Band‑Selective Thermal Meta‑Emitters by Machine Learning. Nature 2025, 643, 80–88. https://doi.org/10.1038/s41586-025-09102-y
2. Khan, S. T.; Moosavi, S. M. Connecting Metal–Organic Framework Synthesis to Applications Using Multimodal Machine Learning. Nature Communications 2025, 16, 5642. https://doi.org/10.1038/s41467-025-60796-0
3. Park, H.; Yan, X.; Zhu, R.; et al. A Generative Artificial Intelligence Framework Based on a Molecular Diffusion Model for the Design of Metal–Organic Frameworks for Carbon Capture. Communications Chemistry 2024, 7, 21. https://doi.org/10.1038/s42004-023-01090-2
4. Serles, P.; Yeo, J.; Haché, M.; Demingos, P. G.; Kong, J.; Kiefer, P.; Dhulipala, S.; Kumral, B.; Jia, K.; Yang, S.; Feng, T.; Jia, C.; Ajayan, P. M.; Portela, C. M.; Wegener, M.; Howe, J.; Singh, C. V.; Zou, Y.; Ryu, S.; Filleter, T. Ultrahigh Specific Strength by Bayesian Optimization of Carbon Nanolattices. Advanced Materials 2025, 37 (14), e2410651. https://doi.org/10.1002/adma.202410651
5. Sun, K.; Li, Q.; Liu, Q.; Song, J.; Dai, M.; Qian, X.; Gummidi, S. R. B.; Yu, B.; Creutzig, F.; Liu, G. Urban Fabric Decoded: High‑Precision Building Material Identification via Deep Learning and Remote Sensing. Environmental Science & Ecotechnology 2025, 24, 100538. https://doi.org/10.1016/j.ese.2025.100538
6. Merchant, A.; Batzner, S.; Schoenholz, S. S.; et al. Scaling Deep Learning for Materials Discovery. Nature 2023, 624, 80–85. https://doi.org/10.1038/s41586-023-06735-9