While the discussion about artificial intelligence focuses on algorithms and chips, it is innovations in materials science that determine the limits of what is technologically possible.
The development of artificial intelligence does not depend only on better algorithms and more powerful chips, but increasingly on advanced materials that enable these technologies to function at all. Each new generation of AI technology demands greater processing power, more memory, better energy efficiency, and high reliability, and each of these requirements directly increases the physical pressures on the materials used in semiconductor manufacturing and data centers. According to a text published in MIT Technology Review on July 21, 2026, produced by Syensqo, advanced materials are no longer just a support for innovation-they define the very boundaries of what is possible.
Manufacturing a single semiconductor chip today involves thousands of precisely controlled process steps with almost no room for error. Small temperature variations or chemical instabilities can create defects that reduce yield and increase costs. With each new generation of chips, manufacturers seek materials that offer higher purity, better chemical and plasma resistance, and stability under increasingly demanding conditions. Innovations in polymers, elastomers, specialty fluids, and other advanced materials are precisely what make each new generation of technology feasible.
Growing AI workloads are also transforming the physical infrastructure that powers them. Increasing compute density is redesigning data centers, creating a need for more sophisticated thermal management, high-voltage power architectures, increased data storage capacities, and fast data transmission. Every part of the system-from cooling and power management to electronic components like connectors, capacitors, and hard drives-is under growing pressure. According to Syensqo, knowledge about fluid circulation from semiconductor and automotive cooling systems can be applied to direct liquid cooling designs for AI servers, accelerating the development of new solutions.
The definition of performance in materials science is changing. Alongside technical requirements, materials are increasingly expected to be developed and produced in a more responsible manner. An example is perfluoroelastomers, materials used for sealing semiconductor manufacturing equipment, operating under extreme temperatures, aggressive plasma, and highly reactive chemicals. Syensqo has developed a new generation of these materials using a production process without fluorosurfactants, aiming to produce a better material in a more responsible way. As the company states, the goal was to ensure that manufacturers no longer have to choose between higher performance and responsible production.
Paradoxically, artificial intelligence itself is becoming a tool for developing the materials that will then enable it. The development of advanced materials traditionally involved a long process of hypothesis, synthesis, testing, and iteration. New digital tools, including AI, help researchers move through these cycles faster. Syensqo uses several AI tools, including Microsoft's Discovery platform, to identify and evaluate promising molecular candidates for heat transfer fluids used in semiconductor manufacturing and data centers. As the company states, "AI does not replace scientific expertise. It helps scientists apply that expertise more efficiently, allowing them to spend less time searching for answers and more time solving the industry's toughest challenges."
One of the key insights from this approach is the ability to transfer knowledge across different industries. As data centers transition to high-voltage architectures and higher power density, the material challenges they face closely align with those from the electric vehicle industry. This cross-industry transfer of knowledge can significantly accelerate the development of new solutions for energy management and thermal processes in AI infrastructure. However, the path from laboratory discovery to qualified material still requires scientific expertise, rigorous testing, and close customer collaboration, no matter how much AI accelerates the initial research phases.
Note: This article is based on material produced by Syensqo, published in MIT Technology Review on July 21, 2026. The text was not written by the editorial team of MIT Technology Review.
