Building the materials foundation for AI

The AI boom has a less glamorous bottleneck than most people think: materials. As chips and data centers strain against limits in heat, power, purity, and reliability, the stuff they’re built from is moving from supporting role to center stage.

Mike Finelli, who leads technology and innovation at Syensqo in North America, says AI is pushing semiconductors and data centers to their physical limits. His shorthand is a pyramid: commodity materials at the bottom, high-performance specialty materials at the top. The more demands pile up — high temperature, high purity, electrical performance, chemical resistance, plasma resistance, long-term stability — the more you climb toward the top.

Syensqo is aiming squarely at that layer. The company is working on materials for high-voltage data center designs, sealing materials for semiconductor manufacturing, and thermal-management products, including fluids for direct immersion cooling. Some of the useful ideas are not even new to one industry. Finelli points to materials first developed for electric vehicles as a fit for the higher voltage and energy-density needs of data centers.

There’s also a shift in what customers want from the materials themselves. Performance still matters, but so does environmental impact, and Finelli says the goal is to erase the trade-off between the two. In his view, sustainability has to be considered at the start of research, not bolted on later like an apology.

AI is now speeding up the search for those materials too. Syensqo is using AI agents to digitally generate millions of possible molecular combinations, predict their performance and sustainability traits, and narrow the field for lab testing. Finelli sees a loop forming: AI helps build better materials for AI infrastructure, which then helps build better AI. That is the real flywheel here, and it is the part worth watching.