Meta is teaming up with building materials company Amrize to develop a new kind of concrete for its upcoming data center in Rosemount, Minnesota. But this isn’t your typical cement mix. It’s an AI-optimized formulation built using open source artificial intelligence models and real-world performance data, aiming to reduce emissions while keeping strength and speed on track.
The custom concrete blend was designed through a partnership between Meta, Amrize, the University of Illinois Urbana-Champaign’s Grainger College of Engineering, and construction firm Mortenson. The result is a high-performance mix that meets the demands of hyperscale infrastructure without the typical environmental tradeoffs.
Julius Kusuma, a research scientist at Meta, explained the motivation behind the project. “We work to design our data centers as efficiently and sustainably as possible, while driving our AI ambitions forward,” Kusuma said. He noted that collaborating with Amrize and the university helped maximize both performance and sustainability for the Rosemount build.
The numbers back that up. According to Amrize, the new ECOPact mix is estimated to reduce the total carbon footprint of the concrete by 35 percent compared to conventional alternatives. That’s a meaningful drop, especially considering the massive volume of concrete used in a typical data center foundation.
Nishant Garg, a professor at Illinois who led the lab testing and data generation, said that using AI in materials science is a game changer. “AI-driven mix design lets us optimize concrete for performance, cost, and carbon in one step,” he explained. As concrete recipes become more complex and introduce new materials, AI also helps forecast things like how quickly the concrete will gain strength over time.
For Amrize, the project represents a bigger shift toward smart construction materials tailored to specific environments. “Partnering with Meta and using AI to develop an innovative concrete mix that meets the unique needs of data centers is just the beginning,” said Jaime Hill, president of Amrize Building Materials. He noted that AI allows the company to tweak concrete properties for performance, energy efficiency, and even thermal regulation depending on the application.
So far, the AI-designed mix has passed strength and set-time benchmarks and is now approved for additional structural areas of the data center. With the success of this trial, Meta could continue expanding its use of AI-assisted concrete across other projects.
While flashy hardware and AI chips usually get the spotlight in data center coverage, the materials that literally support the infrastructure are also evolving. This collaboration shows how open source AI isn’t just powering servers. It’s helping build the floor those servers sit on.
If AI can design better concrete, it raises an interesting question. What other forgotten parts of infrastructure are ripe for reinvention?
