Meta used AI to concoct low-carbon concrete it poured for a datacenter floor

Social media giant Meta has created an AI model to come up with new forms of concrete and used one of the resulting recipes to underpin a new bit barn.

Like any entity that builds datacenters, Meta needs concrete – many square miles of it for the fleet of gigawatt-scale facilities it announced earlier this week. The company is also keen to reduce its carbon footprint, an ambition that concrete complicates because the substance is thought to contribute perhaps ten percent of CO2 emissions.

Meta is not alone in wanting greener concrete – The Register last year reported the Open Compute Project’s (OCP’s) efforts to develop and test low-carbon concrete.

Tests of such substances are necessary because subtle changes to the recipe for concrete can change its performance, and datacenters are demanding environments in which heat and vibration are constant.

Meta observed the OCP’s efforts last year, but was already investigating computer-assisted concrete creation in a 2023 paper titled “Sustainable Concrete via Bayesian Optimization.”

That work seems to have informed a Wednesday announcement that Meta has created a model that employs Bayesian optimization to predict “the compressive strength curves associated with different concrete mixtures.”

“Designing concrete formulas is a complex, multi-objective problem,” explains a Meta post on the topic. “The designer must choose between various types and proportions of cement, lower-carbon supplementary cementitious materials (SCMs), water-to-binder ratios, coarse and fine aggregate types, and admixtures. SCMs’ impact on concrete performance varies by source location and seasonality, requiring long-term tests for validation. Finally, time-consuming tests taking days and weeks are needed to fully validate the performance of new mixes.”

Meta’s model sped that process and, with help from construction company Amrize and the University of Illinois Urbana-Champaign, saw the social networking company pour its new concrete at a datacenter in Rosemount, Minnesota. Local media report the facility will initially draw 10 megawatts and crank up to 75MW in coming years.

• Meta reveals plan for several multi-gigawatt datacenter clusters
• Meta calls €200M EU fine over pay-or-consent ad model 'unlawful'
• More trouble for authors as Meta wins Llama drama AI scraping case
• Meta offered one AI researcher at least $10,000,000 to join up

Plenty of AI pundits have predicted AI will change the way scientists fight cancer or invent medicines.

AI helping to develop new forms of concrete is a little less glamorous.

Big Tech companies know that electricity generated to power their enormous datacenter fleets produce plenty of CO2 emissions and seek alternatives that are cheaper and perhaps more likely to burnish their reputations. With this effort, Meta’s even added some AI pixie dust and shared its work for others by publishing its code under the MIT License.

But while Meta has published info about the strength of concretes it cooked, it didn’t detail the CO2 emissions associated with the concrete mix it used in Rosemount.

Maybe an AI can figure that out. ®