Materials
Amrize, Meta Partner on Low-Carbon Concrete for Minnesota Data Center
Bespoke Bayesian mix design slashes carbon and shaves a week off the placement schedule
.webp?t=1754535416)
Deployment of Amrize’s AI-generated concrete formulation in Meta’s Rosemount, Minn., data center building.
When Meta broke ground last year on its data center in Rosemount, Minn., about 15 miles south of Minneapolis and St. Paul, the social media giant faced a concrete paradox: the material’s carbon burden loomed large even as project schedules left no room for slower, low-cement mixes.
So, Meta approached ready-mix supplier Amrize—the North American unit of Switzerland-based materials giant Holcim that was spun off in June—and researchers at the University of Illinois, Urbana-Champaign to hand the problem to an algorithm. The result was an algorithm-optimized recipe that reaches 4,000-psi strength 43% faster and trims embodied carbon by 35%. The “bespoke” mixture is now being used across the 715,000-sq-ft data center campus and could reshape how hyperscale builders specify concrete.
Cement production accounts for roughly 8% of global carbon dioxide emissions, a reality colliding with the trillion-dollar race to erect ever larger, faster data centers. The Rosemount pilot is demonstrating how marrying Bayesian optimization with ready-mix know-how can cut carbon without slowing placement cycles—an outcome industry analysts say could propel performance-based concrete specifications across infrastructure.
Yet adoption is hardly guaranteed: a white paper by Switzerland-based cloud BIM collaborative Revizto published in June warns that construction’s sluggish embrace of digital workflows already threatens data-center delivery just as demand peaks. The Amrize-Meta innovation has arrived at an auspicious time, serving as both proof of concept and a cautionary tale for an industry under pressure to decarbonize quickly.
Cracking the Carbon Code
The collaboration between the Silicon Valley tech giant and old-school materials manufacturer began in late 2023, when Meta challenged Amrize in its former corporate structure to deliver a greener slab mix for the $800-million Rosemount campus without extending the critical path.
Rather than the standard “cook, cure, break and tweak” loop that can require 200-plus lab cylinders, Meta supplied its open-source Ax and BoTorch frameworks to guide the search. Amrize said researchers at the University of Illinois Urbana-Champaign uploaded hundreds of thousands of historic cylinder tests, regional aggregate gradations and supplementary cementitious material chemistries into the model.
The Bayesian-optimization engine—a self-learning algorithm that runs a trial, updates its probability model, and then proposes the next most promising recipe—treated three objectives: early compressive strength, drying-shrinkage resistance and global-warming potential as a single multivariable problem.
“We provided the materials, Meta provided the algorithm, and the university used hundreds of thousands of data points to optimize these mixes,” said Rodrigo Gallardo, Amrize senior vice president for aggregates and construction materials. In a companion blog post, Meta research scientist Julius Kusuma wrote that the collaboration “maximized the performance and environmental profile of the concrete being used in our Rosemount Data Center.”
Iterations converged after just 18 physical trials—about one-tenth the number required by a conventional design-of-experiments path, noted Nishant Garg, the University of Illinois professor who supervised testing.
The winning recipe replaced 40% of portland cement with a ternary blend of ground-granulated blast-furnace slag and Class F fly ash, leveraged a polycarboxylate admixture package to trigger rapid hydration and pegged the water-to-binder ratio to Minnesota’s swing-season humidity via the American Concrete Institute maturity-method envelope. “AI-driven mix design lets us optimize concrete for performance, cost and carbon in a single step,” Garg said in a release announcing the collaboration.
A brief intellectual property clause governs the partnership: Meta retains ownership of the Bayesian code base, while Amrize controls the final batch-plant formula and licenses it for the tech firm's current and future campuses—an arrangement both sides say speeds replication without bogging down in patent filings.
Meeting the $1 Trillion Data-Center Moment
Full-scale placements vindicated the lab math. The EcoPact mix reached 4,000 psi 43% faster than the baseline, trimmed drying shrinkage 9.5% and cut embodied-carbon 35% versus the National Ready Mixed Concrete Association’s regional average, according to Amrize. "By bringing the use of AI into a concrete mix design, we are getting closer to the optimal balance of sustainability and constructability,” Blair McNeil, campus director for Mortenson Data Center Solutions, the contractor hired for the buildout, wrote in an email. The schedule gain allowed follow-on trades to pull forward rack installation dates, Meta said.
A prototype slab test at Meta’s Rosemount, Minn., data center project site validates the performance of the new formulation. Photo courtesy Meta
Cost—often the stumbling block for greener mixes—proved manageable. Gallardo said upcharges were “not substantial,” which squares with national research: a Rocky Mountain Institute study of three U.S. building case studies found embodied-carbon cuts of 19%–46% could be achieved at cost increases of less than 1%.
That calculus matters, as global data-center capital expenditure is projected to top $1 trillion by 2029. Paying a fractional surcharge for a faster, cleaner mix is therefore “within normal contingency for hyperscale projects,” Dell’Oro Group analysts note in their 2025 market outlook.
Verified transparency helped seal the deal. An Environmental Product Declaration for the Rosemount mix was third-party-verified by WAP Sustainability Consulting and posted in Meta’s materials portal, giving designers real-time carbon tracking for every yard placed.
Meta’s procurement strategy amplifies the signal. Through the Infrastructure Masons Climate Accord and the Open Compute Project, the company now requires suppliers to publish third-party-verified declarations and favors performance-based specifications over prescriptive cement limits.
Those moves echo the American Concrete Institute’s Center of Excellence for Carbon-Neutral Concrete, which is drafting guidelines to qualify data-rich, low-carbon mixes for mainstream structural use. “Rosemount is just the beginning,” said Jaime Hill, president of Amrize Building Materials, adding that digital design allows bespoke concretes to target everything from thermal-mass management to radiation shielding.
Yet scalability faces a material hurdle. The American Cement Association warns—based on presentations at the 2025 IEEE-IAS/PCA Cement Conference—that the U.S. could face a 50-million-ton shortfall of supplementary cementitious materials by 2050, roughly half of today’s consumption.
Gallardo says Amrize is piloting calcined-clay and harvested-ash blends to diversify SCM supply, and the Bayesian optimizer is “agnostic to the binder, so long as we have the right input data.”
Long-term durability testing is also underway. The University of Illinois Urbana-Champaign has cast 56- and 90-day sulfate-resistance specimens, with preliminary results due this fall. Early indicators, Garg said, show no compromises relative to the control mix, and the data will be fed back into future model iterations.
From Pilot to Playbook
The path from headline pilot to industry playbook runs through the job trailer’s server rack. Revizto’s white paper, “The Infrastructure Behind Innovation: Data-Center Construction at a Crossroads,” points to design compressions as short as 10 weeks and rework risks topping $10 million when digital coordination lags. “The rising demand for data centers is a chance to rethink how we build,” Jason Howden, Revizto’s chief innovation officer, wrote, urging owners to insist on integrated delivery and predictive scheduling.
Meta has tried to grease the skids by open-sourcing its “SustainableConcrete” code on GitHub, complete with tutorials for training models on local aggregates and SCMs. “Our aim is to increase the adoption and optimization of sustainable concrete mixes in the construction industry at large,” Kusuma wrote.
Startups are racing to commercialize similar workflows: Concrete.ai offers cloud-based Bayesian optimization for ready-mix producers, while Giatec’s SmartMix platform uses neural networks to predict strength curves from field data.
Standards are beginning to catch up. American Concrete Institute Committee 318 and ASTM Committee C09 have discussed draft language that may allow performance-verified mixes to qualify under structural codes, based on discussions during an Institute webinar in March and an ASTM virtual town hall Q&A in May. The earliest formal ballot would be sometime in 2026, pending draft completion.
Amrize is bullish that the gears are finally meshing. “Owners are asking for carbon disclosures, tech vendors are opening their code, and the codes themselves are moving toward performance,” Gallardo said. “You need all three levers to pull at once; Rosemount proves that can happen.”
Looking for quick answers on construction and engineering topics?
Try Ask ENR, our new smart AI search tool.
Ask ENR →
Bryan Gottlieb is the online editor at Engineering News-Record (ENR).
Gottlieb is a five-time Society of Professional Journalists Excellence in Journalism award winner with more than a decade of experience covering business, construction and community issues, including finance, law and real estate. He has reported for Adweek and the San Diego Daily Transcript and led a community newsroom in Santa Monica, Calif. He served as editor-in-chief of the Detroit Metro Times and as managing editor at Roofing Contractor, where he helped shape national industry coverage. At Engineering News-Record, Gottlieb covers breaking news, large-scale infrastructure and megaprojects, regulatory developments and business trends across the U.S. and global construction sectors.
email: gottliebb@enr.com | office: (248) 786-1591
