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 …
I doubt the output of the 'AI' (likely just pattern identification in known blends to extrapolate new ones without as much physical iterative testing) is being sent straight to the construction site. Concrete/cement specs typically list allowed ranges of core components and additives (no paper, no string, no sellotape) but there is latitude and, at least in the UK, they're being broadened to permit low carbon blends.
Crucially, there's a requirement for physical testing and anything novel (but almost certainly within broad range specs) will be certified for a specific time frame, based upon the content and environment.
Yes, for years various projects about "doping" concrete for various reasons – including keeping the stuff liquid long enough for those insane Arab skyscrapers – has built upon the material science research into steel alloys and ceramics. Concrete itself is essentially "doped" cement with sand and gravel helping form a more open and less brittle lattice. I seem to recall there is even a version of concrete that uses carrots to make concrete stronger/lighter. Though it may just be to help see in the dark!
Interesting fact: the whole "carrots help you see in the dark" legend originated in WWII disinformation. The British didn't want the Nazis to catch on to the fact that their night bombing missions were becoming more accurate due to the introduction of radar, which was then a brand-new and top secret technology. So instead they spread the word around that RAF pilots were getting fed loads of carrots and other food rich in vitamin A, to improve their night vision.
Honestly, this is one of those rare examples where carbon-capture-and-storage (at the cement plant) is about the only sensible answer. Anything else is just splashing around in sand and water.We really don't want AI involved in sussing that one out.
Or perhaps we do. I read a report about the Biosphere-2 habitat project. That had problems with their CO2 levels unexpectedly decreasing to levels where they couldn't sustain their greenhouses. Culprit turned out to be the concrete because that absorbs CO2. Concreting over the Amazon probably isn't an ideal form of CCS though, but I guess where concrete is used in say, urban construction projects, CCS-optimised version might be a good idea. Or at least a way to extract carbon credits or offsets.
"try telling building control that you had AI knock up the batch specs or getting an engineer to sign off on the structural strength"
(dusts off the Civil's cap*). They won't mind: The concrete has to go through QC and testing before it's used regardless.
Issue is: Concrete is exothermic: It has to be kept cooled while setting to stop the concrete dehydrating and powdering. So if they mess about with the mix, is it going to cure faster or slower, and what temperature will it reach as that affects it's strength.
And... as it's AI: Will it really be that good? Or will it be OPC with sand from a beach? If so... that's going to be REAL expensive...
*I know: haven't done concrete technology work for >cough< years. Still remember parts of it, though. Was a really fascinating subject. Mind you, so was boring**. That was incredibly interesting...
** Yes, there's a joke in there: Boring = digging out tunnels. It really is a fascinating subject. And why a certain Muskiness founded the Boring company... something I do have to admit made me smile...
I don't care about anything that Meta does, this is just pure virtue signaling and compared to all the pollution Meta is delivering it is like saving a spit while polluting an ocean. If Meta would be gone completely, nothing would have been actually lost, but Giga tons of carbon-dioxide would have been spared.
Not convinced CO2 is a problem - yet. I still think it's way off that. Currently it's helping green the planet. We are in a warm period geologically and without us would be approaching a peak anyway. Now although a sceptic of all the fuss it makes sense to consider and develop methods not to emit CO2 in case it becomes a problem. One of which, despite Bill Gates refuting it, would be to not decimate plant life and consider how to keep more trees/grass or whatever absorbs the most CO2. It would seem as plants flourish in higher CO2 they will make the best buffer rather than increasing our CO2 on datacenters in order to reduce our CO2! Anyone smell a rat? Meanwhile can we just reduce polluting in general?
@Roland6 “Sustainable Concrete via AI.” is… a Meta hallucination Meta Bullshit.
"Meta has developed an open-source AI tool to design concrete mixes that are stronger, more sustainable, and ready to build with faster—speeding up construction while reducing environmental impact." https://engineering.fb.com/2025/07/16/data-center-engineering/ai-make-lower-carbon-faster-curing-concrete/
Lying wankers. There is the open-sourced code https://github.com/facebookresearch/SustainableConcrete/tree/main. Its not a tool to design concrete mixes, its purpose is to quantify concrete mixtures’ sustainability and strength.
There are two models (not AI), compressive strength model predicts compressive strength over time and global warming potential (GWP) model. Surely GWP is not the only metric defining sustainability, missing metrics will need modelling.
Pull the other one, it's got bells on it!
Seriously, the creation of a new type of concrete is something best left to a structural engineer. When the foundation of their bit barn starts falling apart in a few years time, maybe the happy people at Meta will realise that putting important things in the hands of a hallucinating LLM is a bad idea.
Not that Meta have done anything that calls for fairness but from the title of the paper I suspect this is probably using one of the more practical machine learning approaches designed for real world problem solving, rather than a witless platform for tricking gullible executives into firing all the staff who do the real work at their companies.
Civil engineers have been working with concrete for centuries, optimising properties for different applications, and there's no need at all for AI, simply speak to the cement makers, and look up existing properties tables. I'm guessing that far from "AI inventing new concrete recipes", it's marketing greenwash to both try and convince the gullible that all the money being poured into AI DCs is somehow less bad than other less valuable stuff like homes, hospitals, roads, etc.
"When the foundation of their bit barn starts falling apart in a few years time, maybe the happy people at Meta will realise that putting important things in the hands of a hallucinating LLM is a bad idea."
Some new mixes often turn out to be very poor due to sensitivity of the ratios. Get something just a bit off and it's much worse than an older more pedestrian mix. The lack of experience with the concrete by the construction crews can also be a point of failure. The use of AI for something as mundane as concrete formulations is interesting, but unlikely to be something to crow about until a lot more time has passed.
I'm surprised they have sufficient data to train the AI on. One imagines the amount of data required would need to come from several companies and they wouldn't be keen to hand it over. Also, only time will tell if it is good and I wonder if any failures after time will be get as much publicity and therefore have the data.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
