Science Friday - Secrets Of Ancient Concrete, And... Data Centers In Space?

Episode Date: January 14, 2026

The concrete of ancient Rome is famous for its durability. Just look at the Pantheon and those iconic aqueducts that helped transport water throughout the empire—still standing 2,000 years later.But knowledge about how this concrete was made hasn’t been very solid. Well, scientists have discovered a construction site in Pompeii preserved in the volcanic ash, which might hold clues to how we can improve our concrete today. Concrete researcher Admir Masic joins Host Ira Flatow to discuss the findings.Plus, we’ll look at the infrastructure of the future with engineer Benjamin Lee, who breaks down the recent news of tech companies looking to move their power-hungry data centers to space. They discuss the daunting engineering challenges and possible benefits.Guests:Dr. Admir Masic is an associate professor of civil and environmental engineering at MIT in Cambridge, Massachusetts. Dr. Benjamin Lee is a professor in the department of electrical and systems engineering and the department of computer and information science at the University of Pennsylvania in Philadelphia.Transcripts for each episode are available within 1-3 days at sciencefriday.com. Subscribe to this podcast. Plus, to stay updated on all things science, sign up for Science Friday's newsletters.

Hi, I'm I Refleto, and you're listening to Science Friday. If you've listened to the show for a while, you know that one of my favorite topics is concrete, really. And some of my favorite examples of the long-term durability of concrete are from the Romans. Take the Pantheon and all those famous aqueducts built thousands of years ago and still standing. But how exactly was this concrete made? and what ingredients went into the cement? That knowledge hasn't been very solid. Well, late last year, scientists discovered an actual cement mixing site in Pompeii, preserved in the volcanic ash, and it might hold clues to how we can improve our concrete today. Here to uncover those secrets

is Dr. Admir Mosch, Associate Professor of Civil and Environmental Engineering at MIT, and one of the co-authors on that study. Dr. Moshich, welcome to Science Friday. Thank you for having me. Nice to have you. So you go to Pompeii and you make a discovery there that changes about how they made their concrete? Yeah, so the context of Pompeii is fascinating.

If you think about a city that 2,000 years ago was literally frozen in time by this eruption of Mont Vesuvius and then stayed preserved till we excavated it, offers an incredible snapshot of what they were up to. And interestingly, a year ago, archaeologists excavated an construction site, including very well-organized construction materials like roof dials or bricks and tools. So imagine literally an active construction site that allowed us to walk through and understand from the raw materials how they were mixed and prepared to be then used in construction. So imagine people were building a house at the moment of eruption and literally stopped the operations and froze that scene. then it felt literally like entering into Pompeii 2000. So it was for me, a time travel, literally.

Right. From what I understand, the Romans left all of this concrete behind, but did they leave behind the recipe for making this cement that becomes concrete? Yeah, that's a great question. It's interesting that in that ancient time, there were these scholars, architects and scientists that did document a lot of, of what they were up to. Some of notable ones are, of course, Vitruvius and Pliny the Elder,

and both of them describe recipes for making this magic material. Vitruvius in his de-Architectura talks about magic powder. He says, quote, this substance, when mixed with lime and rubble, not only lend strength to buildings, But even when peers of it are constructed in the sea, they set hard underwater. So they document this recipe of mixing volcanic ash, lime, and then they notice this fantastic property that it hardens underwater. Now, did you discover that the ways that they made their concrete was different than the

ways we make ours? So what did these scholars, ancient scholars, suggest. is that lime, not the lime that we use in our cocktails, but processed limestone. So you take the stone, you create a kiln. And the product of this is quick lime, calcium oxide. And what these ancient scholars suggested is that this quick lime would first be mixed with water to make the reaction and create slake lime and then added to the volcanic ash to make concrete. What we found in Pompeii is something slightly different.

Ancient Pompeians would take this calcium stone, grind it, mix it dry with volcanic ash, and then add water. And that makes a slight difference we call hot mixing because mix heats up because of the reactions of clickline with water within the mixed, temperatures that can go up to 200 degrees in some hotspots. Vitruvius says first sleigh climb and then makes it nevertheless. In Pompeii, there is clear evidence that the raw materials were pre-mixed dry and then water was added, at least in this specific case of the Roman villa in Pompeii. And why is it that these ancient Roman structures are still so able to stand up for thousands of years? What's different about the cement in them than the cement we use?

Yeah, great. And we are talking about a self-healing concrete. Because of this way of hot mixing that when the microcrack is formed, basically are dissolving and recrystallizing in cracks that cannot be met with our modern analogs. That's amazing. I want to ask you what it was like. standing in this workshop that's thousands of years old for the first time for you? Oh, personally for me, was outstanding in a sense that together with my team at MIT,

we came up with this theory based on the research that was done on ancient Roman walls in Little Town in Preverno, of course, analyzing post-mortem concretes in a sense that they were already mixed, you know? And all our crazy hypothesis that indeed Romans use hot mixing, many of my colleagues criticized these arguments simply because it's difficult to think that this process might have been applied by dramas because Vitruvius says, hey, do this. So for me, finding a pile of pre-mixed dry, volcanic ash and quick, climb was incredible. I got emotional and of course my archaeologist friends made fun of me saying Admir, I mean there are beautiful frescoes around you and you get emotional and cry by looking at a

pile of dirt that pile of dirt was, you know, years of waiting to really confirm quite a challenging hypothesis. Well, as someone who loves to talk about concrete, I can see how you can get emotional about that also. Let me return to one question that that's fascinating, about this self-healing quality of the concrete that they made. You say modern concrete does not have the self-healing quality. Is it possible to put that into modern concrete somehow? Yeah, that's exactly what we are up to now

and we were able to patent some of these ideas, and the companies are starting to offer self-healing, Roman-inspired concretes out there. That's great. Thank you, Dr. Mashich, for taking time to be with us today. Thank you very much for having me. You're welcome, Dr. Admir Mashich, Associate Professor of Civil and Environmental Engineering at MIT.

After the break, we're going to move from something concrete to something still on the drawing boards. How realistic is it to move data centers to space? Stay with us. As tech companies build up data centers for AI, they are also searching for new ways to supply the immense electrical power necessary to run them, like reviving old nuclear reactors, for instance.

But lately, tech leaders like Google CEO Sundar Pichai, former Amazon CEO Jeff Bezos, and OpenAI CEO Sam Altman have been thinking not only outside the box, but out of this world. How do we one day have data centers in space so that we can better harness the energy from the sun? So these giant training clusters,

those will be better built in space because we have solar power there 24-7. The world needs a lot more processing power if that looks like tiling data centers on Earth, which I think is what it looks like in the short term, or we do go build them in space. I don't know. It sounds cool to try.

try to build them in space. Yeah, it does sound cool, but you know what? The devil is in the details, right? What resources and money would it take to move these enormous structures into space? And does this mean we could start seeing more of our infrastructure move into orbit in the next few decades? Here to beam us up is Dr. Benjamin Lee, Professor of Electrical and Systems Engineering at the University of Pennsylvania in Philadelphia. He studies microprocessor design and how to make them more efficient. in data centers. Welcome to Science Friday. Thank you so much. Wonderful to be here.

Nice to have you. So, you know, in those clips we just heard, it sounds like that power is the main draw. Is it as simple as that? There's more energy available in space. So let's put stuff up there. That's certainly the starting point. As many of us already know, finding the energy on today's grids is increasingly difficult, especially given the size of our data centers. We're talking about 1,000 megawatts, 2,000 megawatts. And grid capacity is scarce. And certainly, many of the hyperscaler data center operators are having difficulty finding sites that can support those power loads.

The second issue really is about where the energy is coming from. There has been a lot of interest in carbon-efficient energy. So solar is certainly one of those. But solar energy on Earth is clearly intermittent. You can't compute with solar energy in the middle of the night. And launching data centers into space could solve that problem. We could put data centers in Sun-synchronous orbit and get solar energy exposure on those solar panels pretty much continuously. So I think energy is definitely the starting point for orbital data centers.

Well, let's look into those devils in the detail. How big would the solar panels have to be to capture that much power? hour. Right. So we're talking about kilometers of solar panels if we're talking about these giga-scale data centers. If you're talking about a gigawatt or two, those panels end up being fairly significant. And the size of it is one thing. But the launch costs, the weight of those panels, and sending that all up to the space, it would be certainly another. Yeah. And when you have panels or anything in space that big, don't you have risks from, well, let's say space debris for these massive structures? Certainly. And reliability and repair is an ongoing challenge for anything we sent

into space. We know that especially for these largest data centers, components will fail occasionally, and we will need a strategy to replace and address that. And that's true for the hardware components that perform the compute. It's also going to be true for the solar panels. other infrastructure that goes up. Well, you know, on Earth, these data centers are cooled by water. How would you cool them up there where there is no water? That's right. And so the fact that space is cold is one thing, but it doesn't really help our ability

to extract the heat from the processor chips and send it out into the ambient. Normally, on Earth, what we do is we rely on blowing cold air or flowing cool water over the compute and then eventually releasing it out into the atmosphere. Without an atmosphere in space, you would need to rely on radiative cooling. And radiative cooling really relies on yet more panels and larger surface area that would allow us to radiate that heat out into space. So not only are we talking about the surface area for solar panels, we're also talking about surface area for the cooling. Okay, so let's say you've got these data centers in orbit around Earth and they're crunching a lot of data. How do you effectively communicate with them?

speed, sending the data back and forth. Right. We might be able to send on the order of 10 gigabits per second up into space using radio frequencies. But the difficulty is really about moving large volumes of data. So in one of the earlier clips you aired, there was a discussion of training large AI models in orbit. The difficulty is that training requires massive data sets, essentially all of the Internet's data. and getting all of that data up into space would be challenging.

We would essentially have to launch that data up into space along with the compute. There's an old joke in computer science where we say that if we wanted to send a large amount of data quickly, maybe from the East Coast to the West Coast, the easiest thing to do, the fastest thing to do would be to put a disk in the mail. And I think that's certainly true when we're talking about sending things into space as well. Well, you've mentioned all the negative parts about doing this, all the challenges of putting data centers in space. Why not just make data centers on Earth be more efficient in how we make power and how they process data? That's right. And I think right now we know that we are somewhat wasteful in how we are consuming data center resources. We are pursuing

first and foremost, better and better AI models, and we're willing to pay whatever energy cost it takes to get that new capability, that new application. We don't want to worry whether or not, if we had just thrown an additional 100 megawatts of the problem, we could have gotten something we didn't have before. So right now, when we think about the AI models, they are equipped to answer any possible query

that any possible user could pose to it. And that generality really, is computationally expensive. If, however, you knew that you wanted a model for finance or you wanted a model for medicine or for education, you could come up with specialized models. I could give you an answer that is just as good, but require far fewer calculations. The reason why we haven't done this yet is because we don't know what those really compelling applications are, the ones that will change the way we live and work. But once we figure those out, I think there will be plenty of opportunity to reduce the energy costs and improve energy efficiency.

Because what you're talking about, what these guys are talking about going into space, that's not going to happen this year or 10 years from now, is it? And in that time, who knows what kinds of efficiencies we might have? That's right. And going back to your other point, I would also say it's very hard to know what the energy landscape in the United States will look like in 10 or 15 years, whether we will have much more battery capacity, much more solar capacity, small modular reactors. All of that is also possible within the next 10 or 15 years. 15 years, and that may change how we view the relative merits of doing one of those things versus sending data centers into space. Do you still think, though, that we should be thinking about sending data centers into space

as more than just a thought as a possibility? I think replacing these massive terrestrial data centers with orbital data centers is really ambitious and potentially a long-term goal. But in the near term, I think they're a really interesting, fundamental research questions, going back to some of the things we had just talked about, radiation hardening, thermal management, power delivery. I would also say that the challenges associated with data movement also means that there is an opportunity to do more compute in space.

Maybe we have satellites collecting massive amounts of data, image data, or other kinds of data from orbit. Instead of sending that all down to Earth for processing, you can imagine doing more of that compute in space. I think that would be really great application of orbital computing without going to this end goal of replacing massive terrestrial data centers. So you do think we should expect to see more of our infrastructure moved into orbit. Yes, and I think that that is a natural progression from where we are today, where we are seeing with SpaceX and other private space companies significantly reducing the cost of launches,

and also with the advent of these constellations of communication satellites like Starlink. I think putting some amount of compute to complement that communication would be really exciting and would make a lot of sense, but maybe not at the scale of 1,000 megawatts of data center capacity. Thank you, Dr. Lee. Terrific. Thank you so much. I really enjoyed it. Dr. Benjamin Lee, Professor of Electrical and Systems Engineering at the University of Pennsylvania and Philadelphia.

This episode was produced by D. Peter Schmidt. I'm Ira Flato. Thanks for listening.