Scott Caravello: So, we'll first, I think, give a quick background on what quantum computing is, which will help frame up that discussion. But the big idea here is that because quantum computing is supposed to be so much more efficient, that it's going to significantly amplify the compute power that's available. But it's a complicated story. It's not likely to be some magic bullet for our power needs in the near future. It will increase efficiencies alongside the classical computing workloads that power AI today and that we previously talked about on the podcast.
Katherine Forrest: Right. So, let's just sort of back up. You know, I like to sort of do the 101 to remind people who haven't listened to our... We actually did a two-episode piece on quantum computing, both to remind folks that it's there, but also to just, for those of you who don't want to go back and listen, to just sort of give you a high-level overview of what quantum computing is. And so it's a form of computation that uses the principles of quantum mechanics—the principles of quantum mechanics—to process information or to actually perform the compute. So, in the classical world, you know, the way in which computers have always run, there are “bits,” which are 0's and 1's, and those are the smallest unit of information that is used in classical computing, what we're now calling classical computing, the 0's and 1's. But in quantum computing, they're called “qubits.” And, you know, there's a real difference between the two because with classical bits, they're determined. They're either a zero or they're a one. And they're only a zero or a one at any given point in time. So, in other words, in classical computing, which is the computing that we've all lived with up till now, you know, the bits that are actually making our computers run, they occupy a single state, a zero or a one. And we just take that for granted, right? We just take that for granted all day, every day, Scott. You're taking it for granted right now, I bet, right?
Scott Caravello: Indeed.
Katherine Forrest: Okay. But in the quantum computing sort of world, you go to qubits, which are the smallest level here with quantum computing, and they are not just zeros or one, they can actually be both at the same time. And that is a pretty sort of unusual thing. And it's called “superposition,” all one word—superposition. And superposition is the principle of quantum mechanics. And it means that you can have qubits be in multiple states all at the same time. So it's, you know, and let me just give you sort of very much of a simplification. But superposition, which allows a qubit to be everything all at once until there's a certain point in time when things collapse and we'll, you know, we won't actually get into all of that right now. But when you have the ability to have qubits be everything all at once, then certain complex computing operations can be run much more quickly and much more efficiently. And as you have previewed already, Scott, you know, most experts don't expect quantum computing to be moving out of research labs and replacing classical computing right away, but we are now making significant advances in quantum computing. And the most likely, and in fact, what's actually already starting to happen is a hybrid setup, which is quantum processors working alongside today's classical computers. So, I just want to say one more thing about the superposition, think about it in terms of probabilities, that the way that quantum computers work is they're able to calculate multiple instances of probabilities all at the same time, whereas classical computers are determined, you know, in terms of these the 0's and 1's sort of, you know, force a determination at every point along the way.
Scott Caravello: Right, and so kind of operationalizing that further, we can think of the key hardware components of classical computers on the one hand in two broad groups that are relevant to the AI discussion. The CPUs, or the central processing units, which are just general purpose computing chips. And then the GPUs, or the graphics processing units, which we've talked about recently, but they're specialized chips that are doing a lot of math in parallel and which are perfect for AI. And, so, then on the quantum side, you now have the quantum processing units, or QPUs, which would be an add-on in this hybrid setup. It's kind of like how GPUs were added to accelerate AI compute and help developers achieve so many breakthroughs, but when that happened, it didn't just go ahead and make CPU obsolete in the picture. Quantum systems shine when they're handling very hard math problems, but they could still lean on the classical computing machines for certain tasks that the quantum power isn't required for, like moving data around, fixing errors, or storing information.
Katherine Forrest: Right, and it's really fascinating. And, so, just not to confuse our audience: right now, today, AI models are run by and large, and almost exclusively, on classical computers. So, that classical computer setup that we talk about, the ones that everybody has in front of them right now, or in their offices or at home, on their phones, those are all classical computers. We're talking about something different. We're talking about the advent of this far more probabilistic kind of computing, this quantum computing that can actually, in the future, and then very near in future, add huge amounts of capability to the classical computing world. And, so, there's actually a really interesting framework and I want to sort of direct our audience towards the IBM website where you can read all about this. But IBM has been developing a really interesting framework that they unveiled just a short time ago about this hybrid approach that, you know, hybrid and classical with classical and quantum together. And they call it the “quantum centric supercomputing architecture,” or “QCSC” for short. And the idea is that, to create a quantum system where quantum processors, classical processors, and even specialized AI accelerators, all work together, you have one unified compute fabric. And that compute fabric is in this IBM created quantum centric supercomputing, QCSC, architecture. So, it's really fascinating.
Scott Caravello: Yeah, so, again kind of hammering this point about the hybrid nature, right? With this system, rather than thinking about quantum as a separate and siloed technology, IBM's vision is that the quantum pieces, the system would slot right into the existing classical computing, supercomputing infrastructure. And, when we're talking about supercomputing in that context, we're talking about extremely powerful, high-performance computing systems, but still on the classical side of the discussion we've been having. And, so, IBM is designing what is called a system reference architecture, which essentially is a blueprint for how to connect multiple quantum processors together and link them with the classical high-performance computing resources. And so, again, that's the important point on this development, or one of them, is that quantum wouldn't be replacing the classical computer, it would be a part of it.
Katherine Forrest: Right, for at least now…
Scott Caravello: Right.
Katherine Forrest: …because there's also quantum computers, just quantum computers. Those are being worked on, but there are lots of things having to do with stability and some other things that we won't talk about too much right now, maybe we'll just touch on them, that will make quantum computing without this hybrid approach a thing that will happen in the future. But right now, for our current purposes, this hybrid approach is really fascinating. And one of the key elements of this approach is what's called the middleware layer. Middleware—like “middle,” and then W-A-R-E—middleware layer. And that's the software that actually orchestrates workloads across both quantum and classical systems. And so it's the director of the workloads and says, okay, this is going to go to quantum, this is going to go to the classical piece. And the system intelligently decides which parts of a problem get sent to the quantum processor and which stay on the classical side.
Scott Caravello: Right, and, so, then another piece of this is what's called “circuit knitting techniques,” which allow you to break up large quantum problems into smaller pieces that can run on today's more limited quantum hardware, again, because the technology is still really in development. And then it will stitch those results back together, if you will, to, you know, get the full piece. And, so, that would help to get way more capability out of current quantum systems while we wait for the larger, more powerful ones to mature.
Katherine Forrest: Way… more… capability? Is that what we say? Way more capability?
Scott Caravello: I told you, this is why I need the Celsius. I don't know what to tell you.
Katherine Forrest: All right, so okay, so we've got this sort of example, one example of some real advances in this hybrid architecture. So what I really wanna talk about now, let's turn ourselves to the data centers and to the implications for the data centers because as we explained with superconducting qubits in the prior episode, they've gotta be kept really, really cold and they have to be kept at a temperature that's near absolute zero. And that's true even for the hybrid architecture that we're talking about because you still got a portion of that that is quantum. And so you still have to have the ability to actually keep those superconducting qubits at an absolute zero temperature. And, so, what you want to do is you want to have the motion of the atoms almost stop completely inside that qubit. And so that's a really, really cold temperature that is needed to do that. And to further anchor that idea in people's minds, you have to get to negative 273.15 °C and -460°F to get there. So those qubits have to be protected from any stray magnetism and vibration also. So, you've got to not only keep them incredibly cold, but you've got to keep them away from any kind of magnetism or vibration. So they're really delicate little things. And other designs like photonic quantum use light and can operate close to room temperature, but they also require their own special networks and control systems. So in all events, quantum computing is going to demand additional aspects to the data centers that are already in existence to house them. And it's going to require, of course, these cooling rooms and special ways of eliminating or controlling the magnetism and the vibration. So there'll also be new expertise. But there is a possibility, there is a possibility that part of what this could do, because you're going to now take part of the functionality and the requirements of a hybrid classical system, quantum system, and you're going to actually split them between the two that you might end up shrinking down, a little bit, the classical system.
Scott Caravello: Right, so to sum up some of the takeaways from that, first, there's a challenge. And so even when the quantum is operationalized and we're ready to yield the benefits, there are adjustments that are going to need to be made to the infrastructure that we use to power both AI and other resource-intensive technologies. So taking them into account, and the fact that classical computing still has a big role to play, the need for big data center facilities doesn't vanish in the near term. It'll change, sure, but it'll still be there. And we may, you know, start to see them featuring just more of the quantum infrastructure, like the ultra-cold cryogenic refrigerator towers and isolation rooms that will just be sitting next to today's high-density GPU rows used for AI.
Katherine Forrest: Right, and so let's talk about how energy plays into all of this. And energy is a big topic of conversation right now. The energy markets are in some turmoil with what's going on in the Middle East. But remember that the energy that can be used to power AI data centers can include solar and hydro. And so we're not limited to fossil fuels, which is very important when you've got some disruptions in terms of the cost and the accessibility of different kinds of fossil fuels. And as we discussed in our episode, about a month or so ago, on hyperscalers, today's AI data centers require a lot of power. So, you've heard us say that quantum can handle all kinds of complex operations in a far quicker way, and a more efficient way, than classical computing can, but... how does that translate into energy needs and does it actually reduce energy needs? Now, interestingly, it is a little bit too early to say exactly how it's going to play out because, again, you're still going to need a lot of energy for that refrigeration that we talked about. And in the near term, those freezers are going to take a meaningful amount of power. So, even if quantum computing can be more efficient for a specific problem, that doesn't necessarily mean that those efficiencies are going to translate into a one-for-one reduction in the infrastructure. You might have some infrastructure reductions in terms of footprint, but you might not have reductions in terms of what you're pumping into, for instance, a particular AI center because you're going to, in terms of power, because you're still going to need to power the portions that are running the quantum facility.
Scott Caravello: Yeah, and, you know, when you start talking about the problem level, right, about how efficient quantum computing is going to be for specific computing problems, there's a split in the debate within industry. Some expect the quantum computing to improve energy efficiency per calculation. Others, though, are pointing out that the specialized equipment that we've been talking about, like the cooling needs, are going to require significant overhead that is going to really cut into those efficiency gains.
Katherine Forrest: And so then we have what's called “Jevons Paradox”—J-E-V-O-N-S—and it's pronounced like “Devons Paradox.” And it's actually a general economic term. It's not a quantum specific phrase, but the idea is that when something becomes cheaper or more efficient to do, or to use, people might in fact use a lot more of it rather than less of it. It's sort of this paradox. You think that you're going to get sort of suddenly an excess, but you don't get an excess. You might get more usage. So, if quantum does lower the cost of solving, for instance, all kinds of massive computing issues, say through drug discovery, complex, what we call, optimization problems, it may be that we find that we discover ways to simply run more tasks, bigger tasks and that those actually offset some of the power savings, that we actually run more through the system because we're able to.
Scott Caravello: Exactly. So having covered all that information about infrastructure, maybe, Katherine, we can talk a little bit about where we might first expect to see AI uses that are impacted by quantum.
Katherine Forrest: I want to say, Scott, not just because I'm on the ocean, but the world is our oyster on that because we don't exactly know. But the things that we do know about are that because you've got this massive, probabilistic computing power that quantum will be able to do, for instance, in drug discovery, simulate all kinds of molecules, all kinds of materials. Where AI is able to identify the best option among a range of possibilities that we already are starting to see, quantum can massively potentially increase that, which is part of what we are calling this optimization. And you might see fewer rows of GPUs grinding away at that same problem because they're going to be using quantum computers. But anyway, drug discovery is a terrific example of the possibilities of quantum that I think people can really grab onto. But anywhere that you're talking about needing to run massive amounts of data, where you wanna actually play out millions or huge numbers of say Monte Carlo simulations, you're going to have a need for quantum computing that's going to be really, really interesting.
Scott Caravello: Yeah, and so exactly on that point, right? Logistics and scheduling is another good example because it's one thing when you're talking about 10 different possibilities, but if you're talking about 10,000 possibilities all at once for the best way to move something or schedule shipments, AI enhanced by quantum could do that faster. And so it may not matter in a lot of contexts, but where those real time decisions matter and there are a lot of different potential options, it could really have an impact. And then another one, Katherine, that we talk about already with AI is algorithmic trading.
Katherine Forrest: Right, I mean, think about arbitrage as an example where you're trying to sort of take a piece between the delta of two price points. And if you think of arbitrage and you're able to figure out, and discover, all pricing at all moments as a trader, like immediately, the arbitrage capabilities are extraordinary. On the other hand, you're not going to be the only one with a quantum computing capability. So there's that. So people ask, well, when? When are we gonna be able to have access to all of this? First of all, there is some access right now in some specialized places. But to really commercialize it, you need to build some applications. We’ve got now some of the infrastructure, but the applications really have to come. So you hear everything from five years to decades, I'm much more in the near-in sort of solution. And there are companies that I won't name because I realize every time I name a company, I have to go through a risk review, but you know you're out there. And so there are companies that are making really extraordinary leaps and bounds. And as quantum sort of increases its developments, we're going to be able to get there, I think, more quickly. So, I think we're going to see quantum, in my view, in the next… certainly by 2030, in a real way. I think that super intelligence is going to help that because once we have super intelligence with classical just AI running on classical computing it's going to help us solve some of those really complex quantum issues. And by the way, plug for the audience, Scott, just before we go on. My book on super intelligence is now available for pre-order, commercial break, on Amazon. You can pre-order it from World Scientific Press on Amazon called Of Another Mind: [and it’s called] Navigating a New Social Contract with Superintelligent AI. Katherine Forrest and Amy Zimmerman, get your copy now, pre-order. Actually, we're moving up the publication date. It's gonna be in May, it's not gonna be in June. Anyway, point is super intelligence is going to help us, I think, get advances on quantum computing faster.
Scott Caravello: Totally agree. And so I will just provide a quick wrap up because we're running out of time and I'm going to deliberately slow down because again, the Celsius is really raising my heart rate at this point. But will quantum reduce the need for sprawling AI data centers? And our view is at least not in the foreseeable future, won't do so in the aggregate, right? It'll reduce the need for these racks of GPUs that are just throwing brute force computing power to handle certain workloads. But the overall footprint is gonna remain significant as ambitions grow and the complexity of these hybrid systems increases.
Katherine Forrest: Right, so we've got labs that are working on this, but we already have early hubs where you're able to see some of the commercialization happening now. So look around you, dig into it. It's going to have impacts on what the AI infrastructure looks like. I think it's going to have an impact on the footprint. It's certainly going to have an impact on what goes on inside of it. So we'll follow those developments, very exciting stuff. But Scott, that's all we've got time for today. I'm gonna have to go out and hike now.
Scott Caravello: Well, I'm Scott Caravello. Thanks for joining us.
Katherine Forrest: You're not gonna hike?
Scott Caravello: I mean, that's just mean.
Katherine Forrest: Alright, we'll see you folks next week. Bye bye.
