Hello, hello. I am going to apologize for this voice; it's because I was sick, not because I was partying, though it's Roth Conference. You probably will not believe that. Good afternoon, my name is Suji Des ilva. I'm the Semiconductor and Intelligent Systems Analyst here at Roth. We've had a great two days of conference; we're going to wind things down here with a couple more fireside chats. I'm proud to have D-Wave here, one of the innovators in quantum computing. The CEO here, Alan Baratz, is here, and the CFO John is sitting in the audience. Thank you guys both for coming; we're really excited to have you here. Quantum computing is one of the new areas in my intelligent systems world, and certainly kind of coming up, coming up the curve here.
So Alan, quantum computing clearly has been out there, you know, we've had cloud compute, AI, high performance compute, I think it was a continuum with quantum compute kind of being one of the more leading edge ones. Can you talk about where we are in this, the commercialization of this technology?
Sure, absolutely. So first of all Suji, thanks for the opportunity to be here, and thanks for leaving an empty chair between us.
We appreciate that.
You know, in case you all didn't catch it, Suji did say he's not feeling well, so, you know, we'll make sure we have some separation here. Anyway, commercialization of quantum, look, we're absolutely at a watershed point in time for quantum computing. We are at the point where quantum is now moving from research experimentation to actual production use, supporting businesses and helping to improve their operations. But what's interesting about that statement is there's only one company in the world that can actually make that statement, and that is D-Wave, that's our company. And the reason is that, we selected a different approach to quantum from everybody else in the industry. We selected an approach that is much easier to work with, it's easier to scale, less sensitive to noise and errors, and that has allowed us to become commercial today.
So we have companies like Interpublic Group, a large ad firm, or VINCI Energies is a big construction company in Europe, Mastercard, you all know. These are all companies that are working with our system to leverage it to help improve their business operations, and that's happening today. So, you know, you might hear quantum pundits, other quantum companies say that quantum computing is not yet commercial. Some will say we're a few years away, some will say we're five years away, some will say you're 10 years, we're 10 years away, and that's true for their systems.
They really do have fundamental research that remains to be done before those systems will be large enough and, I'll say quiet enough, error-free enough to be able to support real business applications at commercial scale. But with D-Wave, we're absolutely there today. Now, maybe I'll make one more point since I promised Suji that I would do a lot of talking to help save his voice.
It's a great point.
We think about the applications for quantum as falling into two categories, what I'll call evolutionary applications and what I'll call revolutionary applications. Revolutionary applications are the kinds of applications that you always hear about with respect to quantum, developing designer drugs, global weather modeling. You know, I'll even put breaking RSA, public key cryptography into the revolutionary space. Yeah, we're years away from being able to solve those problems with quantum computers, but there are also evolutionary applications. Things like improved employee scheduling, or better autonomous vehicle routing to optimize manufacturing plant floor optimizations, or protein folding. These are all problems that actually are being solved today by businesses. It's just that they're computationally so hard that they're using heuristics to try to come up with what they hope are good enough solutions.
But with our quantum systems, we're able to come in and give them better, if not optimal solutions to those problems much faster right now. So we're improving the performance, in some cases the solution quality, in other cases of those applications that are in use today in the market. And that's real value, real ROI for these customers out of quantum today.
There was a really interesting example you told me about. If FedEx or UPS wanted to optimize their entire U.S. delivery, you gave me some numbers that were interesting.
Yeah, so, with our quantum systems today, we can solve problems with, you know, up to 1 million variables. That may not mean much to you, but I'm going to give you context. With one million variables, we can solve the types of problems that I was just talking about. We can, we can help companies better schedule their employees against a broad set of criteria and constraints. If we wanted to solve the full-up FedEx global routing application, that would require about 50 million variables. So there's still a ways to go before quantum can solve even all the business optimization problems that are out there, but there are a lot of them that we actually can solve today, and we're doing it.
It speaks to the roadmap you have. You talked about, you recently announced a collaboration with Zapata AI. I think a lot of folks are very knowledgeable about the AI landscape, what's going on, but also still just reaching around trying to figure out how players are coming together. So it would help us to understand what the collaboration may mean for you guys, providing an advantage.
Yeah, so that's a really exciting partnership that we announced, but before I address that, let me take just a quick step backwards. The class of applications that we are addressing today are what I characterize as business optimization problems. Okay? These are problems where we have to minimize or maximize optimize some objective function subject to a bunch of constraints. However, the customers that are solving those problems today and that are interested in leveraging our system to improve the solutions that they're getting also need to use generative AI to get what will likely be business operations data for the future so that they can start thinking about how they optimize for the future today. So Gen AI is important in their context as well. So we're often asked questions about whether or not there's an opportunity for us to improve how generative AI is done.
The answer to that question up until a few months ago was yes, but we probably need a few more years before we can get there. Recently we started working with a company called Zapata AI, as Suji mentioned. They've done a lot of work in how to leverage quantum computers to improve model training for generative AI. The concept is that quantum computers can deliver samples for model training that fit a quantum distribution, and I'm not going to get into details about what that means, but quantum distributions have many more degrees of freedom than classical distributions. So in theory, if you're training your model on samples from a quantum distribution, you should be able to get a better fitting model because you have more degrees of freedom to work with than if your samples are coming from a classical distribution.
So this is work that they've been doing for a few years now. They did some work with IonQ, another quantum computing company a couple of years ago, and they got some promising results, but, you know, the number and size of samples that they could get was limited because those systems are small and buggy. They did some work with IBM more recently. Again, got some interesting results, but samples are small and buggy. We started working with them about three months ago. In their first meeting with us, they said, so, you know, if we need a million samples, you know, what will it take, a week or a month for you to generate those for us? And we said, a microsecond. And they were like, what? I mean, that's the difference between a commercial quantum computer and a research prototype.
So, so we established a partnership with Zapata whereby, we are together building an application to take to market that essentially uses our quantum computer on the backend to generate quantum distribution samples for model training, and then we're going to take that to market together. Now, there are two really important benefits to doing that. One, better models. So, you know, as I said, more degrees of freedom allows us to get a better fit. But there's another really important element to this, and this is one that we're maybe even more excited about. It turns out that our quantum computers are very energy efficient. We use superconducting circuitry, and superconducting chips basically draw no power, so the only power requirement is the refrigerator to run at cold temperatures.
What this means is that not only will we be able to generate better models, but we will be able to do it at a fraction of the power consumption that occurs today with NVIDIA GPUs to train the models. So we think that this could be quite transformative in both respects, and that's really exciting. And, and by the way, we're not on a long timeline here. Our goal is to get this to market within three-six months, so we're really trying to move that forward.
Are there lead customers you're working with already, or that's to come?
Zapata is doing some work in the area of protein discovery with some drug companies, and so that is the lead application area.
Okay, you also published a paper recently about quantum supremacy. It sounds like a Jason Bourne movie title, but I would love for you to tell us what that means practically for the investor base.
Yeah, so let me start by saying there's a fair amount of confusion, I think, in the market around terminology. Quantum supremacy, quantum advantage, quantum utility. These are the terms that you'll typically hear. John Preskill at Caltech originally coined the term quantum supremacy. And essentially what he defined that to be is the ability for a quantum computer to solve a problem that cannot be solved classically within any reasonable period of time, but where reasonable is measured in years to millennia. In other words, it's not like you can solve it on a quantum computer twice as fast as you can solve it classically. So essentially you can solve it on the quantum computer, but you can't solve it classically. Right? When after Google for the first time announced quantum supremacy, and then three days later IBM showed that it was not.
But after Google announced quantum supremacy for the first time, some Google researchers came out and said we shouldn't call it supremacy. We should call it advantage. Right? And then some researchers kind of jumped on that bandwagon. I'll be honest with you, I kind of view that as self-serving on the part of Google. We announced supremacy, now nobody else can. Okay, fine. The problem we've got with advantage is that has now come to mean many different things to different people. And that now means everything from I can solve it a little faster to it can't be solved classically. So we've kind of screwed up that term, right? Which is why we wanted to be crystal clear with this result that you cannot solve this problem classically. So what was the problem? Basically what we are doing is simulating magnetic crystals to find their properties. Okay?
So it's a magnetic materials simulation computation. Now, this is a real world problem. This is a problem that scientists regularly work on. Right? What we've been able to do is solve this problem at different lattice sizes and different lattice structures from small to large. On the small size lattices, we run on our quantum computer, and thanks to Oak Ridge National Lab, we've been given time on Frontier, which is the most powerful supercomputer in the world. And on the small lattice sizes, we agree, right? So classically on small lattice sizes, you can get the optimal solution, you can prove it's the optimal solution, and we run and we match perfectly. Okay?
On the large lattice sizes, we're still running in about 20 minutes to solve it, but the extrapolation would argue that at those sizes it would take well over 1 million years on Frontier to perform the computation. So that's what we mean by we can solve it efficiently, it can't be solved classically. So then you might say, well, hey, Alan, how do you know you're getting the right answer? Right? I mean, if you can't solve it classically and you can't guarantee, so you don't know what the answer is, how do you know you're getting the right answer? This is very clever and very important work that we did with partners around the world. This was not done by just D-Wave researchers. It turns out that there are a lot of things known about these crystals. They've been researched for quite a long time.
While you can't get the exact solution, we do know a number of different properties that come out of the exact solution. Once we compute the solution, we can then go and look at whether those solutions have the properties that they must have in order to be the exact solution, and we match perfectly. We know we're getting the exact solution to this simulation. Now, this is a huge deal, not only because we've demonstrated true quantum supremacy on our quantum computer, but this is the first time that anybody has demonstrated quantum supremacy on a real world problem. As I said, Google announced it a number of years ago, and then three days later IBM showed that it could be solved classically. Google then sort of upped the ante and announced it again.
That result seems to have held, except for the fact that it's not a real world problem. It's basically kind of an esoteric computation that can only be run on their quantum computer. You can't even take that same computation and run it on one of their competitors' quantum computers. Right? So it's a very esoteric, not at all useful result versus, in our case, for the first time, a true important real world problem. And that's a huge step forward.
I know I speak for the investors when I say we prefer the real world because that's where real money is. Can you talk about the product roadmap a bit? Because you guys have been bringing the Advantage platform forward, Advantage 2, so love to hear about that roadmap.
Yeah, so our current flagship quantum computer is called Advantage. It's a 5,000-qubit quantum computer. It is the largest quantum computer in the world. Now, I want to be a little careful here because, as I said at the very beginning, we've taken a different approach to quantum from everybody else, and so our qubits are different from the qubits that the other folks are trying to build. But nonetheless, we do have our current flagship product with 5,000 qubits. We just announced and made available through our quantum cloud service a small version of our next generation quantum computer. We call it Advantage 2. This system only has 1,000 qubits. The final product will have over 7,000 qubits, so that'll be up from the current 5,000.
The current system has only 1,000 qubits, but it has twice the coherence for faster time to solution, and it has 40% greater energy scale. Details don't matter other than to say it allows us to specify problems with greater precision, so we get better solutions. This 1,000 qubit Advantage 2 system is more powerful than our 5,000 qubit Advantage system. And in fact, it was the Advantage 2 processor that we did the quantum supremacy work on. We started doing the work on the Advantage processor, but we weren't getting high enough quality solutions. There was error in the solutions. With the Advantage 2 processor, we actually were able to achieve the supremacy result.
As we look to the future, as I said a minute ago, we're going to up the number of qubits in the Advantage 2 processor from where it is today at 1,000 to over 7,000, and there'll probably be a stop midway at around 5,000 qubits that you should kind of watch this space on in the not too distant future. We're feeling quite good about where we are with that. Moreover, our investments are not just in hardware, not just in the quantum computers, but also in software. For many of our customers, we use what we call hybrid solvers, which means that we're using GPUs and CPUs together with the quantum processor that actually allows us to solve larger problems than we can solve natively on the quantum computer, although the supremacy result was done natively on the quantum computer.
So we're constantly investing in the software as well, and in fact, we're working on our next generation hybrid solver, which is already showing amazing results. So just to give you an example, many of you may be familiar with the traveling salesperson problem. It's a kind of a well-known hard computational problem. You're a salesperson, you need to visit a bunch of cities, and you want to come up with the shortest route to visit each city once and only once and end up back where you started. It's an exponentially hard problem, which means that at large sizes you cannot solve it optimally on a classical computer. It would take too much time. So people use heuristics to try to solve it.
With our hybrid solvers, with our current generation hybrid solver, we're able to solve that problem up to about 200 cities, which is beyond what classical can solve. With the new hybrid solver, we can solve it up to about 1,000 cities. So we're constantly improving our software alongside of our hardware, and we're actually really excited about the product roadmap, about the progress that the team has made today. I mean, we're way out in front when it comes to commercial quantum, and we'll continue to innovate.
I had the pleasure of visiting the Vancouver facility, and the technology that goes in is amazing. Throughout today, the audience, if there are any questions.
Just on that last point, what's the cap on the cities for classical computers for that problem?
Yeah, so that's always a challenging question. If you want to solve it exactly, guaranteed exact solution, the cap is maybe around 100 cities. Okay? But there are heuristics that can get you further up. You're just not guaranteed to get the optimal solution, and often you will not get the optimal solution.
What was kind of the technology problem you solved or how were you able to get commercialization while whether it's taking a different technology route or just R&D than IonQ and Rigetti and IBM?
Yeah. Okay, so there are two parts to the answer to that question. First of all, I really do need to reinforce the fact that we took a different approach to quantum computing from everybody else. There are two primary approaches. One is called annealing, and the other is called gate. We decided to start with annealing. Everybody else decided to go down the gate path. Now, there's a reason why that happened. The reason why we decided to start with annealing has to do with what the state of technology was when we started building our quantum computer, which was 15 years ago. So we've been working on these systems far longer than anybody else. Right? At that point in time, there wasn't really line of sight to how to build a gate model system. Right? So we started building an annealing quantum computer.
We knew it could solve many important problems, like business optimization problems, but we knew it could not solve all problems. Seven or eight years ago, when everybody else jumped into the quantum for IBM, IonQ, Google, Rigetti, it was believed that a gate model system could solve all problems. And that's why they all selected gate rather than annealing. Two years ago, everybody got surprised. Researchers in the U.S. and Europe proved that gate model systems cannot deliver good solutions to optimization problems. So we have a bifurcation in the application environment. There are applications that will always require annealing. Those are business optimization problems. There are applications that will always require gate. These are things like differential equations for quantum chemistry.
So it's played out quite well for D-Wave because it means that there's an important piece of the market that only we can address from a quantum perspective because it requires annealing, and we're the only company in the world that does annealing quantum computing. So what were the hard problems that we needed to solve? First of all, there are architectural problems with respect to how you actually design an annealing quantum computer. Then we chose superconducting. We needed a fabrication process that would allow us to fabricate these chips with low enough noise that we could have long enough coherence time to actually solve real-world problems. So we did a lot of work on the fab process. Then there was work on refrigeration. We don't build our own refrigerators. We use Bluefors and Oxford, but those refrigerators are not very reliable.
After about three months, they build up contaminants, and you need to warm them up. You can't have a commercial quantum computer that has to warm up every three months. So we extensively modify to basically run them for three, four, five years without warming up. So we've had to innovate in a lot of different areas.
I would add, they deliver it as a service. It's inside Amazon's cloud services. There's no way Amazon let them put their machine in there without vetting the hell out of it. Clearly, you guys done a great job. With that, we'll wrap it up. Thanks, everyone.