Good afternoon. So, my name is Alan Baratz. I am the CEO of D-Wave. I have no clue who that guy in the picture is. It's a real pleasure to be here with you all today. I thank every one of you who took the time to come down here to the New York Stock Exchange and be here in person. Thanks also to those of you that are joining us via the live stream. There's quite a bit that we would like to cover with you today, but honestly, the Q&A sessions are equally if not more important. We want to make sure that there's plenty of opportunity to ask questions, and so we've allocated three different times for Q&A, and hopefully through that we'll have the opportunity to answer most if not all of the questions.
So, I'm going to start with the obligatory picture of Richard Feynman, although hopefully you haven't seen this version of the picture before, since our excellent graphic designers made it up. Richard Feynman, of course, coined the term quantum computer back in 1981. And he's famous for a number of other interesting and important commentary on quantum mechanics and quantum computing. One that I think is particularly relevant to us, and by us I mean the entire quantum computing industry here today, is the one that we have up on the screen. If you think you understand quantum mechanics, you don't understand quantum mechanics. This is very complex technology. The problem with that is that it's way too easy to say misleading things and to spew hype into the marketplace, and it's hard for all of us to be able to separate all that noise from the signal. And so, part of what I'm hoping to do here today is at least for D-Wave, help you think about how to separate the signal from the noise, and how to think about D-Wave in the context of what we are telling you against, if not the science, at least sound logical thinking.
So, I'm going to start at the very top. What is quantum computing? For me, quantum computing is energy-efficient computing for solving hard computational problems. That's typically not where people start when they talk about quantum computing, but I start there because quantum computers consume very little power. And on top of that, they use quantum mechanics to compute the solutions to problems very, very quickly. And that combination truly allows them to be much more energy-efficient at solving problems than classical computers.
Now, there's more than one approach to building a quantum computer. And this is particularly important in the context of understanding D-Wave and how we are different from every other quantum computing company in the industry. There are two main approaches to building a quantum computer. One is called annealing and the other is called gate model. For those of you that have been following us over the years, you've heard us talk quite a bit about annealing because frankly, we started with annealing and we're the only company in the world today that brings annealing quantum computers to market. As we go through the course of the day, we're going to talk not only about the power and value and road map for annealing, but we're also going to talk about gate model because we are now also a significant player in gate model quantum computing. But before I dive into it, let me just take a minute to try to explain the difference between gate model quantum computing and annealing quantum computing. And I'm going to start by talking about classical computers.
So, typically you'll hear people say that classical computers use bits to store information and a bit can be either a zero or a one. And this means that at any point in time, the information that the classical computer is seeing is one possible solution to a problem. It's how zeros and ones have been assigned to the various bits. That shows you one possible solution. It's like dropping a tennis ball into a landscape and it hits at one point. Now, suppose you want to find the overall lowest point in this landscape, which is the valley to the left of this image. Well, how do classical computers work? They basically perform computations and logical operations that attempt to explore the landscape one point after the next. They're moving that tennis ball around point by point looking for the lowest possible point. Now, if your landscape is simple like a bowl, it's easy. There are very efficient algorithms for finding the lowest point. You can get there quickly. But, if it's a complicated landscape, this is actually an exponentially hard problem. What it means is there's not enough structure in the problem to be able to efficiently find that lowest point. The amount of compute time is exponential in the size of the problem.
And this is exactly the kind of problem that's good for quantum computers, these very hard computational problems. So, how does a gate model quantum computer work? Well, gate model systems use qubits, and a qubit can be zero and one at the same time, which means that at any point in time, you're able to see multiple possible solutions to the problem, not just one possible solution. And you're able to perform operations on those multiple possible solutions simultaneously, computations, logical operations. So, the problem becomes one of not kind of figuring out how to follow the ball to the lowest point, but you've got all the balls. You can see all the points. Now, the algorithms are all about figuring out which one of those balls is actually at the lowest point. It's a different way of thinking about computation and how to program the computer. But, there's one other important point about gate model systems, and that is quantum mechanics was not really designed to support digital computation. So, as we build these gate model systems, we're trying to implement digital quantum computation on quantum mechanical systems. We're pushing against nature, and that introduces errors. That's why gate model systems are so error-prone, and why without the ability to correct those errors, we're never going to be able to solve useful problems on gate model systems. So, error correction is critically important for gate model systems.
Annealing quantum computers work in a very different way. Annealing quantum computers are really all about finding the lowest point in a large landscape. They're about using quantum mechanics to find what we call a low energy state. It's working with quantum mechanics. You can think of it like dropping water into the landscape, and looking at where it pools. It pools at the lowest possible point, because the water is able to tunnel through the hills and valleys, and get to the lowest possible point. Annealing quantum computers use quantum mechanics to do exactly that. They use superposition, they use entanglement, and they use tunneling. They work with nature to find the low point, or solve the problem. Right? And as a result, they're nowhere near as sensitive to noise and errors, and that's why our annealing quantum computers are solving hard customer problems today without the need for error correction. But, both are important. There are problems that require annealing. Specifically, business optimization problems are well suited to annealing. Gate model systems are not very good at solving them. And there are problems that require gate model. Quantum chemistry, for example, materials discovery. These are well suited to gate model systems, not very good at solving optimization problems. Right? You need both annealing and gate in order to be able to address the full market for quantum.
Okay. So, now how do we think about finding the signal within all the noise that's going on out there? Well, we think that to be successful in the quantum computing industry, you need three things. One, you need to be able to address the full market for quantum. All the possible use cases. Two, you need to do that with viable commercial products, not research tools, but viable commercial products. And three, you need to be able to deliver measurable customer value in using those systems. Address the full market with commercially viable products delivering measurable customer value. And we think that D-Wave is right now pretty much the only company in the world, the only quantum computing company, that can address and deliver on all three of those, and I'm going to explain why. Let's start with the first, address the full market.
Okay. So, there are many different TAM analyses that have been done by different consulting firms. TAM, total addressable market analyses. You know, Boston Consulting Group has done it, McKinsey has done it, a variety of other consulting firms have done it, but the one that most of the industry has latched on to for 5 years now is the Boston Consulting Group study. They put it out initially about 5 years ago, they refreshed it about a year ago. BCG puts the total addressable market for quantum at just under a trillion dollars in roughly the 20-year time frame. And they divide it into four different technological segments or use case areas. Cryptography, optimization, machine learning or AI, and simulation. And they assign a value to each of those adding up to the 850 billion total value. Well, if you think back to what I said a few minutes ago with respect to annealing and gate model, gate model cannot address all of these. Annealing cannot address all of these. Gate model, for example, is required for quantum mechanical simulation, drug discovery, materials design. It's also quite good at breaking codes. Right? Annealing quantum computers are required for optimization problems, workforce scheduling, manufacturing plant floor optimization. You need both to address the full set of use cases for quantum computing, and D-Wave is the only company in the world bringing both annealing and gate model to market. We're the only company in the world that can address the full TAM for quantum, the full set of use cases for quantum.
In addition, sometimes the gate model companies will have optimization envy. They vacillate between, oh, we can do optimization as well. Look, we've solved this optimization problem as well as it can be solved on a laptop. See, we can do optimization as well. Okay, as well as it can be solved on a laptop? What about solving it better than it can be solved classically? There's significant both theoretical and experimental evidence that gate model systems just can't beat classical on optimization, but annealing systems can. Then sometimes we'll hear, oh, annealing's just a niche. It only addresses a niche market. Well, is optimization a niche market? First of all, it's a full 25% of the TAM that we just discussed. But think about it. The numbers, 100 billion to 220 billion. Bigger than the global semiconductor market. Bigger than the global cybersecurity market. Are those niche markets? I don't think so. The point being that there's only one quantum computing company that can address the full TAM for quantum. And each of our products, our annealing quantum computers and our gate model quantum computers, are addressing a very significant portion of the TAM.
Okay. Let's move on to commercially viable products. I want to start by defining commercially viable. You will hear every quantum computing company say, we're commercial today. We've sold a quantum computer to this company or this company is using our quantum computer to work on this collection of problems. Right? Research experimentation with a quantum computer is not commercial use of the quantum computer. Commercial use of the quantum computer is using it as a part of your business operations on a daily basis. An integral part of your business operations. That's what it means to be commercial. So, again, lots of noise, lots of hype. Oh, we're commercial. Next time, your favorite quantum computing company other than D-Wave, because I know D-Wave is all of your favorite. But next time your other favorite quantum computing company says, oh, we're commercial. Say, oh, who's the customer? What's the application? How often are they running it? Can I go talk to them? Ask us. We'd be more than happy for you to talk to our customers. Every time we have an event, we actually have customers participating in the event with us. So, you got to cut through the noise. Commercial means in use on a daily basis as part of business operations. Now, the customers that we bring on stage, some of them are commercial, some of them are still in the R&D phase, and we're clear to talk about that as we engage in those discussions.
Okay. So, let's start with our annealing products. Our flagship annealing quantum computer is our Advantage2 system. It is a 4,500 qubit annealing quantum computer. We've also built a complete ecosystem of products around that. We have a quantum cloud service called Leap. This is designed to support business applications in production, building in the reliability, availability, security, privacy needed to be able to support commercial applications. We also have a complete suite of software development tools, our Ocean SDK, for building applications leveraging our annealing systems. We have a suite of hybrid solvers that bring together classical and quantum to solve problems on the quantum computer that are larger than can natively fit onto the quantum computer. And we have a professional services capability for helping our customers understand which applications can most benefit from quantum and how to build out those applications. This is a very comprehensive and very mature platform of products in the annealing arena. We've got one of the largest aerospace companies working with our products, one of the largest airlines working with our products, one of the largest chemical companies working with our products, one of the largest mobile carriers working with our products. We're working with government contractors on national security and defense. This technology space is quite mature today. Still more work that needs to be done, but nonetheless quite mature today. And we've talked about some of the metrics associated with improvements that these customers have seen leveraging our quantum computers. We'll talk a little bit more about that later.
We've also pointed out that our annealing quantum computer, our Advantage quantum computer, is the only quantum computer in the world that has been able to solve a useful real-world problem that cannot be solved classically. True quantum supremacy. We published this in Science a little over a year ago. This is computing properties of materials that were computed on our system in 20 minutes. That's the image on the right. And it would take nearly a million years, billions of minutes, to solve it on the fastest supercomputers in the world. Now, let me just diverge for a minute and talk about a little bit more hype, right? There are many people that would love to say we have not achieved supremacy on the system. And they tried. I mean, you may have seen a headline a week or two ago that a group of researchers just published a paper in Science where they basically have shown that classical can solve the problem that we solved. Okay, go read the paper. Right? The headline and the paper are very different things. They have not solved all the lattices that we solved, all the sizes of the lattices that we solved, all the evolution times that we solved, and all the properties that we computed, all of which is required to achieve supremacy. They've done a small subset of what we have done. The supremacy result has been out there for almost 2 years now. It was published in Science a year ago. It was put on the technical archive a year before that, and it still stands. This is the one true demonstration of quantum supremacy.
As far as our annealing roadmap is concerned, we've talked about the fact that our Advantage3 system will have 100,000 qubits. We haven't really put dates around that previously, but now we are providing dates for the 100,000 qubit Advantage3 system and a smaller 20,000 qubit version. There are two important technologies that are enabling our move from 4,500 qubits, or 4,400 qubits, sorry for rounding up, to 100,000 qubits. They are multi-chip, so basically moving from more qubits on a chip to interconnecting processor chips, multi-chip and scalable control. On the multi-chip front, basically what we're talking about, and it's shown here, tiling in two dimensions our 4,500 qubit processor chips. So, four to five chips to get to 20,000 qubits, 25 or so chips to get to 100,000 qubits. We need to be able to interconnect these things electrically through bonds that are superconducting and preserve quantum mechanical properties. This is work that we've been doing with the NASA Jet Propulsion Lab. We've developed the materials and the process for this, and we've actually been able to demonstrate the ability to bond chips and preserve these properties. So, that was one of the two technical challenges that we needed to overcome to drive to the Advantage3 100,000 qubit system.
The second, scalable control. And this is a really important point because it's going to come back when we talk about the gate model systems as well. So, typically, if you go ask anybody in the industry, how many control lines do you need to control a qubit? You'll hear three to five per qubit. So, that would mean that our 4,500 qubit Advantage2 system would require 12 to 20,000 IO lines. We control that 4,500 qubit processor with 2 to 300 IO lines. Because we've been able to build data pipelining, multiplexing, addressing onto the processor chip, which means we can much more efficiently control the qubits. And that's critically important. Now, when we go from 4,500 to 100,000 qubits, though, we can't have a 25-fold increase in the number of IO lines. So, we've been working on modifying our control architecture to make it even more scalable than it has been up until now. We've completed the design. We've completed the masks for the first chips to basically test out the design, and they're going through fabrication now. And that's the second important component of being able to scale to the 100,000 qubit processor. Both of these technologies are well down the development path, but obviously, we have some time and work to test and really kind of wring out any issues associated with either of these technologies. But, we're quite enthusiastic about the path to 100,000 qubits.
And why is this important? So, if we think about our current 4,500 qubit system, when combined with our hybrid solvers, we can support problems with up to 2 million variables. And that's fine for solving a lot of problems. But, if for example, we wanted to solve full up last mile routing for FedEx or UPS, that would require 50 million variables. So, we still have to continue scaling on the annealing side to solve larger and more complex problems.
All right. Now, I want to spend a little bit of time talking about the gate systems. We announced quite a few years ago that we were going to start working on a gate model quantum computer as well. A lot of the work that we had been doing was looking at how you could take that control architecture that I talked about for annealing systems and move it into the gate model environment. We had not spent as much time on high-quality qubits and the ability to error correct those qubits. But, what you need in order to have a commercially viable gate model quantum computer is you need error correction and efficient error correction. You need scale and you need for the systems to be fast. Right? Nobody wants a slow quantum computer. And so, we've been focused on how to advance the state of the art and our technology development in each of those areas. Efficient error correction, scalable systems, fast computation.
So, we're bringing together to solve this collection of problems two critically important technologies. Relative to the ability to efficiently error correct and have fast computation, we closed the acquisition of Quantum Circuits in January. This was a company that was spun out of Yale University and founded to basically commercialize some very important new qubit technology advances that had been developed at Yale. And in a few minutes you're going to hear from Rob Schoelkopf, who was leading all of that work at Yale and was the founder of Quantum Circuits.
The qubit technology that was developed at Yale, spun out into Quantum Circuits, acquired by D-Wave, is very revolutionary. First of all, it is superconducting qubit technology, which means it is fast. Superconducting qubits run a thousand to 10,000 times faster than trapped ion or neutral atom qubits. These are superconducting qubits. They're very fast, but typically, the trapped ion or neutral atom folks won't argue that they're slower, but they'll say, but our qubits are more natural qubits. They're more efficient qubits. They're much easier to error correct. Superconducting may never be able to error correct. Well, here's the really important thing about the new qubit technology called dual rail qubits. They have the same efficiency as trapped ions or neutral atoms. So, think about it. These are revolutionary qubits that have the speed of superconducting and the efficiency or reliability when it comes to error correction of trapped ions or neutral atoms. The best of both worlds.
Then, we are going to marry that with the control technology that we had developed for our annealing quantum computers to allow us to much more efficiently scale these gate model systems. Essentially, attacking all three of the elements required to truly have a commercially viable gate model system. Efficient error correction, scalability, and speed.
Think about it this way. Trapped ions or neutral atoms are like a bicycle. Simple and efficient. Easy to error correct, but slow. Superconducting, the things that IBM or Google are using, very, very fast, but very complex. Not all that reliable. Kind of like an old piston airplane. D-Wave's dual rail qubits are the best of both worlds. They are fast, but they are also simpler and far more reliable like turbojet engines. The best of both worlds.
Okay, let me just provide some metrics now to help back this up. We think that there are two metrics that it's really important for everybody to start focusing on when it comes to getting to truly commercially viable gate model quantum computers. One is speed and the other we're calling lambda. This is the error correction efficiency and I'm going to spend a minute talking through this. First of all, speed. The chart on the left shows you the time required to perform an error correction cycle on superconducting, trapped ion, and neutral atom. Look at that. 5 microseconds for superconducting. 30,000 microseconds for ions, 200,000 for neutral atoms. They are very, very slow technologies.
Okay, so superconducting fast and I think everybody accepts that. Right? The debate really has more to do with, yeah, but can you actually error correct these superconducting qubits? So, that's where we think lambda becomes very important. Lambda is the rate at which you can reduce errors as the error correction code increases. So, if we look at the chart on the right, okay? At the top, I have surface distances. Surface code is the typical code used to error correct gate model quantum computers. And there are four surface code distances shown there. Distance code three, five, seven, nine. And typically the number of physical qubits you need is the square of the distance of the surface code. Okay? Surface because it's got a length and a width. Okay? So, at distance three, nine, we're saying 17 physical per logical qubit. Distance code distance five, 25, 49. There's a factor of two, which I'm kind of glossing over. 7, 49 times 2, 97. 9, 81 times 2, 181. Okay? So, distance three, distance five, distance seven, distance nine are increasingly larger surface codes for doing error correction. And the larger the code, the better the error correction. At the bottom is the number of physical qubits that you need per error corrected qubit, logical error corrected qubit, for each of those distance surface codes. Now, look at the two lines. Currently in the industry, we're looking at a lambda of two. What this means is that as you increase the surface code distance, you halve the errors. So, you get one half the errors for each increment in the surface code. That's the gray line.
line. Okay. What we're looking at with our dual rail qubits is a lambda of 10. With each increment in the surface code, we get a factor of 10 reduction in errors. This becomes really important if you think about what you need to actually be able to use a gate model quantum computer. You need to be able to perform computations. And you need the errors to be low enough that you can perform all those computations before an error emerges. We think that we're going to need about a million computations to start doing useful work. This means we need an error rate of 10 to the minus six, one error in a million, in order to be commercially viable.
So now look at those two lines. At a lambda of 10, you can get to 10 to the minus six with a distance nine surface code. Roughly 100 to 200 physical to logical qubits. That's the path we're on with our gate model program. That's very viable. All right? The gray line, where the industry is right now, what would it take to get to 10 to the minus six? We're talking a thousand or more physical qubits per logical qubit by the time that gray line gets out to 10 to the minus six. Okay? That's the power in the dual rail super qubits. They are superconducting, fast, with a very rapid decline in error rate as we increase the error correction.
So, for our product roadmap for gate model, we've divided it into two eras, what we call the post-Nisk era, where we start seeing some error detection capability that we can use in the computation, and then the fault-tolerant error era, which is where we're actually at the point where we can really perform up to a million computations and do useful work. So, this is what you can expect from D-Wave over the course of the next roughly 6 years with the goal being that in 2032, we would have 100 logical qubits built on 10 to 20,000 physical qubits with an ability to support up to a million gates or computations. Actually, the beginning of commercial viability.
So, this is what you need to be asking others about. Well, what's the speed of your computation? How rapidly are you going to be able to reduce errors as you increase the error correcting code? And when will you be able to support at least a million computations? Those are the key elements. Okay?
The only other thing I'm going to say about the products is we were really, really pleased to be a part of the US government announcement a week or so ago, basically talking about the fact that the US government is providing D-Wave along with a number of other companies funding to basically support R&D work around the development of quantum computers. This was really important for D-Wave for two reasons. First of all, this was the first time the US government had actually endorsed D-Wave. And secondly, they not only endorsed D-Wave, but they endorsed both annealing and gate. So, now the US government has finally started to realize annealing is important alongside gate. They are both important, and this was an endorsement of both of those approaches to quantum computing. Okay.
Now, what I want to do is transition and start talking about customers and customer value. And I want to start by in a minute asking Lucas Haugen to join me up here on stage. Lucas is from AT&T. To be fair, AT&T is not in production today, but they are working with our quantum computers on a variety of applications. Lucas is an innovative director of data science at AT&T's Chief Data Office, where he leads a world-class team of data scientists, ML engineers, developers, and solution architects to deliver enterprise scale impact. He's recognized as a leader in digital transformation leveraging agentic AI, advanced machine learning, automation, and big data engineering to drive operational excellence and create business value approaching 1 billion annually for AT&T. Lucas and his team began exploring D-Wave's impact on their operation at AT&T less than a year ago, and he's here to talk a little bit about that experience. Lucas?
And in just a minute when Lorenzo comes up, he'll talk a little bit about how we work with customers to help kind of evaluate problems, understand which can benefit from quantum and how to build out proof of technology, a proof of concept to really validate them. But thank you for being here, Lucas. Appreciate it.
Um so, you know, I'm pretty regularly now asked about the relationship between quantum and AI. Um I guess because everybody's kind of got AI on the brain these days. Um so, let me just spend a minute talking about my view on this. I think that AI and quantum are very synergistic. And they work well with one another in two ways. One, they're each good at different sorts of things. And you can use them together to solve a problem letting each work on the part of the problem that it's best at addressing. For example, you might use AI to predict product demand for the future and then use quantum to optimize the supply chain to meet that demand. So here you have the two technologies working side by side to solve a problem, each addressing the portion of the problem that it's best at solving.
The second, quantum can help make AI model training and inference better, faster, and more power efficient. Um we have been working for a little over a year now with a Japanese pharmaceutical company called Shionogi. Uh they essentially trained a model to generate molecular structures for new human drugs. And they trained the model classically and they were getting good results. They retrained the model using our quantum computer. And what they found was that 10x, 10 times more of the structures generated were well-suited to human drugs. In other words, the model was more accurate. It was also trained faster and consumed less power. So here we have a situation where we're actually using quantum to make AI even more efficient and more powerful. And I think over time AI and quantum are going to come together in both of those ways. The latter, potentially very transformative.
And then finally, I wanted to spend a minute on blockchain. Um there's some very interesting work going on right now in the quantum classical blockchain arena. Uh we've been doing some work with a small company called Post-Quantum Labs. They've developed a testnet for a new approach to proof of work. Uh the proof of work computation can be done either classically on CPUs and GPUs, or it can be done using a quantum computer. They currently have over 18,000 people mining in the testnet, and 1,600 compute nodes on the net. Those compute nodes are mostly CPUs and GPUs, and then our Advantage 2 quantum computer. Now, our quantum computer is not online all the time. We put it online for about 5 minutes a day. When it goes online, it wins almost all the blocks. In other words, it's much faster at doing the mining. It's also more power efficient.
And so, what we're trying to do here is demonstrate on the one hand, using a problem that can be solved by either quantum or classical, the value that the quantum computer brings over classical. Another example, perhaps of quantum advantage, right? But, in addition to that, showing that if you use the quantum computer, we can actually do the mining more efficiently and with lower power consumption. And by the way, it's quantum safe. In other words, a quantum computer can't break the system. So, this is an interesting new application area for us. We're quite excited about the early results we're seeing. We have started a more substantial benchmarking phase to really understand the difference between the classical and the quantum on this testnet. And, you know, at some point in the not too distant future, we will publish those results. Okay.
So, before we kind of do the first Q&A session, let me just wrap up by going back to Feynman and just reiterating quantum mechanics is hard. And as a result, it's easy to spew nonsense and convince people of things that just aren't real. And you always have to challenge. Yeah, like I said, customers, who's the customer? How are they using it? Can I go talk to them? Scientific result. Oh, we've been able to solve this optimization problem better than classical. Well, okay, where's the paper? Let me go read the paper and see, are they really doing that? Or, you know, is it kind of like, well, we can solve it as well as a laptop can solve it, not better than classical. Always challenge.
So, for us, we are focused on addressing the full market for quantum. Every use case by bringing both annealing and gate model systems to market. We've proven our ability to deliver commercially viable products through our Advantage 2 annealing system and the technological platform that we have built around that, and we're bringing that expertise into the gate model world as we speak. And we actually have customers that are using us today as a part of their business operations. In Q1, we announced a Fortune 100 company that signed a two-year $10 million quantum computer as a service agreement for us. I can tell you they have moved their first application in production. So, it is now running on a daily basis in production, and we are working with them on several other applications. This was our first, if you like, enterprise site license. With that, I would be happy to take some questions. Well, okay. Let's go to the very back of the room.
