Hello, everyone, and good afternoon. Thank you for joining us for the virtual day of Needham's 19th Annual Technology, Media, and Consumer Conference. My name is Quinn Bolton, and I cover the semiconductor and quantum computing sectors for Needham. It's my pleasure to host this fireside chat with D-Wave Quantum. In August of 2022, D-Wave became the third publicly traded pure-play quantum computing company, and remains the only company that provides quantum annealing systems that are used today to solve real-world commercial problems. With 69% of revenue from commercial customers in the last 12 months, D-Wave is positioned as the market leader in commercial quantum computing systems. Joining me for the fireside chat today are Dr. Alan Baratz, CEO, and John Markovich, CFO. Alan, John, thank you for joining us.
Thanks, Quinn.
Thanks, Quinn.
It's a pleasure to be here. We appreciate the opportunity.
Awesome. Wanted just to start, as there are probably some in the audience who are still coming up the learning curve on quantum computing with some basic questions. Maybe to begin, Alan, can you provide us a quick overview and history of D-Wave for investors?
Sure, happy to do it. So, we are a U.S. publicly traded company, QBTS on the NYSE. As you said, Quinn, we went public, about two years ago. We are a full-stack quantum computing provider, so we provide everything from the quantum computers (we design and manufacture the systems) to the quantum cloud service . That's the vehicle that our customers use to access our quantum systems. We have our own quantum cloud service, designed not just for research experimentation, but to support real business applications in production, building in the privacy, the security, the availability required to support those applications. We have our own suite of software development tools for programming the system. Those tools are all open source, so that developers and our customers can help us to enhance and evolve the systems.
We have a professional services organization that basically helps our customers to understand which applications can most benefit from quantum, and how to build out those applications. We, as a company, are over 15 years old, so we're, in some sense, the oldest quantum computing company. That means that we have been working on building quantum computers for 15 years now, much longer than anybody else in the industry, and we are the first commercial quantum computing company. Our current generation system, called Advantage, is a 5,000-qubit quantum computer. It's in production today. Actually, we have 3 systems in production in our quantum cloud service , and it's being used by companies like Mastercard or VINCI Energies or Unisys, all working on real business applications to benefit their business operations.
Perfect. Now, there are two types of quantum computing systems in the marketplace today, gate model systems and quantum annealing systems. Can you just discuss the difference between gate model versus quantum annealing?
Yep.
Perhaps more importantly, why did D-Wave choose the quantum annealing approach?
Yeah. So, as I mentioned a few minutes ago, we actually started building a quantum computer about 15 years ago. As you pointed out, there are two primary approaches to quantum computing. One is called annealing, and the other is called gate model. 15 years ago, there really was no line of sight to being able to actually build a gate model quantum computer. But it was believed that you probably could build an annealing quantum computer, and so that's a big part of the reason why we decided to start with annealing. However, it also turns out that annealing is a much easier technology to work with. It's easier to scale. That's why we're at 5,000 qubits today, and the rest of the industry, focused on gate model system, is at hundreds of qubits.
It's much less sensitive to noise and errors. We're able to deliver good, if not optimal, solutions to hard problems today, without the need for error correction. It's believed that error correction will be required to solve any real-world problems on a gate model system. And finally, annealing is very good at solving business optimization problems. These are problems like employee scheduling or autonomous vehicle routing for improving manufacturing operations.
Frankly, most, if not all, of the important, hard business problems that businesses need to solve are optimization problems, and annealing quantum computers are very good at solving that class of problems. However, annealing quantum computers cannot solve all problems. For example, they can't solve differential equations problems. You cannot use an annealing quantum computer for quantum chemistry, for example, in drug discovery. For that, you need a gate model system.
Now, everybody else in the industry is building a gate model system, and you might ask, "Well, why did they all decide to go with gate model?" The answer is, when they all jumped in 7 or 8 years ago, the engineering and the science had progressed to the point where it was believed that you probably could build a gate model system. Moreover, at that point in time, it was believed that a gate model quantum computer could solve all problems. So their view was, "Well, if I'm gonna build a quantum computer, and I now think I have line of sight to being able to build a gate model system, and a gate model system can solve all problems, let me do that, rather than annealing, which, while an easier technology to work with, we know cannot solve all problems."...
So everybody else decided to jump in on the gate model side. However, a couple of years ago, something very surprising happened. It was proven by a group of researchers in the U.S. and Europe, and it was also shown experimentally, that gate model quantum computers cannot deliver a speed-up on optimization problems. So we actually have a split in the application environment for quantum.
We have problems that will always require annealing, business optimization problems, and we have problems that will always require gate, differential equations problems, and we have problems that either can address, depending on the specific nature of the application. But that has left D-Wave in a very, very interesting position because it means that there's a class of problems that only we can address, because we're the only company in the world that builds annealing quantum computers.
That, that's great. Another question just for folks: Can you describe the company's Leap quantum cloud service, and what does that provide customers?
Yep. So, as I mentioned at the beginning, we do have our own quantum cloud service, and that is the vehicle through which our customers access our quantum computers. We've designed all of the software for that cloud service. As I said, we designed it to support business applications in production. We don't have our own data center for running the cloud service. We basically buy compute from Amazon AWS, and so the front end of our cloud service runs on AWS servers, and then we are fiber-connected from there to the locations where our quantum computers are housed. We have three of them today in the cloud service, one in Vancouver, Canada, one in Southern California, and one at the Jülich Supercomputing Centre in Germany.
However, it is the Leap cloud service that our customers use to access our quantum computers, as well as a collection of what are called hybrid solvers. Hybrid solvers use classical compute together with quantum compute to compute the solutions to problems larger today than what a quantum computer natively can support. The classical side of the hybrid solvers typically uses CPUs and GPUs, and so we spin those up as needed in AWS, and then they work together with our quantum computer. So when a customer accesses our Leap cloud service, they're basically getting access to all of our hybrid solvers, as well as native access to our quantum computers.
So it sounds like it's a combination of doing some development work through the hybrid solvers, but also when they're going to run, you know, jobs actually on the quantum computers, they access it through Leap.
Yeah, there's actually a programming environment inside the Leap cloud service that hosts all of the software development tools that are available for programming our system. So you can use the Leap cloud service to program your application, and then you use the Leap cloud service to run your application, either in a test mode or in production mode.
Perfect. Maybe spend a minute talking about the company's go-to-market strategy and your focus on, you know, QCaaS, rather than, say, you know, selling, you know, full, complete systems.
Yeah. So, you know, we are currently focused on commercial business optimization problems, and, you know, we have a large number of commercial customers. I think when we last reported, Q1 of this year, we said 75 commercial customers. And, those commercial customers, are working on or have deployed a broad array of different use cases. However, as we've looked across all of the different use cases that we or our customers or academics have worked on, and there are well over 100 of them, what we've, been able to do is to determine which of those use cases will excel on our current-generation quantum computers and which of them will require future-generation quantum computers, and we've actually built out a timeline for all of that.
The use cases that we can address really well today revolve around resource allocation and resource scheduling, primarily in supply chain logistics and manufacturing. And so our go-to-market approach really revolves around that vertical strategy, focused today on supply chain logistics and manufacturing, with this set of use cases that our current quantum computers and hybrid solvers are very good at addressing. And then, as our technology evolves over the years, we layer in additional use cases. Next up is generative AI and machine learning model training.
And in fact, we've just made available an early version of our next-generation annealing quantum computer that was the basis for being able to support machine learning model training, and we've already started down that path and are seeing very, very good early results, both with respect to more accurate model training, as well as more energy-efficient model training, because our quantum computers consume far less energy than GPUs consume. So we're quite excited about all of that. Now, with respect to how we engage our customers, we have two models: professional services, and as you mentioned, QCaaS, or quantum compute as a service. Typically, we will sell a customer a professional services engagement to help them evaluate their applications with respect to the extent to which they'll benefit from quantum, and then build out a proof of concept.
From there, it would move into trial deployment, and ultimately production deployment. The initial phases of this are professional services engagements. So they're typically, you know, 4, 5, 6-month engagements to help customers do an application evaluation and build out a proof of concept. Once the application moves into production, then the customer moves to a quantum compute-as-a-service engagement agreement, where they're basically running the application on an ongoing basis, and it's making calls to our Leap Cloud server to get access to the quantum computers in order to support that application, and that's a monthly recurring revenue model. The professional services engagements, as I said, are typically 3, 4, 5, 6 months long. Quantum compute-as-a-service agreements can be a year, 2, or 3 years long as the customer is actually deploying an application in production.
Perfect. You know, kind of looking at the company's revenue stream today, you'd mentioned is sort of, you know, a combination of the professional services and the QCaaS, the company historically has not focused on complete system sales. What do you think has to happen that might have the company reconsider potentially selling complete hardware systems?
Yeah, I'm gonna say something that you may find controversial, or you may not even agree with, but we actually did sell systems early in the life of the company. We sold three systems. We sold them primarily to government entities, U.S. government entities. And that's the model that all the other quantum computing companies are pursuing today, sell systems to government entities. But here's the problem: There are only so many “customers,” mostly governments, that will buy a system that does not have a proven ROI. And once they take delivery, and they realize that there's not a whole lot they can do with those systems other than play around with them, they're not gonna buy another system.
So we sold three systems early on, but we couldn't sell another system, and so we decided we needed a different model, a model that would allow more customers to get access earlier at a lower entry price point, and that's why we transitioned to the cloud service model. And that's worked really, really well for us and is allowing us to grow the business with a much broader base of customers.
I mean, you know, our competitors that are public, you know, they're highly consolidated, mostly government customers, mostly system sales. Now, as your systems start to deliver a solid ROI, and you're demonstrating real business benefit and real business value, then that may open up opportunity to sell systems. So we're at the point now where we're starting to see those benefits. Occasionally, we'll have a customer come back and inquire about purchasing a system.
We'll evaluate the opportunity. If it looks real, if it looks like, you know, it makes sense to sell the system, we're not opposed to doing it. But, you know, starting with system sales before you've actually proven the value of your system, is gonna lead you to running into a wall. And we-
Maybe maybe putting the cart before the horse?
Uh, exactly.
Okay.
So, you know, I would encourage everybody to start with a cloud model, build a base of applications, demonstrate the value and the ROI, and then if that leads to system sales, so be it, right? Now, by the way, we do know how to do system sales. We've sold systems. We've placed systems at customer sites. We know how to maintain them. So we know how to do this. It's just we want the market to pull us there, rather than us pushing on the market prematurely.
You mentioned earlier that annealing systems are particularly good at solving optimization, job scheduling, you know, type tasks. You may spend a minute discussing why is that the case? You know, what allows annealing to be so good at performing optimization, and are there other applications where annealing is particularly well-positioned? You'd mentioned that there's some applications that can be solved by both gate model and annealing, so maybe touch on some of those applications as well.
Yeah. So first of all, annealing quantum computers are native optimization engines. Essentially, the way an annealing quantum computer works is that they run one algorithm. It's called the annealing algorithm, not surprisingly. And what the annealing algorithm does is it finds the lowest point in a multidimensional landscape. That's really all it does. So our quantum computer does one thing, and it does it really well. It finds the lowest point in a multidimensional landscape. However, it turns out that the classical analog of that is an exponentially hard problem, what's known as an NP-hard problem, so it's one of the hardest optimization problems that we know of. That means that pretty much any other optimization problem can be relatively easily transformed into that optimization problem, which the annealing algorithm solves natively, okay?
So there are two points here: One, programming our quantum computer doesn't require any knowledge of quantum physics. It just requires an understanding of how to do a mathematical transformation of your problem into the NP-hard problem that our quantum computer can solve, okay? Secondly, because the quantum computer natively solves that optimization problem, it's very, very good at ultimately solving your problem. So that's why annealing quantum computers are so good at optimization. And by the way, you know, there are many, many, many problems that fall into this category of optimization, and that can be solved in this way. However, there are other problems that we can solve.
In fact, one of them that I will mention is extremely important because we believe we've just, for the first time, demonstrated quantum supremacy on this problem, and this would be the first time that anybody has shown quantum supremacy on an important real-world problem. The problem is a materials, a magnetic materials simulation problem, so basically simulating properties of a magnetic material. We did this work with researchers around the world. It took us about a year to complete the work. We published the paper on the preprint server, the archive, about 2 months ago, and it's now going through peer review at a respected, highly respected, scientific journal.
However, what we've been able to show is that we can solve this problem on our quantum computer in about 20 minutes, and working with researchers around the world that are experts in all the other classical techniques for solving these problems, whether it's tensor networks, PEPS, DMRG, machine learning, quantum Monte Carlo, and using the fastest supercomputers in the world, we were using Frontier at Oak Ridge National Lab, which is arguably the fastest supercomputer in the world, that the best you can do in solving this problem would be well more than 1 million years. So we're solving it in 20 minutes, way more than 1 million years to solve it with the best algorithms on the fastest supercomputer classically.
Moreover, and this is a really important point, that same problem that we solved in 20 minutes would require well more than the global annual energy consumption, okay? So not only are we very fast, we're also very energy efficient. But this gives you, an example of an important real-world problem that's not optimization, that our quantum computers are also capable of solving.
Does that apply, that simulation example, does that apply to a broader set of simulation examples? You'd mentioned I think it's magnetic materials. You know, can the concepts be more broadly applied to other types of simulation? I know you said you can't do differential equations, so you might not be doing drug development, but are there other types of simulation or subsets of simulations, where you would have that same advantage?
Yeah. So in fact, there are some very important problems that we believe we will be able to tackle as a result of this work. I mentioned a few minutes ago that we're now starting to make progress on machine learning model training. It turns out that there's an interesting relationship between what we've done with this quantum supremacy problem and how you could train models using a quantum distribution. So that's one of the things that's falling out from this. We are also doing some interesting work on hashing functions that we think could be quite compelling as well. So yes, there are other interesting problems that this work will also allow us to tackle.
Great. We wanted to ask maybe for you to share, you know, some examples of the fields that customers are using D-Wave systems today in solving real-world problems, and maybe a couple of examples from some of the customers, the types of applications they're using D-Wave to help, you know, compute.
Yep. So, for example, Interpublic Group and their subsidiary, Momentum, which does promotional tour planning for large companies, they've developed an application that leverages our quantum computer, which can compute optimal tours in about an hour, when previously it was taking them about a day to perform the same computation. So, you know, at the core, you can think of this as the traveling salesperson problem, but you know, visit cities in the optimal order, and you know, visit each city only once and end back up at the same location you started at. But there are many, many more constraints associated with promotional tour planning that need to be addressed as well. Another is work that we're doing with a large European construction company called VINCI Energies.
Basically, we've worked with them on how to design more cost-effective, energy-efficient, and aesthetically pleasing HVAC systems for large buildings. And they're seeing really interesting results across many different dimensions: the cost of the HVAC installation, the number of jogs that you have in the route for the HVAC system, the time to compute, much shorter, so benefits across a broad array of metrics. We're also working with Mastercard. We're working with Mastercard on a few different applications. One of them we're working with them on is customer loyalty rewards program optimization. And so the idea is, you know, you've got merchants that are developing rewards programs, and you want to figure out the best set of customers to offer those promotional programs to, to maximize the uptake.
Leveraging quantum, we can kind of come up with allocations that ultimately increase revenue. We're also doing some work with a U.S. government contractor, Davidson Technologies. We've worked with them on a couple of different applications. One of them is in the area of missile defense. And so, you know, you've got kind of incoming threats, and you've got countermeasures, and you need to determine which countermeasures to target to each of the threats to maximize the probability of taking out all the threats. Turns out to be a very complex optimization problem, and the list goes on and on. It's quite exciting, actually.
Sure. Yeah, no, that, that's, that's great. Thank you. Moving to the roadmap, maybe spend a minute first talking about your current Advantage system and the prototype Advantage2 system. What are the sort of feature differences between Advantage and Advantage2 ? That would be, that'd be great.
Yep. So Advantage is our current flagship product. It is a 5,000-qubit quantum computer. Each qubit is connected to 15 others, and this is the system that is allowing us to address all the problems that I just talked to you about. However, the quantum supremacy work and the material simulation that I mentioned a bit earlier, that was done on our next-generation Advantage2 system.
This system is not complete yet. It is not fully in production yet, but we have early versions of it that we have made available in our quantum cloud service. So we currently have a 1,000-qubit version of the Advantage2 system in our cloud service that people can start experimenting with. Ultimately, this will have 7,000 qubits, so current Advantage is 5,000. Advantage2 will have 7,000 qubits. Current Advantage has degree 15.
Each qubit is connected to 15 others. Advantage2 will have degree 20. Each qubit is connected to 20 others. And Advantage2 has significantly increased coherence time, increased coherence time by a factor of 2, and with error mitigation, by a factor of 20, so significantly increased coherence time. What do I mean by increased coherence time? What I mean by that is that the length of time within which we can perform calculations while the qubits are still coherent, not impacted by the external environment, has over doubled. And the importance of this is that we know that, if we run the annealing algorithm while the qubits are coherent, we get to the optimal solution much, much faster, okay? And so that's the importance of increasing qubit coherence time.
But we've got an early 1,000-qubit version of Advantage2 in our cloud service today. We have finished fabricating 4,500-qubit versions of Advantage2 . They're being calibrated today. We'll make one available in our cloud service as soon as it's been calibrated on the path to the 7,000-qubit system, which we expect to have, you know, within a, you know, relatively short timeframe going forward.
So we're really excited about Advantage2 . It was the increased connectivity and the increased coherence time that allowed us to perform the quantum supremacy result. So it was that smaller 1,000-qubit version of Advantage2 that we did the supremacy result on, which means arguably, even that small 1,000-qubit version of Advantage2 is more powerful than our 5,000-qubit Advantage system. But it can't solve problems as large as a 5,000-qubit Advantage system 'cause we need the additional qubits.
Got it. And you guys had, I think, just recently announced the Fast anneal capability. Is that really only available on Advantage2 ?
Mm-hmm
... or does that also apply to Advantage, today?
... Yeah. So fast anneal is available on all of our systems, Advantage and the Advantage2 prototype. So what is fast anneal? As I said a few minutes ago, if we run the anneal within the coherence time of the qubits, then we are actually able to get better solutions faster, okay? So, until recently, within our cloud service, the anneal times were fairly long because we needed improved electronics and control and filters in the system to be able to run shorter anneal times. We've had it in our lab for about two years now, and we've been doing a lot of research on it, and we did the quantum supremacy work on it, but we have now made that available to our customers through our Leap cloud service.
Now, there's one other thing that's pretty exciting, actually, about the combination of the increased coherence time and the fast anneal, and this is just research underway, so not a product, not, you know, you know, it's just research that we're doing. There's some theoretical evidence that if you can anneal within a certain amount of time and then do what's called reverse anneal within that same amount of time, and then anneal again, and reverse anneal, and anneal, and reverse anneal, if you can run that protocol, we sometimes internally call it bang-bang .
So anneal, reverse, anneal, reverse, and you can be coherent within each phase, coherent within the anneal, coherent within the reverse anneal, coherent within the anneal, coherent within the reverse anneal, then there's some theoretical evidence that you can have the same effect as if you were coherent across all the iterations of anneal, reverse anneal. So essentially, with relatively short coherence times and very fast anneal times, so we can anneal coherently in each of those segments, that would be just as good as if we had coherence times across the entire set of anneal, reverse anneal, which means we can potentially get infinite- the effect of infinite coherence with relatively short coherence. Now, this is research underway. It's very exciting. We're having some fun with it. We'll see, right?
No, that's what I was gonna say, the extended or infinite, you know, anneal times, or not anneal times, coherence time, certainly sounds like that would be a huge step forward for you know, the company, but also for customers. I know we focused a lot of the talk on your annealing systems, but I don't wanna sort of sell you short because I know you're also, you know, have development, or in development, doing some work on a gate model computer. And so maybe spend a minute or two just giving folks a sense of your gate model program, where you are, and what you hope to achieve with your gate model system.
Yep. So, we did announce a few years ago that we were going to also build a gate model quantum computer. Why? Two reasons. First of all, many of our customers have use cases that require both annealing and gate. I've already talked about the fact that there are use cases that require annealing, use cases that require gate. There are customers that require both.
I mean, think about, for example, a drug company that wants to design a new drug, put it through trials, put it through manufacturing. Some of the elements require gate, some of the elements are optimization, they require annealing. And so we wanted to be able to support all of the use cases, for our customers. Secondly, a lot of the technology that we develop for our annealing quantum computer is directly applicable to a scaled, error-corrected gate model system.
So for example, multilayer qubit fabrication. Everybody else in the industry puts their qubits on a single layer, which means they have very few qubits on a chip. Our 5,000-qubit processor is a single-chip processor. Why can we get 5,000 qubits on a chip? Because we have a multilayer fabrication process, a three-dimensional fabrication process. Moreover, we put control on the same chip as our qubits. Nobody else does that, okay? So that tight integration, that density gives us, A, the ability to build much larger processors because we can put more qubits on a chip, and B, the ability to actually do efficient error correction on a gate model system 'cause we've got control on the same chip as the qubits. So, you know, it's a very, very important technology for building gate-scaled, error-corrected gate model systems.
There are other technologies we have as well that are directly applicable, so we decided, you know, let's be the only company in the world that has both annealing and gate, can support all of our customers' use cases, and properly leverage the technology that we develop for annealing into the gate model space.
Perfect. John, I wanna get you involved here. You've been waiting patiently. Thank you. Maybe just spend a minute discussing sort of the latest set of customer metrics. Again, you know, you guys differentiate yourself from the gate model folks by, you know, having a significant percentage of revenue coming from commercial customers, and so maybe, you know, kind of review for us the stats from the first quarter about the number of customers, number of commercial customers, Fortune 2000, percent of revenue, et cetera, et cetera.
... Sure. So, every quarter, Quinn, we report a set of customer metrics on a trailing four-quarter basis, and then compare it with the immediately four quarters prior to that. So through Q1, we had a total of 128 customers over the last four months, which is an increase over 113 customers over the prior 12 months or four quarters. That included 75 commercial customers, which increased from 69 customers over the prior period. And of the 75 commercial customers, fully a third of them, or about 25% of them, were Forbes Global 2000 customers. We have other metrics that measure commercial traction.
For the same comparison, trailing four quarters over four quarters, the revenue from our commercial customers increased by 51% on a year-over-year basis, and then commercial revenue as a percentage of total revenue increased from 63% to slightly under 70%. And then the revenue from the Forbes Global 2000 customers increased by 50% over that timeframe, and constituted 27% of our total revenue.
Perfect, thank you. Can you discuss the bookings trends? I think you're now eight quarters in a row of year-over-year bookings, and I believe bookings hit a record in the March quarter. But maybe spend a minute just discussing booking trends and, you know, I don't know if you wanna comment on, you know, sort of how you see bookings continuing through 2024, but I... I'll ask.
Sure. You're right, our first quarter represented our eighth consecutive quarter of bookings. No, we haven't given any guidance on future bookings.
Okay. Lastly, maybe just talk about your 2024 targets for adjusted EBITDA, and then maybe touch on just, you know, what's the status of the balance sheet and your outlook for funding the business going forward?
Sure. So the guidance that we've provided for adjusted EBITDA on the year is less than $54.3 million. And in our Q1 earnings that we just reported earlier this week, we announced as of May 10th, we had $33 million of cash on the balance sheet, which is up substantially on a year-over-year basis. We also announced that we had two separate S-3s go effective on April 12th. One was for our current equity line of credit program. We have had a $150 million commitment with Lincoln Park in place since we went public. $82 million of that was still available as of the end of the first quarter. And then we also filed a $175 million shelf registration statement that also went effective on April 12th.
Perfect. We've got about 30 seconds left, so Alan, I'm gonna try to squeeze one more in for you. Why should investors be considering investments today in quantum computing rather than waiting on the sidelines, seeing how things develop, and you know, seeing the companies getting close to breakeven or cashflow positive?
Well, they, they should invest in D-Wave today, because we are commercial today, and actually growing the business. We're the only quantum computing company that is doing that. Relative to other companies that are kind of still in the R&D mode, I don't know, maybe they should wait on the sidelines.
Well, perfect. I think that brings us to the end of the time for our session. Alan, John, thank you very much for joining us at the Needham Conference, really appreciate your attendance, and thank you to all the participants watching the webcast. Thank you, everybody.
Thank you, Quinn.