Orchestrate all the Things - The past, present and future of quantum computing. Featuring D-Wave CEO Alan Baratz

Episode Date: November 21, 2022

Quantum computing could be a disruptive technology. It's founded on exotic-sounding physics and it bears the promise of solving certain classes of problems with unprecedented speed and efficiency. The problem, however, is that to this day there's been too much promise and not enough delivery. D-Wave is the company that pioneered quantum computing. In this exclusive interview, D-Wave CEO Alan Baratz talks about about quantum computing fundamentals and how this is related to the market’s current state, real-world clients and use cases, and what the future holds for this space. Article published on VentureBeat

Welcome to the Orchestrate All the Things podcast. I'm George Amadiotis and we'll be connecting the dots together. I hope you will enjoy the podcast. If you like my work, you can follow Link Data Orchestration on Twitter, LinkedIn and Facebook. Quantum computing could be a disruptive technology. It's founded on exotic-sounding physics and it bears the promise of solving certain classes of problems with unprecedented speed and efficiency. The problem, however, is that to this day there's been too much promise and not enough delivery.

D-Wave is the company that pioneered quantum computing. In this exclusive interview, D-Wave CEO Alan Barrage talks about quantum computing fundamentals and how this is related to the market's current state, real-world clients and use cases, and what the future holds for this space. George, it's a pleasure to meet you and thanks for the opportunity to talk. Likewise, a pleasure to meet you as well. And well, I've known D-Wave for a long time and well, it's the first time I get to talk to someone from the company. So, well, I'm glad that we made it happen. I am as well. Great okay so let's start

from the beginning actually and since obviously as we just mentioned it's the first time that we connect I thought it would be a good way to start to just ask you to just say a few words about yourself and your background and how you came to work for D-Wave and what you do there exactly in this kind of general intro. Let's see, and we'll take it from there. Sounds good. So again, thanks for the opportunity to talk today. I joined D-Wave about five years ago, originally to run R&D, and then I took over as the CEO about three years ago. And we've made a lot of progress with the company over the last few years.

I'm excited to have the opportunity to spend some time today talking about that. If we go a little further into my background, I was the first president of JavaSoft at Sun Microsystems. So I was responsible for bringing the Java technology to market, building the developer ecosystem, growing the revenue. A lot of what I did there actually is similar to what we're doing now at D-Wave as we're creating a new industry and building a new ecosystem here as well.

I've also been a senior executive at a number of large companies. I was responsible for all products at Avaya and all software at Cisco. I'd been a small company CEO three times prior to joining D-Wave. I sold all three of those companies and now I'm the CEO at D-Wave and I'm thrilled that we've just taken the company public and are now listed on the New York Stock Exchange under the ticker symbol QBTS. That was quite an accomplishment for the team. And finally, I've been a venture investor.

I was a managing director at Warburg Pincus, and I opened their first Bay Area office. So I've had an opportunity to see the tech industry from the small company side, the big company side, as well as the venture investing side. Okay, well, interesting. You have quite a diverse background, then, I would say. And, well, even though, you know, taking, well, Java and quantum computing are quite the distance, it sounds like, well, you do have some experience in, and there is some common ground, at least in what you want to achieve

with with deep wave so yeah go ahead yeah yeah george it's all about uh bringing new technology to market uh developing it bringing it to market commercializing it growing the business that's what i love to do and you know that's what we did at sun Sun with JavaScript. That's what I'm doing now at D-Wave. Great. So I've been going through the list of latest news about the company. And one of the things that caught my attention was the fact that you seem to have a big part of your offering now on offer on AWS Marketplace. And so I thought that's an interesting development. And obviously, I guess it's part of the plan to take things mainstream, basically. And so I was wondering if you'd like to do like a quick walkthrough of what are the parts

of your offering that are now on offer on AWS. And one particular thing that caught my eye there was the fact that you seem to be also offering professional services on AWS, which is something that I hadn't seen before. And I was wondering, how does that work exactly? Sure. So in order to answer that question,

let me take a step back for a minute. At D-Wave, we are basically focused on the development and delivery of quantum computers, software and services. In fact, we're the first and frankly, currently only commercial quantum computing company. And so while everybody else in the quantum industry talks about government research grants as revenue and national labs and academic institutions as customers. We talk about companies like MasterCard or PayPal or GlaxoSmithKline, Johnson & Johnson, Volkswagen, BASF, Deloitte, Cervantes, and the Port of LA. These are all our customers and they're working on real business applications to help improve their business operations. The reason why we've been able to become commercial

so much faster than anybody else in the quantum industry is because we took a different approach to quantum from everybody else. There are two primary approaches. They're called annealing and gate. We decided to start with annealing because it's much easier to scale, much easier to commercialize. And that's brought us to this point. Of course, we're now also building gate model systems so that we can address the full market for quantum. The reason why you need both to address the full market for quantum is because while gate model systems, what everybody else in the industry is doing, they're very good at solving problems like quantum chemistry or computational fluid dynamics,

but they're not very good at solving optimization problems. Things like employee scheduling or shipping container loading and offloading or feature selection for machine learning. They're just not very good at that. But annealing is very good at solving optimization problems, while not very good at solving quantum chemistry and computational fluid dynamics. So in order to address the full market for quantum, you need both. We're the only company in the world that does annealing to address the optimization portion. Now we're doing gate as well. So we'll be the only company in the world that can address the full market for quantum. We take our products and

services to market through a number of different channels. You mentioned AWS Marketplace. AWS Marketplace is one of the channels that we use. So previously, you could access our quantum computer through AWS Bracket, which is the AWS quantum offering. And frankly, through Bracket, AWS customers can access many different quantum computers, ours being one of them. However, you can no longer access us through Bracket. Instead, you access us through AWS Marketplace, which means you buy directly from D-Wave through the marketplace. And you can purchase not just access to our quantum computers, but also access to our hybrid solvers, which combine classical and quantum to solve larger problems than quantum can solve alone. And you also get access to a variety of different services from us, including our professional services. We do have a professional services

organization, and they help our customers to identify which applications can most benefit from Quantum and how to build out those applications. So now going through AWS Marketplace, you get access to our complete set of offerings, than when you went through bracket where you could only access our quantum computer. Okay, I see. Well, thanks for taking that step back. And actually, by doing so, you sort of picked into a number of things I had in my list to ask you in there during this conversation. But again, let's take things from the start. You already touched upon, well, something which is quite a fundamental distinction in quantum computing. So the divide, let's say, between annealing and gate model. And I have to be honest with you, I'm only vaguely familiar with what those terms mean.

And I'm guessing that most of the audience will probably not even be vaguely familiar with them. So would you be kind enough to explain as simply as possible, what are the fundamental principles between this divide and what's actually the fundamental principles driving quantum computing in the first place.

And by doing so, we may be able to better grasp. So why is this distinction important? And what's the principle behind this rough guideline that you gave? Well, annealing is good for optimization problems and gate is good for everything else. Yeah. Okay. So first of all, all quantum computers use quantum mechanical effects, things like superposition or entanglement or tunneling. And, you know, you don't really have to know what those things are, but they are quantum mechanical effects.

And all quantum computers use those effects to solve hard computational problems faster than they can be solved classically. So that's what the quantum industry is all about. As I said a few minutes ago, and as you reiterated just now, there are two primary approaches to quantum, two different ways of using those quantum mechanical effects. One is called annealing, and the other is called gate model. So let's start with annealing. What does it do? Basically, it uses those quantum mechanical effects to find the lowest point in a multidimensional landscape. With a classical computer, if you wanted to do that, you'd have to traverse the hills and valleys of the landscape looking for the low point.

And that can be a very long process. With quantum computers and annealing quantum computers, we use entanglement and tunneling to basically tunnel through the peaks and valleys to find that lowest point. And as a result, we're able to find it much faster than with classical computers. So what's so important about finding the lowest point in a multidimensional landscape? Well, it turns out that that problem itself is one of the hardest optimization problems. In fact, any other optimization problem can be fairly easily mapped into that problem. That's how you program the annealing quantum computer. You take your problem, you map it into that landscape problem, and then you let the annealing quantum computer solve it. That makes annealing quantum computers actually fairly easy to program.

You don't need to know anything about quantum mechanics and quantum gates. Gate model systems are very different. Gate model systems are a little bit more like classical computers in the sense that you program them by specifying the sequence of instructions needed to solve the problem, like algorithms on classical computers. It's just that those algorithms, those instructions are very complex quantum mechanical gates. And as a result, it's very difficult to program gate model systems. Moreover, gate model systems are much more sensitive to noise and errors than annealing systems. That's why annealing systems are a little bit easier to build, and we've been able to move

so fast in commercializing the annealing quantum computer. Finally, and this was new news as of about a year ago, about a year ago, it was actually proven mathematically as well as demonstrated experimentally that while annealing quantum computers are very good at optimization and not very good at solving differential equations, which is what you need for quantum chemistry and computational fluid dynamics, it was proven that gate model systems, which are good at differential equations, are not good at optimization problems. In fact, you'll likely never get a speed up on optimization problems out of gate model systems. You need annealing for that. So that

means that in order to address the full market for quantum, you need both annealing and gate systems. And just one last point on this. You said annealing can solve only optimization problems. That's not quite true. There are actually four, if you like, technological categories of problems that quantum computers can solve. There's optimization. There's linear algebra. This is basically machine learning.

There is factorization. that's cryptography, and then there's differential equations. That's computational fluid dynamics and quantum chemistry. Okay. Annealing quantum computers are very good at optimization problems. They can also solve linear algebra for machine learning and factorization for crypto, but they cannot solve differential equations problems. Gate model systems are very good at differential equations problems. They can also attack linear algebra for machine learning and factorization for crypto, but they cannot address optimization problems. So there's a set that only a nailing can solve, that's

optimization. There's a set that only gate can solve, that's optimization there's a set that only gate can solve that's differential equations and then there's a set that both can solve that's linear algebra for machine learning and factorization for crypto okay great thank you that's uh that was quite uh enlightening actually and i guess you implicitly also addressed uh something that was also uh in in the list of questions I had for you. So I was wondering how come since D-Wave basically pioneered the annealing part of quantum, let's say, then why did you choose to also address the gate model? I think it was about a year ago, actually, that D-Wave started doing that. But I think based on what you just said, I can infer the answer, which is that you wanted to expand the range of the applications that you can address, right?

Yeah, that's exactly right. A year ago, what we concluded was that our annealing quantum computers had achieved commercial status. They were capable of solving real business problems at commercial scale. And a lot of, if not most, of the hard underlying technological problems had been solved to get to that point, which meant we had some bandwidth, some deep research and physics capabilities and bandwidth to apply to another class of hard problems. And so we decided to go ahead and initiate a gate model program. And that would allow us to be able to address the full market for quantum and frankly, be the only company in the world

capable of addressing the full market for quantum because we're the only company that can deliver the annealing portion for optimization. Yes, to the best of my knowledge, at least that's exactly the case. I don't know of any other company that does both. I think actually you're the only one that does annealing and everyone else does gate. That's correct. And now we're doing gate as well. True. So why is that?

I mean, why is it that only D-Wave set out from the get-go to address annealing? I'm sure there are a number of technical issues around that and I may be at least again vaguely familiar with some. So people talk about the need to have like deep freeze stage so that the quantum computers can work as error free as possible. They actually also talk about what is probably to the best of my knowledge again the most difficult thing to address when working with quantum, which is the error rate and how you can possibly minimize that. the value of having more or less of those, or actually some other people also talk about, well, it's not all about how many of those you get, but how many of those error-free you can

actually get. So yeah, I'm waiting to hear how, well, why, how, basically let's start from this. How come you're the only ones that do un annealing? Yeah, that's a great question. And it's, in some sense, a fluke of history that worked out really well for us. And what I mean by that is D-Wave was the first company to try to build a quantum computer. We started over 15 years ago. Now, at that point in time, it was not actually believed that you could build a gate model system. The science and the engineering had not yet progressed to the point where it was really believed that you could build a gate model system. But it was pretty widely accepted that you could build an annealing quantum computer. So, you know, we decided to go ahead and build the annealing system because that was something that, you know, we believed we could do.

Now, we knew that while annealing would be very good at addressing a certain class of problems, we also knew that it could not address all problems, but that was okay because, you know, we felt that it was important to get started, to bring a quantum computer to market. And since this was the only thing that it was really kind of understood that you could build everybody else decided to jump into the quantum space, the science and the engineering had progressed to the point where at that point it was believed that you probably thought that a gate model system could solve all problems. So everybody else concluded that they might as well build a gate model system because, you know, they believe they could, and it could solve all problems, whereas annealing, it was known, could only solve a subset of the problems. So everybody else jumped into GATE. What happened was a year ago, everybody got surprised, us included, because that's the point in time at which it was proven that GATE model systems can't really deliver a speed up on optimization problems. The reason is, in order for a gate model system to solve an optimization problem, it requires classical overhead. And we now know that the classical compute required to get the speed up on the gate model system far outweighs any of the benefits that the gate model system can provide on the optimization problem. So what that meant was that

you really do need both annealing and gate, annealing for optimization, gate for differential equations. But at that point, we were the only company in the world doing annealing. We now have a 15-year head start. We have over 200 US-granted patents and 100 in process worldwide. So you have huge patent mode. So it'll be very difficult for anybody else to come into that space, certainly anytime soon. So it's just worked out extremely well for us and allowed us to create a very valuable company. Okay. I see. Indeed. Again, based on what you just said, it sounds well, like a combination of, well, foresight and a little bit of luck, to be honest, because, well, I just said nobody could possibly have known when you started that things would turn out that way.

Hey, if I have a choice between being lucky and not being lucky, I'd rather take being lucky. So, yeah, a little bit of luck is not a bad thing. Absolutely. I mean, who would not make that choice? Exactly. Okay, so let's also talk a little bit about the hard parts about building both annealing and wave models. So, you know, so kind of naive question for anybody who's not into quantum. Why do you need the deep freeze state? And how is that related to the infamous error rate that you get from qubits?

OK, when you say deep freeze state, you mean why do we need to run at cold temperatures? Yes. Yeah. Okay. So first of all, when it comes to building a gate model system, there are many different technological approaches that are being worked on. There is superconducting, which is what requires the system to run at very cold temperatures. But there's also trapped ions, which runs at room temperature, and photonics, which runs at room temperature. So some of the technologies must run cold. Some of the technologies do not need to run

cold. With annealing, since we're the only company in the world doing it and we decided to use superconducting annealing runs cold okay now it's not that you need to be cold in order to have low error rates and high coherence times because ion trap systems do not run cold photonic systems do not run cold and it's believed that you'll be able to get low error rates and high coherence times in those regimes as well. Although it hasn't been fully proven yet because it's still early on in the development of those systems. However, the reason why superconducting systems, whether annealing or gate, need to run cold is because that's how you get to superconducting operation. In other words, it's not about error rates. It's about actually getting the conducting material to to kind of become superconducting.

And, you know, currently there are different materials that become superconducting at different temperatures. But we use low temperature superconductivity. We run at about 10 millikelvin, which is very close to absolute zero. That would be zero kelvin. OK, I see. And where does the error rate come into play? And how does it affect people wanting to develop applications leveraging quantum? And what can you as a maker do to make their lives easier, basically? So errors impact computation on both annealing quantum computers and gate model quantum computers. By the way, they also impact classical computers. That's why we have things like parity error and error correction. In other words, you know, we don't even think about it anymore.

In the old days, we did. We don't think about it anymore. In the old days we did, we don't think about it anymore. But even with your classical computers, whether it's storage or compute, every so often there's a bit error, right? The bit's supposed to be one and it's showing up as zero or it's supposed to be zero and it's showing up as one. But we have algorithms that correct that in classical computers because it's very rare. And so it's easy to detect and correct. With quantum computers, though, it's not so rare. The qubits, the unit of storage, are much more sophisticated with classical computers than the bits,

sorry, with quantum computers than the bits in classical computers. And so there are many more ways that errors can be introduced. And it's typically by interacting with the environment, right? You know, magnetic or radiomagnetic interference. You know, I mean, EMI, there are many issues that can cause errors to get introduced in the system. And with quantum computers, you know, they're far more sensitive to environmental factors creating errors. And so errors occur much more frequently. Now, gate model systems are very, very sensitive to this. And the reason is when you start doing a computation on a gate model system, you're basically applying instructions to the qubits. Like with classical computers, you're applying instructions to the bits Okay, as soon as an error gets introduced if it's not Corrected your computation falls apart and that's kind of obvious

Right and since these errors occur so frequently without error correction You can't get through more than 20 or 30 instructions without the introduction of an error and the computation falling apart. But for many of the gate model algorithms, you need tens of thousands, hundreds of thousands, or millions of gates instructions. So you can't do very much with a gate model system without error correction. And that's why it's so important to get that error correction, which will give us qubits that can have high fidelity through long computations. With annealing quantum computers, it's different. With an annealing quantum computer, remember, you're searching for that low energy or that low point in a multidimensional landscape. So you start running the computation, right? If an error gets

introduced, you actually can stop the computation and then restart it from where you are in the search. You can't do that with gate model systems. With a gate model system, once there's an error, you lose everything. With an annealing system, once there's an error, you actually settle into a solution, a point on the landscape. You get that back and you can restart the computation from there. So that makes annealing quantum computers a bit more robust against errors. And that's why with our annealing quantum computers, we're able to deliver good, if not

optimal solutions to hard computational problems today without error correction. Okay, so you don't actually apply error correction in the annealing? In the annealing systems, we don't. There are error correction schemes that have been created for annealing systems. They could improve the computation, but not enough that it's currently worth the investment in the hardware to introduce those schemes. Okay, interesting. So I guess that only goes to reinforce the argument about, well, it's not really, it's at least it's not only exclusively about how many qubits you can get, but it's more about how many of those qubits you can get error-free.

It is, but I'd say it another way. Again, for annealing, we're pretty robust against errors, so we don't worry too much about it. For gate model systems, really it's about how many gates you can get through before there's an error, right? How large an algorithm can you run before there's an error? And that's what we need to focus on. It's not about the number of qubits. It's not about the connectivity between the qubits on gate model system. It's about circuit depth. How many gates can you actually use in your computation before there's an error and the computation falls apart? And of course, it's really error rates that are currently limiting the circuit depth to such a small number of gates. And until that gets addressed, there's no evidence you'll be able to do anything useful with a gate model system. You know, there are people that are saying, you know, there's evidence that you can do useful things today without error correction. No. The results that show statistical significance, they're very weak.

And there's really no strong evidence that you can do anything useful with a gate system without error correction. Then you'll hear people say maybe three years before you can do something useful with a gate system. I think it's at least seven years because you just are not going to be able to do useful work without error correction. And we're at least seven years away from error correction, maybe more like 10 years away. So I think we've got a ways to go on gate model system. But but, you know, for us, that's

okay. Because we've got a kneeling today, we're commercial today, we're building our business today, while developing the gate system for the future. Yeah, yeah, I can see that. So I was just wondering, again, on the topic of error correction for Wave, does then the error correction have to also work preemptively in a way? Because otherwise, if you, based on what you said, that, well, if you have an error on the gate model, then you basically just lose everything. Don't you have to sort of act in advance and catch the error before it actually happens or make projections and take appropriate action?

Yeah. So there are different schemes that have been proposed for error correction on gate model systems. One of them basically kind of allows you to get awareness of the errors as they're occurring, but you actually don't have to correct them until later in the computation. And then you can come back and fix what had happened in the past. So, you know, it's not so much about when you correct the errors, but it is important to detect them as they're occurring. And that's what these schemes all allow you to do. Okay. So then let's shift the focus back to annealing because, well, frankly, it sounds like where most of the action is at least today.

Today. So one of the things you mentioned earlier was the fact that when errors happen in annealing, you at least have the chance to sort of apply what sounds to me like saving your current state, let's say, and then resuming from where you left off. So how does that work exactly? And again, something else that you mentioned earlier is that, well, the way that people use quantum is in combination with some other system,

which typically is some classical computing system. Does that help in that case? So is one of the things that you use the classical computer system for to be able to save state and resume when that's needed? Yeah, so yes, but it's a bit broader than that. So, first of all, let's talk about just the annealing algorithm for a minute. So, you know, there's really only one problem that the annealing quantum computer solves.

That's the one I mentioned a minute ago, finding a low point in a multidimensional landscape. And the annealing algorithm typically will run for on the order of a microsecond. So it actually runs very fast. Depending on the noise in the systems and the errors, if it's a perfect system, no noise, no errors, at the end of the microsecond, you have the optimal solution. You're at the low point in the multidimensional landscape. If there have been some errors introduced, right, or some lost coherence in

the system, which is what errors basically result in when we're talking about qubits, then at the end of the annealing algorithm, you might not be at the optimal solution. OK, so what do you do? You restart. You run the annealing algorithm again from where you left off. OK, so that's kind of what I mean by, you know, the ability to, you know, restart the computation from from where you left off in some sense. Not exactly the way it works, but good enough for description. And yeah, at that point, you would use classical software to look at the solution,

determine if it's good enough, or are you at the optimal point, or do you need to run some more? Okay, but hybrid and classical does more than that for us today. So, um, currently, um, in order to run the annealing algorithm on your problem, the entire problem has to fit in the quantum computer. So this is where we start to worry about not just the error rates, but also the number of qubits and the connectivity between the qubits. Because

the more qubits and the more connectivity, the larger problem we can fit into the quantum computer to run the annealing algorithm on it. So the second thing that hybrid does for us is it allows us to solve problems that are larger than what can fit natively in the quantum computer. And the way it does that is that the classical part of the hybrid solver takes the full problem and it looks for the hard sub-problems that require the quantum computer. And it sends those problems off to the quantum computer to solve. And then it iterates. So you're taking smaller subproblems out of the big problem that require quantum.

You're sending them over to the quantum small enough to fit in the quantum computer, and then you get the results back and iterate. That's the second thing that hybrid does. Currently, our hybrid algorithms can solve problems with up to a million variables and 100,000 constraints in an optimization problem, which is large enough to solve many real world commercial problems. There are still problems beyond that. For example, if we wanted to optimize FedEx routing, full up, right, backbone and last mile, that would take tens of millions of variables. So we're not there yet. But we're currently at a point where we can solve

a number of important real world problems. And of course, we continue to enhance our systems, both our quantum computers, we've announced our next generation advantage to computer, which will have more qubits and more connectivity, as well as even lower error rates and allow us to solve larger cores and enhancements to our hybrid solvers. Okay, great. Well, you have to excuse my curiosity, and maybe we went a little bit deeper than you expected in those things, but well, you don't get the chance to hear those things firsthand every day. So I thought I'd take advantage of that.

But what you just said is a great segue to shift gears a little bit and focus more on, well, what kind of people, what kind of problems people can solve with using quantum and how do they actually do that? And, well, actually, you just mentioned something called the hybrid solver. And I'm wondering whether, well, to what extent this sort of breaking up your problem, basically, that you described, is something that people developing the applications have to do? Or whether this is something that, well, the hybrid solver can do for them. Yeah, the hybrid solver does it for them, is the short answer to the question. We've designed

our system so you can submit your full problem to the system, to the hybrid solver, and the hybrid solver will worry about determining what goes to the quantum computer and what gets solved classically. Now, as far as what kinds of problems are we solving today, frankly, it's a broad array of problems that are all focused on improving operational efficiencies, reducing cost, driving growth. Let me give you some examples. MasterCard. We're working with MasterCard on a number of different applications, but one of them is customer offer allocation. So, you know, MasterCard kind of develops offers with vendors. And then a decision needs to be made with respect to which offers to make available to which cardholders, right? And you want to offer those to the customer that's most likely to take advantage of it, okay? So now you've got all the cardholders, you've got all these different offers, and you need to figure out which offers to assign to which cardholders to maximize the kind of uptake in the offers.

That's a very large, hard optimization problem. That's one we're working on with MasterCard. Another example, BASF, a very large chemical company. We're working with them on job shop scheduling to basically optimize the workflow through their various chemical labs. Another example, really good one, work we did with a company called SavantX and the Port of LA. So, you know, supply chain logistics is, you know, a hot topic. The idea is how do you reduce the time to load and offload containers from ships and reduce the time for vehicles to pick up the goods and move them out of the port, right? Well, working with our quantum computer, Savantex and Port of LA have been able to see a 60% improvement in the performance of the cranes loading and offloading the containers

and a 12% reduction in the time for vehicles to basically pick up goods and move them out. Okay. That's another example. You know, we've worked with Volkswagen on both global traffic management, as well as optimizing the scheduling of painting vehicles in order to reduce waste through the process. So, you know, it's a very broad array, but there are three verticals that we are currently focused on. Manufacturing and logistics is the first, finance is the second, and pharma is the third. And we're focused on those three verticals because those are the industries that have really important and very hard optimization problems that need to be solved.

Okay, well, I get the impression that, well, if you want to solve one of those problems as an application developer, let's say, then what you should probably be doing is not thinking about, well, an algorithmic way to approach it, but rather what is the best possible way you can declaratively describe your problem? What are the parameters? What's important and how they are interrelated? So I'm guessing that this is what people have to do, actually, in order to be able to use quantum. To use our system, correct. But we've tried to make it very accessible to them by giving them the ability to specify the problem in a way that they're accustomed to specifying it. So in many cases, these problems are specified as what's called a linear programming problem or a quadratic programming problem. This is the language of these kind of optimization engineers.

So now we allow them to take that specification and feed it directly to our hybrid solvers. And we translate it into the right form for using the quantum computer. OK, I see. So it's probably it's not something like an API that people would be able to embed into their traditional, let's say, programs directly. It's more and again, based on what you just said, it's really an optimization problem. So not something that's, well, the average software developer's cup of tea, probably. It's something that's addressed to a different role in an organization.

Exactly. It's addressed to the data scientists and the data analysts. That's really, these are the people that deal with things like this. Okay, I see. And well, to tie this back into where we started the conversation, so which parts of your offering on AWS right now can these people use and how does this make their life easier so to speak yeah so um we have a few different offerings on aws marketplace today um first of all you can buy access to our leap

cloud service so you know in addition to the quantum computers and the hybrid solvers and software development tools we also have our own quantum cloud service. It's called Leap. And that's where you go to access all our capabilities. So you can buy access to our Leap cloud service through AWS Marketplace if you kind of want to manage kind of problem formulation and submitting the jobs to the system yourself. Or you can buy a professional services engagement with our PS team where they'll do all the heavy lifting for you. They'll understand the problem that you're trying to solve, and then they'll map it and they'll manage access to the systems and solvers. We also have one more targeted offering on AWS Marketplace,

which is very interesting, and that's a feature selection offering. So one of the most important elements of machine learning is called feature selection. The idea is that when you do machine learning, you typically will have a lot of characteristics or classifiers that you want to try to learn on. OK, but if you try to learn on all of those, it could be hundreds or thousands of them. You'll end up trying to overfit the model. You'll create a model that's not very good. So what you want to do is kind of a pre-processing step on machine learning is try to identify a much smaller set of those characteristics that are most representative

of what you're trying to learn and then build a model on that set. Well, finding that small set of very strong classifiers from a big set of weak classifiers is a very hard optimization problem and one that our system is very good at. So we have that targeted offering on Marketplace as well. If you're doing machine learning and you want to do a better job with feature selection, you can buy, if you like, a feature selection capability from us through AWS Marketplace. So you could buy access to our quantum computers and solvers. You can buy access to our professional services organization. You can buy access to a targeted solution for optimizing your machine learning.

Great. That's also another thing that I've been meaning to ask you. So whether you get many machine learning practitioners using your systems and what for exactly? So I think you already gave at least one part of the answer. So people use D-Wave for feature selection. Do they also use it for other parts of the machine learning pipeline? Not currently. And the reason is neither our quantum computer nor any of the gate model systems

are yet at the point where they can beat classical GPUs at the core learning process, the core model building process. However, there are components of machine learning where our quantum computer can deliver value. Feature selection is the first of those and the only one that we have a targeted offering around. But we are looking at some of the other elements that themselves are optimization problems that can aid in the machine learning process. But, you know, when we talk about feature selection, for example, we, you know, one of the

problems that often comes up in the finance industry is fraud detection. And machine learning is typically what's used there. And we have worked with financial companies on feature selection for fraud detection as an example. Okay. I was also wondering, again, what you do sounds like could be leveraged by a discipline which is somehow related to machine learning, even though it doesn't have the same mindset. It's a discipline called operations research. And while it's mostly about optimization, so I'm wondering if you have operation research practitioners using the Wave as well. We don't have an operations research practice, but you're right.

I mean, OR is heavy optimization. And so while we don't call it operations research, we do work with customers on the optimization problems that need to be solved in that context. Another thing that sort of makes sense, if you think about it, when I hear you mention your list of clients, basically, it sounds like most,

if not all of them are basically from the top tier, I don't know, Fortune 500, 2000, 5000, whatever it is, you want to call it. And, again, it makes sense in the because of the fact that, well, you do need a number of things to do to be able to, to reap the benefits, let's say, of using D-Wave. You need, well, first, large enough problems. You need a certain degree of sophistication and people who can actually model those problems and understand the benefits of applying something like D-Wave to solving those. And I'm guessing that, well, you also need a substantial budget, actually, even though I'm not familiar with the price range that you offer. This is my estimation, at least. So I'm wondering

if besides this list of clients, you also have other more like mid-market, let's say, companies using your services. Yeah. So at the end of Q3, we just announced our Q3 earnings last week. In that call, we announced that in the first three quarters of this year, 2022, we had over 100 customers, over 60 of them commercial, the others government and education, and over 20 of those global 2000. Okay. So what that means is that there are still 40 or so that are commercial customers that are not Global 2000. So we are addressing more than just very large corporations. And let me give you an example. One of our customers is a Canadian grocery chain called Save On Foods.

They operate mostly in the western part of Canada. They have in production today the use of our system for e-commerce grocery delivery. And they are working on a second application that we hope to help them move into production within the next few months. So it's not just the large global 2000, although they are important customers of ours. Okay. All right. Well, speaking about quarterly results, another thing that I wanted to ask you is, well, the rationale behind taking the company public, basically, and well, a brief, if you can be

as kind as to offer like a brief overview of your results from being public so far, starting with Q3, since it's obviously the latest one. Yeah. So look, Q3 was a good quarter for us. You know, we announced over 30 percent quarter-over-quarter revenue growth. We announced over 30%, you know, first nine months this year compared to first nine months last year, customer growth. And so we're quite, and we announced, you know, progress with our gate model quantum computer as well as progress with our annealing quantum computer. One of the things that was really exciting as a part of this Q3 announcement was that we actually

were able to demonstrate and prove that our annealing quantum computer is using large-scale coherence. What I mean is coherence across most of the system, not just directly connected qubits, which is really important because large-scale coherence is what allows you to move quickly toward the optimal solution. And this was the first time that we were actually able to provide a proof that our system is doing that. That was published in Nature Physics. So, you know, a lot of good progress as a public company from a revenue perspective, from a customer account perspective, from a product delivery perspective, from a contributions to science perspective. You know, as far as taking the company public, you know, it's another financing exercise, right? I mean, by taking the company public, we raised cash and we opened up a variety of new funding sources for the company. Obviously, we want to move as aggressively as possible toward building revenue and moving to profitability, but it's still going to take some time to get there. And so, you know, we wanted to make sure that we had access to as broad an array of funding

sources as possible. And as a public company that opens up access to new capital. Okay. I see. And okay. I think we're actually closing to the end of our time. So let's wrap up with one broad question. I was wondering what your take is

on the broader ecosystem in quantum, both in terms of hardware, which you already sort of touched upon in the beginning of the conversation, but also in terms of software, which to be honest with you, was something that I only realized myself recently that is a thing, actually.

Yeah. So, look, I mean, when it comes to the broader quantum ecosystem, from a hardware perspective, it's really about the two approaches, annealing and gate. Annealing is commercial today. Gate is many years away from being commercial. As I said previously, at least seven years, maybe more like 10 years, all having to do with error correction. In the meantime, from a software perspective, obviously on the annealing front, we continue to enhance our hybrid solvers. We continue to enhance our tools.

We continue to work with customers to build applications to benefit their business that can then be brought forward to other companies as well, similar applications and so on. So taking one customer and being able to kind of replicate that with other customers or taking one customer and being able to upsell to a second or a third application. In the gate model space, it's still about research experimentation. The software that's being developed is basically what I'll call attempts to show evidence that you can do something useful with a gate model quantum computer, but typically these are very small scale problems, you know, really just trying to show some benefit from quantum because the systems just aren't yet large enough or capable enough to do any more than that. So the software is really focused on tools and research experimentation.

Okay, great. Well, thank you. It's been a mostly productive interesting conversation i learned a few things which is one of the great things about doing this actually thanks a lot for your time and i guess well unless there's any other closing statements you'd like to make we may as well wrap up here all i'll say is thank. I really appreciate the time and it was a fun conversation. I hope you enjoyed the podcast. If you like my work, you can follow Link Data Orchestration on Twitter, LinkedIn, and Facebook.