Please welcome to the stage, President and CEO, Alan Baratz.
Good morning, and thank you all for taking the time to be here with us today. It's my pleasure to be able to spend the next two hours with you. I just gave the production team a heart attack. I'll do my best not to take two hours, but we have a lot of amazing content to share with you. We're going to talk about everything from our product roadmap to the amazing work that our customers are doing leveraging our technologies and our services, to the recent quantum supremacy demonstration that we published in Science. We've got a lot to share with you, and we're going to try to have some fun along the way.
So, we've been busy over the last few months. Everything from publishing in Science, the first demonstration of the ability for a quantum computer to solve an important, useful problem that can't be solved classically — quantum supremacy on a useful problem. And then a week later, we showed how you could use that computation to basically create a quantum hashing function, the basis for blockchain and cryptocurrency. And we took that and actually built a blockchain prototype out of it, running across four of our quantum computers as one of the first distributed quantum applications. We'll be talking a bit more about this in a few minutes.
We also sold our first Advantage system. We sold it to the Ulm Supercomputing Center in Germany, and we're going to be spending some time talking about why they acquired a system and how they're planning to use it. Financially, we had a great FY24: record bookings in Q4 and record bookings for the year. We also announced that we will have record revenue in Q1 of this year, and we have over $300 million in the bank, more than enough cash to achieve profitability. So, we're very well positioned right now.
But despite all of that, there are still too many leaders in big tech that simply don't understand quantum computing and are spewing misinformation into the industry and the marketplace. And so we're also working very hard to ensure that the proper messages are getting out, that they're getting out to the marketplace, to the financial community, to the academic community. And to that end, we developed a framework that we call Quantum Realized.
This is a three-part framework that anybody can use to determine if a quantum computing company can actually deliver value. And so whenever you or anybody else is looking to invest in quantum, you should ask yourself if the company you're working with meets all three of these criteria. The first: has their quantum computer been able to solve a problem better than classical computers can solve it? And I'm not talking about has a simulator of their quantum computer been able to do something as good as classical, but is there a quantum computer that can beat classical? Second, does the quantum system have the reliability and the availability needed to actually be able to use it to get useful work done? And finally, does the company have relationships with customers that are using the quantum systems either in production or on the path to production? And those are the three criteria that for us define Quantum Realized. And we believe that D-Wave is the only company thus far that has actually achieved all three of those goals.
So as we go through the remainder of my presentation this morning, I'm going to talk about each of those three areas. Can the quantum computer solve problems better than classical, beyond classical? Do the products have the performance, the reliability, the availability needed to support whatever applications you want to run? And then finally, are there customers that have actually used the systems and services to some benefit? So, why don't we go ahead and jump in?
I mentioned that we had for the first time ever demonstrated the ability for a quantum computer to solve an important, useful, real-world problem that cannot be solved classically. This was done on a small version of our Advantage 2 system, the 1,200-qubit version of our Advantage 2 system. But we were able to solve this problem in minutes that would otherwise have taken classical nearly a million years. So what was it that we actually did? The problem was in the area of magnetic materials simulation, basically trying to compute properties of a variety of different types of magnetic materials. So magnetic materials are basically lattice-structured molecules, and they can have many different types of lattices underlying them. So we performed these computations on three different lattices: cubic lattices, diamond lattices, and bcc lattices. And our computations were done on all three of those lattices.
Second, we performed the computation on varying sizes of those lattices, ranging from a small number of qubits to a relatively large number of qubits. For example, in the case of the 3x3x3 cubic lattice, we went up to 6x6x6. Subsequent to that, on our larger 4,400-qubit Advantage system, which I'll talk about in a minute, we've actually doubled that up to 12x12x12, but a variety of different sizes. Next, what we're looking for when we're simulating these magnetic materials is how they behave within various magnetic fields. And we need to look at how they behave over time. So we looked at a variety of different evolution times from a nanosecond up to 40 nanoseconds, and finally we computed a number of different properties on these magnetic materials. So all of that is what embodies the quantum supremacy demonstration that we published in Science.
And in order to be successful with this, we needed to be able to perform all of those computations. And all of those computations we were able to perform in minutes on our Advantage 2 quantum computer for each of the computations. And for one of the hardest of the lattice structures and the largest size of the lattice structures at roughly 20 nanoseconds, computing each of the properties, we were able to do each of those computations, as I said, in minutes, whereas it would have taken nearly a million years using the best possible classical algorithms on the fastest existing supercomputers. We used the Frontier supercomputer at the Oak Ridge National Lab, which is a massively parallel GPU system.
Moreover, the energy that we consumed in doing those computations was about 12 kilowatts. It would have taken well over global annual energy consumption to be able to perform that computation classically on the fastest supercomputers in the world. So both very fast and very energy-efficient.
Now, when you do something like this, you tend to get a big target painted on your back because everybody that is working on classical algorithms to do these sorts of things wants to be able to show that they can do as well or better. And we've had some papers come out with overreaching claims that state, 'Oh well, you know, we've been able to perform some of these computations, and as a result, we can perform all of them classically better than D-Wave.' Well, that's kind of like running a triathlon. So, I'm the best in the world at the triathlon. Somebody comes along and says, 'Hey, I've been able to run the first quarter, run the first quarter of the running portion of the triathlon faster than you ran the first quarter, but that means I can beat you in the triathlon.' That's the kind of things that we're hearing right now.
And that's fine. We really do want to understand how well classical can do so we can understand the boundary between classical and quantum, but it's really important that everybody be clear about the work that they've done and the relationship between that from a classical perspective and quantum. These results are just not possible classically, and that's quantum supremacy. And why is this so important? I said that this problem was an important, useful problem. The reason why this is an important, useful problem is because magnetic materials are used everywhere. They're in your cell phone. They're part of the speaker. They're part of the microphone. They're part of wireless charging. They're also used for sensors. Many of the medical devices that we use today are built out of magnetic materials.
And what's interesting about magnetic materials and the use of them as sensors is that they become most sensitive when you operate them close to what's called a phase transition point. This is like ice freezing. Well, magnetic materials go through a similar phase transition when put in a magnetic environment. And so what you want to understand for these materials is at what point do they go through this phase transition and how sensitive are they just prior to that point? Because if you can do that, you can explore a variety of new magnetic materials, understand how good they are going to perform as sensors without ever having to fabricate any of them. That's a materials discovery platform, and that's what this computation allows us to do, and that's why the computation is so important.
Okay, I want to switch gears now and spend a little bit of time talking about our products and our product roadmap. We showed a chart similar to this at Qubits last year, but we have since significantly extended the product roadmap from both a hardware and a software perspective, as well as the types of applications that are enabled at each step along the way. So at the top of this chart is our hardware roadmap: Advantage 2, Advantage 2 performance update, Advantage 3, Advantage 3 performance update. Right below that are the software advancements that we're focused on: NL solver, NL solver feature expansion, improving ML model training, and so on. And then at the bottom of the chart are the applications enabled by each of those classes of technology or each of those deliverables.
And there's some really exciting things in this roadmap. For example, Advantage 2 — we've talked about this before — increased connectivity and coherence. I'll talk a bit more about that in a minute. Advantage 2 performance update: novel annealing protocols, cyclic annealing. We talked a little bit about this at Qubits last year, but the point is that with cyclic annealing, we can effectively extend coherence times for problem solving beyond the physical coherence time, and as a result compute better solutions much faster. Advantage 3: digital-analog computing. I'm going to talk a bit more about this in a minute, but it's basically bringing together annealing with some digital gate capabilities that will allow us to actually solve optimization problems significantly better even than we're solving them today. And really exciting, Advantage 3 performance update: it's all about scaling. I'm going to talk about our roadmap to 100,000 qubits.
So, our current flagship product is Advantage. It is our 5,000-qubit annealing quantum computer. We currently have four of them in production in our Leap quantum cloud service that many of you are using today in support of your applications.
Advantage 2 is the next-generation system that will be made generally available later this year — not too much later this year, actually. And it's a big leap over Advantage. It has more connectivity. So each qubit in Advantage 2 is connected to 20 others, whereas it's connected to only 15 others in Advantage. This means we can solve larger and more complex problems. We double the coherence time in Advantage 2 over Advantage, which means we can solve problems faster, and we've increased the energy scale by 40% in Advantage 2 over Advantage. Energy scale is like the precision with which you can specify problem parameters, and that gives us better solutions.
We've had a 1,200-qubit version of this system in our Leap cloud service for a little while now. There are a lot of jobs that have been submitted to that, but the final product version will be released very shortly with 4,400 qubits. So, as we look to the future, it's always about more qubits and connectivity, greater coherence, and increased energy scale.
But what we want to do now is really accelerate the drive to larger systems. Up until now, with each generation of quantum processing unit, we've increased the number of qubits by fabricating more qubits on the processor chip. And we still have some headroom there. But to really accelerate growth in the number of qubits, we have to move to multi-chip, interconnecting multiple chips. And so we have aggressively started working on that approach to scaling our systems. But it's not just about being able to interconnect multiple chips. When we do that, we need to make sure that we are also preserving coherence and that we are continuing to increase connectivity.
So in this roadmap to 100,000 qubits, we are targeting at least double the connectivity over Advantage 2, so going from 20 to 40, and 10 times the coherence. Then we need to be able to control these large processors with 20,000, 50,000, 100,000 qubits with a relatively small number of I/O lines. When Google demonstrated Willow, they had four I/O lines per qubit. That doesn't scale. Today, with our Advantage system, we control 5,000 qubits with 200 I/O lines, but even that's too much. So as we scale to tens of thousands and hundreds of thousands of qubits, we need to be even more efficient with addressing and control. And then of course we need to ensure that as we interconnect the multiple qubits, we are preserving things like, you know, quantum properties, entanglement, tunneling across the chips.
All of this is a significant focus for us right now as we drive from Advantage 2 to the Advantage 2 performance update, Advantage 3, and the Advantage 3 performance update. And then I commented that we are also pursuing integrated analog-digital computing. And what I mean by that is the ability for our annealing quantum computers to also be able to process some gate-model gates. And in fact, what you see in the picture on the right is that we've got a chip and we are able to excite one of the qubits. This is like a one-qubit gate. And then we're able to pass that excitation through two-qubit swap gates around the perimeter of the chip. Right? So here we've got an annealing fabric, and within the annealing fabric we are processing some gate-model gates.
And we're quite excited about this because we think that this is going to give us a tool to be able to further improve how we perform on optimization problems by giving us another vehicle to escape local minima. And we're quite excited about this. It gives us one other thing, and that is the ability to read out in more than one basis, which is foundational for things like provability with respect to hard computational problems.
And then there's our gate-model program. So we are working not just on continuing to improve our annealing quantum computers, but we are also focused on our gate-model program, and in fact now we are accelerating the work on that program. So the focus is first of all to be able to demonstrate that we can use our cryogenic control — the control that we use for our annealing quantum computer, with some modifications — to control gate-model qubits. Then we're going to take that and demonstrate the ability to perform one- and two-qubit gate operations. Again, leveraging the cryogenic control. By the way, cryogenic control is something that no other company in the world has. Only D-Wave has that capability right now. And ultimately, it's going to be required for every superconducting gate-model program. And then two interacting, and finally multiple interacting logical qubits. Our target for the first set of logical qubits will be distance five, which is essentially 50-qubit logical gates, and we're going to move as aggressively as we can on this roadmap.
Of course, it's not just about the hardware. It's also about the software. And we're always continuing to evolve and improve our hybrid solvers. Our current best hybrid solver is called the NL solver. It's a nonlinear solver. It excels at solving nonlinear optimization problems. And we have just announced that we have now introduced the ability to also support continuous variables within the nonlinear solver. And continuous variables is really important for problems where the solutions are not on discrete boundaries. For example, 3D bin packing for cargo loading, where, you know, the boxes may not sit on discrete boundaries within the container, and there are a whole host of other problems, satellite placement, for example.
And of course, as we continue to drive the hardware roadmap and we continue to drive the software roadmap, we are always doing it in a way that ensures that our systems remain reliable and available. Our current systems through our quantum cloud service have greater than 99.9% availability. And in fact, we now offer service level agreements to our customers who are moving applications into production to make that formal commitment that if you're willing to base your business applications on our quantum systems and services, we will ensure that you will have the uptime and availability required to be successful with those applications. And of course, there are a host of other things that we do within our quantum cloud service and to our systems, for example, SOC 2 Type 2 compliance for things like security and privacy, and so on.
Okay, with that, it's demo time, and so I'd like to ask Alex Gandillo to join me. Alex is our Vice President of Algorithms, Performance, and Tools. Alex, thanks. For those of you who have been at Qubits before, you will know that Alex is the person that I always bring up to do demos, and every year I try to get out of it. And every year it's always — and most of the time he's successful with a demo. Oh yeah. Yeah. But you know, you never know. The interesting thing about our demos is they're not actually demos. We're actually running real applications against our Leap Cloud service. And so a lot has to go right for this to work. The main thing in a facility like this is the internet connectivity. But here we go.
All right. Awesome. Thanks, Alan. Hey everyone. Glad to be here. This is actually my fifth Qubits in a row doing a live demo for you all. And historically, our demos have always followed a particular pattern, which is that we've introduced a benchmark problem and then showed how awesome our hybrid solvers are at solving those benchmark problems. And they're always awesome at solving those problems. But one of the things that benchmark problems — while they're quite easy to understand and they're quite difficult — they don't often capture the same complexity that, like, a real-world customer application-style problem has. And so given that this morning we announced that one of our customers, Ford Autos, has gone into production with an application, we thought it'd be really fun to show you all a demo of that application. So this is a customer-focused demo.
So, this looks complex, and that's because it is. The problem is this: Ford Auto has a body shop. They have a set of vehicles that need to come in, and those vehicles have doors that need to be installed. There's a couple sources of complexity for this problem. So, the first one is in the varieties of vehicles that need to come in. There's buses, there's vans, there's so-called jumbos. And then it also comes in in the sense of which door needs to be installed. Sometimes it's the left door, sometimes it's both doors, sometimes it's the right door. And then the vehicles come in different colors. This is a commercial fleet, so many of them are painted white, but some of them are painted special, which means like decals or that kind of thing. And then some of them are painted other colors than white. So there's this big complexity in the set of different things that you need to do.
But in addition to that, in order to have a valid schedule, a valid order in which to install these doors, they have some rules. These rules are again coming from the customer. So for instance, they want to do two buses in a row always. Not one, not three, two. They want to be doing four consecutive jumbos in a row. Exactly four. Why? I don't know. The customer told us that's what they want. So I've actually gone ahead and run a problem already at this scale to give you a sense of what this looks like. So this was run this morning in the interest of time because we're all trying to move along. And as you can see, this meets that requirement. So if I filter by jumbos, you get this sort of four in a row. Okay.
So let's go ahead then, and now that I've introduced the basic one, let's go ahead and make this problem a little bit more tricky. So I'm going to take this up to this 350-vehicle scale, and I'm going to go ahead and kick off a run of our hybrid solver. So, at this 350-vehicle scale, things start to get really complicated. The number of vehicles has grown quite a bit. By the way, the different background color is a different background color for all possible combinations. So, you can really start to get a sense of how many different combinations there are. I have to scroll for quite a while to get through all of this. And so, just the scale of the problem and the number of constraints that have come in are what are driving the complexity of this problem.
So, as I said, this is running live in our hybrid solver. Ah, and here comes the solution. That's good. The internet here cooperated with us. And here's the solution returned by it. So again, I can go ahead and filter by jumbos. And I see that I get these groups of four, and they have the equivalent separation. It's worth mentioning that that by itself is better than the solution that Ford Auto was using before was able to do. They weren't even able to express all of the constraints and get solutions that met all of them.
But in addition to that, they had an additional requirement, which is that they wanted to minimize the spikes and troughs — a prolonged period of extra work and a prolonged period of little work — because different vehicle combinations have different amounts of effort that go into it. And so here I show a comparison between a solution that's found classically using a greedy approach. And you see this big long spike in the top and this big long trough at the end, whereas our solution was able to find a nice even, as even as it's possible, a nice even sort of work gradient, where it's averaged over those last 20 vehicles.
So just to summarize, because it's always nice to put a summary on this: the thing that they were most interested in is our ability to encode all of their business-specific constraints to their problem. We were also able to make a nice even schedule. And by the way, just for fun, their old solution, the one that didn't work that well, took 30 minutes, and ours only took five. So, we were able to deliver quite a bit of value to them, and that's why they're moving into production with this application. And with that, I'm going to turn it back to Alan. Thanks.
Thank you. So, I was watching Alex here on stage about an hour ago, like frantically at the keyboard, and I walked up and said, 'Are you okay?' And he's like, 'I don't know.' So, we were having internet problems, but fortunately the demo gods were with us today. Good job, Alex. And this is really important because, as Alex said, all the demos that we've shown up until now at our Qubits conferences have really been benchmark demos, not the actual production application that one of our customers is using in production. This is the first time we're actually showing a complete application, and that's pretty exciting for us.
Okay. I want to talk now about a few more application areas. So I commented earlier that the quantum supremacy work is opening up some new application areas for us. One, and I spent a little bit of time on this, is in fact materials discovery, and there are various groups around the world that are looking at how to take that supremacy computation and turn it into a materials discovery platform. For example, there are researchers at Los Alamos National Lab that are looking at how you could create this kind of materials discovery platform.
But this computation is also finding its way into other application areas. And I commented on this as well a little while ago. The fact that the same computation that we used to demonstrate quantum supremacy can actually be used to create the first-ever quantum hashing function, which in and of itself is very important and potentially very powerful. And then the ability to take that hash function and use it to do proof of work and validation of proof of work in a blockchain, which opens up a whole new category of application opportunity for us as we start thinking about blockchain for general ledger or quantum proof of work underlying cryptocurrency.
At one point, you may have heard Elon Musk say we should put the US government on blockchain. Well, there's no way that could be done with classical GPUs. The energy consumption required to do a blockchain of that magnitude would just make it unrealistic. But now with quantum proof of work rather than classical proof of work underlying the blockchain, we're actually able to do this with very little power consumption. So what we're talking about here is a very energy-efficient blockchain. And we have built this, and we do have it running as a distributed application across four of our quantum computers today, where any one of the quantum computers can compute the hash function, any other of the quantum computers can validate the hash function, and this is used to actually build the blockchain.
And the image that you're seeing on the right is actually an example of how the blockchain gets built, where, you know, you may end up with different threads being created, and then you always converge onto the one authorized thread to create the blockchain. And this is work that we put on the archive and an application that we actually have running today across our systems. So, we're really excited about the opportunities we think that's going to open up for D-Wave.
And then, of course, there's AI and machine learning. And we also have a fair amount of work going on in this area. For example, we are looking at enhancing our Ocean toolset so that the AI/ML work we do could be integrated with the PyTorch software library, which is the Python AI/ML library. We are also looking at how you can use quantum and classical together to solve a particular use case. For example, you might use generative AI to predict product demand in the future, and then you'd use quantum to optimize a supply chain to meet that demand. So this is an example of quantum and AI working together to solve a problem.
particular problem. And then finally, and this is the one that in some sense we're most excited about, is the ability to basically insert the quantum processor in the middle of AI model training to be able to use it to build better models as well as to build models with far less energy consumption. And as you know, if you ask anybody what's the big problem with AI today, they'll say energy consumption. Like, you know, I need to buy a nuclear power plant. But with quantum, we believe that we will be able to dramatically reduce the energy consumption. For example, in the blockchain that I talked about a minute ago, that's about a 1,000x decrease in electricity consumption over the classical proof of work that's being done today. And we believe we can bring even more improvement in energy consumption to AI model training. And we're quite excited about that. But it's not just D-Wave that's working on this.
Um, I wanted to ask um uh Dr. Masaru Tateno to join me on stage. Uh he is a uh he is the chief scientific officer from Japan Tobacco. Please or maybe not. Ah there he comes. He was in the green room. Hey production team. Thank you. Are you [Applause] please? Okay. So, uh thank you for being here. I really appreciate taking the time to come uh all the way over from Tokyo. Um so, look, what we're interested in is trying to understand what you've been doing with the D-Wave systems and I know that you've been working on generative AI for molecular discovery, but maybe you can just spend a few minutes talking about the work that you've been doing.
Okay. Um in our drug design work um um we have to actually create the molecules, new molecules, um but uh um new molecules should be um satisfy the uh should satisfy some um molecular properties. Uh it's very strict because it's pharmaceutical drug. So um we have to um create the new molecules with um good molecular properties. So um to create the new um molecules we have to uh explore the very very huge molecular space compound world. So um quantum computer, quantum annealing computation um is very useful to uh efficiently and effectively um search for the um very huge chemical space.
So so the idea is that by leveraging the quantum computer you're able to come up with more and better options for um molecules in the drug arena. Yeah. And what's next? Where where are you going to take this work next?
Mhm. Yeah. Yeah. Um um to create the molecule is the um one issue for us and um at the same time we have to um satisfy much more um conditions. For example, um the molecule created should bind to one protein but not uh bind to other proteins. It makes the side effect. So um we have to um induce the facilitated selectivity. It's also a combinatorial optimization problem. So we can apply other programs.
Right. Well, Tateno-san, thank you very much for being here. It's great work and we're really looking forward to working with you as we move forward. Thank you very much. Thank you very much.
Okay, let's uh switch gears and talk about uh optimization and quantum optimization. Um from our perspective, optimization represents most of the important hard problems that businesses need to solve. Um this is everything from workforce scheduling to cargo loading to last mile routing. Even protein folding is an optimization problem. And we think the future of optimization is quantum optimization because these problems are all so hard computationally that you can't get typically the optimal solution. So most businesses are using heuristics to try to come up with what they hope are good enough solutions. But with quantum optimization in many cases we can provide better solutions faster. And we've worked with a variety of different companies uh over the years on problems in this area. Everything from Pattison Food Group for workforce scheduling which we've talked about in the past. Uh Alex just showed us the work that Ford Ottumna is doing on automotive body assembly and this is the newest application actually moving into production, um to NTT Docomo working on optimizing cell tower uh resources. We'll talk a bit more about that in a minute. Um but a broad array of applications are optimization applications and our annealing quantum computers are quite good at addressing those applications. And now with the improvements in our NL solver, there are a set of new applications that are enabled as a result of that. Everything from much better portfolio optimization to better offer allocation. And offer allocation is where you know you you know you've got promotional offers and you want to try to decide who to off which customers to offer them to to maximize the uptake, um to maintenance, repair and overhaul. Basically ensuring that you have the parts that you need to maintain your systems to keep the fleet running. And we're actually working with customers in every one of these areas as new optimization applications. But what I'd like to do now is call three people up on stage to talk about their experiences in working with D-Wave on uh quantum optimization applications. So uh first of all Jay Lol, who's uh principal senior technical fellow at Boeing. Jay [Applause]
Uh second is uh Dr. Shuichi Fukuda. Yes. Yeah. Did I get that? Okay. Please have a seat. Uh data scientist at NTT Docomo. Uh and then finally is Thomas Ree. Thomas, thank you. who is vice president uh of product and industry marketing at SAS. So each of you is at a different point in the quantum journey. Um Jay at Boeing is just getting started. Uh Thomas uh has been working with us and our systems for a while now but not yet in production. Uh and then Fukuda-san uh is from NTT Docomo and actually has progressed quite far uh with the quantum uh application and is actually using it today uh as part of their business operations. So really what I wanted to do was spend a little bit of time just getting some perspective on um you know what the journey has been like for each of you. So in your case Jay, you you're really just getting started. So maybe you could just spend a few minutes telling us what you're up to.
We're getting started and we've been in this for a long time. We're getting started with D-Wave looking in depth at the class of optimization problems that are affecting our business. These range from manufacturing examples such as you've shown earlier. Um although there are a few more parts in an airplane than there are in a van, um and manufact you know optimizing the roughly 1 million parts that come together into an airplane into a coherent supply chain that is robust against disruptions, delays and the like, um is really a vastly more complicated problem than we've ever taken on as an optimization problem within the company. And one of the reasons we're working with D-Wave is this depth of background in optimization and deep understanding of how they're formulated, where you can make good progress, and where you're going to have to kind of hold off and and maybe take a a longer term view. This is one of the things that we really came to work with D-Wave on, and we're really interested to see how that work develops over the next year.
Yeah. And and you know, I don't want to um kind of put words in your mouth, but but I know we got off to a little bit of a rocky start because early on, I think you had experimented with a variety of different quantum systems and just really weren't sure that they were ready for prime time yet. And we spent some time trying to work with you and your team to help help you get comfortable.
At the time we first spoke, we were sure they weren't ready for prime time for the kind of for the class of problems we were interested in. What our discussions with you have helped us learn is that the ceiling is higher than we'd expected and the floor is lower than we'd expected. And so that those discussions really are why we're here today.
Great. Great. And Thomas, you've uh you've been working with us for a little while now. Maybe you can talk about what you're up to.
Yeah, sure. So, we've been uh we've actually been um really focused on the quantum AI space for a couple of years now, but our um really our intent is is finding real life use cases that our customers can benefit from very similar to what you talked about earlier today. And um the work that D-Wave and SAS um are doing together is around a proof of concept that we have completed uh with a consumer products uh company. Now, this is a long-term uh longtime SAS customer um that we've been doing a number of things with around the area of forecasting, optimization, other advanced analytics and AI. Um in this particular case, um we took some SAS optimization capabilities um together with uh with D-Wave and and the quantum annealing uh capabilities that that you have there um to to help them solve a factory setup challenge. And um the results have been very impressive. Um as such um we're we're looking at um really doing some additional work together um as we move forward with this.
And and what I really like about this particular partnership is that um it's a three-way partnership, right? I mean, yeah, we're spending time working with you and your team, but then you're also working with your customer and, you know, that opens up some really interesting opportunities to kind of expand this in a much more significant fashion.
Yeah. Yeah. For sure. Yeah. The project is um the project entails um a a hybrid classic quantum um approach where we are um addressing a um integer linear modeling um to help optimize this production um very important production um setup uh process and what we've been able to do is you know take the the the SAS optimization capabilities together with the quantum annealing and um really just seen as I mentioned earlier a great result across really three areas. Um and that would be the quality, the feasibility and the speed. Um and in particular on the speed we've been able to take this process from over 10 hours down to minutes. Um and and that's quite impressive and and it it's going to result in some very significant financial uh savings and opportunities as they're able to more proactively uh set up for production runs. But more importantly is the quality had to be there. The accuracy had to be there. You know getting there quicker wasn't going to help us if we couldn't get to the precise number. Um and that's what we've been able to prove out is all three areas.
Great. And Fukuda-san. So, um, I love what you've done, although I don't love it, uh, as as much as I would if we had helped you get there. The point is, you actually built this application entirely on your own. D-Wave wasn't really involved in helping you build the application. And so that's fantastic because you know you were able to quickly get comfortable with our systems and services and the ability to actually build the app and build something that created value. On the other hand, we kind of like to sell our customers professional services to kind of help them build their applications. So um it's great what you've done. Uh and maybe you could just kind of spend a little bit of time talking about that. I'm I'm interested in hearing how you got started. Uh you know you know how difficult was it to actually start using our system and then what results did you obtain?
Uh I I'm working in the Docomo and it's a Japanese mobile operator and we are using the web server in comm uh telecommunication field and it is very interesting for me to optimize the base station setting uh because base station has a a lot of parameter to optimize the end for the customers experience, mobile experience. So it's really important for us to optimize the setting. Yes.
Mhm. And so what's next for you as you look to the future?
Uh we are sorry uh we are using the li server and next step uh maybe we are working for the optimization five optimization in telecom telecommunication field and uh many of them are going to be commercial networks. Yes.
And and Jay, how about you? Where do you see this going next?
Well, we've we've just started narrowing down the problem sets that we want to work on. Um I think the two most three most promising really are uh global satellite network um optimization. How do we optimize the the uh the constellation and array of satellites, the delivery of services to be disruptive or against disruption of ground services. Um that's one application that we've been looking at. Materials optimization, which is the one that we that got us started in our discussion, is still a as we've learned a tremendously rich area um as we expand it well beyond the toy problem that we'd looked at earlier. And then this manufacturing optimization, how do we make sure that the materials and the work packages follow all the constraints that are associated with each as we go and put all the different pieces and components of our airplane together during final assembly. Um, the thing I like about our team and our relationship is we're we really have a pretty strong optimization team within our own uh quantum we have a strong optimization team at Boeing. We have a strong quantum computing group at Boeing. We're bringing both of those teams to bear. And likewise, you're doing the same. Bringing a strong math optimization team, a strong team that understands the quantum technology. These are the kinds of the kinds of collaborations and and co-research activities that we feel, you know, really lead to tremendous outcomes and help us better understand that roadmap and journey that we're going to take in the future. Our focus is on solving those long range very difficult hard business problems and with that seeing the way to implement the easier kind of more tactical or quicker uh quicker implementations. I'd rather us work on the hard things because those are the things that create completely new capabilities for us as a company.
And you and I had a phone conversation a week or so ago and you said that one of the things you were interested in trying to understand is whether the quantum systems could actually help in understanding which of the input parameters the optimal solution is most sensitive to or least sensitive to.
When when we manufacture our airplanes, one of the most typical things we have, we have a production plan, but that production plan is rarely if ever actually followed to the letter. Um, something always happens. Somebody's sick for a day and so work gets done slower than it should. And that has a a follow-on effect, a wave of activities beyond that that are affected. Understanding and building optimal manufacturing flows that are disruption tolerant and under more to the point understanding what are the things that they're truly sensitive to and what are the things we have to mitigate or put controls in to ensure that they don't happen in order to maintain a more reliable more repeatable schedule from the company's perspective we would like to get every airplane out the door as fast as we possibly can that is safe and manufactured 100% properly. If we can if we can't do it the very fastest way, but we can guarantee that we'll get all of them out at 98% of that pace, we'll take that every day, every time.
And one of the things I liked about that comment is that our quantum systems are actually quite good at quickly providing a variety of solutions that are very close to optimal but quite diverse. And my my thought process was this could be actually used as you look across those samples to try to understand okay which of the parameters actually is the uh optimal solution most sensitive to and this is something that classical computers really struggle with.
