Quantum Computing is Real & Commercial Today, says D-Wave CEO

WEBVTT
00:00:00.150 --> 00:00:00.820
Hello, everyone.
00:00:00.820 --> 00:00:03.463
Welcome to another session of OptoSessions.
00:00:03.463 --> 00:00:08.147
I've got Alan Baratz on the call today, CEO of D-Wave.
00:00:08.147 --> 00:00:10.900
How are you doing, Alan?
00:00:10.971 --> 00:00:11.601
I'm great.
00:00:11.601 --> 00:00:13.617
Thanks for the opportunity to be here with you.
00:00:14.249 --> 00:00:17.623
Now we're here to talk about obviously D-Wave and quantum computing.
00:00:17.623 --> 00:00:25.821
A lot of people probably know about quantum computing from the news or something like this, but probably don't have a deep understanding of it.
00:00:25.821 --> 00:00:35.250
So if we just start from the beginning of, if you could give us an introduction to D-Wave and also just quantum computing in general and where it is today, that would be great.
00:00:35.856 --> 00:00:36.226
Sure.
00:00:36.226 --> 00:00:39.509
starting with the second question first, what is quantum computing?
00:00:39.509 --> 00:00:49.878
It's very simply using quantum mechanical effects to solve hard computational problems faster than they can be solved using classical computers.
00:00:49.878 --> 00:00:54.764
And these quantum mechanical effects are things like superposition, entanglement, tunneling.
00:00:54.764 --> 00:01:03.042
All of these quantum mechanical effects are used within our D-Wave quantum computers to help solve important problems.
00:01:03.042 --> 00:01:10.367
As far as D-Wave and what we do, we are a full stack quantum computing provider.
00:01:10.367 --> 00:01:26.177
We provide everything from the quantum computers, we design and manufacture them ourselves, to our quantum cloud service, which is currently the primary mechanism that our customers use to access our quantum systems to run their applications.
00:01:26.177 --> 00:01:42.712
to a complete suite of software development tools, all open source, so that customers can basically help us to evolve and improve the tool set through to professional services to help our customers build and deploy applications.
00:01:42.712 --> 00:02:04.701
But what really makes us unique in the quantum industry is that we are the first and only quantum computing company that is actually supporting customers today with applications leveraging our quantum systems in production as a part of their business operations.
00:02:04.701 --> 00:02:16.293
And so these include companies like Patterson Food Group or NTT Docomo or Ford Autoson that are all using our quantum systems and services today as a part of their business operations.
00:02:16.965 --> 00:02:17.776
Brilliant.
00:02:17.776 --> 00:02:30.841
And so when you say full stack, your basis, you make both the hardware and you have software that runs on top of it that people can use to build applications for in quantum computing.
00:02:31.146 --> 00:02:31.937
That's correct.
00:02:31.937 --> 00:02:35.461
So we design and manufacture and make available the quantum computers.
00:02:35.461 --> 00:02:38.824
Our current generation system is called Advantage.
00:02:38.824 --> 00:02:45.830
It is a 5,000 qubit quantum computer, the largest and most powerful quantum computer in the world.
00:02:45.830 --> 00:02:58.789
And as I said, we have a complete suite of software development tools that allows our customers to build and deploy applications that leverage those quantum systems.
00:02:58.789 --> 00:03:10.134
And I should also point out that those software development tools include access to what we call hybrid solvers that are part of our quantum cloud service.
00:03:10.134 --> 00:03:28.123
And so this is where we use both classical computing, CPUs and GPUs, together with the quantum computer to solve hard computational problems where we're basically allowing each processor to solve the part of the problem that's best suited.
00:03:28.123 --> 00:03:29.860
to the capabilities of that processor.
00:03:29.860 --> 00:03:33.645
And you mentioned there 5,000 qubits, I believe.
00:03:33.645 --> 00:03:38.731
Can you put that in just a comparison to traditional computing?
00:03:38.731 --> 00:03:42.087
What are we talking about in the order of magnitude difference?
00:03:42.087 --> 00:03:54.558
Well, first of all, let me point out that our quantum computers are much larger than any other quantum computers that are kind of being developed today.
00:03:54.558 --> 00:04:00.862
All of the other quantum systems are in the tens to a few hundred qubits.
00:04:00.862 --> 00:04:07.268
So our systems are in order of magnitude larger than any other quantum computer.
00:04:07.268 --> 00:04:15.356
Now, To put it in context versus classical computing, let me do it in terms of a specific application.
00:04:15.356 --> 00:04:41.341
We recently published in Nature a demonstration of our quantum computers solving a magnetic materials computation in minutes that would take nearly a million years to solve on frontier at Oak Ridge National Lob, which is one of the largest and most powerful supercomputers in the world.
00:04:41.341 --> 00:04:53.865
So at a few thousand qubits, our quantum computers are performing computations well beyond what's possible with classical computers, even the most powerful supercomputers in the world.
00:04:53.865 --> 00:05:01.646
Okay, so this is basically the next iteration of technology on top of supercomputers.
00:05:01.646 --> 00:05:03.375
So that's old technology now.
00:05:03.887 --> 00:05:06.759
Well, we like to think about it that way.
00:05:06.759 --> 00:05:10.591
Yeah, supercomputers are last year's technology.
00:05:10.591 --> 00:05:12.192
They're very capable.
00:05:12.192 --> 00:05:24.197
There are very important problems that can be solved on classical computers, whether they're CPUs or GPUs or massively parallel GPU systems.
00:05:24.197 --> 00:05:30.257
But there are also problems that cannot be practically solved on those systems.
00:05:30.257 --> 00:05:34.831
And that's where the next wave of quantum computing comes in.
00:05:34.831 --> 00:05:43.274
And I should make another point as well, which is that it's not just about the time to solve the problems.
00:05:43.274 --> 00:05:46.636
It's also about the power consumption.
00:05:46.636 --> 00:05:53.850
So as I think we are all aware now, GPUs are massive power hogs.
00:05:53.863 --> 00:06:07.923
And we hear about it every day that big tech is looking to buy nuclear power plants to kind of power these hyperscaler compute infrastructure environments.
00:06:08.663 --> 00:06:12.744
quantum computers are much more energy efficient.
00:06:12.744 --> 00:06:25.218
So that computation I referred to a few minutes ago in the area of magnetic materials, it took us about 12 kilowatts to perform that computation.
00:06:25.218 --> 00:06:33.653
If we were to perform it on frontier, it would take well over global annual energy consumption.
00:06:33.653 --> 00:06:39.290
So this is a much more energy efficient approach to computing as well.
00:06:39.302 --> 00:06:40.819
How does it achieve that?
00:06:40.819 --> 00:06:41.944
What makes it possible?
00:06:41.944 --> 00:06:47.307
So that's really in the way that quantum computers operate.
00:06:47.307 --> 00:06:56.492
As I said early on, quantum computing is all about using quantum mechanical effects to solve hard computational problems.
00:06:56.492 --> 00:07:02.954
And let me give you a very high level explanation of what I mean by that.
00:07:02.954 --> 00:07:08.240
If you think about current computers, information is stored in bits.
00:07:08.240 --> 00:07:13.745
and a bit can be either a zero or a one at any point in time.
00:07:13.745 --> 00:07:24.173
So what that means is that at any point in time, a classical computer is evaluating one possible solution to the problem.
00:07:24.192 --> 00:07:37.761
And the way classical computers work is they basically very intelligently iterate from one potential solution to a problem to the next, searching for the best possible solution.
00:07:37.761 --> 00:07:46.408
And algorithms are the term associated with that intelligent search process.
00:07:46.588 --> 00:07:51.913
Quantum computers use what we call qubits and superposition.
00:07:51.913 --> 00:08:02.420
And what this means is that at any point in time, qubits can be combinations of zero and one.
00:08:02.420 --> 00:08:22.684
So what that means is that at any point in time, A quantum computer is able to evaluate and operate on multiple possible solutions to the problem at the same time, allowing it to iterate its way to the best possible or correct solution much faster.
00:08:22.684 --> 00:08:24.516
And that's the way our quantum computers work.
00:08:24.516 --> 00:08:29.127
I want to dig into that more, the technology must be, I mean, it's obviously very advanced.
00:08:29.127 --> 00:08:35.006
Maybe I know I've seen that one of at least the things that you focus on is quantum annealing.
00:08:35.287 --> 00:08:39.265
Could you touch on that a bit and why you've chosen that?
00:08:39.716 --> 00:08:40.177
Yeah.
00:08:40.177 --> 00:08:45.763
So there are two main architectural approaches to quantum computing.
00:08:45.763 --> 00:08:56.351
One is called quantum annealing, as you just mentioned, and D-Wave is actually the only company in the world that provides annealing quantum computers.
00:08:56.351 --> 00:09:01.856
Everybody else is focused on gate model quantum computing.
00:09:01.856 --> 00:09:06.171
Now, there are some important differences between the two.
00:09:06.171 --> 00:09:14.953
Perhaps one of the most important differences is that annealing is a much easier technology to work with.
00:09:14.953 --> 00:09:20.115
It's easier to scale and it's much less sensitive to errors.
00:09:20.115 --> 00:09:38.263
And as a result, our quantum computers are an order of magnitude larger than any other quantum computer and they are able to deliver good if not optimal solutions to hard computational problems today without the need for error correction.
00:09:38.503 --> 00:09:46.965
On the gate model side, the technology is much more challenging and it's very, very sensitive to errors.
00:09:46.965 --> 00:09:59.370
In fact, there's no evidence at all that a gate model quantum computer will be able to solve a useful commercial problem without error correction.
00:09:59.370 --> 00:10:05.100
And we're still many years away from error correction on gate model systems.
00:10:05.100 --> 00:10:09.102
And that's why D-Wave is commercial today.
00:10:09.102 --> 00:10:14.886
That's why we have customers that are using our systems as a part of their business operations today.
00:10:14.886 --> 00:10:27.619
And that's just simply not possible with gate model systems because they're still at a very immature R &D stage where really all that's going on is research experimentation.
00:10:27.619 --> 00:10:31.143
So why haven't those companies chosen quantum annealing then?
00:10:31.143 --> 00:10:37.466
So there's a really interesting historical answer to that question.
00:10:37.546 --> 00:10:41.798
D-Wave began building a quantum computer over 15 years ago.
00:10:41.798 --> 00:10:45.370
The first company in the world to start building a quantum computer.
00:10:45.571 --> 00:11:00.383
At that point in time, the science, the engineering had not yet progressed to the point where there was line of sight to how you could build a gate model system.
00:11:00.383 --> 00:11:06.405
But there was line of sight to how you could build an annealing quantum computer.
00:11:06.405 --> 00:11:10.326
And so we decided to start with annealing quantum computing.
00:11:10.326 --> 00:11:14.091
In some sense, it's the only thing we could do at that point in time.
00:11:14.091 --> 00:11:21.495
Now, we knew that annealing would be very good at solving business optimization problems.
00:11:21.495 --> 00:11:26.962
Frankly, most of the important hard problems that businesses need to solve are business optimization problems.
00:11:26.962 --> 00:11:29.364
And we knew that annealing would be good at that.
00:11:29.364 --> 00:11:32.475
But we also knew that there were things it could not solve.
00:11:32.475 --> 00:11:40.967
For example, you can't use an annealing quantum computer for molecular discovery for quantum chemistry, for example.
00:11:41.307 --> 00:11:44.889
OK, so we decided to start with annealing.
00:11:44.889 --> 00:12:01.583
Eight or nine years ago, when everybody else decided to seriously pursue developing a quantum computer, the science and the engineering had progressed to the point where it was believed you probably could build a gate model system.
00:12:01.724 --> 00:12:17.009
And at that time, and this is a really important point, at that time, it was believed that a gate model quantum computer could perform all computations, could basically solve all types of problems better than classical.
00:12:17.009 --> 00:12:26.004
And so everybody else sort of adopted the point of view, well, We know that annealing is good at some things, but not all things.
00:12:26.004 --> 00:12:29.085
We believe gate model is good at all things.
00:12:29.085 --> 00:12:32.267
We now think we know how to build a gate model system.
00:12:32.267 --> 00:12:33.148
So let's go there.
00:12:33.148 --> 00:12:37.009
Now, fast forward to a couple of years ago.
00:12:37.009 --> 00:12:42.815
And this is a really kind of interesting twist of fate, at least for D-Wave.
00:12:42.815 --> 00:12:56.145
A couple of years ago, it was proven by researchers in both the US and Europe, that gate model quantum computers will likely never deliver a speed up on optimization problems.
00:12:56.145 --> 00:13:02.630
That you basically need an annealing system in order to address optimization problems.
00:13:02.630 --> 00:13:07.355
So now it's not that one is more general than the other.
00:13:07.355 --> 00:13:08.937
It's not that at all.
00:13:08.937 --> 00:13:13.620
It's that there are some applications that will always run best on annealing.
00:13:13.620 --> 00:13:17.061
And there are other applications that will always run best on gate.
00:13:17.061 --> 00:13:19.052
So you're always going to need both.
00:13:19.052 --> 00:13:32.605
But because of this kind of interesting twist of fate, D-Wave is really the only company that's doing annealing and can address that, frankly, very important portion of the market.
00:13:32.605 --> 00:13:38.330
So it's actually worked out quite well for Yeah, very good place to be in.
00:13:38.471 --> 00:13:41.875
If I can just go back to the range of offerings you've got.
00:13:41.875 --> 00:13:47.311
So you have the software that sits on top of the hardware that's optional, it?
00:13:47.311 --> 00:13:51.826
Because you also have a cloud solution that uses your own hardware that you have somewhere.
00:13:51.826 --> 00:13:52.775
Is that right?
00:13:52.815 --> 00:14:02.844
So the primary way that we and our customers access our quantum systems is through our cloud service.
00:14:03.085 --> 00:14:16.756
Now, we have recently sold one of our current generation advantage, we call our current systems advantage, one of our current generation advantage quantum computers to the Eulic Supercomputing Center in Germany.
00:14:16.879 --> 00:14:28.821
And we did that because they have an interest in tightly integrating our system with their new Jupyter exascale supercomputer.
00:14:28.821 --> 00:14:32.644
This is the first and only exascale supercomputer in Europe.
00:14:32.644 --> 00:14:43.375
And they want to tightly integrate it and explore a variety of workflows and use cases in the areas of AI and optimization.
00:14:43.442 --> 00:14:57.826
And in order to do that, they actually needed to own a system, have it on site, and be able to have control over more of the operating parameters of the system than we make available through our cloud service.
00:14:57.826 --> 00:15:05.669
But for customers that are simply interested in running their business applications, the cloud service is the way to go.
00:15:05.669 --> 00:15:09.591
It's a much more cost-effective approach to accessing our quantum systems.
00:15:09.591 --> 00:15:12.578
So we actually now have two...
00:15:12.578 --> 00:15:16.011
components to our go-to-market model.
00:15:16.011 --> 00:15:18.653
One of them is the cloud-based service.
00:15:18.653 --> 00:15:21.075
We call it quantum compute as a service.
00:15:21.075 --> 00:15:27.418
And that's ideal for commercial companies that are just simply interested in running their business applications.
00:15:27.418 --> 00:15:29.860
And the other is system sale.
00:15:29.860 --> 00:15:40.280
And that's more for supercomputing centers, government labs that are really interested in tighter integration with other systems.
00:15:40.280 --> 00:15:43.703
and controlling more of the operating parameters of the quantum.
00:15:43.703 --> 00:15:49.489
And do your customers have to build applications specifically to sit on top of the quantum computing?
00:15:49.489 --> 00:15:55.152
And at the same time, I believe you offer your own applications that people can use as well.
00:15:55.779 --> 00:15:59.043
So we actually do not sell applications.
00:15:59.043 --> 00:16:02.403
We sell access to our quantum systems.
00:16:02.403 --> 00:16:04.524
We sell our quantum systems.
00:16:04.524 --> 00:16:14.609
And then we sell professional services to help our customers build and deploy applications that are of value to them.
00:16:14.609 --> 00:16:26.543
And so as a result, we've helped many companies evaluate applications with respect to what kinds of benefits they could achieve by leveraging our quantum systems.
00:16:26.543 --> 00:16:37.052
And then for those where they see a sound ROI on the application, we've helped them actually build and benchmark and deploy those applications.
00:16:37.052 --> 00:16:53.927
So for example, Ford Autoson, which is a joint venture of Ford Motor Company and Koch Holdings in Turkey, we worked with them to develop an application for scheduling the assembly of automotive bodies.
00:16:54.248 --> 00:17:11.782
And what they found was that by using our quantum computers, we could significantly reduce the time to do the scheduling, which meant that they could adapt in real time to changes throughout the day.
00:17:11.782 --> 00:17:13.670
And that was important to them.
00:17:13.670 --> 00:17:16.038
is now moving into production at Ford Autosign.
00:17:16.038 --> 00:17:21.094
And do these applications typically use AI as part of what they're doing?
00:17:21.094 --> 00:17:23.288
And if or not.
00:17:23.926 --> 00:17:25.318
So they don't.
00:17:25.318 --> 00:17:39.394
The applications that are in production today at customer sites are basically using quantum systems to solve hard optimization problems.
00:17:39.394 --> 00:17:48.740
However, we are working on ways to integrate our quantum systems within AI environments.
00:17:48.740 --> 00:17:51.864
And there are two areas that we're looking at.
00:17:51.864 --> 00:18:02.441
One is basically using the two capabilities together, letting each solve the portion of the problem that it's best at solving.
00:18:02.441 --> 00:18:12.964
For example, you might use AI to predict future product demand and then use quantum to optimize the supply chain.
00:18:12.964 --> 00:18:14.556
to meet that demand.
00:18:14.556 --> 00:18:23.182
Here, we're using AI and quantum together, each addressing a different portion of the use case, the portion that it's best at addressing.
00:18:23.282 --> 00:18:40.844
The second area that we are working on, and this is potentially very, very transformative for AI, would be to insert the quantum computer into the AI model training and inference workflows.
00:18:40.844 --> 00:18:42.413
Why would you do that?
00:18:42.413 --> 00:18:47.596
This is back to my comment earlier about energy efficiency.
00:18:47.596 --> 00:18:51.898
So what's the biggest issue right now with AI model training?
00:18:51.898 --> 00:18:53.640
It's power consumption.
00:18:53.640 --> 00:19:03.684
However, what we've started working on, and we have early evidence that this could be a very transformative capability.
00:19:03.684 --> 00:19:21.688
What we started working on is using the quantum computer to perform some of the computations in model training together with the GPUs, where the two working together can use dramatically less electricity to perform the model training.
00:19:21.688 --> 00:19:26.633
And we think that this could be very transformative for the AI industry.
00:19:26.633 --> 00:19:30.894
And we're quite excited about Awesome.
00:19:30.894 --> 00:19:34.098
Are you exploring anything in that realm yet, or is that still early stages?
00:19:34.416 --> 00:19:37.609
no, we actually have work going on in that arena.
00:19:37.609 --> 00:19:42.733
We've actually developed some of the early capabilities.
00:19:42.733 --> 00:19:47.537
We've used it on smaller data sets and we're seeing good results.
00:19:47.537 --> 00:19:56.473
We're now in the process of basically starting to work with much larger data sets to essentially prove out the capability and the value.
00:19:56.518 --> 00:19:56.868
Amazing.
00:19:56.868 --> 00:20:00.382
I wonder if we can just touch on a few more of the real world applications.
00:20:00.382 --> 00:20:08.805
You've already talked about a couple already, but you've announced quite a few major partnerships recently from insurance optimization to autonomous agriculture.
00:20:08.805 --> 00:20:13.288
What are some of the most compelling commercial use cases that you're working on today?
00:20:13.765 --> 00:20:23.413
Yeah, so first of all, let me take just a brief step back and provide some context for all the commercial work that we are doing.
00:20:23.413 --> 00:20:39.557
So there's a lot of confusion in the marketplace with respect to whether quantum computers can actually deliver value today on important real world applications or not.
00:20:39.557 --> 00:20:54.669
Unfortunately, too many of the quantum computing companies are dramatically overstating what they are capable of delivering in that arena.
00:20:54.669 --> 00:21:01.473
So what we wanted to do was to help customers separate the hype from the reality.
00:21:01.473 --> 00:21:09.078
And as a result, what we did was we developed a framework that we call quantum realized.
00:21:09.325 --> 00:21:19.906
basically what you should look for in a quantum provider if you really want to have confidence that you're going to be able to get value.
00:21:20.228 --> 00:21:22.809
And there are three components to that framework.
00:21:22.809 --> 00:21:35.338
First, have the systems provided by the quantum company been able to solve problems better than they can be solved classically?
00:21:35.759 --> 00:21:36.820
Simple thing.
00:21:36.820 --> 00:21:47.201
But the truth of the matter is you probably won't find a gate model quantum computer where you can affirmatively answer that question.
00:21:47.260 --> 00:22:00.901
But for D-Wave, the problem that I described to you a little while ago in magnetic materials, this is a problem that's useful, important and useful, that we are solving in minutes that essentially can't be solved classically.
00:22:00.901 --> 00:22:07.534
So does the vendor have quantum computers that can solve a problem better than it can be solved classically.
00:22:07.534 --> 00:22:19.701
Second, are the systems from the vendor reliable and available enough for you to actually be able to depend on those systems as a part of your business operations?
00:22:19.701 --> 00:22:28.267
And in our case, when you access our systems through our Quantum Cloud Service, we have 99.9 % availability.
00:22:28.267 --> 00:22:32.509
And in fact, we offer service level agreements to our customers.
00:22:32.614 --> 00:22:38.276
We believe we're the only quantum company in the world that actually offers service level agreements.
00:22:38.276 --> 00:22:55.000
And in fact, if you go out and look at the public data around reliability and availability of other quantum systems, you'll see things like 30 % availability, 70 % availability, one week queuing times before a problem will be run.
00:22:55.000 --> 00:22:59.688
mean, our problems run in real time with very high availability.
00:22:59.688 --> 00:23:01.556
So that's the second part.
00:23:01.556 --> 00:23:13.624
And then the third point in the framework is, is the quantum company actually working with customers on real commercial applications and or that are in production?
00:23:13.684 --> 00:23:20.830
And we've talked about Ford Utterson, for example, which has an application on AutoVisit Assembly in production today.
00:23:20.830 --> 00:23:29.173
So those three elements we think are what any company should be looking for and asking about.
00:23:29.173 --> 00:23:31.474
when they start talking to a quantum company?
00:23:31.474 --> 00:23:34.797
Can your system solve problems better than they can be solved classically?
00:23:34.797 --> 00:23:37.859
Are your systems highly reliable and available?
00:23:37.859 --> 00:23:43.604
And do you have examples of where you've worked with commercial companies to be able to deliver value?
00:23:43.604 --> 00:23:46.645
And we can answer yes to all three of those questions.
00:23:46.645 --> 00:23:48.527
Now, some examples.
00:23:48.527 --> 00:23:50.268
I talked about Ford Autosign.
00:23:50.268 --> 00:23:53.010
Another example is NTT Docomo.
00:23:53.010 --> 00:23:59.734
We've worked with NTT Docomo on optimizing cell tower resource utilization.
00:23:59.734 --> 00:24:02.276
And there are two elements to the results.
00:24:02.276 --> 00:24:13.079
One, we've been able to perform this optimization computation in 40 seconds, whereas it had been taking them 27 hours.
00:24:13.421 --> 00:24:32.207
And two, leveraging the solutions that come out of our quantum systems, each cell tower can support up to 15 % more cell phones, which means lower infrastructure costs for NTT Docomo, which is a significant benefit to them.
00:24:32.946 --> 00:24:42.346
Another example is Patterson Food Group, a Canadian grocery chain, which is basically leveraging us for two different applications.
00:24:42.346 --> 00:24:55.888
One is for workforce scheduling, taking a computation that was taking 25 hours a week down to minutes a week, and then last mile routing for e-commerce delivery.
00:24:55.888 --> 00:25:02.662
So these are some examples of customers that are actually using our systems today in production.
00:25:02.662 --> 00:25:05.722
There are other customers that we are working with.
00:25:05.722 --> 00:25:08.144
For example, BBVA.
00:25:08.144 --> 00:25:16.989
We're working with them on financial portfolio optimization, leveraging both AI and quantum computing.
00:25:16.989 --> 00:25:46.220
We've talked about some work that we did with the pharmaceutical division of Japan Tobacco, where we've been able to help them enhance their large language model training in a way where they are now able to basically compute better new molecular structures than with the models that were developed without quantum.
00:25:46.220 --> 00:25:49.721
So a fairly broad array of application areas.
00:25:50.397 --> 00:26:02.778
And it's the main benefit at the moment to do with you've just got the power now to do things that may have taken years before can actually be done in a reasonable amount of time.
00:26:02.778 --> 00:26:05.337
Is this the main benefit?
00:26:05.337 --> 00:26:08.857
So things they wouldn't even try to explore, they can now do.
00:26:09.836 --> 00:26:19.912
So I think of the benefit as falling into two categories, what I call evolutionary applications and what I call revolutionary applications.
00:26:20.814 --> 00:26:24.957
Evolutionary applications are applications like the ones I've just been describing.
00:26:24.957 --> 00:26:28.921
These are computations that companies are performing today.
00:26:29.560 --> 00:26:44.670
It's just that they're computationally so hard that simplifications of the problem and heuristics are being used to try to come up with what the company hopes are good enough solutions, but they're not getting to optimal solution.
00:26:44.670 --> 00:26:50.872
And they have no way to determine how close to optimal the solutions actually are.
00:26:50.893 --> 00:26:56.795
But with our quantum computers, we can deliver better, if not optimal solutions.
00:26:56.795 --> 00:26:59.037
And so there's a business benefit to that.
00:26:59.037 --> 00:27:02.819
So this isn't about attacking problems that can't be solved classically.
00:27:02.819 --> 00:27:08.334
It's about solving current problems better than they are being solved today.
00:27:08.815 --> 00:27:10.996
Then there are the revolutionary applications.
00:27:10.996 --> 00:27:14.636
These are things that can't be addressed today.
00:27:14.636 --> 00:27:32.648
So for example, computing properties of magnetic materials so that I can do materials discovery computationally without needing to go actually fabricate the material and test it to allow me to iterate more rapidly.
00:27:32.648 --> 00:27:35.872
to materials that may meet my needs.
00:27:36.253 --> 00:27:48.901
This is an example of a computational area where we can't even attack it today because classical doesn't have the horsepower to get anywhere near good enough solutions.
00:27:48.901 --> 00:27:54.579
I'm going to your business momentum, et cetera, in a second.
00:27:54.579 --> 00:27:57.890
I can't believe more companies aren't using this yet.
00:27:59.031 --> 00:28:02.055
Well, I will say that we're making really good progress.
00:28:02.055 --> 00:28:10.644
uh For example, we just had our yearly customer conference last week, actually, in Scottsdale, Arizona.
00:28:10.644 --> 00:28:15.269
Attendance was roughly double what we had had the year before.
00:28:15.269 --> 00:28:18.516
We laid out our product roadmap for the future.
00:28:18.516 --> 00:28:19.935
which is very exciting.
00:28:19.935 --> 00:28:21.576
We're at 5,000 qubits today.
00:28:21.576 --> 00:28:26.634
We talked about how we're gonna get to 100,000 qubits in the relatively near term.
00:28:26.634 --> 00:28:33.756
And then we had a whole host of customers talking about how they are using their systems today.
00:28:33.756 --> 00:28:40.935
And so the hallway discussion was very different from our customer conference last year.
00:28:40.935 --> 00:28:47.059
Last year, the hallway discussion was kind of along the lines of, well, how does this really work?
00:28:47.059 --> 00:28:50.708
Do you think that there's a way that I could get some value out of these quantum systems?
00:28:50.708 --> 00:28:56.528
Whereas this year, the hallway discussion was, okay, how can we put an agreement in place to get started?
00:28:56.528 --> 00:29:06.143
Do you believe we're at an inflection point where people, you know, value behind quantum computing is starting to be realized?
00:29:06.143 --> 00:29:11.891
I mean, maybe just by your company that's started to offer these sort of benefits to companies.
00:29:12.219 --> 00:29:21.928
I absolutely do in the sense that we have a rapidly growing pipeline of customers and customer opportunities that we are pursuing.
00:29:21.928 --> 00:29:31.055
As a result, we have started investing in growing our go-to-market team and our professional services team to meet that demand.
00:29:31.055 --> 00:29:40.407
And then in addition to that, we are working on some very interesting and powerful transformative application.
00:29:40.407 --> 00:29:43.079
So I talked about the supremacy result.
00:29:43.240 --> 00:29:46.762
I talked about AI and improving model training.
00:29:46.762 --> 00:29:50.066
That's very transformative application area.
00:29:50.066 --> 00:29:57.594
We're still probably a year or two away from being able to actually deliver a product that provides value in that arena.
00:29:57.594 --> 00:29:59.817
But another area is blockchain.
00:29:59.817 --> 00:30:10.445
Based on the computation that we perform for quantum supremacy, we have been able to create a quantum hashing function.
00:30:10.445 --> 00:30:14.890
So hashing is kind of at the core of blockchain, right?
00:30:14.890 --> 00:30:20.742
It's all about computing hash functions and then proof of work is validating the hash functions.
00:30:20.742 --> 00:30:25.156
And that's how you basically perform transactions against the blockchain.
00:30:25.156 --> 00:30:30.070
And what we have running today, Prototype distributed quantum application.
00:30:30.070 --> 00:31:06.222
This is probably the first distributed quantum application across several of our quantum computers where we are using them to basically implement a blockchain where we are computing hash functions and validating hash functions and as a result adding adding things to the blockchain as They should be added with a high degree of security and most importantly very energy efficient because that quantum proof of work uses about one one thousand the electricity of current blockchain classical proof of work.
00:31:06.222 --> 00:31:19.445
And so we think that, in addition to all the commercial stuff that I've been talking about, we have a couple of very, very transformative applications on the horizon in the area of blockchain and AI.
00:31:19.445 --> 00:31:24.319
And you mentioned going from 5,000 qubits to 100,000.
00:31:24.569 --> 00:31:27.805
What sort of problems does that unlock that you can solve?
00:31:27.805 --> 00:31:31.440
Because that's a dramatic shift in the power.
00:31:31.799 --> 00:31:41.084
Yeah, so first of all, it's probably going to be much easier for us to see that kind of rapid growth than anybody else for the following reason.
00:31:41.084 --> 00:31:53.508
One of the things that we have developed for our annealing quantum computers, and it took us 15 years to get there, is what we call cryogenic control.
00:31:53.508 --> 00:31:55.868
So we use superconducting technology.
00:31:55.868 --> 00:32:04.416
So that means that our quantum processors live in dilution refrigerators and operate at near absolute zero.
00:32:04.617 --> 00:32:08.038
Now, you need to control those chips.
00:32:08.079 --> 00:32:17.726
So you need control signals coming from the outside world at room temperature and going down into the refrigerator and controlling the chips.
00:32:18.686 --> 00:32:30.122
What everybody else working on superconducting quantum computers is doing is they're performing all the control functions at room temperature and then sending the data down.
00:32:30.122 --> 00:32:35.247
What this means is they need multiple IO lines per qubit.
00:32:35.666 --> 00:32:38.098
Two, three, four IO lines per qubit.
00:32:38.098 --> 00:32:39.710
You cannot scale that.
00:32:39.710 --> 00:32:42.951
We, on the other hand, have on-chip control.
00:32:42.951 --> 00:32:46.756
We have digital analog converters on our chip in the refrigerator.
00:32:46.756 --> 00:32:48.417
We have on-chip addressing.
00:32:48.417 --> 00:32:50.519
We have on-chip data pipelining.
00:32:50.519 --> 00:33:00.367
And as a result, we are able to control our 5,000 qubit processor with only about 200 IO lines, okay?
00:33:00.367 --> 00:33:04.448
Not 10,000, 15,000, 20,000 IO lines.
00:33:04.448 --> 00:33:11.508
And this is critical, because as you scale to 100,000, you cannot have one, two, or three IO lines per qubit.
00:33:11.508 --> 00:33:24.567
We believe that with the current cryogenic control capabilities that we have with some minor modifications, that we can support the 100,000 qubits with still just 200 IO lines.
00:33:24.567 --> 00:33:35.048
So first of all, we have already most of what we need in the way of control structures to be able to control these larger processors.
00:33:35.048 --> 00:33:42.807
What we need to do is to be able to build the processor that has that number of qubits.
00:33:42.807 --> 00:33:44.448
Well, how do we do that?
00:33:44.448 --> 00:33:49.471
Up until now, it's always been about more qubits on a chip.
00:33:49.471 --> 00:33:54.071
went from 500 qubits to 1,000 qubits to 2,000 qubits to 5,000 qubits.
00:33:54.071 --> 00:33:56.092
And we still got some headroom there.
00:33:56.092 --> 00:34:00.311
But what we are now focused on is multi-chip.
00:34:00.332 --> 00:34:07.951
Basically, taking our, if you like, 5,000 qubit processors and gluing them together.
00:34:07.951 --> 00:34:09.532
So think about it.
00:34:09.532 --> 00:34:15.824
If I glue four 5,000 qubit processors together, I'm already at 20,000 qubits.
00:34:15.824 --> 00:34:21.070
And I already have the control structure to control that number of qubits.
00:34:21.070 --> 00:34:31.039
So the challenge for us is really in how we interconnect the qubits and maintain the quantum mechanical properties across the chips.
00:34:31.039 --> 00:34:33.481
But we believe we know how to do that.
00:34:33.481 --> 00:34:37.096
And we're working on prototype versions of that in our lab today.
00:34:37.096 --> 00:34:38.556
I didn't.
00:34:38.996 --> 00:34:42.655
And you mentioned that the deal, your pipeline of deals is growing.
00:34:42.655 --> 00:34:44.476
think I've got a stat here.
00:34:44.476 --> 00:34:48.715
It's grown over 500 % year on year, which obviously a huge signal.
00:34:48.940 --> 00:34:57.559
Well, to be clear, that 500 % was Q4 over Q4 bookings growth.
00:34:57.559 --> 00:35:06.791
So if we look at our bookings in Q4 of 23 versus our bookings in Q4 of 2024, we had 500 % growth.
00:35:06.971 --> 00:35:08.451
So bookings are growth.
00:35:08.994 --> 00:35:18.106
And why, because you mentioned we're at this inflection point, but why do you think suddenly you're getting more customer bookings come in?
00:35:18.427 --> 00:35:25.574
Is it just because you've been more real world use cases, people starting to see the examples of how it's working and it's just building on that momentum?
00:35:25.786 --> 00:35:31.076
Part of it is that we have quantum systems that are now capable of solving these important useful problems.
00:35:31.076 --> 00:35:37.963
And part of it is that as we are demonstrating success with customers, word is getting out.
00:35:37.963 --> 00:35:44.630
So is your technology advanced recently, quite significantly, that's opened up these opportunities?
00:35:47.373 --> 00:35:52.318
And you currently serve a mix of commercial, government and research customers.
00:35:52.338 --> 00:35:57.373
Which segment do you believe will be your primary revenue driver over the years to come?
00:35:57.690 --> 00:36:02.210
Well, to be clear, most of our business is commercial.
00:36:02.490 --> 00:36:06.911
So we do disclose a number of customers.
00:36:07.731 --> 00:36:15.771
In our last earnings call, we disclosed that we have over 100 total customers.
00:36:16.170 --> 00:36:22.451
Over 70 of them are commercial versus government or education.
00:36:22.451 --> 00:36:27.570
And of the 70, over 20 are Forbes Global 2000 companies.
00:36:27.844 --> 00:36:32.961
So we are far more commercially oriented right now.
00:36:32.961 --> 00:36:44.411
However, as we start moving into system sales more aggressively and looking at supercomputing centers and national labs, we'll start to pick up more of a kind of government component as well.
00:36:45.235 --> 00:36:53.070
And in the commercial sector, are there certain sectors or industries that you see get the most value back from your systems?
00:36:53.070 --> 00:36:53.360
Yeah.
00:36:53.360 --> 00:37:03.646
So our quantum systems today are most well suited to problems in the area of resource allocation and resource scheduling.
00:37:03.766 --> 00:37:09.489
And they fall primarily into the supply chain logistics and manufacturing verticals.
00:37:09.489 --> 00:37:13.233
So that's kind of where our primary focus is.
00:37:13.233 --> 00:37:32.800
mean, obviously we have customers that come in and from different areas that are interested in working with us and we will work with them, but Our primary focus on outreach is in supply chain logistics and And I wanted to ask a question about the recent tariffs from China and America.
00:37:32.880 --> 00:37:35.880
How has that impacted your business?
00:37:35.880 --> 00:37:39.840
you see this getting resolved over the near term, possibly?
00:37:40.309 --> 00:37:44.929
So it has not impacted our business and will not impact our business.
00:37:44.929 --> 00:37:48.068
And the reason is, first of all, we don't sell to China.
00:37:48.268 --> 00:37:52.568
Quantum computing is a sensitive technology for the US government.
00:37:52.568 --> 00:37:56.489
And so we just don't sell to China.
00:37:56.668 --> 00:38:08.644
Secondly, if you look at the bill of materials associated with our quantum computers, less than 10 % of the cost is parts from China.
00:38:08.644 --> 00:38:11.226
These are low tech parts, things like connectors.
00:38:11.226 --> 00:38:18.034
So even if there was 2X, 3X, 4X tariffs, it just would not change the economics for us.
00:38:18.034 --> 00:38:21.237
Plus, if we had to, we could pretty easily design them out.
00:38:21.237 --> 00:38:24.262
So it's just not an issue for us.
00:38:24.730 --> 00:38:27.938
And is it, am I right saying everything's built in the USA then?
00:38:28.746 --> 00:38:31.150
So we're built in North America.
00:38:31.150 --> 00:38:36.615
All of our processor fabrication work is done in the US.
00:38:36.615 --> 00:38:40.840
However, assembly of the systems is done in Canada.
00:38:40.840 --> 00:38:50.027
And the reason is the company was originally founded in Canada 20 years ago, and we saw the significant R &D presence there.
00:38:50.027 --> 00:38:54.371
We are now a US company headquartered in Palo Alto, California.
00:38:54.393 --> 00:39:06.909
But we still have manufacturing presence, an R &D and manufacturing presence in That's really good, really impressive to see that you can build that in North America nowadays and still make this amazing technology.
00:39:06.909 --> 00:39:18.329
And we've touched a little bit on the future, but maybe if there's other things we haven't particularly talked about yet, what's your vision for quantum computing in the next 10 years?
00:39:18.329 --> 00:39:21.009
How's it going to be?
00:39:21.009 --> 00:39:23.028
Where is it going to be in the enterprise?
00:39:23.028 --> 00:39:26.255
Do you see us across most of the big companies in the world?
00:39:26.255 --> 00:39:37.737
So first of all, for D-Wave, obviously, we're focused on continuing to execute against our roadmap, which includes always larger processors.
00:39:37.737 --> 00:39:42.510
I talked about a roadmap to 100,000 qubits, more connected processors.
00:39:42.510 --> 00:39:45.893
In our Advantage system, each qubit is connected to 15 others.
00:39:45.893 --> 00:39:49.534
In our Advantage 2 system, each qubit is connected to 20 others.
00:39:49.628 --> 00:39:53.949
More qubits, more connectivity means we can solve larger and more complex problems.
00:39:53.949 --> 00:39:57.070
We're also always focused on increasing coherence time.
00:39:57.070 --> 00:40:03.534
Coherence time is the time that the system remains in the quantum mechanical state, unimpacted by the external environment.
00:40:03.534 --> 00:40:11.978
And our Advantage 2 system has two times the coherence over our Advantage system, which means we can solve problems faster.
00:40:11.978 --> 00:40:14.759
And then increasing what we call energy scale.
00:40:14.759 --> 00:40:19.248
This is like the precision with which we can specify problem parameters.
00:40:19.248 --> 00:40:26.170
And Advantage 2 has a 40 % increase in energy scale over Advantage, which means more accurate solutions to the problem.
00:40:26.170 --> 00:40:37.961
So it's always about larger and more connected processors to solve larger problems, longer coherence times to solve them faster, and higher energy scale to deliver better solutions.
00:40:37.961 --> 00:40:44.865
And we've got a multi-year roadmap for Advantage 2, Advantage 3, Advantage 4 that we'll be executing against that.
00:40:44.867 --> 00:40:48.371
Then we are also building a gate model.
00:40:48.371 --> 00:40:49.461
quantum computer.
00:40:49.461 --> 00:40:54.746
The reason why we're doing that is because, as I said up front, there are problems that do require gate.
00:40:54.746 --> 00:40:56.447
There are problems that require annealing.
00:40:56.447 --> 00:40:58.108
There are problems that require gate.
00:40:58.108 --> 00:41:03.333
And we want to be able to address the full set of use cases for our customers.
00:41:03.333 --> 00:41:11.380
And much of the technology that we develop for annealing quantum computers is applicable to a scale error corrected gate model system.
00:41:11.380 --> 00:41:15.413
For example, all that cryogenic control that I talked about before.
00:41:15.413 --> 00:41:23.806
We're the only company in the world that has that capability, and that's going to be required for a scale, error corrected, superconducting gate model system.
00:41:23.806 --> 00:41:28.070
And so we are also working for that, and that is also part of our roadmap.
00:41:28.070 --> 00:41:34.773
And these things will roll out over the course of the next three years, five years, seven years, 10 years.
00:41:34.773 --> 00:41:51.445
But in general, I think that if we look out over the 10 year time frame, that obviously annealing will continue to kind of be critical infrastructure for the class of problems it's good at, especially in the optimization area.
00:41:51.445 --> 00:41:57.731
We'll finally start to see some gate model systems that can do useful commercial work.
00:41:57.731 --> 00:42:05.130
And then I think we will be amazed at some of the things that quantum computers are helping us do.
00:42:05.130 --> 00:42:10.753
particularly in the area of blockchain and AI and dramatically lower energy consumption.
00:42:10.753 --> 00:42:11.344
Well, thanks, Alan.
00:42:11.344 --> 00:42:15.414
That's it's been really, really interesting to talk about quantum with you today.
00:42:15.414 --> 00:42:33.155
And I'm sure it's going to be interesting for the audience as well, because a lot of people still don't know a lot about it, because it is quite relatively nascent in terms of visibility across, you know, even just the if you're just the main company offers it today, you know, it's it's great to have you on the show and been able to share those insights with everyone.
00:42:33.155 --> 00:42:36.273
So I don't know if there's anything you'd like to leave the audience with before we wrap up.
00:42:36.273 --> 00:42:37.922
But thanks again for your time.
00:42:38.581 --> 00:42:40.842
So first of all, thank you for the opportunity to be here.
00:42:40.842 --> 00:42:42.603
I really enjoyed the conversation.
00:42:42.603 --> 00:42:50.226
The only thing I will kind of reinforce is that quantum is real and commercial today.
00:42:50.226 --> 00:42:57.447
We have a framework called Quantum Realize to help anybody pick solution providers to help them with that.
00:42:57.447 --> 00:43:01.088
And we're excited to be leading the charge in this environment.
00:43:01.581 --> 00:43:02.614
Thanks, Alan.
00:43:02.717 --> 00:43:03.519
Speak to you soon.
00:43:03.519 --> 00:43:04.400
Bye.
00:43:04.518 --> 00:43:05.400
Bye-bye.