Before we get started, I'd just like to read the research disclosure. For important disclosures, please see the Morgan Stanley Research Disclosure website at www.morganstanley.com/researchdisclosures. Here today at the Morgan Stanley TMT conference, we have IonQ. To start things off, can you give the audience an introduction to the company, maybe starting with its history and then the differentiated approach the company takes to quantum computing?
Absolutely. First off, my name is Thomas Kramer, I'm the CFO. Quantum computing is a kind of computing that promises to solve problem sets that heretofore were either impossible to solve or not possible to solve in a timeframe that makes it viable. Examples are, solar cells can capture roughly like 23% of the sun's rays and convert them into energy. A simple plant can capture roughly 80% of that energy and do it much more efficiently. We have this answer already existing in nature, but we don't know how to harness it. The reason is because we can't actually model photosynthesis, because the problem set is too complex for classical computers. That's the advent of quantum computing. Today, there are two basic technologies, architectures that do this.
There is superconducting, and there is ion trapping. Superconducting is taking the a classical computer. It's like all of our laptops here. It's silicon substrate. You run some lines and power, and you compute in it. Wouldn't it be great if you can take those 60 years of learning and just drop in some qubits and thereby have a quantum computer? Turns out that that is hard because qubits are notoriously fickle. A qubit is the basic building block of a quantum computer, just like a transistor is the basic building block of a classical computer. It's a bit or a qubit. The difference is that a transistor can hold a value of zero or one, and only the value or of zero or one.
A qubit can famously hold the value of zero and one at the same time. In fact, in reality it holds all the values between zero and one, and then it has a probability distribution among them. That means that you can perform much richer calculations. The way our company got initially started, the spawn of our company is that Christopher Monroe who's working on the world's first atomic clock, realized that this can be useful in computing. He used that same technology that today is how we all keep time because it's very, very stable, and he performed the first quantum gates, and that became the advent of what we are doing. Now our two founders, Chris Monroe and Jungsang Kim, together have 25 years of quantum technology experience.
Great. You know, now on the product side, can you give us an idea of where you guys have come from as far as the product roadmap and then, you know, what we should expect over the next couple of years?
Absolutely. We used to, like Moore's Law of ever more computing power, cheaper, and today there are hundreds of thousands of transistors in all of our computers, actually in your iPhone. You don't need that many qubits for a computer to actually a quantum computer to be actually useful. Most people consider that if you have 70 functional logical qubits, that's the equivalent compute power of today's leading supercomputers. Currently, we have the industry's leading number of qubits, so useful qubits. We call them Algorithmic Qubits. Over time, these will be known as logical qubits. We have 25 of them functioning in one computer, and that is many times more than what you will see in anybody else. Our goal is to go from the 25 we delivered last year in fiscal '22 to deliver 29 Algorithmic Qubits by the end of fiscal '23.
I wanna stay with algorithmic qubits for a minute. You know, I think something IonQ does extremely well is, you know, helps the average person understand that not all qubits are created equal. We often say that once we get to a certain qubit count, we can do things like beat the world's best high-performance computer. You know, some of your competitors have published research saying we're at a certain count that maybe surpasses that, but we haven't achieved the performance. You know, give us a better idea of how you guys approach error correction, you know, and how seriously you guys take this idea of algorithmic qubits or qubits that are, you know, actually computationally available.
Right. You need to have a way to compare what a quantum computer does, and you're exactly right. Algorithmic Qubits is that basic measurement. If you look at different quantum computers, and you run very common problem sets on them, and then you just add one more qubit at a time, after a certain point when you add more qubits, you will not get better output. In fact, you will get output that's indistinguishable from noise. That number of qubits before it starts turning into noise, that's the number of algorithmic qubits that you have. The only way to improve the output of a computer is to have more algorithmic qubits. That is what we are doing. Today, we are doing that by improving the fidelity, the quality of the qubits themselves.
In the future, what you will do is you will pool many physical qubits to construct one logical qubit that you can compute on, and that's what's referred to as error correction. However, how much error correction overhead you need, how many physical qubits you need to create one logical one, is directly linked to your fidelity rate, or conversely, the error rates of the qubits you do have. If you have high error rates, you're gonna need many more qubits to create 1 good 1. We published some research, which was actually we did error correction on a live quantum computer and showed that we could do a 13 to 1 overhead to create a logical qubit.
That means that with 13 physical qubits, we use nine for the error correction part and four to - for just control functions, so that with 13 physical qubits, we could create 1 logical one. The other player who's published anything on this is Google, and in various parts of their research, they say they need 1,000 to 10,000 to sometimes 100,000 to 1. That is staggering because nobody's been able to put more than 127 chips or qubits on a chip anyway. If you're gonna get to hundreds of thousands of qubits, that means that if you pursue this path, you are years and years away from delivering a functioning quantum computer of any scale.
We often hear peers saying, "Hey, for our company," or sorry, "For the industry to hit this milestone, we need this many qubits. It's gonna take this many years." When we hear that, we tell them, "Stop saying that," because it's not productive. Every company has their own kind of fidelity rates of their qubits and the quality of the qubits. Every company really has their own timeline to hit those milestones, and it varies. It might take Google X number of qubits, IonQ Y number of qubits. It's not apples to apples in terms of quality.
That makes sense. As we think about technological breakthroughs as well, I think there's some common misconception out there that I think some people have, where they say, "Okay, by the time, you know, I read about it in the news maybe that there's some breakthrough in physics, that's when quantum computing, you know, can finally be achieved or be, go mainstream." You know, the way I understand it, there's some engineering around substrates, there's some engineering around, lasers that are commercially available and even software that help to achieve these milestones. Can you speak to that?
Yeah. There are - w e don't consider that there are any breakthroughs needed in terms of delivering quantum computing. There is incremental improvement needed, and for us, it is about delivering, just gradually improving the quality of the qubits so that you can eke out, a few more, logical qubits, before you start using error correction. The next big step is to implement error correction at scale, which we have. The first step for us was to show that we can actually make it work and show that in a functioning computer. We've done that. We are not willing to implement it directly in our chips right now because the cost of doing so would be more than we can deliver by just improving the fidelity of the qubits themselves.
After that, the next logical steps will be what's often referred to as photonic interconnects or connecting several chips together so that you can network them and create a larger computer by just having several cores in your machine. After which, the next logical step will be to network several computers together.
Got it. Now thinking more so on the, you know, revenue generation, customer engagement side, you guys announced your bookings number yesterday for year-end 2022. You know, obviously, you guys are well ahead of schedule where you plan to be with quantum computing sort of in its infancy. Can you talk to some of the applications you guys have addressed over the past couple of years?
Absolutely. This is a very promising field, and it's very fun to see all that you get to do.
It's even more fun to think about all the things we will get to do, of course. One of the most exciting projects that we announced last year was work that we do with Airbus, where they wanted to figure out a way to optimally stock the cargo hold. Every time a plane comes in, it gets loaded with stuff, not just people. How many, like how many boxes can you put in there? Which one should you load first? Which one should you put where in terms of weight that should be over the wings, et cetera? This is an amazingly complex optimization problem, one that has not been able to solve via a classical computer today.
Very similar to the traveling salesman problem of saying how many stops can a salesman make when he's out on the road, or easier for most of us to think about paths is a UPS truck. It can make 120 stops per day. That's more or less always true. FedEx, UPS, Amazon. You'd think that they get into work in the morning, and they just get a list of go here, and this is the fastest way you can do it. A computer can't actually compute that on time before all the packages have been delivered because this is what's known as a factorial. When there are 120 possible stops, all the possible combinations are 120 minus 1. 119 times 118 times 117.
The number of possible solution is 6.6 times 10 to the power of 198 or something like that. It's a vastly large number. It's larger than the age of the Earth in nanoseconds. Now, the cargo hold can hold many more packages than 120 and you have not only the dimensions of what goes first. You got first three dimensions in terms of where you put it in the hull, and then you have the weight. You got four attributes. In order to do this, you need just vastly more compute power than we have today. Think about how many planes are being loaded every day, every hour. That's a very exciting project.
Another one we did with Hyundai was for autonomous vehicles, where it's for them, it's gonna be all about, okay can they really drive alone? How can you think about the edge cases? We all know about the Tesla that ran right into a truck that was parked across the highway, that blended into the background because of the setting sun, there was no modeling for that. It turns out that quantum computers are very good at machine learning and figuring out from data what can be done because they're inherently not digital. They're made out of real atoms. What we did for Hyundai was doing image detection, what is it that you're seeing?
We're able to prove that by using a quantum computer and having lower resolution images than with classical, you can actually have a higher detection rate and correct image classification in faster speed.
If we think about some, you know, going back to sort of the earliest times we saw quantum computers first in the market, you know these giant machines and dilution refrigerators to now, you know, sort of the focus of IonQ is shrinking down that quantum processor, combined with what we see now with AI and machine learning and offload processors in the data center. How do you see the future of quantum computing developing sort of along that same path?
Computers need to be small, and that is a truism. If we all had ENIACs, we wouldn't all have computers 'cause we couldn't actually fit them anywhere. Everybody here has laptops and like, your watch is now a computer. The goal for us is to make a computer that eventually will be as small as today's computers, and we are well on the way, which is why we also think it's important not to use dilution refrigerators 'cause they are so big. If you need a couple of missile silos to house your computer, that just limits the amount of people who can have a computer. Quantum computing will not kill classical computing, but it will be an important aspect of computing, and it means that it needs to be available everywhere. We're going, like, we're going for miniaturization.
This is something that is, like, it's a long-term game, but it's also already happening today. We run our computers at room temperature, and we require a space that is not much larger than two, like, fridges that you have in your kitchen, and they're getting smaller every day.
Yeah. I think that, you know, to exactly the point you're getting to, it really brings up this idea of, like, hybrid compute across different types of compute. What type of applications do you think that develops once you can get quantum processors in a data center, in a server, you know, allow workloads to sort of work across different types of compute?
Right. We announced a partnership with Dell recently, where Dell will take us out to their customer base and their prospects for people who want exactly that. They want to have the ability to work with quantum but they also have a very heavy investment in just regular enterprise compute today, and we need them to work together. What you will see is that today, your operating system very quickly and correctly offloads problem sets between the CPU, the FPU, and the GPU, and it knows, like what you need to do where. We are not there with quantum computing, but we will be. The game will be to have available quantum compute power and know when to use it.
The first ones to do this will be large corporations who can afford to invest in both and afford to integrate them.
Switching gears a bit, you know, we've seen you guys become very active in 2023 in opening a new development facility, manufacturing facility rather, doing the company's first M&A. Can you just talk to the strategy behind those two deals?
Absolutely. The first is we are opening an office in Seattle. Right now we have our main office in College Park, Maryland. Although thanks to COVID, having an office somewhere means that you of course, have lots of people all over the place. We also knew we wanted to have another office, and we want to have it in a place with talent, which means that you're looking at San Francisco, New York, Boston, Austin, et cetera, also at Seattle. Turns out that Seattle has a lot of space available after Boeing moved out. They also have a lot of skilled workers who are working in high technology in companies that have developed workers to many other technology companies like us. There's a labor pool that's available. There's also real estate that's available.
We wanted to have access to this labor pool, and we also need to produce computers that aren't just put together by very careful scientists and are working because of all the customizations that happen. We need to be able to stamp out product that are consistently the same. For that, we wanted to have a new facility where we could focus on that mission, having customer-ready hardware being produced. We're going to do that in Seattle. When it comes to our acquisition of Entangled Networks, since we brought Jordan here, who did the deal, I thought that he could describe it.
Sure. Entangled Networks is a small startup based in Toronto that we acquired and announced in January. The company is focused on examining quantum algorithms and thinking about how to run them optimally on different quantum hardware. Their original premise was to say, you have a problem. How could I divide that problem between maybe two or three different quantum computers of different types, maybe coming from different companies? We saw the opportunity here to bring that skill set in-house and to help run algorithms on our IonQ computers. It chiefly helps us in three ways. The first is we can take algorithms that we see today and chunk them up and look at different ways to run them on our current hardware and run them more efficiently.
The second is that we can look at near-term architectures that we've announced and have put into future systems, including multi-core, so having multiple chains of ions on a single core, single trap, I should say, and multi-QPU, which is to say you have multiple ion trap chips in a single system. We can use the expertise of Entangled Networks to take problems and decide, hey, you should really run this problem, this part of the problem on one chip and the other part on a different chip and combine them. That is kind of the second way is it helps us in our near-term architectures. Then the third way that it helps us is it informs our long-term architectures as we're designing new systems.
When we're trying to figure out what to build next, how might I design a system that will run really well for certain types of algorithms or that will give me the best performance possible? This is a team that has all that quantitative skill set, is particularly talented on software and compiler technology, and we're excited to be partnered with them, to take IonQ to the next level.
Thank you. I think we'll pause here and see if there's any questions from the audience before we continue. Sure. Right in the back.
Just on that compiler and software - t hanks. Just on that compiler and software point you brought up there, the other leading trapped-ion business, from what I can understand, is JV'd with CQC. Clearly, Qiskit is class leading or is used widely, maybe not class leading. Do you see yourself moving evermore towards building that quantum stack where you'll have in-house software as well as the hardware as well?
It's a great question. Will we go full stack? The answer is really that we are a full-stack company today. There's probably only one thing that we do that's not full stack, which is we do not develop our own SDK, a software developer kit. That's because there are plenty of them out there that have done a great job. We want come one, come all, take whatever SDK that you like as a developer, and you can run all the major ones on IonQ. In terms of other parts of the stack, like the operating system or applications, those we've been building for a long time now. We really do look more like a full-stack company in every way except for the SDK.
Thank you. I'm quite new to the story. I want to ask maybe on what we're seeing with the hyperscalers investment in supercomputers and getting more predictive models with AI. First of all, how are you seeing the hyperscalers besides Google that you mentioned investing in quantum computing, first of all? Second of all, does it live side by side with supercomputing now as predictive models become more and more efficient?
We see a lot of investment in quantum, which is good. The challenge with the superconductor investment is that that type of quantum compute has very high error rates, which means that currently the compute power it delivers is lower than from ion traps. That over time, if you try to overcome the deficiency you have in the higher error rates by using error correction. The overhead cost is very, very high, between 10,000 to 100,000 to 1, which means that the time to deliver the capacity of even having this error correction is longer. Which is also why you see some of them saying that quantum computers X number is far away. It isn't that far away, but we welcome all the investment in this industry because it will all be needed.
By the time you get to roughly 70 logical qubits, so these are fully error-corrected fault-tolerant qubits, most people reckon that that is a time where a quantum computer will have the same compute capacity as the leading supercomputers. That is not very long time from now. That's only a few years. You will still see that supercomputers will continue to exist, and you will just partition the workload of what is best delivered through a classical compute, be it supercomputer or other clusters versus a quantum computer.
Any other questions from the audience? Darren, I'll just wrap up with one more from me. You know, as we think about your go-to-market strategy, you know, three clear channels is Quantum cloud, Quantum as a Service, and then system sales. Can you talk about which you see as really the future growth driver of the business?
I think this goes in waves. So far it's been access and Quantum as a Service. What will happen as quantum computers actually become commercially available for purchase, that will become the largest channel, mainly because these things are very expensive. We're talking about tens of millions of dollars per computer. You don't need to sell that many of them before that overtakes as the main channel. Long term, you will more than likely see more of what we have today in classical compute, where a lot of it goes to AWS and a lot of the value will be delivered over the software stack. Software will eat the world, so too in quantum. It's just further away for quantum.
One more real quick, actually. On the M&A strategy, granted that you now do have that first deal under your belt, you know, where would you like to add to the business if deals do come up opportunistically?
We would like to add, in the near term, we would like to add, businesses that accelerate our roadmap, our technological roadmap. You know, 10 to 20 years from now, that's a much more complex answer.
If there's no more questions from the audience, we'll wrap up here. Thank you.