IonQ, Inc. (IONQ) Earnings Call Transcript & Summary

Welcome to the last session of Cowen's 50th Annual TMT Conference. Thank you for sticking with us. Very pleased to be joined by Chris Monroe and Thomas Kramer of IonQ, who are going to give us a presentation on the company. Thank you.

Well, thank you all for being here. Okay. So we have some time since it's an intimate crowd, we're happy to take questions at any time. But I want to introduce you not only through the quantum computing field in general, but IonQ in particular. So we are a full stack quantum computer company to make -- we make devices that will, in the near future, displace many of the applications that are run on high-performance computing now and then some other applications. Now full stack is an important idea, a notion, in the field. We build the hardware. Our base is building computers out of the best components, but we're also engineering controllers around that, software on top of that and an application interface at the very top. Full stack means we have to really lead the industry in all of those aspects. Quantum computing needs to be codesigned. The hardware needs to be codesigned to the application. So we have lots of interesting metrics measuring the might of our company in the quantum computing space. Our quantum systems are the best quantum computers as measured by industrial metrics, I'll point more to that in a little while. In particular, on these points here, we have a very top-of-the-line, world-class investor base, having gone public about a half a year ago. We have many partnerships with users in the application space from -- not just universities, but interest in finance and also the automotive industry and many more to come there. We built several generations of quantum computer systems, and we're on track to build many more in the coming few years. Our systems are the only ones that are on all 3 major cloud providers at -- Amazon, Microsoft and Google, Google Cloud platform. And of course, all of the user software packages can be run on our systems. So the leadership of the company spans many different industrial bases, our CEO, Peter Chapman, was most recently a lead engineer on Amazon Prime. I'm co-founder of the company, along with Jungsang Kim. We have academic background, as is befitting in this field. Jungsang is a seasoned engineer who has experience with photonic networks and switches. I've been in the field since the beginning in -- about 25 years ago, and my work back then demonstrated the first quantum logic gate in any physical platform. Thomas Kramer here, CFO, he's taken companies public and has great experience on the financial side of things. And he's also quite adept with the technology that we're using, and he understands the total available market that we will be tapping into in future years. Tom Jones, the most recently at Blue Origin, he's our Head of People. And Laurie Babinski, who's our General Counsel. So this is a picture of one of our most recent -- the guts of one of our most recent quantum computer system. It's called an ion trap because this chip that you see, the center there, it confines atoms above the chip, they're suspended in a small vacuum chamber, and that's the whole thing. It's about the size of a deck of card. And our qubits, they might sound exotic, these atomic qubits being in a vacuum chamber, but this is what makes a quantum qubit system high performing. It's isolation from the environment. And so these atoms really have nothing but each other to interact with. So this is a very mature quantum system in the sense that all of the physics is done in this system, and the challenge to scaling it and deploy it for real-world use is an engineering challenge. The atoms are all identical. And therefore, the system is arbitrarily scaled to as many qubits as you need. The challenge is all of the engineering controllers, the control system that is surrounding. And so with this hardware, we are indeed winning the quantum space race in these early-stage quantum computers. Now this is based on an analysis by BCG a few years ago, and they looked at the quantum computing space as having 3 phases of development with correspondingly larger available markets. The first phase where we're in now, and that is quantum systems are fragile and we have to reduce the passive errors of these quantum systems. And we should be able to -- in this phase, we should be able to get hints of applications that make value and actually hit some of those applications. Now Phase 2 is when we need to deploy error correction. And what this means is that we can operate with larger numbers of qubits and execute deeper circuits. It requires overhead. We need lots of qubits here, and this will allow us to hit an even greater array of application spaces. And then finally, in Phase 3, to scale, we need a modular architecture to think about fabricating quantum computer systems that we can manufacture the individual components and wire them up. And IonQ systems are really the only ones that -- where we have a very well-defined strategy of going through all of these phases. Now BCG also said Phase 2 is 10 years away, Phase 3 is 20 years away. And I think that was -- and we're telescoping that way down. We think that this is going to be not a decade away, but much sooner than that basically because of our technology. So here's a little schematic of our technology. There's a chip down on the lower right. There's nothing quantum about that chip whatsoever. It's basically composed of gold or any metal that's on top of a silicon or a glass chip. And these are just electrodes that hold atoms above the chip. So those dots that are floating above the chip, those are individual atoms. Yes, we can see individual atoms one at a time because there's nothing else there. It's in a small vacuum chamber, so there's no air. And when -- our eyes and ears on these qubits are laser beam. That's our controller. Now quantum systems, you cannot wire them up directly because then they will lose their quantum character because wires are messy objects, no 2 wires are exactly the same. So our qubits, we control them with optical beam, so it allows us to have a very light touch on the quantum system, which is absolutely required. I mentioned before, these atoms are absolutely identical, so we can replicate with perfection, and that's not an exaggeration. They're individual atoms. We don't manufacture them, so there's no manufacturing defects, and this gives us the recipe for scale. Now in Phase 2, when we do error correction, this is going to -- like any error correction, it's going to require some overhead on the number of qubits that we're using in order to correct the errors. We do this with classical computing all the time. Wireless communications, we redundantly send information to correct errors. Well, in our platform, the native errors are so low that the overhead for doing error correction is modest, maybe just 16:1. We demonstrated a 13:1 code very recently in collaboration with the University of Maryland and Duke University. Now this is significantly better than other technologies, especially superconducting technologies, that requires many thousands, maybe up to a million to one overhead. That is to get on error-corrected qubit, you may need 100,000 or 1,000,000 physical qubits. So that obviously is a huge task to even think about, putting a million qubits on a chip and having them all behave in a way that we understand. And so I would say in those technologies, there's no -- there's a lot of physics that's going to happen, let's put it that way. I think they're going to rely on breakthroughs in material science and so forth. Not so for our platform, our individual atoms, we can scale them and have very efficient error correction. Now how are we going to pack all these atoms in to scale to thousands of physical qubits? Well, beyond error correction, we have 2 attacks on the hardware to make a modular large-scale quantum computer following that third phase of the BCG analysis. The first mode is to actually have what we would call a multi-core architecture, where individual atoms are split into groups on a single chip. And the communication between those cores is it's brute force. We're moving atoms with these electrodes. We can just move atoms back and forth between the cores. And with that, we ought to be able to get many groups of these qubit chains on a single chip. And we've done -- that slide, by the way, depicts a demonstration that we outlined a few months ago in the press release. So this is our mode one of scaling. The outmode of scaling in a few years will be using photonics. These qubits can -- they can -- we can move their information on to light and then move their information off chip to another quantum processing unit and then, sort of in manufacturing mode, stamp out those CPUs and this starts to look like a data center. And so this is -- really, IonQ has the only architecture where we can credibly offer a scaling strategy, and all of this is based on engineering, photonics. You can sort of see the challenges here are not in fundamental physics, and this is why we're very comfortable with this road map. Okay. So this slide shows that road map over the next several years as measured in a metric we call algorithmic qubit. So algorithmic qubit, it's a little bit subtle. If you have lots of qubits but they're very noisy qubits, you're not using the power of all those qubits. In particular, if you have some number of qubits, you need to perform some number of coherent quantum gate operations on those qubits that scales with the number of qubits you have. If you have 10 qubits, you probably need a order of 100 operations. It roughly scales like m squared. With 100 qubits, you need 10,000 operations. So algorithmic qubits basically takes the -- not only the raw qubit number, but the performance of those qubits into account. So even if you have 1 million qubits on the table, you may only have an algorithmic qubit number of 5 because you're not able to run very deep circuits. So we are projecting the number of algorithmic qubits in our system will be growing substantially starting in '24, '25 and '26 when we deploy error correction. And what's happening in the interim period here is that we're increasing the performance, the passive performance of our qubits, while also making way for error correction by adding more physical qubits and not algorithmic qubits. And then when we deploy error correction in the year 2025, we can enjoy 64 algorithmic qubits. And that's important, because 64 algorithmic qubits -- to be able to run a circuit on 64 algorithmic qubit is something you cannot simulate on even in a high-performance classical computer. So this is where value will really start to turn on. And then the modular architecture and volume photonics will happen a year later, and then we can really go to our races and start scaling and finding applications for these systems and they're sort of outline what -- where we believe the early applications will be. Now we're very confident in this road map because it involves engineering paths that are well known, mainly optical systems. And they will get miniaturized over time and even higher performing as they're miniaturized. And so in our systems, in our road map, if you have an application and you know how many algorithmic qubits you want, you can just look at the chart and see when that application will start to create value. In the out years, you can see we'll be modeling chemical processes and materials. There's a lot of interest in those types of applications even now. Even though it's 5 years away, our customers are finding they need to get up to speed right now on how to use these systems. And as we scale into their -- into production and hit a level of performance where it creates value, then they will be the frontrunners. So we've recently announced our latest generation system and this is a little bit technical, but the way we program our quantum computers is through optical means, beams that are shining on atoms. Our earlier generations on the right had fixed beams that had to be aligned to these atoms, one beam per atom. It's a tricky alignment, but we're still able to get good performance there. Our Forte system doesn't do that, instead it has 1 beam -- actually 2 beams that are movable. So we can move the beams across the atom. And this is all controlled in software. And that's important because we can deploy it on the fly, decide what algorithm needs, which beams to shine on which atom. And these are our sort of wires, if you will, in our large-scale architecture. And also, it allows you to calibrate errors. So we expect better performance out of forte than we have in our current system. So a few applications over the last few years. We have a partnership with Hyundai, the auto manufacturer. They're very interested in many aspects, but in particular, they're interested in lightweight solid-state batteries. And to do the materials research and develop new materials requires models that we can't solve using high-performance computing. And so there's an ongoing project with our applications team to use our quantum computers to help understand models of solid-state chemistry that would be appliable to battery. They're also interested like all auto companies and many other things involving autonomous vehicles. The aerospace industry is also interested in quantum computers for many applications like this. In the financial space, with QC Ware, a quantum software small company, we've worked with Goldman on a quantum computer -- family of quantum algorithms that performs Monte Carlo sampling. And there's a related project with Fidelity Research on sampling from copulas that is correlated random variables. So the last slide here, I want to give you a snapshot at the performance of our systems, our highest performing system compared to some of the others in the market. So this data was compiled by the U.S. Quantum Economic Development Consortium, which is sort of run by a nonprofit that basically exist to benchmark quantum computers. And so we had algorithmic qubit number across the different companies. The 2 on the left are ion traps, IonQ and Continuum, used to be Honeywell. And on the right, IBM and Rigetti use superconducting quantum system. And these -- these squares, denote the success of algorithms, blue is good, red is bad. Red is noise, blue is perfection in a quantum computers output. And you can see, if you make the quantum computer big enough, you demand too much out of it, you start to hit red. And the IonQ system is able to work with 20 algorithmic qubits effectively, meaning that we can do many hundreds of operations. That's a factor of 256 over Quantanium. That is -- their number is, I think, 12, 20 to 12 qubits. And then IBM and Rigetti are sort of stuck in the very low numbers of algorithmic qubits. Again, because superconducting qubits have to be manufactured and they're very noisy to begin with, it's not clear how those technologies will scale. So with that, we have some time if there are any questions from our generous audience members here. So Thomas, did you want to add anything? And also Jordan Shapiro is our VP Finance, here from IonQ.

I promised Laurie that we would show the last slide. There's one more. But yes, so we have already announced in our last earnings call that we raised our guidance for the midterm and for the rest of the year. For the full year will be $25 million. We remain confident in our plans. We deliver on our technical milestones. And while this is a young industry, we are already doing a lot. And the way to evaluate the young industry is on the output and seeing like what goals are being set out and are you meeting them? Questions?

[indiscernible]

Yes, your statement is almost certainly true that certain modalities will be better for certain application areas. And also, maybe even certain physical platforms might be better applied. Now we've concentrated on what's called the gate-based model of computing because it's universal. You can do everything with it. But often when you can do everything, it means that you're constrained and maybe it's too small. So our road map concentrates on scaling the gate model, but we're constantly aware that we can do annealing in our system if we want to. The problem with those other modalities is that they're not universal. That's okay. I think in class to computing, we have application-specific integrated circuits, ASICs, that do certain tests, and that's fine. I think because quantum computing will be applied to so many different types of problems, that's going to come -- they're going to come out of the woodwork in the coming few years. We feel that the gate-based model is going to be best. It's the best strategy to be able to attack any problem as it develops. So even within the gate-based model, we're working with customers to understand how to compress the circuit to make it more efficient, because we're all resource limited in the number of usable qubits and the error rates. I think there's a lot of attention paid to annealers and so-called quantum simulators. These are very restrictive, they can only work on certain problems. And so I think that's a crutch right now in the field because it's so wide open. But I think we'll probably -- after we push through with our universal gate-based systems, then the ASICs come. I think they come later. Just like in classic computing. I think we had to develop a general-purpose computer, and then there were specialties after that.

And then in terms of [indiscernible]. Does it all operate at room temperature? Or is there a -- there's a correlation that [indiscernible].

All right. Let me go back to the deck of cards thing. So this can operate at room temperature and we don't need anything to be cold, and it's because it's in a vacuum chamber. The atoms are sitting there and there's nothing else there. They're not in thermal contact with the wall. So the walls can be at room temperature, no problem at all. It's -- some of our systems, we do cool down only to get the pressure better, it has nothing to do with the -- we don't need low temperatures.

[indiscernible] don't have anything to do with [indiscernible].

No, no, no. The qubits sit there, they don't talk to their environment. That's the idea. [indiscernible] to say yes. When it comes to scaling these systems, whether we operate at low temperature or not, the cooling technology doesn't get any harder. It's not -- we're not -- let's put it this way, we're not cooling the quantum system itself. Because they're individual atoms -- we do use something called laser cooling, but don't compare that to a refrigerator. It just means a single laser beam that we have anyway, when you tune it right, it keeps the atom at rest. But there's no refrigeration needed. That's important for the scale.

[indiscernible]

We need it. But it's very low power, it's almost an afterthought. We need lasers anyway to execute gates and other lasers very similar to that, actually much lower power, not demanding at all. They keep the atoms in one place.

[indiscernible]

Yes, that's right. Yes, it's easy to multiplex those laser beams. Like I said, the system at a high level is an optical system. All of the advances going forward are going to be based on shrinking, integrating, making photonic circuitry, and we're pretty confident about that technology.

When do you expect to start using [indiscernible]?

Yes, good question. It is true that when you scale anything -- well, think of multi-core classical processing unit, CPU. We don't make chips as big as a football field for the simple reason that you can't connect -- there's too many connections needed. So we break it up into smaller modules on one chip. And what you're doing is you're limiting connectivity by doing that, but at least you get to build it, you get to build the system bigger. So there will be a small price to pay when we connect photonically. And that -- so that graph on the right, it's a little hard to explain, that's 8x8. We have 8 -- the yellow lines are denoted -- are denoting the full connections within a group of 8 qubits. And that's the typical way we operate gates now. It has full connectivity between any pair. But by sending these photons through an optical switch, we can have, with a fixed cost of having to move the information to the edge and then get it into a photon and then hit the edge and then bring it to another, there's a fixed cost of 2x operations. We can connect any pair of qubits to scale. So it actually is fully connected. And that's really -- that's remarkable. There's no quantum architecture that has full connectivity to scale, except ours.

And when do you [indiscernible]?

This, we will -- if you looked at our road map, I think that corresponds to '25, I believe. '26, right? That's a big jump from 64 to 256. I wouldn't take that lightly, but this for us is going to be manufacturing issue. If we can do it once, we just have to multiplex it a few times. In this case, it's 4, and then we'll go from there.

How does that compare to some of the [indiscernible]? How does this [indiscernible]? What's a [indiscernible]?

So in the superconducting platform, this is a solid-state quantum platform and has to be cooled nearly 0 degrees, and the superconducting qubits are edge connected to their neighbors. So if you want to move information from one corner of the chip to the other, you have to spend very extensive time and fidelity on swapping in the quantum way all the way across the chip. It's not fully connected. And frankly, the scaling of superconducting circuit, I think it's sort of the football field approach, right? They talk about converting to optics and using this modular approach. But even that transaction from a superconducting qubit to optics is very researchy right now. It's not very efficient. Nobody really knows how to do it yet. It's a wonderful research, it's happening at universities now. But right now, they're thinking about just printing chips with lots and lots of qubits on them. So I'm not sure there's a real strategy there in the long term for scaling.

[indiscernible]

400 [indiscernible], yes.

400 gigs [indiscernible].

Oh, 400 gig. The speed of the photonic interconnect is actually -- it's slower than the speed of the nearest neighbor gates that we're running right now. Slower by about a factor of 50, I think, right now. But that will improve with better light collections and so forth. The -- what's tricky is we need an optical switch network. So it's sort of like the old-style telephone operator, if you have, say, 16 fibers here and 16 fibers there. You want to configure the system to switch -- to connect any 2 fibers on either side. And that's what allows you that full connectivity I mentioned before. So there is a lot of technique, you can buy that now in the telecom wavelength. We need to do it in the visible. But again, this is sort of a straightforward engineering task. And I'll say you should also talk to my cofounder, Jungsang Kim, who made the world's biggest optical switch in Bell Labs back in the day.

What's the size of the footprint?

Physical footprint of our current system like the Forte system, it's a few cubic meters, and that's dominated by space. These are optical systems and the footprint of the atom -- the atoms are less than 1 millimeter, so they take no space at all. The chip is the standard chip. Back-end chamber is the size of the deck of cards. It's all the optical structure that brings it in. Eventually, we're going to integrate those optics on the chip. That's going to not only make things smaller, lower cost per qubit, but the performance is going to be better as everything moves together. A little bit sensitive to vibration now, not so bad. But a lot of this meter cube, it's bigger than it needs to be. And in the future, as we grow, as we scale our system, the physical footprint will get smaller. Not just because it's cute, but it will give us a better performance and we'll be able to deploy it on the edge in airplanes or people who want to have these things. It's not just one accelerator like CERN in Switzerland, where we have -- the world had one quantum computer that costs $10 billion. This is something we're mindful of cost per qubit to scale. And our scaling strategy uniquely allows the cost of qubit to go way down.

[indiscernible]

So the $25 million in booked contracts will be a mix of the 2 -- or actually no. So we have not yet sold any hardware. What we're selling is access to compute time on our hardware. And in addition, we do professional services, joint application development. The joint application development is not a significant part of our revenue just yet. But it is there. But right now, most of the revenue that we get, comes from computer access time.

[indiscernible]

AES-256, that's 256 bits. And I think you need maybe 8x that working space. That's with knowing our correction. Sorry, I made the math in my head. I think you probably need 100,000 qubits, that's out there. So we have a path to a few thousand. I think that will -- interestingly, these early application spaces will pave the way and paper the way to get to hundreds of thousands. It has to be modular though. It's going to look like a data center. And every extra qubit you add is like adding a whole data center, classical data center.

With that, thank you very much.

Yes, [indiscernible]. Thank you.