My name is Quinn Bolton. I cover the semiconductor, semi cap, and quantum computing sectors for Needham & Company. It's my pleasure to host this fireside chat with IonQ. IonQ became the first publicly traded quantum computing company in 2021, and is a leader in the gate model quantum computing industry. IonQ's quantum computers are based on ion trap technology, which has a number of advantages, compared to other types of qubit technologies that are out there in the market. Based on these advantages, IonQ systems outperform competitor systems on benchmarks based in real-world applications. Joining me from the company today are CFO Thomas Kramer and Jordan Shapiro, VP of FP&A and Investor Relations. Thomas, Jordan, thank you for joining us.
Thank you, Quinn. I also thank you to all of you who are here. Whoever is paying you to come get up this early, you got a good team.
Maybe they're awake and asleep.
There is that.
Maybe starting off, because there may be some in the audience who are still new to quantum computing, I wanted to ask some more introductory comments, but can you provide just a brief introduction to IonQ for the investors?
Absolutely. IonQ was founded in 2015 by Chris Monroe and Jungsang Kim, who between them have 25 years of research into quantum computing. We have leveraged what they had done in an academic setting and taking it into a commercial phase where we're where they proved that you can actually make a quantum computer. We're now making it and selling it. I often get a question, what is a quantum computer? Just in case there are some people here who are new to the topic as well, in cocktail conversations, I like to say that we shoot lasers at individual atoms to make the computers of the future, which sounds really cool, also true. We use lasers, we target individual atoms to encode them with information, and then we compute on them.
This is the difference between a quantum computer and a traditional computer, where you have transistors which can have the value of zero and one, and then you put a lot of them together, like in an iPhone, and you can make them do lots of stuff. There are certain things that an iPhone cannot do or a supercomputer cannot do, or it cannot do it in sufficient enough time. There are a set of problems They're small problems, but they have explosively large solution rooms. Factorials is a group of them. If you think about a UPS driver or FedEx or Amazon or any of these, most of them can make 120 stops per day, and that's what all of their performance metrics are set up to do.
You would think that in order for them to deliver these 120, they come in, their truck is already stacked, and they just follow the delivery map. That map has been calculated to be absolute perfect way of delivering packages, but they have not. The reason for it is that in order to evaluate what is the best route when you have 120 packages is a factorial problem set, which means you take 120 minus 1, so 119 times 118 times 117 to get the number of possible combinations. That answer is, let's see, 6.6 times 10 to the power of 198. It's a number so large, it's the...
The number is the same as the age of the Earth in nanoseconds. Like, it's just unbelievably large. If you have a traditional computer calculate it'll take weeks, and the packages will all be delivered, so there's no point in actually calculating the best route. A quantum computer of sufficient size could take all of the options and evaluate them at once, thereby reducing the compute time. There are a number of problems like this. If you wanna do weather patterns, climate change, none of our traditional supercomputers can do this at a sufficiently detailed level. That was a little long, but sorry.
It sounds like the power of the quantum computer is the ability to compute simultaneously or evaluate solutions simultaneously rather than sequentially.
That's right.
I won't ask you to get.
I should have just told him to say that.
There are a lot of different qubit technologies on the market.
Mm-hmm.
IonQ has chosen ion trap. There's superconducting that we're getting, IBM and Google have chosen. There's photonics. Can you spend a minute, sort of, you know?
Uh, uh-
go through the options, the different, you know, technologies that are out there in the market?
Absolutely. There are a number of qubit solutions out there. There are some still being discovered, and there are many that are being worked on right now. Most of them will be for research purposes, where, you know, scientists, physicists will go and just research things. Most of the computers that are working right now are of one of three types, either photonics or superconducting or ion traps. We're representing ion traps, so I'll go to that last. Photonics is interesting because it's light and light speed, that would be very fast. The challenges with photonics is that you have no way of storing the memory, which is a challenge when you're a computer. Also actually achieving entanglement with photonics is not easy.
It is an approach, and we welcome it in the market. The other most common next to ion traps is superconducting. This is what IBM and Rigetti is using. Google has a project doing this as well, where essentially you take 60 years of experience in semiconductors. You take a silicon substrate, and you jam in some qubits into there, and qubits are the basic building block of a quantum computer, and then you run operations on them. The challenge with a superconducting approach is that you are manufacturing your qubits. Instead of taking atoms from nature, you build them, and that introduces error rates. And that also means that the cost of producing these goes up. More than that, though, superconducting qubits have a native error rate that is higher than for trapped ion.
Quantum computers are very, very fickle. They exist at a moment in time, and they collapse, and then you have to reboot them. With the superconducting approach, they collapse a lot faster, and they have also a lower fidelity or the quality of the output over time. That's the challenges they're working on. When it comes to ion traps, ion traps are where we take atoms. We use ytterbium, YT and barium, BA. We buy these from national laboratories, and then we shoot off one electron to make it ion, so it can hold a charge, and then we put it into our computers. The beauty of this technique is that we don't have to manufacture them. We don't have to think about how they're made. We just take them, and nature's made them. They're identical.
They're similar. They're perfect, as we call them. They also have the benefit that they're very stable. The ion trap computers will stay coherent for much, much longer than the superconducting computer, which means you can think of it as you don't have to reboot it. The coherence time for a superconducting is measured in either milli or microseconds before you have to reboot the computer and put it back together. At rest, our computers can stay up for weeks. That combined with the fact that it have a much higher fidelity rate, that is the quality of the output, it made us pick that as a technology. It is very alluring to think that, yeah, we should be able to just take what we've been doing for 60 years and put qubits in there.
If that worked, or if we thought that would be a better approach, we would have done that. However, with 25 years of research, we think that, going the ion trap approach, which is also a very tried and proven technology, it just hasn't been used for computing. This is what runs all of our atomic clocks, and it's how all of you keep time. All of your iPhones use that.
That was a great overview of the three different sort of major technologies today in the quantum world. Ion traps have, as you say, better coherence times. You can use them for longer. They've got much lower error rates or much higher fidelities. I think the one criticism has been scalability and how many ions you can get into either a trap or how many traps you can put in the system. Maybe spend a minute. We'll get into sort of the full roadmap in a minute, but just talk about what you're doing to try to scale the number of qubits in your quantum processors.
Absolutely. I have to apologize to all of you who are following this industry, and you have to listen to everybody's gobbledygook, and people are saying, "Well, we have these X, Y, Zs, and they're all obviously better." It's hard to compare the different technologies. When it comes to scalability, it's what does that mean? It just means can you build a larger and larger computer, which IonQ has proven over time. We have the most powerful quantum computer in the world, and we have for the past two years since they have been measured, and we will continue to deliver that. The other thing to talk about in scalability is can you actually make more than one of these, or are you just making R&D or museum pieces that only scientists can use?
On both of those counts, we keep on increasing the algorithmic qubits count. For those of you who are into quantum technology, algorithmic qubits are similar to logical qubits. What it stands for is just a qubit that you can compute on. If you can't compute on your qubit, it doesn't matter how many you have. When certain manufacturers say that we have 127 qubits, my question is, how many of those can be used on computation before the output resembles just noise? There are ways to measure this. There's an industry group called QED-C that has come up with a set of algorithms that are typically used in quantum. They will run them over and over with just one more qubit, one more qubit, and then they measure the output.
At a certain point, because the error rates actually multiplies between them, a certain point, once you get over a hump, you're gonna put out a random noise. Then, if that stops at six, and that's the last one, that the superconducting people have put up metrics for, then if you add another 100, you don't get anything. You can do the clock speed super fast, but if you don't get any compute out of it, there is no point. For us, we are focusing on bringing out computers that have more useful qubits in them, and you only need a couple for it to have a dramatic impact. Each time you add one algorithmic or logical qubit, if you want, you double the compute capacity of that computer. You double, literally.
You only need to get to. In the classical way of thinking, most people in the field would say that 70 qubits, logical qubits that are good to compute on, is enough to equate among the best supercomputers out there today. You only need 70. We have posted results from our computer that does 25, and our target for this year is 29. Going from 25 to 29 is actually two to the power of four better, so it's 16 times better than the 25 qubit one. We don't need to talk about having 1 million qubits in your chip. We actually think that having 1 million qubits is detrimental to performance because you have to manage them as well as be able to use them in terms of their computation.
The reason why you hear this number, 1 million qubits, so much is because superconducting qubits have so high error rates that when you deploy error correction, which is the act of putting a bunch of physical qubits together to create good ones, the error correction rate of 1- 100,000 means that you need a ton of qubits just to go a little bit further in terms of computing.
That's probably a good segue into sort of the company's current product families and the roadmap. Can you spend a minute just discussing the key specs on your Harmony and Aria systems that are deployed today and where you expect to go with Forte? We'll get to.
Absolutely
... further out roadmap in a minute.
Absolutely. We actually made our Harmony class computer available on AWS at the tail end of 2020.
Mm-hmm.
At that point, it had a record-breaking number of six algorithmic or good qubits. Other computers at the time had one-two , even if they were claiming a little more. That has just continued to be a massive performer for us, that we have customers working on it today. It's still up on all the three major clouds. Last year, we actually announced that we had managed to increase their algorithmic qubit counts from six- nine, and that means, again, that you have increased your compute capacity by eight times. Last year, we also announced the general availability of our Aria class computer, which when it came out, had 22 algorithmic qubits.
Sorry, $20.
20. We kept increasing it, and it went up to 23, and actually then it went up to 25 at the tail end of last year, which still is by far the most powerful quantum computer out there. What does it mean to have 25 algorithmic qubits? Good question. Typically, it means that you can perform much larger instruction sets. A 25 qubit count quantum computer can operate instruction set up to around 1,000 instructions. These are not the instructions that you find in like the programming languages that we all know. It is what the programming language gets deconstructed into before the computer performs the operation, so it's more like assembly language, if you want.
The, if you have s algorithmic qubits, you can do about like 16-20 instruction sets in one application, which is severely limiting. It means that you can't really solve much. It's just a step function to get to the larger ones, of course. This is also the way to compare quantum computers. In reality, you don't need to hear about like how many different chips are in there, how many lasers, how many lines of communication wire. What you need to know is what can you do with it? A good measure of that is, okay, how many lines of code can I put in one program?
As you go up in AQ count.
Mm-hmm
... you not only have more usable qubits, but you can do almost exponentially more instructions on those qubits.
Not just almost, we can do exponentially more. It's a true exponential power growth curve. That is why we're focusing so much on the algorithmic count as opposed to the physical count. The Aria class computer has 32 physical qubits in it. The next class computer, which we have announced but we haven't put up for general availability, it's on private beta right now, is called Forte. This one will have 40 physical qubits. We're planning to use 32 of those slots again. We're also planning to drive the algorithmic qubit count up to 29 on that one.
Longer term, you'd mentioned that as you get to 70 or so logical qubits or AQ.
Mm-hmm
... you start to rival the world's fastest supercomputers. Can you talk about your longer term AQ roadmap as you look out over the next three-five years?
Absolutely. Our plan is to keep adding qubits to each chip so that, for this year, our goal is 29. For next year, our goal is 35. For the year following, the goal is 64. For the year following that, I believe is either 96 or 128.
Yes.
We follow the typical computer numbers. But once you pass 70, you're in. You put the world supercomputers to rest. They will still be working on the stuff they're working on now, but now we have a class of computers that can work on new problem sets that we can put into production, and that is when you will see that quantum computers, you just can't get around them because now we can solve things that you can't do today, or at least not without waiting months, sometimes years. The way to do this is we start by putting more and more qubits into 1 chip, and you do that by increasing chip size, the technology, you do multilayer chips, and then you put more chips into the same computer.
You can network the computers and have them all work together to be one giant quantum computer. The compute overhead for distributed computing is much less in quantum than it is in traditional computing. In classical computing, the overhead is about 40%. In quantum, it's more like 20%. The reason for it is the scariest part of quantum, because it's what very few people understand. It's the entangled nature of the qubits. The qubits, which are the basic building blocks of the computer, can not only communicate with each other, they are aware of each other. Without you sending signals to them, as long as they're entangled, they will act in tandem, and reading one of them, you can read the others. That's what's so fascinating about it, and that's what allows for this parallel processing aspect.
Maybe this is a good segue into the acquisition you announced yesterday morning of Entangled or the assets of Entangled Networks. Can you spend a minute just explaining what Entangled Networks was up to on the quantum networking side?
Absolutely. Actually, since Jordan spearheaded this deal, why don't you tell us?
Sure. Entangled Networks is focused on technology that optimizes how algorithms are run on quantum computers. For us, we view this really helping us in, I'd call it four ways. The first is that today they can take an algorithm and think about how to segment it and how to run it on our current systems and run it more efficiently. We expect to have near-term advantage in terms of how that impacts today's algorithms with our customers. The second would be that as we split into multiple cores on the same chip, as Thomas was talking about, putting more qubits onto the same chip, Entangled Networks has built a compiler that helps you to optimize which qubits you run on so that your algorithm is optimally configured.
The third is that as we expand to multiple chips, they can do similar math and tell you, "Hey, you might wanna run this part of the algorithm on this side of the computer and the other half of the algorithm on this side, and this is the best way to structure that algorithm." The fourth way that they help us is that as we look to scale, Entangled can help us see around corners on architecture. When we are designing our next generation quantum systems, we will have better visibility with the software and hardware expertise that they provide to design computers that are more able to run at high performance.
It's more of a software compiler almost. I was thinking it might actually be the physical quantum networking capability that I know you're separately developing internally.
That's right. It's key architecture and software expertise and some hardware expertise alongside that enables us to do the scaling in a more methodical way.
Got it.
It plugs into the networking bit. They're not the ones who are gonna enable us to do networking. They will enable us to do better networking.
Better networking and optimization-
That's-
... which qubits can be used.
Right
... multi chips are Network. That's right. Got it. Okay. Can you say how large was the team that will join IonQ? It doesn't look like you've announced sort of the purchase price of the assets, but was it material in terms of your balance sheet?
This is not a material transaction. This is what is lovely, lovingly called a tuck-in. We did this because we are working on, well, so many things. We're making a full stack compute platform, and we have a roadmap that we can execute on, and we can hire the people we need, and we can also do most things internally. There are not that many experts in ion traps, ion trap technology that know more than we do. We also only have 24 hours, and if we can hire somebody who can help us gain an edge, we will do that.
We're actually, we're very happy about announcing the opening of our Canada office since North America is more than the U.S. and with, you know, everybody's having issues with H-1B, and it doesn't seem that we are going to solve that at national level. It's nice to have an offload factor where if we can't bring them to the U.S., we can have them in Canada. There's also a market up there. There's a thriving, actually, quantum technology scene in Canada.
Great. speaking of key talent, can you just spend a minute speaking about how active Dr. Chris Monroe and Jungsang Kim are in the day-to-day operations of the company and to the extent that they're still conducting research at the university level, how does the company benefit from their research?
Absolutely. Both Chris and Jungsang are professors associated with Duke and Maryland. They only take PhD students. They don't really teach classes, but they have some students that they guide through their thesis and that help do research for us. Sorry for them, but we benefit from that research. Jungsang is probably the most overscheduled person in the entire company. I think he must have been featured in the Harry Potter movies because he has this time turner, so he can be in multiple places at once. I've never seen anybody work that hard, and I've worked in technology for 25 years. Chris is also working every day and I have meetings with him.
He is more of the visionary who says, "Okay, this is where we're trying to go in the next five years." Jung Sang says, "This is where we're gonna go in the next five minutes and 15 minutes and month, and then six months." They fulfill two sides of the coin, and they're very, very much actively engaged in the company every day.
Perfect. maybe just sort of turning to the financial side of the business, can you share the various sources of revenue, for the company, including development contracts, private cloud, and your public, cloud engagements?
Absolutely. We have a four-pronged strategy of go-to-market. The first and easiest way for users to interact with our technology is to log on to one of the public clouds. With, you know, a credit card and $20, you can start running quantum code on AWS or Azure or Google Cloud. There are some limitations inherent in that setup, though, because unlike other AWS services, it's not an instance that's spun up and it's immediately there. Your jobs are queued, and then they're run. If you happen to submit your job right after a college class that has homework due on Friday, you will have to wait. That may not work if you're in a corporate setting where you just need your stuff done.
For those clients, they will typically contract directly with us on our private cloud, and then you can schedule your jobs to happen when you need them to happen. At the same time, you may want to try out this quantum thing. Most Fortune 100 companies today have a quantum strategy. Granted, for some of them, the strategy is to wait and see, but for many of them, it is to actually start setting plans for what to do when, and that involves also buying things now, and getting the experience needed to execute when the technology is useful to their business. That means that they're buying from us, but they don't always have the talent available that can actually utilize this technology.
We will work with them with our professional services team, and we will create quantum algorithms with them and run it on our hardware. Those are three. There's like public cloud, private cloud, and then there is professional services, so joint application development. There's a 4th category, which we anticipate will be very meaningful, but we haven't yet today done any full system sales, but we will. When we went public, we said, "Oh, this will happen sometime in the next five years." However, almost immediately after going public, we started getting inquiries about what will it take to buy one. If we buy one, what's the upgrade path? In essence, we're talking about buying two from the same customer. After having enough discussions about this, it is clear that this is going to happen.
It's going to happen in the next 12 to 18 months. We don't know when, but I don't want to show up in an earnings call and say, "Hey, guess what? I sold a full system for tens of millions of dollars last quarter." That's while a nice surprise, we should say that this is anticipated. This is also why we're already setting up production facilities in Seattle. 'Cause up till now it's been enough to create a computer in our lab and ensure that it runs, and if it starts faltering, we can just apply a physicist and scientist to it, and they will fix it.
If you're selling a system, you need to plunk it into somebody else's data center, and they need to be able to run it, and that puts a whole different level of requirements to a finished setup. We have a facility in Seattle. We just signed a lease last quarter and, basically, we're planning for that to be able to stamp up multiple units of the same reference design. Those are the four ways that we're going to market. We are at the midpoint of our guidance for last year's $25 million. We will be announcing our results shortly, but we haven't done so yet. We will also come up with our projections for next year, when I say next year, for fiscal 2023 on our Q4 call.
Now it's important to note that while our numbers are small, they are admittedly small. $25 million is not what you will hear coming from the mouths of many public company CFOs and say, "This is our results, and we're really happy about it." The fact is we are really happy about it. When we went public, we said that in 2021 we expected to do roughly $5 million in booking, which is also a modest number. We did 16. Actually, we did above $16 million. We tripled our own expectations. I can tell you, at the beginning of the year, I was scratching my head to figure out how we can get 5. This field is developing very, very fast, and there is enough interest in the market to sustain a lot of commercial deployments.
25 is quite the step up from five, obviously, but really it's a step up from 16. When we went public, likewise, we said that for fiscal 2022 we expected to do $15 million in bookings. we have surpassed our own expectations. In reality, we're also doing these sales to seed the market. It's good to get cash, and it's good to train our organization to actually deliver things to customers. We want customers to start working with this technology so that when it comes into its own to be a dominant form of computing, people won't have to learn everything overnight. It's important that people start working with this technology today, and that is what we're seeing.
We are seeing interest from a wide swath of companies and industries, from Airbus to Hyundai, and we also have the Air Force Research Labs and UMD. Like, there's a lot going on there. As in any new technology, any revolutionary new technology, the early stages of it has often been funded by government. We have taken very little in government funding, and we don't have any grants today. We're not seeking grants. We are seeking sales. We are now selling to several governmental bodies and academia, which often is a governmental body. On our way to where we think that the market will be going is that for, like, computers, for a long time, you sold them to research institutions, and then it flipped, and it went into enterprise.
We're gonna see that same development happening to quantum.
We've got about five minutes left. I just wanna pause and see if there are any questions from the audience. Otherwise, I will continue with my questions.
Can you talk about supply chain issues going forward for scaling? You know, the specific topics, that's relevant to you is lasers, cost, reliability, going forward. There are clearly identified FTQC as a critical component to scaling.
Mm-hmm.
What's your approach to that?
Our approach-- First off, you're right. And not just about lasers, but lasers is an obvious part because it is one of the most expensive ingredients into an ion trap computer. Our solution is to work very closely with manufacturers of lasers on a worldwide basis. I went and visited with one of our suppliers, actually 6 months even after I started. I went to Europe and visited with manufacturers to see both, what they could do, 'cause Our desires for what the future of lasers should be, and also to secure the supply. What we will do is that we will share our manufacturing goals and plans, and then instead of when we know them for sure, send them, we will adjust them.
That means that they have a longer lead time to make what they need to make and to buy the parts that they need to make. It's fascinating to see how far this has come and how there is a renewed interest in lasers now. The lasers we use are pretty much run-of-the-mill. We don't need to build a reference design for a laser, give it to a manufacturer, and have them make it. We can take what they have, and we can make a couple of tweaks to them if we need to, and that is what we're doing today. Our lasers are about the size of two shoeboxes. If you actually go and visit with these manufacturers and go to their back room, they have lasers that do the same thing, and they're this big.
That they were worried to bring these to market because, well, we worry that people won't want them. They're not as flexible as the big ones. I'm like, "Well, okay, we'll take them." That is fun. That's fun work to see that you can actually do this. It was also great to look at the light in their eyes and say, "Oh, people will buy this." You know, laser manufacturers aren't necessarily the most commercially attuned people in the world. They're not making, you know, pop records. There are a number of very established laser manufacturers in the world today whom we're working with and that we have tight connections with. There are also a bunch of newer folks who are focusing on miniaturization and pushing the limits of what lasers can do.
I think that we will see radical development in laser technology over the next two-five years.
We take a similar approach, whether that be optics, chips, technology, across the board with partnering on working with our vendors.
That's right.
Well, looks like we're sort of up against the end of the session. Thomas, Jordan, thank you very much for joining us at the Needham Growth Conference. We really appreciate you spending the time with us today.
Thank you, Quinn.