Quantum + AI: Could This UK Startup be the Next NVIDIA?
Gerald Mullally, chief executive of Oxford Quantum Circuits, talks about the Oxford spinout's quantum computers in commercial data centres in London, Tokyo and New York, the last built with NVIDIA and Digital Realty. He explains the move from physical to error-corrected logical qubits, the energy case for quantum, and how quantum computers could generate synthetic data to train AI. Stephen Horn and Laila Rizvi conduct the interview, with Autria Godfrey in Washington.
Transcript
For years, quantum computing has promised to transform everything from drug discovery to cybersecurity. But the question has always been when. Oxford Quantum Circuits believes that the answer's now. Backed by partnerships with NVIDIA and Digital Realty, OQC is scaling from research lab to global player, building hybrid quantum AI data centres in the UK, Japan, and also in New York. Its new CEO, Gerald Mullally, isn't shy about ambition. Some even call OQC the next NVIDIA. So what will it take for the UK's homegrown quantum startup to deliver on that comparison?
Welcome to "Agents of Tech." I'm Autria Godfrey in Washington DC and I'm thrilled to be joined today by my co-host, Stephen Horn and Laila Rizvi, coming to us from across the pond in London. All right, you guys had quite the interview, you spoke earlier today with Gerald Mullally about whether quantum computing is finally ready to make a difference outside the lab. Did you get a clear answer from him?
What we do know is that it's not an either or with AI versus quantum. And I think that's very much been the question on people's mind because people are very panicky about where they're putting their money right now. Should it stay in AI? What's the ceiling for AI? Should they move it to something like quantum? What's going to be the next kind of AI as well? That's what people are looking for an answer for.
I thought it's really interesting, Autria, because you know, to some extent that it's still exactly as you laid it out, which is it's going to change the world, but not yet. You know, it's not changing the world right now, but it's being used right now. And you know, what he was able to do was really set out not just the roadmap if you like, but kind of when we were going to get to it and how we were going to get to it. And he also was very quick to sort of point out, you don't need to know how a quantum computer works to be able to make use of a quantum computer.
So, it was really good. He laid out the investment case for quantum, laid out the time scale for quantum and, as Laila just said, what almost most fascinating of all, which is how AI and quantum sort of combine to be the next biggest thing. And, hey, that is a big thing.
Not needing to know how it works to be able to work it is music to my ears, that.
We'll get you a quantum computer for Christmas, Autria.
There we go. Alright, before we get to this really compelling interview, we do want to hear from you. We want to know what you think: is quantum computing finally ready for the real world? Do you think that Oxford Quantum Circuits really could become the next NVIDIA, powering a new era of AI and quantum? Weigh in in the comments, sound off. And while you're there, don't forget to like and subscribe. Now, here he is, Gerald Mullally, take a listen.
First of all, thanks ever so much for coming and talking to us. We appreciate it.
Yeah, thank you very much for having me.
Lovely to see you. Now maybe we kick off, tell us a little bit about Oxford Quantum Circuits.
Sure. So we spun out of the University of Oxford in 2017. We build quantum computers, we integrate them into commercial or co-location data centres, and then we support our customers to solve their hardest problems. We were the first commercially available quantum computer in the UK, the first quantum computer as a service company in Europe, the first company in the world to integrate quantum computers into commercial data centres. And we've now done that a number of times.
We have one in London, we expanded to Tokyo with Equinix, which is one of the world's leading and largest co-location players. And just last month, we launched the industry's first quantum AI data centre right in the heart of New York and the first ever quantum computer in New York City, that was with our partners NVIDIA and then also Digital Realty.
That sounds so exciting.
Yeah, yeah, it was, it really was actually. I got to meet Jensen, which is great, and you know, learn some things in terms of what I need to do into the future, but it was cool.
So are these real quantum computers?
Yep, very much so. We're now on our fourth generation of system and maybe if I step back and just explain where we are and maybe where the market might go. Hermann Hauser, who's, you know, the founder of ARM and kind of a true pioneer more generally in compute. He's a lovely framework for explaining how quantum is evolving.
We've now left what's known as the physical era. And the physical era, you literally got a chip and you printed physical qubits, quantum bits, kind of similar to what transistors are in classical compute, on a chip. Now the problem with that era is you increase the number of those physical qubits, which you have to do, increasing the scale of the computation, the performance of the chip got worse rather than better. In other words, the errors were just too much overall.
And I think what's happened in the last 12 to 18 months, and maybe why there's so much more interest in quantum today, we can get into, there's a, you know... AI is definitely another reason for that in addition. But from a technical point of view, we've now crossed a bit of a threshold where we can increase the scale of the systems. In other words, increase the number of physical qubits, but decrease the errors at the same time. And that crossover is now known as entering the logical era. You take a number of those physical qubits to produce a logical or an error-corrected qubit and that's what's useful for computation.
Now there's a couple of key things within that, is that there is then three big milestones all of the industry is going to go through. Today, the kind of leading systems globally can run thousands of quantum operations before error takes over. We then go through three big milestones, millions, billions, and trillions of quantum operations. Now the important thing overall, and definitely for, you know, listeners more general, and we'll probably get to, you know, hype and not hype and so on later, is that at thousands of quantum operations you can still classically simulate that system. So in other words, today it is not more powerful than a classical computer, but once we head into millions of quantum operations, that's the first time the quantum moves past the capabilities of classical compute. Billions is well into it and trillions is really the transformational side.
So when we hit into millions, there are already quantum algorithms that exist that corporates and researchers have developed that provide known advantage over classical compute. So when we get a question on is quantum three years away or is it 10 years away? It's actually both. And I think it's important not to overhype that, that this is going to be a process over a period of time.
Got you. So are we at a stage of quantum applications maybe as opposed to quantum computing?
I think people are running real applications today. And the reason they're doing that is, it was a couple of reasons. So if you take, let's say, some of the leading banks or some of the leading pharma companies or energy companies, they have small, very skilled teams who are taking their proprietary classical algorithms, developing quantum versions of those algorithms, seeing is there a performance improvement. And sometimes there's not, you know, the classical algorithms that will be better, but where there is, they then plotted on that graph that I mentioned, that millions, billions, and trillions, and then understand actually, okay, this is coming in three years time at millions of quantum operations.
I now need to understand fully who has the best performance system, who has the best roadmap for scaling faster to millions? And really importantly, can I actually use this quantum computer? In other words, is it within my digital infrastructure, is it within the data centre? And then also is it something actually more generally than outside of just me as the quantum expert who can programme a quantum computer? Are they building tools that allow a non-quantum expert data scientist or an analyst to be able to actually use and run a quantum computer? So there's a lot of real stuff happening today, but it's not stuff that runs into production for corporate, it's much more on the R&D side.
Demis Hassabis, who's the founder of DeepMind, he has quite a controversial take about quantum computers versus classical computers, where he says that classical computers are actually massively underappreciated and have untapped potential in the respect that they can actually compute huge combinations of systems like climate systems and protein folding. So for example, with AlphaGo, a chess board has potentially ten to the power of 170 different moves that you can make, which is more than the atoms in the universe. So what do you think of that thinking of classical computers potentially being able to do the same thing as what people in quantum computing claim that quantum computers can help us with?
Yeah, and I think there's no doubt he's correct in the sense that there's a lot more to come from classical compute. I think the question that we would always ask is at what cost? And that's both from an energy point of view and then also a monetary point of view. Like if you take the power consumption of classical compute, an average supercomputer is anywhere between 10 and 40 megawatts of power. While a quantum computer is just 35 kilowatts of power. And that's the difference between powering 50,000 homes per year versus nine household kettles or six hairdryers.
Now the key thing there, it's not just differences in the orders of magnitude. A key difference, and this is the most important point, is that quantum computers break the link between energy consumption and computational performance. And what I mean by that is that, you know, the chip is the kind of the key part for us. The architecture of the chip is key, but we then put it in, effectively, if you've seen those beautiful chandeliers, that's a dilution refrigerator, it's basically a fridge. We turn that on, it goes cold, but we can actually then replace the chip that's in there with a chip that's orders of magnitude more computationally performant, but it's actually the same energy consumption, it's the same fridge.
Now that will change slightly over the years as the generations move forward. But we're looking at about a hundred kilowatts of power in total. But the key thing is that in classical compute, we're now on the pathway towards nuclear power to power our data centres. This could be the biggest potential of quantum compute, that actually that link is broken, that we can get orders of magnitude more computation from approximately stable levels of energy consumption.
I do think there's a huge potential in furthering classical compute, but still there are problem sets that are truly intractable to classical compute regardless of AI. And a nice one that's usually used is the breaking of encryption. Classical compute will take 2 trillion years to break encryption. So you can certainly do it, but at what cost, what energy consumption? For a quantum computer, when we have something between billions and trillions of quantum operations, an OQC system will take eight hours.
But then the question is how many of our problems actually require brute force? And I think that's more Demis's point, where actually a lot of phenomena you can see patterns and structure in. So you can train classical computers with that kind of premise. So do we really need brute force? Do you need brute force for, you know, an AI chess algorithm to beat a grand master? No, not necessarily, and I think that's what I think is the question at hand.
And I think as well if you look at the position of the big tech players, like Google are investing really heavily in quantum in addition to really heavily in AI, and I think the future is very likely to be quantum AI. And how much is AI and how much is the quantum side I think depends on the problem set overall, but we definitely don't see it as adversarial. Like we have a very close relationship with NVIDIA. We're doing a lot of work to figure out what the true relationships are between quantum and AI.
And I think there's training acceleration, augmented inference and efficient data generation are the kind of three big areas where quantum will power the future of AI. But two of those, it will be probably smaller impacts. So training, acceleration, augmented inference, they're likely to be smaller, but important improvements. It's really efficient data generation is the true potential of quantum supporting, empowering AI.
And the reason we say that is that AI's already consumes most of the world's data to be trained on, it now needs as this new sources of data, and this critiques about, okay, the AI is then producing that data that's then training AI, and there's challenges with AI, and this is the classical compute challenge. There's challenges with AI efficiently producing data that is like life, because life is quantum systems. So if you take something like molecular interactions, A, there's not many sources of that data. And they exist usually within biotech companies and within pharma companies, which is not really easy to access or certainly not train global models on.
Although Larry Ellison would like to.
Wants to do that. Yeah. Exactly. Which is quite cool overall. But quantum will be very good at synthetically generating that type of data because it can perfectly simulate a quantum system. And if that's possible, in other, other words, you don't need the types of investments that we're seeing in Oxford to create those sorts of institutes. If you're not able to access the small data pools that exist within biotech companies or pharma companies, a quantum computer will be amazingly good at synthetically producing that data that AI then will be trained upon. And that's the true power of quantum and AI and in that relationship. And I think that's a really good one for the future of AI.
Let me talk a little bit about your relationship with NVIDIA. And one of the things that we've been really into in the podcast for the last few shows is looking at the global state of AI and the role for countries like the UK in that, you know, because we've got the big blocks of China and the US, but obviously, you know, we are in it too. And so it'd be really interesting to us to get your view on that, on how the UK competes on that global scale and your relationship with NVIDIA.
Yeah, so it's a really important question and I think the first one is quantum is really important to the future of the economy and national security. We did a piece of work with Oxford Economics and the impact on the economy is about 7% productivity improvements, which is a big deal, certainly in terms of what we're striving for.
Could really do that now, couldn't we?
We could do that now, that's medium term. So you know, it's towards 2040 before we see that. In other words, you're well into the trillions of quantum operations before you see that sort of productivity improvement. But that has a value to the economy of about 212 billion and that's the entire region of Liverpool. So it's really important for the future of the economy that we get this right.
The second element is the security impacts we've spoke through in terms of the kind of future breaking of encryption. But on the defence side, there's lots of applications of quantum and some very serious ones in terms of future defence and security for the UK. But it's really important on that basis that we, A, get the ecosystem right. So you've got all of the big players here in the UK competing in the UK, supporting our corporates and our government and our civil society to improve what they're doing through the use of quantum compute.
But there's also an argument for it to be sovereign. In other words, if it's the case that we're actually dependent on other people's technology, is that a good thing for the UK longer term? Do we see the proper maybe spillover benefits that we would or could if we had a global giant that was a British company, not just an American company with a subsidiary or part of its company in the UK? And that sovereign argument I think is really powerful and that ultimately means how do we get that right?
Yeah, because DeepMind was originally a UK company and now it's part of the US, so it would be interesting to see how we actually get past that kind of barrier of exiting to the US market. Which I agree is a bit of cultural issue. I just wanted to loop back round to your relationship with NVIDIA and I know Jensen was quoted saying that you would be potentially the next NVIDIA, which is a huge kind of yardstick to measure up to given that it's the most valuable company.
We have a way to go yet though.
Yeah. But NVIDIA obviously scaled, you know, aggressively to become the most valuable company in the world within 20 years. So what do you kind of say to that, and do you see quantum achieving that same scale so quickly, and do you think you'll get there?
Yeah, so I think there's kind of a couple of things we learn from NVIDIA and I think there's nice parallels which we definitely look at from a scaling point of view and a business model point of view. I think the first one is just their core tech. You know, it's the GPU itself and what they did with that over many years. We have the fastest speed ever benchmarked from a QPU point of view. We have the industry leading efficiency from physical to logical qubits that I mentioned. So we two physical producing one logical, and that ultimately means we can just get more from a single wafer, right? That's the kind of the easy way of understanding that.
So we have an incredible core technology and that's what Dr. Peter Leek, who's our founder and chief scientific officer today, that's what he invented in 2017. And that's what we keep working on, what we keep building. So I think from an architectural standpoint, we're in a great shape overall.
I think the second thing, and I think a real learning from NVIDIA, is the extent to which they integrated, they're masters of integration. And I think that's something we've learned from them on. So if you take another good example for a company in the world to integrate quantum into data centre, no one's done it since. So we're still out in front on that basis. And that means if you're a company or a corporate or a customer, you'll be able to access our quantum compute securely so you can run real workloads. So this is not about sending data to an insecure lab environment, this is about sending it to what you do with your classical compute. And then we're tightly integrating with NVIDIA in addition, which means that we're really exploring what a CPU-GPU-QPU relationship will look like in the future and what this quantum AI focus will mean overall. So I think the integration side is probably one of the biggest learnings from NVIDIA.
But the third one is the true genius, I think, of NVIDIA, which was it's CUDA, it's building a set of tools that actually formed a bit of a moat around their core technology, because developers just loved using CUDA and obviously CUDA then was running on their GPUs, and that platform that was built around it, and it's a developer ecosystem now, that's a really powerful, and that's the hardest thing to produce, but it's also I think a key reason why they're in the position they're currently in. And that's self reinforcing cycle of we create an ecosystem that then reinforced our core technology and onwards and onwards.
So we'll be launching actually really soon, at the start of the new year, an AI agent approach to the use of a quantum computer. In other words, this is AI for quantum. So it's a set of tools that ultimately mean that you don't need to be a quantum expert, but you don't actually need any real skills to be able to explore and utilise a quantum computer, and that should open up the market. We estimate there's only about 4% of corporates that even have one employee that can programme a quantum computer. In other words, that is really quantum skilled. If it's possible for anyone in that corporate to explore the potential of quantum, just using an AI agent, that's quite powerful.
Rishi Sunak said in his piece in the "Sunday Times" that the workforce of the future, people will be managing AI agents to manage the, the work, if you see what I mean. So that's really fascinating what you say about that integration with AI and quantum then, shows that you're going to be able to get a lot more out of these systems without knowing what a qubit is.
Yeah, and I think the future, you know, if I can say something slightly controversial is that...
No, we love it.
Yeah, yeah. Well, it probably will be that controversial. So you know, if we look forward a bit, kind of my hope for quantum overall is that it's boring. You know, the hype is kind of gone. It's people just going and probably not realising or caring what sits behind what they're doing. They're just want to solve their problem, and one part of that, and might be a small part of that overall workload, is actually running on our quantum computer within their secure data centre infrastructure, fully integrated into their stacks. And I think that's the hope for quantum, that you move out of this kind of hype phase and into something that is actually just about solving our customer's problems.
Well, thanks so much for coming in and talking to us today. We've really enjoyed it and you've really helped us reframe some of these issues. It's going to give us a lot of food for thought. So thank you and good luck.
Thank you very much for having me.
Thank you so much. Thank you.
Really compelling interview there. Laila, it sounds like he made some, you know, comments and statements that might have been a little surprising to you.
Yeah, I mean I definitely, I was saying to Stephen, I came into it as a quantum sceptic, because there's been hype every year about quantum, that okay, yeah, this is going to be the year, this is going to be the year that changes everything. 10 years ago, in 10 years it was going to change everything. And you know, do you remember Web3, then crypto hype, as in there's been lots of these, and AI has actually kind of gotten out of that gravitational pull. So I wasn't sure if quantum would as well.
But I think Gerald, for me, the case that he made in terms of quantum systems and simulations feeding AI in the form of the data it needs in order to scale and to be trained. And we've used up all that data. The internet's been used up and accessing private data is not easy. So he did convince me on that side and I'd never thought of it in that way.
I think it's interesting, you know, because I take what Laila says, I mean, you know, we've been following quantum for a long time, haven't we, at WebsEdge, you know, we've been at the American Physical Society for 15 years, you know, and each year you go it is a little bit like Laila says, it's like a group of the aficionados, but that you've felt that change over the last two or three years and you've found a much bigger kind of ecosystem coming together with bigger companies, with more applications.
And I think today it was that laying that out for the future with that roadmap of kind of where you got to with the kind of understanding that, you know, quantum is not something you can ignore. You know, AI will not do this on its own. Supercomputing will not do this on its own. The planet cannot support AI doing this on its own. And the role of quantum is laid out there almost for the first time for me and I don't think we will look back from this point.
Okay, quick final thought from the both of you on this. Do you think that them being called the next NVIDIA is a correct term? Yes or no?
I think basically if the quantum computing or the quantum chips basically result in a significantly better model, such that people don't want to use the models but which don't use them, then yes they will become the next NVIDIA, because every single AI model will have a quantum chip behind it, if that's what people want and that's where the demand is. But if that doesn't happen, then no.
I'm all in, Autria. I think there's so much demand and I think the way that it's going to be met is by having chips of all different types and sorts in all kinds of different types of devices, probably controlled by AI. So I'm all in, I think we're going to hear much more of this.
Alright, and all in from Stephen Horn, you heard it here. All right, thank you both so much and a huge thank you to Gerald Mullally for joining us. If you've got thoughts on when quantum computing will finally hit the real world and whether OQC deserves that next NVIDIA title, drop them in the comments below. And while you're there, make sure to hit subscribe, like, and share, and we will see you right back here next time on "Agents of Tech." Stephen, Laila, always a pleasure to see you guys. Until next time, goodbye.
See you, take care. Bye. Bye-bye.