Joe Ghalbouni is interviewed by Yuval Boger and describes his journey from quantum communication research and teaching in Lebanon to leading quantum and AI initiatives at Point72 and founding Ghalbouni Consulting. They discuss how to educate non-technical audiences, identify sector-specific use cases, and build roadmaps for quantum, AI, HPC, PQC, and QKD, including quantum risk assessments and migration to post-quantum cryptography. Joe highlights high-value financial use cases in optimization and quantum machine learning, the role of quantum-inspired methods on classical hardware, and the practicalities of deploying production systems with SLAs, security, and hybrid cloud/on-prem access. He argues that algorithms and domain understanding are today’s main bottlenecks, gives an optimistic timeline for quantum usefulness.
Transcript
Yuval: Hello, Joe, and thank you for joining me today.
Joe: Hi, Yuval. Thank you for having me.
Yuval: So who are you and what do you do?
Joe: I’m Joe Ghalbouni. I have a PhD in quantum communication. I’ve been in actually in, I obtained it in 2013. Then I’ve been an associate professor of physics for eight years at the Lebanese University at the Faculty of Sciences. And then I made my research. I moved to the industry, starting with the Point72 hedge fund, the Point72 asset management. I was working with their innovation team that is based out of New York and Stanford, but by being located in Paris. And I’ve stayed there for almost four years, and I have very recently started my own company called Ghalbouni Consulting.
Yuval: Wonderful. And in Point72, I think you dealt with quantum. So was it too early?
Joe: So that’s a very good question. I would say the approach at Point72 when I joined was to really look at innovative technology very early. So the momentum, if you remember in 2020 for quantum really took off. And at the time, the strategy was to say we want to be part of that moment. We want to be part of that quantum moment. And we want to see how does that apply for us in finance? and how can that give us a competitive edge?
Yuval: And in retrospect, I mean, what were you able to achieve without, of course, without revealing any state secrets?
Joe: Absolutely. So what I can mention is, as someone who came from academia, who didn’t really have any experience in corporate, that was a first challenge, first of all, understanding the corporate mindset. And second, it was to, how can I approach a technology that complex with an audience that is not, you know, an audience that is studying physics, just like it’s been the case at university. And the good news about it is the enthusiasm that I have seen internally of people wanting to learn about this technology. So what we were able to achieve first was really starting from an education perspective. What is a qubit? What is superposition? What do we mean by quantum computing? What does it exactly do? Is it there to replace what we currently have? Is it there to complement it, et cetera? And so that’s why it’s really been a great journey starting from this point. And then once people started to get a little bit more familiar with what quantum is and more importantly understanding what it is and what it’s not, because I think that was really a task that we needed to do. And I think it’s valid everywhere in sectors that are not directly. directly dealing with quantum computing, people started, you know, coming to me and telling me, well, I have this problem, which I think might be interesting for quantum. And to me, that was a great exchange because I had the quantum expertise, but I did not have the sector expertise. So I did not really know what were the use cases that they would be interested in dealing with. And from there, that exchange started, which led to doing, you know, more and more targeted sessions where we would identify use cases and then start to do POCs and start to find solutions together. And I’m really speaking about early days. So back then, we were in 2021 and I stayed at Point72 until 2025. So we really had a long time to look at the different roadmap of quantum technologies.
Yuval: And today you have Ghalbouni Consulting. I hope I said that correctly. What does Ghalbouni consulting do?
Joe: Yes, so based on my experience that I’ve gained at Point72 and since now I have a corporate expertise as well as my academic expertise, I now use the two in order to do the following. First of all, I go to clients who want to adopt emerging technologies and not just quantum, but also AI, because I’ve had the privilege of heavily working in AI, especially generative AI when I was at Point72. And so this really helped me strengthen that field, as well as high-performance computing, which, as you can see now, is quantum is an integral part of HPC. It’s not something on the side. So I help companies adopt these technologies, help them do their roadmap, help them understand what’s the best way for them to have a return on investment on these technologies, and make sure that they can leverage the technology without having to hire, let’s say, 50 quantum computing. scientists and you know maybe they don’t have the means to do it does that mean they can’t leverage the technology well of course that doesn’t mean that so that’s where I really help them I help their internal staff actually understand how they can benefit the technology on the other side I also help companies on the other side of the sector so the AI quantum and HPC in case they need especially advice on how to go to market because what I’ve seen being from both worlds I’ve seen that there’s still you know that bridge that needs to be built where vendors actually understand what the sector wants. There is an expectation a little bit that we’ll go to the to the sector and tell them this is what our solution does. But the sector needs to understand why this solution works for them. And frankly, the different sectors have been doing great efforts in understanding how these technologies can be leveraged. But I believe that there is an additional effort that needs to be done on the vendor side. And here I’m especially speaking quantum vendors because this is a very emerging technologies compared to AI where they understand okay if we go to a financial sector today let’s not just talk about portfolio optimization which remains an umbrella of various problems underneath but really start to show effective use case understand how they compare with what people do today classically so have proper benchmarks and then explain to them in their terms this is your expected their ROI in a year, two years, three years, et cetera.
Yuval: Let’s assume that there was a hypothetical company, 0.73. So it’s like Point72, but you don’t know the politics or you don’t know the people. And they came to you and said, okay, we hear about this quantum thing. Should we get started? And then, if so, how do we get started? What would you tell them?
Joe: That’s a great question. And that’s really the question that I have to answer, as Ghalbouni consulting. So the first thing I would do is really try to understand what is it that they already know about this technology. Because again, we don’t want always to start from scratch if they already did some work on their side. And if they are starting from scratch, that’s perfectly fine. But I’m just trying to understand where they are at the moment. And then we start to understand how this technology actually applies to them. I’m not going to a sector and tell them you should do quantum if they don’t need to do quantum. Maybe quantum is not for them at this moment, or maybe there are not enough use cases for them to actually address. And also being very transparent in telling them your sector might benefit from quantum, let’s say, in one year, in two year. While another sector, I might tell them, well, quantum is for you in five years. But that doesn’t mean you shouldn’t be prepared. So based on what we have actually identified, I’ll tailor a specific roadmap for that. 0.73 hypothetical company. But also another aspect I’ll be looking at is the cybersecurity aspect, which I think everyone should be looking at. And given my background in quantum communication and my knowledge in quantum cryptography and cryptography in general, that’s where I would do a quantum risk assessment, where I would be looking with them on how to actually, what are the vulnerabilities that you have at your company? What are you using? What are the different cryptography mechanism. And then we would put a strategy in order to move to PQC if that hasn’t been done already. But also, I will always insist and tell them to keep in mind that QKD is about, it will happen. Like there’s no way you can say I’m not going to do QKD. I have heavily been advocating for QKD for a very simple reason is that the security of quantum key distribution lies at the hearts of quantum physics. It’s inherent. The security is there. It’s not a conjecture where you are supposed to have a mathematical problem that is intractable. And then you find an algorithm and you break it. No, there’s really a security at the heart of these quantum communications. And that’s why to me, QKD is definitely the path forward in the end. It’s not just PQC. PQC is a transition.
Yuval: When you speak about security and communications, that cuts across multiple vertical industry. I mean, of course, financial services have security, things that they need to solve, but also logistics or pharma or government or others. But when you go to the specific quantum computing problems, they seem to be a little bit different between different sectors. So do you feel that you’re able to advise, say, a logistics company or a government institution on quantum computing?
Joe: So that’s an excellent question. and I think this is really a very important point to take care of when you go and approach a specific sector. Of course, I’ll be very transparent and honest with you. It’s way easier for me right now to go and address problems in the financial sector because I have been exposed to that sector. Now, if I go to the logistics, I have the background with my PhD for the math and the physics behind it to understand how this problem should be addressed. but there is homework on my part that I need to be doing. If I really want to be a good advisor, I really need to understand their work. So there’s definitely at least a couple of months or one month where I need to be up to speed with what it is that they are doing and understand their work. For healthcare, it’s even bigger because there’s biology, there’s chemistry in it. So I’ll definitely need to sit down with them and understand what it is they do. But once I understand how these problems face, a computational bottleneck, that’s where I can really take over. But yet, to your question, I can’t approach every sector from a quantum computing use case in the same manner.
Yuval: So let’s go back to financial services. You’re right that a lot of quantum computing vendors are attracted by financial services for obvious reasons. But then when you start looking at individual problems, you ask yourself, okay, well, is the classical solution good enough? I mean, is there a problem that really needs to be solved? is the amount of data or the speed in which something needs to happen is too high for quantum computing. So you can start paring down the relevant problems. In your experience, what are relevant areas for quantum computing in financial services?
Joe: Okay. Awesome. Great question. So mostly if I can consider two categories of problems and then break them down a little bit, which benefit financial services are obviously optimization problem and machine learning problem. Now, optimization problem, we have to be very careful what we mean by optimization and where are the different algorithms that make sense. If you take, for example, a portfolio, right, and you need to allocate a specific budget to different assets in your portfolio, you really need a very powerful algorithm like an HHL. And the reason behind it is because you have, first of all, you’re not going to beat a convex solver. So you’re not going to take a convex formulation and try to beat it quantumly. You might be able to do so, but that’s not really the point, because convex solvers are extremely optimal in classical computing. Where it will be interesting is when you start to add constraints that are going to get your problem non-convex. That’s one thing. Second thing is the whole discretization. It’s very complicated classically to actually do a discrete distribution. while for quantum it’s inherently discrete. That’s a big advantage that people sometimes do not pay attention to where quantum can really be a game changer. And so when I go to such problem, I want also integers in terms of value. I’m not looking for binary allocation. That’s why I don’t consider QAOA to be the best algorithm when it comes to portfolio allocation. However, if you do index tracking, where you are selecting a subset of your S&P 500, for example, to understand how the index evolves, that’s where QAOA can be interesting, because then a binary return is really what you want. You want to make sure, do I select this asset or do I not? And so, when you look at your problem in that form, that’s where the optimization in those particular areas start to become interesting. So we’re not ditching altogether the classical solution. Rather, we’re making it more efficient using quantum parts of it, or we are even addressing for some cases where it becomes classically intractable, we are adding the quantum component. And for the second, for machine learning, where we really see a lot of interest is really twofold. One of them is being able to train a model with less data. And today data is extremely costly. When you say, I want financial data, that costs a lot of money, actually. And so if you want to train a model to detect anomalies in a time series, or you want a model for predictive output, if you have less data and you are able to get accurate results, that’s a lot of value for finance because you are saving a lot of money. And that’s where quantum machine learning is very interesting. Another aspect where it’s also interesting is in terms of actually being able due to the quantum feature mapping in machine learning, in quantum machine learning, to give you patterns that are classically undetected. So think of it as anomalies in a time series. You might miss them in a classical algorithm. But with a quantum algorithm, there’s a chance you might actually catch these anomalies. Why? Because your whole pattern of detection is being done differently. Another aspect, which is interesting for finance and for many different sectors, and that’s where there’s that convergence that goes back with AI, is let’s say, the fine tuning of a model. So you can fine tune a large language model with a quantum algorithm with much less data for the fine tuning than you would do for the classical fine tuning. And creating data for fine tuning is very time consuming. It needs a lot of curation. So being able using less data to train or fine tune, sorry, your model as accurately as a classical way or even better, that’s a big game changer for quantum. And that’s also where finance can benefit from.
Yuval: Do you think QAOA and other variational algorithms are here to stay or they’re just a temporary solution?
Joe: So I cannot give a definitive answer on that one, to be very honest. But I kind of feel that QAOA has been a good introduction at the start. I might see it useful for problems like index tracking. But again, in terms of the scalability of QAOA, when we go to the fault-tolerant era, I’m not sure that this will really scale well with the type of problems that we have. And VQE, I mean, we have seen a lot of example in quantum chemistry where VQA is a good toy problem, but the scalability becomes really questionable. And I think these algorithms were good when we didn’t have the machines we have today, and also comparing to the machines we’re going to have in two to three years. So I would say yes or no. I would be surprised if they remained the most important ones, let’s say it that way.
Yuval: Let’s assume that you and I developed an algorithm. We went to a financial services company. They tested it on a toy problem. They tested it versus the classical solution. They feel that, you know what, it’s good. And we should take it to production. How does production look? Are they comfortable in your experience working on a quantum computer that’s on the cloud? If so, with what security? Or if not, do they need it in-house?
Joe: That’s a brilliant question. And a point I actually address as part of my consulting offering when we go from POC to production. So going to production and here I’m going to particularly focus on financial services because that’s the one I know the most. It’s an extremely delicate process because for one reason, when you say that you have asset managers or even banks that are operating with a system that is in production, you expect a 99.99% of uptime. That’s really a first condition. You expect proper SLAs where you understand if there’s a downtime. When is it going to happen? And what’s the reason? You expect to have what we call postmortem in case one of the system goes down. I mean, we’ve seen what happened with Cloudflare a week ago, right? It affected the whole internet. So if you are relying on a device, of course, people are going to be more comfortable if the device is up the whole time than if there’s a downtime. And it comes back to your question. Do we want it on the cloud or do we want it on premises? The best option, in my opinion, is a hybrid approach. Why? Because if you’re on the cloud, you know that you don’t care about the maintenance yourself. all the software updates are done. You know that there’s a company that has signed a contract with you for an uptime. You know that you can have more capacity in the cloud. You know that you can have the latest devices on the cloud as soon as there’s an update to the hardware. So it’s a good solution. But when it comes to cybersecurity, yes, that’s going to be very critical because companies usually either want a private link or what they call a direct connectivity to a center. So that would mean more quantum computers in those HPC centers. Now, for quantum vendors, that’s a very good news because that means they need to sell more systems. And selling on-premises is a good option as well. First of all, if you need your cybersecurity or if you have very low latency requirements, where you need something in your own HPC that you have built. So if you rely in a system, in a quantum system in production, My personal advice to a company would be to have both systems. Now, of course, we’re not talking about this happening this year, but eventually on a roadmap, on a three to five years roadmap, that’s what I would strongly advise. Do not rely on a single point of failure.
Yuval: What do you think is holding the industry back today? Is it that computers are not sufficiently large? Is it that the algorithms are not sufficiently developed, the software environment, the education, I mean, and please don’t say all of the above. What’s the most pressing problem in your opinion?
Joe: So first I will say all of the above, but then I will say what are the most pressing. In my opinion, the most pressing is the algorithm. The reason being that even with today’s machine, there are very interesting things that we can do. But people do not necessarily know how their use cases actually map into these algorithms or what algorithms we can have that take into consideration these use cases. And I really think there’s a big disconnect. In my opinion, if you want, if you want to properly have a quantum solution for a sector, if you want to propose it, you need to understand 10 over 10 the sector, 10 over 10 the quantum part. And that’s very rare, sorry, to have these two profiles together in one person. And not trying to do any advertising for me in that regard. But that’s really where I think I bring an added value because if I go to a sector and I don’t understand it, I can advertise my algorithms for 24 hours. No one will listen. No one will understand what I’m saying, even if I break it down. So to me, that’s really the most important point. The hardware, yes, it’s still something that is extremely important and crucial and we need to address it, but we already have roadmaps for the hardware. We already know where we’re going. going. It’s not like we don’t know. But for the software, yes, because I’ll be very honest with you. We might reach a point where we have a very powerful quantum computer, but we just don’t know what to do with it. And frankly, we’ve seen that with AI. We have very powerful AI tools, and we’re just learning how to use them. Like, for example, prompt engineering became a thing because now you have access to the model, but you need to know how to properly ask questions. And has everyone really gotten the proper training to ask an AI chat, but what it is that they want? No. We’re going to face the same problem with quantum at an even higher complex level, because simply we have the machine, but we don’t know how to use it. So to me, the algorithms are the most important part we should be addressing.
Yuval:. Given that you have both the industry experience as well as the academic background, So you can read beyond the press release and the hype. What do you think is the most compelling results or use case that you’ve seen in the financial services industry?
Joe: So to me, I have very much been following the work done at J.P. Morgan in terms of the financial services. Because the approach of J.P. Morgan was not only to list what can be done on fault-tolerant devices, but rather on NISQ devices and then scale from there. So that’s one approach that I really like. Another approach that I found extremely relevant and for the financial service are all the quantum inspired approaches. The reason being that today, a financial service or any sector for that matter, cares about gaining computation efficiency. If you tell them, I have found a way that uses a quantum approach but that executes on GPU, they’ll be all ears. They don’t really buy the solution because, of the quantumness of it. They buy the solution because of the efficiency behind it. So if you found another way which people will call classical and that’s fine, tensor networks, they want to call it classical, they can call it classical. But it pushes back the limit that we have, at least it’s a stopgap. You’re going to sell. And we’ve seen companies offer that. We’ve seen multiverse do it, for example, and there are other companies that have been looking on that aspect. So to me, these These are the works that have been the most relevant for the financial sector.
Yuval: One question that gets asked all the time, I really should check what Polymarket says about it, is how long before quantum computers become commercially useful? Some people say, oh, it’s going to be 10 years, some five, some two, some three. What’s your guess?
Joe: So I’ve always been more optimistic than pessimistic, and I’m going to to say why in a second. But certainly I would never listen to people who have been telling 10 years for 10 years. Like 10 years ago people used to say 10 years and those same people tell you 10 years. No one has a clear understanding of that roadmap for one simple reason, not because of the hardware, but because of the mapping of the use case. They still do not understand to this day how a use case properly scales. They do not understand what is the relevancy of that use case. compared to what they do classically. To me, I wouldn’t be surprised if I find quantum advantage next year. Now, people are going to tell me, no, this is way too early. Quantum advantage is not, I’m not saying a quantum advantage in the sense of over all the use cases that we know, but it could be one simple use case where we’ve seen there’s more efficiency. And from there, the momentum will pick up. The reason I’m very optimistic is because if we take a look at this year’s Nobel Prize with Michel Devoret from Yale University, I’ve actually had the tremendous chance of meeting him in person when I was doing my PhD because he graduated from the same school where I got my PhD. And so he visited at the time our lab and he saw my experiment and they were extremely happy at Yale University that year, that’s 2011, if I’m correct, because they were able to conceive a four-qubit superconducting chip and applying a first two-qubit CNOT gate. So look where we were and look where we are today. So when people keep telling me it’s in 10 years, 20 years, 30 years, like what’s your argument behind it? It’s simply because you expected it to happen in five years and it didn’t happen, so you’re pushing back the boundaries even more. So I don’t see it that way. I see that what will also accelerate it is the integration in HPC centers. And that’s why the NVLink, in my opinion, is a very strategic piece of the puzzle because that’s where we now have that close intersection between HPC centers and quantum. And that’s where we’re going to see, in my opinion, the magic happen. And if we look at what NVIDIA CEO, Jensen said, like one year ago, he said one time quantum is in 30 years. And then the next month he was having a panel with a lot of quantum CEOs. So again, there’s no strong foundation to say the timeline is very far away. we’re closer than we think we are. I’m definitely not in the pessimistic kind of people who say, no, this is going to take years and years before it happened. What might take years and years is having a global quantum advantage across all sectors for all use cases. Yes, that will take more time, obviously. That’s not going to happen quick. But the quantum usefulness is going to happen use case per use case. It’s not going to happen overnight for everything. So is it going to be in 10 years? no, my personal guess would be, I wouldn’t be surprised if it is in a couple of years.
Yuval: So speaking of Nobel Prizes, but not limiting ourselves to that, I wanted to ask you a hypothetical. If you could have dinner with one of the quantum greats, dead or alive, who would that person be? Oh, my God, that’s extremely difficult to answer.
Joe: There are so many that come to mind. But probably Richard Feynman. Definitely Richard Feynman. Yeah.
Yuval: Very good, Joe. Thank you so much for joining me today.
Joe: Thank you, Yuval. Appreciate it. Thanks for having me.