Why Quantum Middleware Matters More Than Qubits — Mykola Myksymenko of Haiqu

Quantum computing is no longer waiting on physics. It's waiting on software to stop lying. Quantum computing has raised billions, produced headlines, and delivered almost no durable utility. Hardware improves, claims escalate, Results remain narrow. My guest today operates in the one layer where quantum computing either becomes infrastructure or quietly fails. Mykola Myksymenko is building the software stack that decides whether today's quantum machines ever matter. Welcome to Startuprad IO, your podcast and YouTube blog covering the German startup scene which with news, interviews and live events. My guest is Mykola Myksymenko, co founder and CTO of HiQ. Trained in theoretical physics at the Max Planck Institute for Complex Systems at Dresden, University of Gottingen, University of Magdeburg and the Weizmann Institute. He later led large scale industrial R and D before founding HiQ. His company focuses on a problem the industry prefers to avoid. Quantum hardware exists, but the execution layer required for real applications does not. HiQ builds hardware, wear middleware designed to minimize noise, extend usable circuit depth, and make current quantum processors economically relevant. The company has already worked with IBM, Airbus, BMW, HSBC and Capgemini, released open source tooling adopted by the ecosystem, and recently raised an $11 million seed round while pushing its first product to market. Welcome Mykola hi Joe, it's my pleasure to be on this podcast and it's my pleasure to have made it through the intro without buttering your family name. Your early work in theoretical physics trained you to think in terms of fundamental limits rather than incremental improvements. How did that mindset shape the way you look at computing problems today? I think important moment that we are living right now is that there are many exponential technologies which kind of intercept and accelerate the pace of technology development that we are facing today. So on one hand there is AI that everyone is talking about, but on the other hand there is kind of underlying current which is quantum computing. And as a scientist you tend to understand like what are the kind of fundamental processes which might drive some processes around us. And that's what helps also to understand how these different currents actually can accelerate each other. And while quantum computing was seemingly too far away just like around a decade ago, today we see it already like giving some value in various scientific applications and we see how much it accelerated over just like last two years. And while being in this community and in this scientific endeavor for a while, I can already kind of predict that within next years we will see quite a lot of progress in quantum computing. Such as we may not even predict that in the incremental way. You then move from academia. We already talked about it into industry R and D leadership. Where did you first see the gap between impressive research result and systems that actually work in production? Little disclaimer, the systems never work in production as they normally should. That's true. But one thing that kind of made me shift from academia to industry was particularly that I wanted to work on something which is immediately relevant or immediately can be applied to some hard problems in real life and in academia. My background was in quantum condensed metaphysics. I worked a lot on like predicting and trying to manipulate quantum states of matter in order to discover new materials or new properties of materials. And while it's very fascinating and very interesting field of research, it also has quite long horizon in terms of applicability. So some application of what I was doing may happen, for example in real materials or be discovered in real materials in tens or twenties years from now. And then I shifted to industry where the things were on completely different pace. And here what made me to make this shift was particularly the progress of AI. So while working on the complex, so called the field of complex physics, of physics of complex systems, I was also interested in the physics of neural networks. And apparently a lot of physics theory and methods can be applied into like theory of learning. That what made me being able to naturally transit into that field and made quite actually a good career in industry where we built actual production grade system based on the state of the art research which might just appeared in the literature. And we picked it up and built some systems around that. That was actually very exciting. And I'm still kind of using that on one hand, fundamental understanding of the processes and interest in fundamental science. But at the same time in industry I got a lot of these interesting methodologies and approaches which allow now build very quick and rapid iterations of research in the startup environment such that we can actually quickly arrive into some relevant results which can be applied already today to some specific problems. I see you said you had you built quite a career in the industry. So when did you decide to find high Q? You factually bet that software and not the hardware would decide the fate of quantum computing. What convinced you that this was the layer where real leverage actually sits? Yeah, that's exciting story. So as I shifted to industry, I was always fascinated about quantum computing. Quantum computing is something that I did my master's degree. So I started with quantum computing, then shifted into condensed metaphysics, still worked a lot on non equilibrium physics and methods of simulating quantum matter and quantum states and then worked on AI for a bit. And what I seen in AI was very similar to what I see today in what happens in quantum computing. So we have systems which already are capable of doing interesting results in the very narrow domain. So, for example, you can apply quantum computers, modern quantum computers, to specific simulations of states of matter or some specific optimization problems which might be very hard or nearly impossible to solve on the classical computer. However, there are still classical counterparts, but in some cases, solving some of these problems on the quantum computer can be already faster and sometimes even more convenient. And that's exactly what I seen in the beginning or early stages of deep learning, where these systems were already discovered. And people seen a lot of interesting, fascinating applications in computer vision in like image recognition or signal processing. However, it took around five to 10 years before these systems became like massively productized. And one application for that and one background process which contributed to that embrace of technology was particularly building a bunch of the infrastructure around the research artifacts, for example, like PyTorch system or TensorFlow systems, which allowed a lot of people to build these complex and more and more complex neural networks easier and train them at scale, reproduce the results at scale, such that the whole research community accelerated dramatically. So we kind of lacked that in quantum computing. So running something on the quantum computer two years ago was very hard. Like people experimented literally with like small prototypes of 4 to 10 qubits. Today we already see community which runs applications on hundreds of qubits and hundreds of qubits. This is already a scale which is often hard to reproduce classically. So you need to be an expert in numerical techniques in classical high performance computing in order to reproduce those results. So it went from extremely toy problems to already more or less state of the art high performance computing. So I'm very optimistic about where it all goes. And I think this intuition of combining different experience between fields helps me today to shape our research agenda and prioritizing what to focus on. Out of curiosity, you've been talking about that quantum computing can now deliver what high performance computing as state of the art computers can deliver right now. So I was wondering, if you extrapolate when quantum computing has its full capability, what could this mean? Especially everybody is talking right now about the energy usage of all those data centers that the big AI companies or the hyperscalers in general are right now building, what could that mean? Would it mean like 100% more computing power per what or where would quantum computing go there? And then can you give us a very tiny idea if you have the current LLM models running on a much more capable quantum computer? Yeah, that's probably something which is important to dig in. So many people see quantum computers as just like another GPU or another large computer which runs bigger and faster models. This is a little bit different story. So quantum computers are different kinds of computations and it's suitable perfectly to some specific kinds of problems. And it can be completely useless to different problems. And running classical computations on the quantum computer, it's not the way forward. So it's not enough just like to take your favorite classical computing algorithm and run it on the quantum computer. You need to completely rethink the problem and rethink the mathematical approach to your algorithm to actually use all the capabilities of quantum bits there and then tunnelement the ability to parallelize those computations using quantum computers. So this is different way of computations. And that's importantly to realize that it's probably not immediately applicable to running large LLMs or running like conventional machine learning applications. Actually, if you think about applying big data problems, one of the crucial bottleneck in quantum computing is actually loading data in the quantum computer. So one of the breakthroughs that we did at Haiku was particularly unlocking this capability of loading large scale data sets, which are of the industrial scale number of features into quantum computers. Before that it was just impossible. And there are like small, small improvements here and there which probably enlarge the number of problems that we can solve with quantum computers. But before, before we go into like ubiquitous era, we will be focused first on the very narrow problems. And some of them are designing new molecules and simulating new molecules. That's a very hard problem for classical computers. But this is something where quantum computer can actually help because this is a natural system to tackle with qubits, it maps almost one to one. We can think about the problem that I was solving in my scientific career like a problem of materials. Like if you want to simulate superconductivity or magnetism in some specific new types of metals, then quantum computers can potentially help with these problems. Or computational fluid dynamics. That's another interesting problem which is very widespread and industry. So companies in aerospace or automotive run a lot of computational fluid dynamics simulations and you very quickly actually kind of hit the glass ceiling there with a grid that, that you can simulate on the classical computers. While on the quantum computers theoretically you can go to much more denser grid of your simulation and have much more precise results. However, it's very hard right now to map this algorithm directly. And we also working on that helping some companies who are partners to actually make this transition. So there are A bunch of problems like this optimization simulation, natural systems and multiphysics simulations. These are probably those which will be on the kind of early, early low hanging fruits, horizon and middle to long term. What would it mean to everybody who's now thinking about investing in data centers who are already there, announced in vast, vast numbers? Do you think that'll change something over time when quantum computing really becomes productive, especially in terms of space needed and energy for computing of some form? Maybe still LLMs of artificial intelligence. Look, even from the current perspective, if you want to simulate some process in nature, like for example, a molecule, you can do that on the classical computer to some extent. And I used to work, for example, on some of the largest supercomputing centers in Germany. And those simulations cost a lot. They can cost easily like $200,000. And in order to simulate some material or some specific physical effects while. And you can run those simulations sometimes for weeks in order to get some reasonable result. And we already witness some interesting types of these simulations where we can simulate the same system, still simulatable classically, but we can simulate it already today on the quantum computer, and we can get results maybe in 10 minutes. So even from that perspective, like running large supercomputing cluster for weeks, or running a quantum computer for a few minutes, I can already hint that you actually probably spent much less energy on that. When we will be able to run machine learning simulations at scale, I would say that that probably will transit also to machine learning simulations. I need to be honest that machine learning is still hard to solve directly on the quantum computer. But there are very specific problems, for example, anomaly detection, which potentially quantum computers can actually solve more better than classical computers. And we already see few of those examples here and there. But ubiquitous machine learning, where we, for example, apply it to LLMs, it's probably some years away from now. Some years. Okay, I see. Let's get back to the original questions I envisioned for this interview. For founders watching quantum from the outside, noise, not qubit count, is the real reason most pilots fail quietly. What actually happens inside these systems that makes noise such a dominant constraint? Yeah, so you should think about quantum computing as basically early stages of classical computers. Like try to remember what happened in classical computing in between 40s and 60s. So these were like, these were large machines. Like sometimes they were like of the size of a room and they were very hard to operate. They did not have error correction at the time. So basically they were operating with valves. And those valves can very easily overheat or become malfunctioning. So Literally the first bugs were in those computers, just because some things did not function well. So you should be a high level hardware operator and understand the low level algorithmic theory in order to run something on these machines. And that was for a while before error correction was introduced to these machines. And today we are in very similar era in the quantum computers. So these machines already exist, they have hundreds of qubits, hundreds of qubits are hard to simulate. And we already have evidences from Google, from IBM and other companies that you can simulate on these machines some very complicated states of matter or quantum states, which are impossible or will take extremely long time to simulate classically. So we have those evidences. Now the question is how we can use these machines for something practical. And that's where the bottleneck is. Because in order to run something on the quantum computers, you need to be very much an expert in how to fight noise and imperfections in these machines. They are not perfect, and in order to apply error correction to them, you need thousands of qubits, and we are not yet in that regime, and better fidelity of those qubits. So we are in the low qubit regimes and noisy machines. But apparently even in this regime you can extract something useful. And in order to do that, a lot of expertise is needed. What we try to introduce is to some extent democratization of ability for a wider community of scientists, of algorithm researchers, of material scientists, quantum chemists use quantum computers for their own simulations and try to discover something useful. And in order to do that, we abstracted away all those low level manipulations with noise, with improvement of performance, with reducing the impact of noise on the algorithm performance, such that scientists can purely focus on science and on the mapping their problem to the quantum algorithm, and they can completely forget about what happens under that layer. And we think that that is crucial for today, because today we need more and more scientists to work with quantum computers and discover those applications. The more people will run on quantum computers, the sooner we actually will discover userful applications, as we did with classical computing, as we did with deep learning applications and other use cases. You often say the quantum software stack simply doesn't exist. Yet when companies start experimenting with quantum today, where do they usually underestimate the integration complexity? Yeah. So when we interact with industries, what we frequently see that many companies start setting up the quantum programs. They understand that this is a technology of the future, and sooner or later this will dramatically transform of what they do in many of the heavy high performance computing simulations. And in practice, this is not only about building Software. It's also about deep, low level expertise in happening something to run on the real hardware. And that's where we see real gap. So I know few companies who actually have some expertise in running simulations on real hardware at the state of the art scale. Most of the simulations which people do are of the toy scale. But you can run more. And in order to do so you need deep expertise in, for example, tailoring noise such that you can use various symmetric properties in order to simplify the noise that is in the hardware. Then you can do a bunch of tricks in order to suppress its effect on the algorithmic performance. You can also work on the algorithm itself in order to reduce the number of operations that operate in that algorithm and drive you to the resulting state. So all of these tricks are complex in nature. And frankly speaking, you need to be expert in just doing this for some very short period of time while these computers are in this state of performance. And I don't think anyone in industry should actually spend significant amount of time focusing on this low level optimization. So this should be done automatically. And that can be done by a middleware stack. And in quantum computing it's almost non existent. So if you think about again, like parallel to classical computers, we have seen software driving the adoption revolution several times already. For example, when we had the first personal computers, this kind of middleware revolution was happening because of adoption of BASIC language. So that was introduced by today's Microsoft, such that more and more quantum enthusiasts were able to actually program computers in a much easier way than they did before. Then the same kind of revolution happens recently with introducing of GPUs and CUDA language or framework to program those GPUs. So in this case, again like a middleware stack, in this case, CUDA helped a lot of scientists and experts in AI, for example, to use GPUs for their computational needs. And currently we are in this very same kind of stage with quantum computers. So right now I would say there is no CUDA or no BASIC for quantum computers. It's still very hard to program them. You literally work on the quantum assembly level. And it takes a lot of effort to run something and we want to abstract that away. Giving people instruments and tools such that they can think about the problem, can think about like what would be the actual use of quantum computers in their particular problems, or how to decompose or change the algorithm itself, rather than thinking how to optimize it to run on the hardware so that that we can take and abstract away from the user, we're talking middleware here that sits exactly in a layer where responsibility shifts from the hardware isn't ready to the execution model is wrong. What does that layer actually do inside the quantum workflow? Maybe you can dumb it down for non physics majors. Right, so imagine that we have hardware and apparently a lot of hardware providers, they don't focus that much on the software layer just because the hardware is so hard to build. There are so many problems that you need to be laser focused in order to make your qubits better, more stable, scale them to a larger quantity and so on. So there are a lot of challenges there in hardware stack and then there are people who worked on application sites. So we still need to discover a lot of quantum algorithms in a lot of different algorithmic areas and then understand which actual real life problems map to those quantum algorithms. So there are just a handful of those today. So I would say there are two communities. One is low level on the hardware, another community works more on algorithmic side and they do not overlap that much. So that where we sit. So it's not only Weeb, there are other companies working in this space, but the focus is to provide the middle layer between low level hardware which operates with pulses and low level assembly languages, quantum assembly languages and higher level abstractions for algorithm definition. And that's apparently not that easy. And in the future we need to understand that the whole stack is being changing as well at the same time, because as computers get better, we introduce new and new paradigms. For example, in few years from now we will have full tolerant quantum computers, so they will operate with error correction loops in the middle in order to first encode more qubits in a single logical qubit and then correct errors if those appear. So that's already another kind of algorithmic subroutine which need to be somehow integrated in this low level stack. And so in order to do so with many of these components, we need some kind of orchestration layer which will orchestrate when and which tools we should use. There should be ability to run on one hand real time error mitigation and error suppression, on the other hand error correction cycles and finally some finalized post processing of still logical errors which might still happen in those computers. So there are many, many complexity points here and this becomes more and more complex actually field, but the stack itself kind of grows in the real time right now and today it's not that complex. But I imagine that in 10 years from now that will be very similar to what we see in classical computers today, which is very complex Actually middleware stack, which we even don't know, like typical programmers don't even know what's happening to their algorithm after they compile. I had to smile because there was a setback for the IBM stock price a few days ago that because Claude the AI model could start working COBOL code. And so I was wondering at one point in 10 years if there's somebody who's trying to get COBOL running on quantum computing. So every. But my understanding is basically every system, every software out there that is running on the current silicon based wafer technology will need to be redone for the quantum computing age. Not really. So we don't need to push every software to the quantum computers. So not all of the problems need quantum computers to be solved. So these problems that are mostly required quantum computers are those which are complex in nature, which require enormous amount of memory in order to run some transformations in data structures or enormous amount of compute. And these are specific classes of problem and other problems like for example, you don't need quantum computers on your phone or like a quantum browser or whatever. So this is completely different space. And in principle, in the future we will see basically these two technologies cooperate together. And we already have those hybrid quantum classical algorithms where you have some subroutines running on the classical computer, while those more complex algorithms can be decomposed and run separately on the, on the quantum computer where you get the result and then integrated back into your classical workflow. So these, these technologies will cooperate. So I would say that that's the future that we will see. For example, like on the phone, every day you call your Uber. Imagine when you do that, you solve the complex optimization problem such that your driver needs to navigate the shortest path to your location. In the future, if quantum computers will be ubiquitous, that tiny optimization problem will probably be delegated to a quantum computer to solve maybe fraction of a second faster. Okay, but we took a little detour. My frequent listeners already know that. But we're still talking middleware here. And by operating in this layer, haiku forces customers to confront whether they are buying, learning, signaling or real performance. What kinds of motivations do you actually see when companies start exploring quantum computing? Right. The challenge in any of the exponential technology is that initially it seems like nothing happens. The progress is so slow that it seems like that the real value will be years ahead and you don't need to invest to it today. So you can just sit and observe what happens. On the other hand, the nature of exponential technologies says that at some point you have a Very rapid growth of technological adoption and capabilities of this technology. And I would say that's what everyone sees in AI today, but that what we will see to start happening in quantum very soon. And the reason for that is that it's not. First of all we have a pace of technology which is accelerating. And as being in this part of the industry I literally see that three years ago it was almost impossible to run anything. So there were just few people, handful of people who were running something on the hardware. Today there are many people in the world who run on the IBM's largest machines for free via cloud. And today you can immediately access largest scale quantum computers in the cloud. And that's still not enough to immediately give you the business value. But that's something, if you see this perspective, that's already something quite impressive. Another thing that we see from our perspective just few years ago, running something on these largest machines could cost you easily tens to hundred thousand dollars if you want to get truly interesting new results. With the help of technologies like ours, we can reduce this cost to tens of dollars or up to hundreds of dollars. Such that we dramatically democratize the level of accessibility of these technologies. Such that now any student can afford running so called utility scale experiments on the quantum computers available to him or her today. And that again accelerates everything. Because now there will be more experiments, now there will be more discoveries, now there will be more algorithms and the whole cycle accelerates. And we will see the adoption getting faster and faster and faster. At the same time there is a hardware progress which we see in parallel which is again moving pretty fast. Companies actually hit their milestones in different qubit modalities. In superconducting qubits, we see growing number of superconducting qubits grids. We see error correction codes already applied on the superconducting computers. We see trapped ion computers actually also growing in number of qubits and in the speed of running operations on these computers. Neutral site atom computers also move forward very fast, like soon. There are a few companies who promise photonic quantum computers. So there are a lot of different kind of underlying currents which accelerate the whole pace of this technological progress. So I'm very positive and like very optimistic about the future of quantum computers. Me too. But would have also understood is that many current quantum initiatives may be premature. From your perspective, when should companies pause experimentation rather than accelerate it? Right. So I would say I probably had to say it in the previous question, but one of important aspect, like why it's important to do something today. Suppose tomorrow there are quantum computers which operate at scale. We can run any algorithm we wish. The challenge would be to discover those algorithms. There are just handful. And not every problem is so easily mappable to quantum algorithm. So it will take another few years to just discover algorithms for their particular business problems. Then it takes typically in enterprises, it takes few years in order to adopt those algorithms into enterprise software ecosystem, integrate them, test them, make sure that they produce the right values in some specific toic conditions. So like stress test them in different scenarios. So all that takes typically time. For example, in finance, if you want to implement a new algorithm which would help, for example in some decision making, the roadmap to implement this algorithm can take sometimes few years. That's why it's important to start developing these algorithms today. Imagine that your competitor already did it today. And then in the year from now, quantum computers suddenly start to showcase value. So you still will need to to take few years in order to build, test, integrate the workflow in your organization. So you will be already losing. And since it's exponential technology, you will be losing a lot. So it's very similar to what we see today in AI. The same story can repeat itself in quantum computing. So stay in the game. It may cost you, but if you're not part of the race, it may cost you more. Okay, this is the point where founders decide whether quantum computing is a distraction or an execution problem they will eventually have to face. For Bayer evaluating Haikyuu, the real value is not speed ups in isolation, but making quantum experiments behave less like bespoke research projects. What changes when your software sits in the execution pipeline? Yeah, so when, when we integrate our software, the very first effort that you see, suddenly you can run much more experiments for the same cost. So typical enterprise programs are not that expensive on the scale. So these are typically few people and few hundred thousand dollars, up to a million could be in compute costs. But that limits you in the number of experiments that you can run and various use cases that you can check and empirically experiment with. So with our software we can not speed up, but reduce the cost of those experiments by partially also speeding them up in the factor of ten or even hundreds. And that basically leads you to a situation when for the same budget you can run much more, and that can much faster bring you to better intuition, better understanding of the technology, and better and better algorithmic pipeline for your specific business use cases. So it's not immediately that you will solve all the problems, but you will get much more intuition and operational knowledge. Of how to run something on the quantum computers. And hopefully with few customers with whom we are working today, we also will bring them to the edge of quantum usability and userfulness when they can start up operating these systems along with classical high performance computing systems. You framed the buying decision around reducing the total time of learning rather than just the costs per run. How should companies think about the economics of experimenting with quantum computing today? It's quite expensive. We don't think about the total cost. But what other economics? Future payouts, future risks, current risk, future payouts. Right. So it seemingly running something on quantum computers seems like a lot, right? So like one hour of execution on the quantum computer can cost up to $50,000 or even more, depending on the technology and the company who is providing that machine to you. But in practice, the time spent on the machine and time spent for example in high performance computing clusters, these are different types. And for example, some problems you can still solve on the quantum computer, like let's say in one hour or two hours. So it seemingly costs a lot, but these problems might be completely impossible to solve on the classical computer. And for example, if you're a chemistry company and you need to create new molecular structures to have specific properties, you need to run those simulations daily and a lot of them and eventually you are literally limited by the size of molecules that you can simulate. And if quantum computers will be of a particularly good quality, you will be able to simulate them in minutes or maybe hours instead of weeks or months on the classical computer. Or sometimes you cannot do that at all on the classical computer. So it's eventually we will just bring get to the economy where you either can run something or you can't. So that's, it's not even like in terms of economics, it's just like there are some applications which are just impossible to will be impossible to run classically. You're doing very, very great. But I'm afraid we are way over time here. So I think we are now running around 30, 35 to 40 minutes in maybe 45 minutes in total. But usually that's the length of one of our interviews. So I'm afraid we do a little cutting here and go into part number two. That's all folks. Find more news streams, events and interviews@www.startuprad IO. Remember, sharing is caring. Sat.