Episode Summary
The episode explores the intersection of quantum computing and computational engineering software with a focus on how these technologies can transform product lifecycle management (PLM) and engineering processes. Hosted by Fino, the podcast features Juha Riippi, CEO and co-founder of Quanscient, and Valtteri Latinen, CTO of Quanscient. The company specializes in developing quantum algorithms for computational fluid dynamics (CFD) and multiphysics simulations to enhance design and simulation workflows.
Key insights from the conversation include the fundamental differences between classical computing and quantum computing, particularly how qubits enable superposition and entanglement, allowing for solving complex problems more efficiently than with traditional methods. Additionally, the discussion highlights practical applications of these technologies in engineering software, such as automated workflows that streamline design changes and simulations.
For PLM and engineering professionals, the episode underscores the importance of embracing quantum computing now to prepare for future advancements. The integration of quantum algorithms into existing simulation tools can significantly enhance productivity and innovation, making it a strategic investment for companies looking to stay ahead in their respective industries.
Full Transcript
FinoWelcome to the AI Across the Product Lifecycle podcast. This is like episode number three or four. And I'm joined by Quanscient a very exciting Finnish quantum computing startup. And I have a CEO, Juha Riippi. And I'm also joined by CTO of Valtteri Latinen. I think I pronounced this correctly. Do you guys want to introduce yourselves? I'm really excited to have you on the podcast today.
JuhaSure, Michael. Yeah, happy to be here. Thanks for extending the invite. So I'm Juha ⁓ Riippi, ⁓ CEO and co-founder of Quanscient Background more in classical software development. So I'm not the quantum physicist here by far, but I know how to build software products and kind of, of course, more involved in the business side nowadays. that's my background in a nutshell. So long, long history in creating various software products, 16 years of that before founding Quanscient
FinoAnd living in a wonderful Finland.
JuhaFinland in wonderful city of Nokia, which was actually the founding place of the company with the same name back in the 18th
FinoBack when they were doing farm equipment and not telephones.
Speaker 3Yeah, hi everyone. Thanks, Michael, for the invite. I'm excited to be here. So yeah, I'm Valtteri Lahtinen, chief scientist and co-founder of Quanscient. And yeah, my background is then in science and academia before we founded Quanscient. So I've been doing research on multiphysics modeling, on mathematics of that, as well as, of course, quantum computing for an extended period of time before we found it. consigned in 21. and living in the forest of Lempäälä, Finland.
FinoNice. I've got a friend that's got a cabin up there. I'll have to come visit sometime. ⁓ So I think that ⁓ what might be good to start with is the what is quantum computing? Because I think that like a lot of other terms, AI and blockchain, know, a lot of terms get thrown around and it's not always clear to the layman what we're talking about. when you're, what is quantum computing? Val Terry, I think this is yours.
Speaker 3Right, sure. So essentially, it's a different computing paradigm from classical computing. And ⁓ maybe some analogies could work. So I mean, if you think of classical computer with bits, so the classical computer bit is kind of a light switch. So it's either on or off. Quantum computing bit, a qubit, is kind of a light dimmer. So ⁓ in that sense, It can be on or off or something in between. And this is what we call the superposition principle. ⁓ it can be sort of in a combination of on and off at the same time. And then.
Finoshorting your scat time thing, right?
Speaker 3Yeah, exactly. Exactly. And then there are other kind of quantum mechanical effects at play in quantum computing. So entanglement is kind of where another part of magic sort of happens. So if you think of these qubits again, they can be entangled, which classical bits cannot be. And what that means is that, okay, if one qubit would be a coin, you have a magical coin that is somehow linked to another coin. ⁓ You throw the one coin, you get heads or tails. Okay, it's heads. By that, by the virtue of getting heads in this particular coin, you actually know simultaneously that the other one will be tails, for example. So they are kind of magically linked in this sense. And this all comes from wonderful quantum physics studies. That is what we know. And ⁓ now using this entanglement and superposition, What it enables us to do, in fact, is to solve certain types of problems that are exponentially larger. especially if we think of multi-physics that is what QuantScient does. So for example, computational fluid dynamics problem. So with each qubit that we add to the computation, ⁓ the size of the problem that we can encode is actually doubled. So you can think that with 100 qubits, you are already at quite large problems. And then because of the superposition principle, what we can do for these exponentially large states is that we can actually perform certain computations simultaneously on all of these states. And this is what kind of makes it possible to solve certain types of problems way faster than would be possible in any classical computer and exponentially large problems.
FinoSo maybe you can also explain like, so quantum computers are obviously different even in nature than classical computers, right? I classical computers are on the silicon chips and they're at room temperature more or less, but I think quantum computing has a whole different physics to it, right? I mean, that's why in Finland you guys even have a quantum computing manufacturing company, right? Because it's nice and cold over there.
Speaker 3That's very true. yeah, actually Finland has a very long tradition of cryogenics and yeah, it's cold in Finland. So maybe it makes sense. But yeah, mean, yeah, for example, with superconducting qubit modalities, mean, there are of course different types of quantum computers. So there are superconducting qubits, there are ion trap qubits, for example. And these sort of different modalities actually require different levels of, for example, temperature. But OK, so superconducting qubits, are typically at very low temperatures of the order of some millikelvin. And this is because at that point, you only start to get these quantum effects out of these superconducting circuits. And so they actually start to behave quantum mechanically, and then you can harness this power of quantum computing.
FinoSo in other words, you're just barely above absolute zero. Like everything slows down to almost just still and you just have those particles barely moving, right? Because you reduce the energy to almost quantum levels rather than normal levels that we have.
Speaker 3Exactly. you kind of get some thermal noise.
FinoYou're gonna do that at me? right. So you could actually do if you were do a quantum computer in space, it would be, I mean, then you would have an absolute vacuum, right? ⁓
Speaker 3Yeah, exactly. Exactly. That's pretty nice. But let's see if we have at some point, know, on spacecraft.
JuhaThat's also the reason when you look at typical pictures of quantum computing, it's like a chandelier. Pretty much what you see is the cooling equipment, the cryogenic cooling and the quantum processor is actually tiny, it's like chip level thing. Of course, you have all the wiring and stuff, but the chandelier dominates the picture.
FinoRight.
JuhaTypically, if people have seen quantum computers, that's what they see.
FinoYeah, I them at a Web Summit. IBM rolled them out back in 2017, 2018. They were pretty impressive. But I think they were like it was like a two qubit machine. Maybe you can explain like the size because we're only in the eight to 10 qubit era, right? And what we're looking for is to get to the 80, 100 qubit era. Like you just mentioned, if you get to that point, you can do exponentially more calculations. Correct?
Speaker 3Yeah, that's correct. Although, ⁓ I mean, especially in many of the superconducting quantum computers of today, you actually have more than 100 qubits even. So for example, IBM's certain devices have more than 100 qubits. However, we are still in this kind of noisy era so that we cannot compute a very long simulation or a very long computation because each operation that you perform on the qubits will add to the noise and very soon the noise kind of takes over and the end result is just garbage. So we still have a lot to do in terms of reducing the noise of the qubits and the quantum gate operations. That's so that we can actually carry out this very long computation.
FinoSo, believe also another key difference is the way you write programs, right? You're not using C sharp or Java to write this stuff. It's another way of thinking, isn't it?
Speaker 3Yeah. So it's a very, very low level of programming because we are still kind of in the, it's very similar to, you know, what you had, what you had back in the day, like with punch, punch and what not. Yeah, exactly. ⁓ so, so you're really kind of just actually putting these quantum gates together. So, you know, similarly as you have in your traditional computers, you have these and gates and not gates and whatnot. There are similar types of gates for quantum computing and you are actually putting these programs together by adding gates one by one sort of. And of course you have a language to do that or several different programming languages to do that. But essentially what you're doing is just compiling these quantum circuits that actually then produce a quantum program. And in terms of sort of rethinking the problem, so if you think of again, for example, the fluid dynamics problem that we are trying to solve with quantum computing. So not all methods are very quantum native. Not all methods are very amenable to quantum speed up either. So if you just take naively your finite volume method, which is, you know, a traditional method for solving CFD and try to put that into a quantum computer, you don't really necessarily get anything useful out of it. But rather you kind of have to rethink it in a quantum native way. Like what's the most quantum native way to express this particular problem? That's what we are trying to solve with our quantum lattice Boltzmann methods, which we have found out to be extremely quantum native.
FinoAnd so ⁓ another interesting thing that might be interesting for the people listening is assuming that you've got a, what's the relationship between AI, you know, the big buzzword right now and quantum computing? Is there a relationship? Are they feeding off each other or are they completely different things? I think that there's actually a relationship there, right?
Speaker 3I mean, they're of course, sort of feeding each other so they're complementary. that I mean, quantum will help AI in terms of, you know, increased computation power. Now, AI is also helping quantum. we are actually, for example, many quantum programmers are using AI methods to ⁓ shorten the quantum programs, for example, and get rid of the noise that's currently kind of ⁓ kind of the bad thing for the quantum computing. So we cannot really solve anything meaningful yet in that sense. but yeah, these are very complimentary and really feeding off each other.
FinoSo in this series, I like to understand in your journey of creating quantient or sorry, quant science, I guess you'd prefer that I say. You started with an opinion in 2022 what AI was and it's probably not the same opinion you have in 2025, right? mean, it's been just an amazing, incredible thing happening there. And I'm sure in the quantum world, you've had equivalent revolutions or not, but. I'm just wondering like how, when you started out, cause you guys, when did you found, when did you guys start working on this? wasn't that long ago, right? Yeah. Right in 2021. how did you already see like back in 2021? So even before OpenGPT, did you already think like, wow, we're really going to need AI is going to be really important here? You were just like, no, it's just CFP and quantum. I mean, was there already an inflection point at that point?
Juha- There's really been kind of a, it's been an evolving story, but I'd say kind of if we look at engineering AI, of course, like the recent breakthroughs with Gen.AI, LLMs are making an impact, kind of making the engineers work easier, kind of with prompt based engineering and stuff like that coming up. Then if you look at kind of what we do in the grand scheme that has been the very same idea since the start is massively increasing simulation throughput with a combination of cloud and quantum computing. really huge speed ups for simulations and increase the scale of simulations. that of course, kind of, if you think about simulation tools today, they're treated mostly as a validation tool. So engineers design something, run a handful of simulations and then iterate on that. And that's a lot because the of the traditional tools out there. Usually everyone can be expensive, slow and also possibly require a huge local computing cluster. So in practice, often less than 1 % of the design space is ever explored. Now we kind of started with the hypothesis of what if you could explore all of it? And that's also where certain AI methods come in. So things like generative design, course, can feeding simulations with lot of design variations and variations that an engineer can't even fathom because the exotics and the topologies might be so different. the other half, that's something we didn't foresee because back in 21, at least I couldn't imagine what Gen.ai is doing today. But then the other idea that we has been a part of our hypothesis from the start, which is more traditional AI is kind of, okay, what other things can this throughput enable? And what about kind of training AI based on simulation data? People do that to a degree in kind of limited scopes, like in fluid dynamics, it's been done for a while, but for multi-physics where you have to combine different areas. So physics like electromagnetics, acoustics, thermal mechanics, and so forth. not so much and our hypothesis on that is that really the big limiting factor there is availability of data. All these like, ⁓ tender networks, ⁓ and kind of, neural operators, that technology out dates, LLMs. It's been out there for a while, but nobody's really, it's because nobody has access or access to tons of multi-physics simulation data, nor they have the ability to generate that. And that's really part of our thesis. kind of delivering that power. you can, I mean, We've done reasoned examples where we run 50,000 simulations in a matter of a couple of minutes, train a neural operator based machine learning algorithm on that and actually use that to solve inverse problems in the industry, which does it in milliseconds. that's already doable with kind of classical cloud compute and the novel algorithms we have there. So what I'm kind of going with this is that quantum also interplays with that because AI is notoriously bad in extrapolating. So, you know, what if you'd like to train it with higher fidelity, more accurate, more complex simulation data with more scale? That's what quantum can deliver to kind of this three part picture where you have cloud solvers, quantum solvers, and know, AI being trained with all that. ⁓ that's really been in our minds, ⁓ pretty much right after founding the company. So well, What if you had practically unlimited throughput for simulations? What things could that enable? And we find that quite interesting and definitely something we're putting a lot of effort into.
Really exciting. So then how did you use AI in the development process? I'm sure you were probably doing, I mean, a lot of people use AI for brainstorming when you're coding, but maybe that's a lot different in the quantum world. Or did you have to teach the AI how to do quantum programming and then it could kind of help you do it? I mean, maybe it's a naive question, but I'm just wondering how, as developers, how did AI come into your workflow? as you are developing questions.
JuhaValtteri can comment more on how it's used under FUNJ. But of course, in our product development, which is obviously more like day-to-day software engineering, ⁓ we use it as much as any of the other ones. we generate specifications ⁓ using AI and from those specifications, accelerate the early stages, generate all automated testing and things like that. Just accelerating things in the development. pretty standard popular engineering software development. Yeah, cursor is one of the go-to tools for us.
Finostuff like that.
Speaker 3And similarly on the software side, I mean, we also use for example cursor and whatnot. So really kind of, because it's software development in the end as well, even though it's quantum software development, but it's software development. These tools can be utilized. ⁓ But then as I mentioned also, we kind of use some middleware tools that is kind of between the hardware and software. where AI can actually help us reduce the noise for example in this quantum computer so that we can actually get better results out of the real devices of today.
FinoSo it's a give and take relationship in the case of the quantum. I mean, you're using the quantum stuff, the AI is helping you do that. And it's also helping you reduce the noise, which helps you get better results, which then can be reinterpreted by AI for the next run. it's sort of ⁓ a, what's the word? It's just a benign circle, right? It's all helping each other. But did you have to actually teach the AI how to think quantum? Obviously, the LLMs have no idea what that means until you tell them,
Speaker 3I mean, they're pretty good at it already. ⁓ So in terms of, know, software engineering stuff, quantum software engineering stuff, they're pretty good at it. So nothing that is now, of course, these specific tools that I mentioned in terms of the middleware, for example, I mean, they are of course, like highly engineered and highly specific so that they have been trained to do exactly what they do. But ⁓ yeah, in terms of, know, software, we can use pretty much similar tools.
FinoAnd you're using like a customer of Quanscient would have ⁓ a Quanscient LLM that was already trained on the data or they would be using OpenAI or Anthropic. Just trying to understand how do you end up with your own LLM that's trained specifically on the stuff so that ⁓ it doesn't end up in the context and then trains the rest of the LLMs in the world because Anthropic just stole all your prompts, right? I'm just wondering, I mean, that just dropped two weeks ago and I'm wondering how that impacts everybody. I'm sure that everybody's like, my God, Rag, we've got to do that because we don't want Anthropic and OpenAI to steal all our IP that we've been using in these prompts. So just that's the kind of question to you guys.
JuhaYeah, when it comes to that kind of what LLM features are available for the user. So yeah, we have a chatbot that's effectively a helper kind of from the same
Finostandard thing now, right? You kind of have to have that.
JuhaYeah, exactly. It's kind of a must have nowadays. We've trained it with all our documentation, all the kind of the API ⁓ code that we have. So it's pretty handy in generating the code. We use a rag method ⁓ that's been optimized quite a bit by us. I mean, yeah, it's kind of a must have. Nothing super special. Some of the other use cases we've done with Gen.ai is For example, anomaly detection for user parameters. So for example, in material parameters, it's super easy to misplace a comma or something like that. And then all your results are completely wonky and you spend days debugging what on earth is wrong.
FinoThat never happens
JuhaSo anomaly detection, it has been trained on tons of our internal simulation demo data. So it actually detects these things and it tells you that, are you sure about this parameter? It looks a little off. ⁓ And also generating material properties. LLMs are actually great in that. So you can at any time go, chat GP, give me material properties for this. typically, up there. that's one feature we're building on, integrating that directly to the product.
Speaker 3Exactly.
FinoYeah.
Juhajust saving engineers times. I think you can generalize that these LLM GenAI features, they're typically kind of, they're starting to be a must have. And they are things that streamline an engineer's day to day, just make the product that much easier to use. And I would dare claim also that when you're looking in engineering landscape, CAD products, computer-aided engineering tools, ⁓ We're starting to be at a point where that's no longer a competitive advantage. It's more of a must have. if you don't have those in general, you're kind of looking around at what.
FinoWhat's the deal? So like how, and I guess another question I had because there are, you you guys aren't the only company out there doing solving CFD problems on, you know, in a, in the startup world, which is interesting too, because of course, traditionally it's been the ANSYS either universities or ANSYS, you know, the, big simulation companies. It's cool to see there's a lot of startups now, physics AI, we talked about when we were introducing our Offline we were talking about that or I found out so Navi.ai. So how do you guys differentiate yourselves from the other physics solvers out there, the CFD solvers that are in the startup world?
JuhaWell, one of things kind of to highlight is that we're not a CFD solver. So for the quantum solvers we're developing, that's a starting place. We're starting with fluid dynamics physics, but the cloud-based solver we have, that's a multi-physics solver. So in the sense, it's very similar to the philosophy that, for example, COMSOL multi-physics was designed on. So it's a strongly coupled multi-physics. You can do electromagnetic acoustics, electromagnetic waves, ⁓ fluid dynamics, mechanics, thermal, all in a strongly coupled single simulation. So, and we're working with a lot of energy sector companies, semiconductor manufacturers and such who are using our product to accelerate their R &D. So the big difference with like PhysXX, Navier AI is that I think they're more on the side of researching novel AI methods to accelerate simulations, where as we are building novel numerical solutions. For example, in the simulation AI examples that we've used our own data to generate impressive results, actually, our AI, it's not to downplay it, but we are not the world leaders in that. ⁓ But if you have the ability to generate tens of thousands of simulation data points, extremely cost efficiently, extremely fast,
FinoOkay.
JuhaYou can use pretty traditional AI methods to actually get great insights from that. And that's kind of our thesis more. And we do believe that in engineering, again, AI and numerical methods, they're not competitive. They are complementary. You will always need both. And both of these technologies must advance. And our chosen path is taking to numerical methods to their limits with existing classical compute and in the future with quantum compute.
FinoNice. About the same thing that you'd say, Val Terry, where...
Speaker 3Yeah, exactly. You have put it really well. And yeah, obviously we also differentiate by the quantum side of things so that we are actually building these quantum algorithms for CFD. And we are definitely the world's leader in that. ⁓ quite recently, we, for example, ran the first in the world 3D CFD simulation on a real quantum computer of IQM. ⁓ At least to my knowledge, has never been done before. This is enabled by these quantum native methods that we are building because they are more efficient, more native to the quantum computer than your regular CFD methods would be.
FinoThat's really exciting. ⁓ now that we're four years into Quanscient and you've already looked at AI and you've seen other things, how do you see AI now? How do you think that going forward, it's hard to even look six months in the future because it changes every two minutes. if you're looking forward, so guess the next thing for you guys is when you get the next quantum leap, the next zero behind the numbers of qubits that you can do, which is going to increase your capacity to calculate. But ⁓ how do you guys see the future? Will AI surpass the acceleration of AI? Will quantum catch up to it? And then the two of them will advance together? Because sometimes you might feel that quantum is a bit stalled. We've been at a couple of qubits for quite a long time. jumped to another zero yet, whereas AI, we've already been through, Gen. AI, now MCP, and it's kind of, I'm curious on how you guys perceive that going forward. Like what's the future?
Speaker 3I can maybe comment on the quantum side first and Juha can then continue on that. ⁓ So if you look at the roadmaps of the big quantum computing companies right now, I think there's a lot of excitement around those right now because for example, I IonQ just released their roadmap where they are promising several millions of qubits by end of 29. So several millions of physical qubits, would then entail basically thousands and thousands of logical qubits that are fully error corrected. And this means like full blown quantum advantages for so many fields at that point. And it's not very far. It's only four years from now. Now, of course you can question whether that's doable or not, but that's their roadmap. And many other companies are claiming like error corrected machines with plenty of qubits by the end of the decade. And I think the first quantum advantages also in the multi-physics side and CFD side of things will be tied to AI. so that actually the first sort of quantum supremacy simulations that we can do that are kind of intractable to classical methods and classical HPC, they will not necessarily be very long simulations so that you could do a lot of of time steps because again we are still kind of in the noisy era and the noise will build up. But at some point we are at the stage where we can actually do that so that it's a large enough simulation that it couldn't be solved. on a classical HPC, but not necessarily very long. But with that, we can already teach AI, like a foundational physics AI model, something about these problems that it's just not possible to teach with any classical method. And here you have quantum advantage.
FinoPretty cool. So you are here, obviously agreeing.
JuhaI definitely agree. I think in the broader lens, if we kind of look at engineering workspace, if we look a few years ahead, so how product design could look like. So think about you, first of all, all these generative design ⁓ technologies are improving. I mean, engineers just kind of need to plant the seed, which they can use. Generative design software create tons of variations from that design, which then again can be really quickly initially validated. Let's say a combination of numerical and potentially foundational physics models that have been trained with tons of simulation data, perhaps including quantum, ran numerical simulation data ⁓ for quick validation and feedback loops to kind of explore a very broad design space. Once you're able to kind of narrow that down based on what you actually want, you can dive deeper with very sophisticated numerical methods, which you want to do at one point because they're always deterministic kind of way. The engineer, want repeatable results and AI can't offer you that. So you can dive deeper and you can basically land on an optimal design that matches your specifications extremely fast. or even start from an inverse problem. Start from, want this as an output. What's the kind of hardware that produces that for me? Which could again be achieved by very large data sets of numerical data and then using kind of AI that's been trained on those and actually kind of do it backwards. Because effectively, if you think about it, that's what engineers want. They want a certain result from their device. as long as it's feasible to fabricate. ⁓ Well, other design parameters that you might have, it doesn't really matter then what the design is like.
FinoYeah, it's really cool. It's interesting. I'm glad you mentioned the determinism because I was talking when I talked to Karen Talati about first resonance, I was saying, well, how do you get deterministic results from AI? he said, well, if you think about manufacturing, it's actually very probabilistic. This machine could break and things could go wrong and you might end up with not enough stock. mean, there are There is a bit of probability even if we'd like it to be terministic. So I thought that was an interesting perspective. ⁓ And it makes me think so. ⁓ Usually at this point in the interview, I'd like to talk about like when you take this into the real world, like how ⁓ we already met, you already mentioned how important it was to have data, right? You need data, huge data sets to train so it understands these things. And, you know, the companies you go into are not necessarily all that mature in terms of data. In fact, usually the simulation guys are somewhere else and very far away from the guys doing manufacturing and the guys drawing the parts in CAD. And so I'm very curious, like when you go in and use, because you guys have been successful selling this, I look at companies of having kind of a spectrum point one to five in terms of digital maturity, right? Like one is basically email and Excel, right? And that's a lot of companies are there. And then you five is like autonomous virtual twins, you know, with AI and MCP and you know, Barely humans, know, human not even required other than, you know, turning it on and off and, you know, maybe do some vacuuming around it. Obviously, very few companies, if any, are at five, right? And even at four, there's probably almost nobody. But what do you guys observe when you go in? Is that one of the first things you look at is how mature they are? Because you're going to need that data. You're going to need things to train your stuff. Otherwise, it's not going to be that interesting, right? So I would just like your feedback on that initially.
JuhaSo yeah, at the moment we definitely look at that kind of how mature they are and what's the readiness. mean, if they've never touched multi-physics simulations or are just kind of starting on that, it's a pretty long way to start pitching like massive design exploration and huge simulation through because they don't have any comparison points. So typically our customers are ones who have a lot of experience in simulation tools. And there might also be these pain points and frustrations on the limitations of the existing tools where we can really come in and help. on the long term, ⁓ we do think that, well, all these Gen.AI, LLM, engineering AI features will reduce the barrier of entry for simulations because you no longer need a physics PhD to understand what's going on in the solver because you don't need to understand what's going on in the solver because you can just interact by prompting. Think of Jarvis from Iron Man, one of my favorite action heroes. You're just talking to your computer-aided engineering software.
FinoYeah. ⁓ So is a bit of a democratization that's happening then. I mean, eventually that simulation department and those design departments are going to be a lot closer than they are today, right?
JuhaI think the way the technology is going, yeah, definitely it's bound to happen. There's a big transformation in engineering going on as pretty much anywhere else with all these new technologies. yeah.
FinoAnd I'm also curious because I can see definitely that data problem from a classical computing point of view. But Valtteri, when you're looking at quantum computing, that's a whole other ballgame, right? In terms of democratization, because people don't really understand it, is one of the goals of quantum making these tools relatively easy to use so that someone that doesn't have a PhD in quantum physics can actually still get results out of it? Is that one of the goals that you guys have?
Speaker 3⁓ Absolutely. I think the user, the customer shouldn't have to be worried about, know, what's the hardware that the problem is solved on actually. I mean, they can interact with the tool just as they would interact, you know, with any other engineering tool. And the solver itself can actually decide, you know, whether this is good for quantum computing, whether this is better to be solved on a classical computing. And this is kind of our vision so that The customer or the user doesn't really have to care that there's actually a quantum computer behind that. So they don't definitely have to know anything about quantum physics necessarily.
FinoSo at the end, what we're going to have is really hybrid, right? It's going to be a hybrid and where you'll take the best of breed, the stuff that you, it's better on quantum will do there. And that'll also compress the entire cycle of design to manufacturing, won't it? Because you'd be able to do a lot more things in parallel rather than waiting for this part of the design to complete. You can already do a bunch of what ifs and, you know, variability testing to figure out what's the right option and then re-add those parameters and recalculate, right?
JuhaThat is the goal. exactly. mean, yeah, any startup kind of needs to pick their niche, but ⁓ maybe if we go only to quantum physicists, it's a little bit too narrow.
Speaker 3In terms of enabling huge data creation, so just how quantum helps in that as well. it's not only that we can solve these huge problems necessarily, but also we can use the superposition principle again to, for example, ⁓ solve the problem at the same time for millions and millions of initial conditions. using the superposition principle so that we're actually generating at one go a huge amount of data in this sense.
FinoSo your what-if scenario range could almost be infinite, right? You can say, we'll calculate this for all the ranges of this variable from 1 to 1,000 in groups of five and just do it all in one shot rather than iteratively having to go through to that, right? That's one of those advantages.
Speaker 3Exactly, Got it.
FinoThat's pretty, that's super exciting. I remember also we were talking about ⁓ the, because ⁓ I think another issue that's gonna, well, another thing about these new kinds of computing is what Jensen Wong of Nvidia has been talking about is power consumption, right? And the fact that these AI machines, the Nvidia that are just very, very, very... But basically he was saying how every factory will have an AI factory and the AI factory will have a nuclear power plant to power it. But I think we were talking about that. You guys mentioned some really cool things about fusion already that I didn't know. And then you mentioned that quantum actually ⁓ is far more power efficient ⁓ at the end of the day. So, any of you guys want to talk about that?
JuhaWell, like on fusion, so I mean, we're not a fusion energy company, but some some of our customers are.
FinoYeah, that's what you're telling me.
JuhaYeah, and of course, mean, if mankind can solve that problem, it's pretty much endless clean energy for humankind. It's one of the biggest problems we can solve. And there's been a lot of progress in Fusion. ⁓ Our customers like Proxima Fusion, for example, based in Germany, are doing lots of ⁓ cool things with that.
FinoYep. Forever.
JuhaOne of the elements that's really enabled this recent progress is evolution of computational physics. designing these complex, for example, stellarator shapes, which have very exotic ⁓ geometries and topologies, and how they're supposed to contain the plasma is computationally a hugely complex problem. And more advanced simulation tools definitely help speed that up and help make breakthroughs there. ⁓ it's really inspiring to be even a small part of that and working with the geniuses.
FinoThat must be super exciting. it's interesting, guess, that we might in our lifetime see all three of these things. We might see AGI, might see quantum, and we might see fusion. That would be pretty awesome, right? It'd be almost like being there when the light bulb was invented. ⁓ I had another question that kind of takes a different direction for Valtteri, because ⁓ I attended a Capgemini Engineering Summit, and there were actually I was surprised there were actually several, they had a whole team at Capgemini Engineering that were quantum physicists and quantum, there was a quantum chemist that gave a nice talk. And then there's one guy, they talk about the quantum apocalypse. And I was like, what? He's like, yeah, the year, you know, we had the year 2000 bug and this is actually worse because every cryptography, every method of cryptography based on prime numbers is basically transparent when you've got a quantum computer. I mean, you can solve those, all those keys basically in half a second rather than tens of thousands of years even for really, really long keys. ⁓ So I don't know, I felt like, ⁓ it seems a bit paranoid, but it's true that we use a of prime numbers. I mean, there are other encryption methods. I just wonder if you guys, know, could you guys educate me on that and educate the audience on, you know, some of those kinds of problems that might be introduced, the Y2K problem from a quantum point of view, right?
Speaker 3Right, mean, it's a real problem. know, eventually it will happen that quantum computers will break that encryption. Now, how far are we from that? It still remains to be seen. But okay, if we get to these millions and millions of qubits with tens and thousands of fault-tolerant logical qubits, I mean, we are pretty far in that then at that point. ⁓ What is of course kind of the scary thought is that right now, know, malicious actors are possibly stealing data or storing that. And then, you know, they're just waiting for the quantum computers to mature enough so that they can break the encryption and use that for malicious purposes. So, yeah, I mean, it's a real thing. And we definitely, you know, as societies, we should... kind of get ready for that and start using, you know, quantum safe encryption methods.
FinoSo like what I think is it AES that's quantum safe? think I think there's a couple of methods like that are far safer than other ones,
Speaker 3There are a couple of methods that are at least supposedly safe from the methods that we know right now in quantum. Of course, can we guarantee that they are forever safe from quantum? don't know. Cryptography is not really something that I'm super expert on, but this is definitely a real thing. As said, we should be kind of... varied and start doing something about it.
FinoYeah, kind of freaks me out when my password for my bank is only six numbers. That kind of freaks me out already. I'm just like, come on guys, you know? So like now you're taking conscious. Well, it's sort of like the, as you're coming out and saying, we're going to steal your data. it's our data center, we don't care because that U S law goes over everyone else. So suddenly is there going to be a movement back to on-prem because proud isn't safe anymore. ⁓ I wonder about, well, that's, that's another question for you guys. mean, what did you, how did you take that? Well, cause I felt there were two, well, there were actually three big announcements recently, right? There was Microsoft saying that us law trumps every other law. So if the NSA, the CIA, the FBI wants the data from a European as your server, they get it without any questions asked. Number two was Anthropic saying, remember what we said about not using your prompts to train our LMS? Well, we lied. We're actually going to use, we're already using it, right? And OpenAI has admitted that too. So that's also a problem. And then the third one is that, ⁓ you know, we only have one AI unicorn in Europe, right? We have Mistral.ai and it was like this close to being bought by Apple and then ASML stepped in with, you know, $4 billion and kept it in Europe. ⁓ How do you guys see that? Is that maybe has no impact on you at all? Or maybe you guys are thinking about it. I'm just thinking those seem to be for us sitting in Europe, those are kind of huge things and they're challenging our sovereignty. They're challenging our privacy and they're also trying to link us as a independent economy, right? From an AI point of view.
JuhaSo absolutely. And it definitely does have an impact on us. So we have not really been shy about the fact that our service is running on AWS. yeah, there is this problem. I think definitely Europe should come up with their own kind of hyperscalers, own cloud services, ⁓ get some independence on those own AI methods. it's, of course, there's a lot of ground to cover compared to the mega corporations, kind of US based ones. So yeah, it's a difficult problem. I really don't have a big solution, but it definitely does impact us. ⁓ But there are some upsides too. So, you know, we were talking about quantum cryptocracy, for example. if you're thinking about future scenarios like ⁓ Who will be kind of leading most likely the quantum safe cryptography implementation? I would dare claim it's the hyperscalers who are already investing billions into data security instead of like on-prem data centers. if you completely disconnect from the internet, the only way to hack them is like social hacking, ⁓ get into the data center with a memory stick or something. But even that happens all the time.
FinoOkay All the time.
JuhaSo I mean, but I would also there claim that's way more difficult to do with AWS data centers because they have basically armed guards and the security is quite strict. So there are pros and cons with that approach. Having said that, some of our big customers are asking about on-prem solutions. So yeah, I think we can apply. our software to be run on-prem in large compute clusters. It's also, it's not science fiction to convert it into some other hyperscaler. It's actually quite simple to do. It's just take some work. So going multi-cloud and being able to run our platform in different environments is definitely something we want to do because it gives us way more options, both in business, but also in this security matters.
FinoBut we do, there are a few cloud providers. mean, there's OVH here in France, you the Gaia X initiative, but I guess there's still not quite there in terms of the number of GPUs you need to do what you want to do,
JuhaYeah, availability of resources. So kind of one of the big advantages of AWS definitely is this next to infinite scaling. So go big on your jobs. So yeah, but it's kind of chicken and egg problem. We won't have a European hyperscaler if everybody's kind of thinking, well, I'm probably better off with the Americans. Take my money to the European ones. So it's a tricky.
FinoYeah. Yeah.
Juhaany problem.
Speaker 3And in terms of computing, mean, Europe is playing its stakes quite okay, I think. I mean, we do have several great quantum computing companies that can become kind of the winners of the race at some point. Now, of course, in US, it's the same thing. There are many, good companies and also China is a big player. But at least, you know, I think Europe
FinoThat's very w-
Speaker 3Europe sort of up to this quantum computing thing a bit earlier, I think. then there's a clear strategy for Europe and its strategic advantage to actually build these European quantum companies because that way we can utilize it for the European good and European defense as well. ⁓ it's also a defense asset.
FinoAbsolutely. And it's cool to see Finland take the lead there, right?
JuhaYeah, have to agree as a Finn. yeah, I think we're kind of punching above our weight, ⁓ definitely in quantum. there's quantum companies per capita or quantum scientists per capita, we're probably one of the leading countries. I don't have exact statistics on that, but I would dare make that claim.
FinoGreat. Well, let's hope that it's a Finnish guy that comes up with the... That's awesome. ⁓ You mentioned IQM, that's the Finnish company or that's the Finnish one, right? So they're the ones talking about million cubits. Wow. So it might be Finland that takes that crown.
Speaker 3That was actually IonQ's roadmap. that's a US based company, I mean, IQM also has a roadmap towards tolerance. Yeah. Yeah. They all sound the same, right? But yeah, I mean, they also have a very credible roadmap towards this. ⁓ I have no doubts that they will do great things.
FinoI am sorry.
JuhaYeah, it's an unwritten rule that any company involved in quantum has to have a Q in the name, we won't even fall into it. We started with one.
FinoSo last thing to of wrap it up. So like when you bring Quanscient in and people start using it, is there an epiphany? Like we were talking about the digital maturity. there, when they start using it, do they realize, wow, if we need to, you know, we need to do that multi-dismant stuff. Or maybe you were saying, you are that people already did in order to use Quanscient efficiently. They've already thought about multidisciplinary work rather than having all these siloed engineering groups, right? I'm just wondering, it bringing in questions, does that create the opportunity for them to be better at AI, better at data maturity and better on their digital roadmap? Or is it, you know, is there a ripple effect or slow ripple effect, or does this just happen in a bubble in the simulation department and nobody else even sees it?
JuhaWe're hoping it'll be a ripple effect. of course, kind of we ⁓ were on our fifth year now, now starting. So ⁓ most of the work we're doing is involved with simulation engineers and kind of where, you know, classically computer engineering products are used these days. But that's definitely something we want to expand on that, expand on the multidisciplinary users. But there's there is still ways to go in this democratizing of simulations. It's not a feat that can be done very quickly to make it user friendly enough so that somebody who's never used simulation products can just pick it up and start doing useful stuff.
FinoBut I suppose you still are. But if we're talking about the designer tweaking a CAD tool, that should be the first one to be closer, right? Because then they can do a design change, throw it over to you to do some what if and say, no, you need to have it at this angle and then your turbulence will be lower, right?
JuhaYeah, and what we're already doing there is, for example, automated workflows with our clients. So there is an integration from the design software and certain simulations are confirmed automatically. So basically, the designers are working, doing their work and a lot of that is streamlined because our solver can be run completely programmatically, of course, with APIs.
FinoThat's cool. ⁓ Well, we're almost at the top of the hour. Do you have any concluding comments, Valtteri, to encourage everybody about quantum computing?
Speaker 3Yeah, well, maybe, maybe just that, you know, I think the right time to start kind of investing to quantum is now because as I mentioned, the roadmaps of the hardware companies seem like quantum advantage is not very far. And when quantum computers are there, our algorithms will be there as well to actually facilitate the quantum advantage. In our case, in CFD and Multiphysics.
FinoAwesome. You are any closing comments? fun.
Juhanothing to add on the quantum side. ⁓ I just say that it's I wasn't working in this field as I established kind of in the introductions, I come from more generalized software. I just find ⁓ engineering software to be such a fascinating field where there's a lot of opportunities still yet to come, which we probably can't even imagine yet with already the existing technology, but also all the new technology coming up. I think we're starting to see a big shift in. engineering software, how people interact with design and simulation software.
FinoI think it's really exciting. It's a very vibrant ecosystem. I think I've identified 275 companies now. And then those are probably about 40 that are just physics and CAE and CFD, which is really, really exciting. Well, this has been a fantastic conversation. really appreciate you both taking your time out of your busy schedules. I hope that one of these days we can actually meet and have a beer or some vodka and a herring together or whatever. It's been fantastic. Thank you. What's that?
Speaker 3We can do both beer and vodka.
FinoOkay, that would be great. ⁓ So ⁓ are you guys going to be attending any conferences? people want to actually meet you guys, you're going to be doing any conferences between now and the end of the year? Just throwing that out there because maybe...
JuhaNot so much before the end of the year. for any of our customers active in the MEMS industry will be in January in Salzburg in MEMS 2026. ⁓ But I think the rest of the year is sort of slowing down in terms of conferences. Of course, there are some startup events like Slush in Helsinki, Bits and Pretzels in Munich next week where I'll be and we'll also be at actually This week Thursday in AWS kind of industry future summit in Stockholm.
FinoVery cool. Very cool.
Speaker 3And right now in Portugal there is this UCAS conference, European Conference on Applied Superconductivity. we have a booth there. If people are around, go and check us out.
FinoAwesome. I hope they will. And I hope that maybe in a couple of months we can talk again because, you guys will call me when that big computer comes out, a big quantum machine. It's been a real pleasure. Thank you. thanks to everybody that joined. please leave us your comments and drop a line to Juha and Valtteri and we'll talk to you next time for the next podcast. Thank you.
Speaker 3Thank you. Thank you so much. Cheers.