Photonic Computing: Breakthroughs, Challenges and Insights with Denis Kalinin

[00:00:00] Rahul: Hi Dennis, uh, thank you so much for joining me

[00:00:02] Denis: hi, Rahul. Yeah. Thank you so much for having me.

[00:00:05] Rahul: Yeah. So, you know, I was reading about optical computing and then, uh, I didn't realize this. I didn't realize that there is a potential for optical processors to operate at terahertz or, you know, petahertz speeds. That's like order of magnitude faster than what we have right now. And, The, the one alternate computing platform that I have looked into briefly on the podcast as well as quantum computing and I feel like optical computing is more promising and more ahead in terms of like, technology maturity.

[00:00:38] Denis: um, optical computing is kind of maybe, maybe considered as a more short term. in terms of the, promise that and, and, the commercial traction that, it has, but of course, what would be more impactful is still a question mark. I, either [00:01:00] quantum or, photonics or optical computing, because both topics are really large and, they, actually solve, a bit different problems. Optical computing is mostly focused on solving,deficit of compute, power, and, energy efficiency problems. while, quantum computing really focuses on, um, Solving the problems that were impossible to be solved previously, like the end body problem, like the salesman problem, because it uses, quantum mechanic principles, and, essentially make a lot of, sophisticated challenges, not exponential, but linear because, due to, quantum effects, they can perform, multiple Thank you.

[00:01:46] Denis: Thank you. calculations at once in parallel. and that's why quantum is really good for solving extremely sophisticated optimization problems. Or another quite promising use case for [00:02:00] quantum is, quantum simulations, like creating a very,complicated.digital twins, for example, for manufacturing, I can, I can dig deeper into some of the use cases of other companies, but essentially that's quantum and, uh, optical computing is focused on basically doing the same as the classic computing does, like as GPU as NVIDIA GPUs do, but just way more faster and way more efficient.

[00:02:31] Denis: So that's, that's the difference in, in a nutshell.

[00:02:33] Rahul: Yeah, yeah, yeah. So other than, so this is essentially a different way of computing, right? the classical computing is, electrons and this is like light. So, I would love to know more about, the basics of, photonic computing and, also, like, why has this become so important for now, because you've been talking about the [00:03:00] potential of photonic computing for a long

[00:03:02] Denis: yeah, yeah, sure. well, first of all, a small disclaimer because I'm not a physicist. So, I may probably, so there may be some more technical, people and scientists could explain it more accurate, but from the VC perspective as a, as VC investor, I would explain it as the following. So, Traditionally, the compute has been done, used, by using electrons.

[00:03:28] Denis: And electrons are, on the one hand, they're quite... useful for transmitting and, uh, data and, uh, making the computations, but they have several problems. one of the problems is that, they do not travel as fast as we like them to, to travel. and, the second problem is that they interact with, a lot of other materials, like for example, when they, are transmitted through, cables.

[00:03:54] Denis: or just being, used in, in a chip, and that creates losses. and, [00:04:00] we've seen a similar, situation of, electronics being replaced by optics, in, data transmission, like, in, uh, Cooper. Cables, which were all around, and were used by networking companies. just some 30, 20 years ago, and now almost, all the data transmission is done by, optical fiber, and this, this is for exactly these two reasons, light.

[00:04:29] Denis: Travel way faster than electrons. light is the fastest, matter that can,with, with a high speed and, light does, does not interact, with other, materials as electrons. So this basically makes, optics, um, a viable alternative to, electronics. and that's what happened in data transmission, area.

[00:04:52] Denis: And, um, if we even look at the history, it's even interesting that, comparing the [00:05:00] history of, electronics and history of optics, optics. Is following electronics by around 50, 50 years back, with a, with a 50 year, years, lack, because the first, Fleming Wells. were invented in like 19, um, 1904.

[00:05:20] Denis: and, the first research on optics was in, in the sixties. and then, the first amplifier. like the, the, the, the first, electronic, transistor was, made by, bell slabs in, later forties while the optical amplifier, which basically kind of has similar function, can be compared with transistors, was, invented in, and implemented in, 1980s.

[00:05:48] Denis: In the 1980s. and it kind of. By following the trend of data transmission, we can, project and forecast that the same will happen in the compute space as well. [00:06:00] Uh, we saw the first microprocessor by, by Intel,in the, in the 80s, um, around 80s. And basically we think that. In, uh, five to 10 years, we may see a viable replacement of, the compute, like the viable, um, implementation of, new compute architectures, based on optics.

[00:06:27] Denis: So this is, this is how we see the history. And this is how I think we can draw some parallels, why optics can be good, architecture for compute.

[00:06:38] Rahul: Yeah, interesting. So, In terms of transmission, it's, it's already, like, like a fully matured technology almost everywhere, even the internet, uh, we use optical fibers. But for compute, what was the challenges, in terms of not being able to make[00:07:00]

[00:07:00] Denis: So, the, the first technology in, the first attempts to, do the compute, based on, photonics. Was in, the space called, silicon photonics, um, like light matter, light, intelligence and some other companies, attempted to do it and still trying to implement it. It's basically, using both optical, technology, but also the traditional CMOS.

[00:07:27] Denis: Architecture for silicone, trying to kind of connect them together, basically, to, implement the interconnects between ships, you know, as photonics, like fiber for photonics, architecture, and, uh, it worked to a certain extent, the biggest challenge is that, uh, you, as Transistors, become smaller and smaller.

[00:07:53] Denis: optics cannot basically keep up with this space because, the ability to, [00:08:00] mean, miniaturize the optic architecture is just. very limited for, for the reason that, the wavelength of light is, larger than the wavelength of electrons. That's why you just, you just cannot, uh, make, uh, chips or interconnects based on light as small as, four electrons.

[00:08:23] Denis: And that's the biggest, that's the biggest problem. That's the biggest bottleneck, that, silicon photonics. is experiencing currently and, it's, I would say the problem is it hasn't solved yet with, for this particular, stack for this particular kind of technology. and that's why kind of maybe also getting a little bit of head.

[00:08:43] Denis: we as, Runa Capital, we haven't invested. in this space, really, although we looked also on, a lot of companies and, integrated photonics and silicone photonics space, and we made a, bet on, the so called the new way for optical computing, [00:09:00] which I also share a little bit, but that's not, that's not silicone photonics.

[00:09:05] Rahul: That's, that's for sure. Okay. So, so you can make a, uh, photonic. Compute, that is one way and the other way is, you know, you don't, you still operate with the silicon compute, but then all this interconnection can be photonic, right? So, I was listening to another interview by the CEO of Light Matter and he, I think they are more focused on this interconnect, between compute.

[00:09:34] Rahul: And he has this argument that, uh, a lot of compute. Even now, with, with, for, for a lot of AI applications are not being fully utilized because the data transfer between the different compute instances, uh, is so slow. So a lot of time the usage is like very minimal 5 to 10 percent or something like that.

[00:09:56] Rahul: and yeah, that, that's why the focus is there.

[00:09:58] Denis: yeah, this is [00:10:00] basically, well, light matter, as far as I know, they started from, uh, making, uh, integrated photonic chips for, um, AI applications. So it was a bit of a different, uh, area and then they pivoted towards, uh, optical interconnection chip, uh, which is, uh, quite an interesting area as well. We invested in,

[00:10:21] Denis: in a company, called, in Lytra, it's, it's a company out of, Switzerland, and they basically developed a multicolor laser, which is like you have, um, light source, and then you split the laser into, hundred parallel, light beams, by the technology called, frequency comb. And, uh, then you allow to perform the calculations through.

[00:10:48] Denis: Every single beam by increasing, and this is how you increase the efficiency by a hundred times, for data transmission. And this is, I would say it's, it's, it's [00:11:00] focused on the similar, use case on similar application as light matter, because they also target data centers. And, uh, but the technology is different.

[00:11:10] Denis: Like they, they also use, one of the architectures, also using, photonic chips, but the main IP and the main kind of, uh, technology stack is laser itself. And how it transmits, the data for data centers. Um, so this technology is quite interesting. They've been, already, tested by NTT, the Japanese telco, which also is looking at, photonic space quite actively due to, very huge strategic interests.

[00:11:37] Denis: They are one of the largest data center operators in the world, and also providers of equipment for data centers. So, that's an interesting one in, Interconnect and kind of data transmission, but it's still that's one segment, I would say. And, uh, if we talk about the market, um, it's still limited, like it's, growing [00:12:00] quite fast because, the data centers are increasingly becoming a bottleneck, in terms of like how effectively you transmit the data, how much data you have.

[00:12:16] Denis: In general, the amount of data is growing, um, rapidly and will grow even, more, uh, even faster in the next 10 to 20 years. and that's why the market like is going for, is growing from like several dozens of millions right now to several billions. In the next, 5 10 years. So it's still an interesting play, but that's, I would say, quite limited, in some point.

[00:12:43] Rahul: okay. So there is interconnect. And then, so what are the other segments and where are the other opportunities? I think there is one is the photonic processors, the compute sort of a

[00:12:55] Denis: basically there are, well, um, photonics [00:13:00] has a lot of... different segments, like photonics is a huge industry. And, like one of the applications is LIDAR, for example, for, um, basically generating like, um, 3d objects based on the sensors and it's used for autonomous vehicles, for example, which we haven't touched yet, but that's one segment.

[00:13:22] Denis: Then there are many other. Applications like some applications are even used in healthcare, like one startup in the UK is using it for, um, improving the, function of, uh, healthcare device called, called OCT, device, which is optical coherence tomography, for scanning, um, eyes, and they use photonics as well to make the, this technology more efficient.

[00:13:50] Denis: This is. Yeah. Like all the, all the different, all the different applications, which are not necessarily connected with computer at all. Um, but if we talk about [00:14:00] optical computing, uh, that's where, the market is growing right now. And that's a huge market that we will see will reach like several hundred billion dollars.

[00:14:13] Denis: and there. We kind of see two major approaches. One is analog computing, uh, computing for specific applications, mostly AI due to, very rapid growth and generative AI and a lot of demand for AI training, AI inference. And then, digital computing and digital computing is something which is extremely sophisticated to implement for the reasons that I just mentioned, when I was talking about photonic, uh, chips, and they're, they're just, we think that it needs to be implemented in a different way, not by using chips, but [00:15:00] basically creating like the full, fully optical, uh, processors.

[00:15:05] Denis: using only put only optical components, and this is extremely sophisticated because it's hard to recreate the whole infrastructure, the whole architecture based only on photonics. Like you have many different challenges that I think, well, it does make sense to go really deep into them in detail. But, so far, there are no companies that.

[00:15:28] Denis: done it successfully. We invested in one early stage company out of Germany, which is currently attempting to do this, and we think that they have quite good, chances because the, the idea in this space is really strong, and they have. Well, they're very close to having a prototype, but it's still, it's still very early, um, and this is, but essentially if they succeed, or if any other companies in this space, [00:16:00] in this segment succeed, this technology may potentially, replace the, the.

[00:16:04] Denis: Course, like the CPUs of, our PCs that we, that we, that we use, every day. And, this market is huge. This market is like, so far it's,300 billion, but it will be growing of course, very rapidly because, but, the promise of this technology is that. It can increase the clock speed, like, the basically the speed of, processing units by like a hundred times. and this is, of course, for, such, mass, mass market application species is a, is a huge number, uh, is, is a very significant improvement. Um, but this is, I would say it's still very early. Uh, on the other hand, if we talk about analog computing, which, uh. Has different applications, but we, specifically look at AI related applications.

[00:16:58] Denis: We invested in one company [00:17:00] called, uh, Lumai out of Oxford. they. Do what matter and light intelligence, tried to do in the beginning. Um, but they use different, technology. They, they basically said that, we think that integrated photonic chips is not a viable, um, tech stack is not viable architecture because it's really hard. Scale, photonics on such a small size transistors on A2 d uh, dimension. that's why they basically, um, abandoned the, the whole idea of chips and decided to make processors. And the processors basically like it's a box. That sits on the rack at data centers and uses, the so called free space, optics or 3d optics, where light travels basically inside this box freely [00:18:00] and, and you basically encode the information in the, in the photons.

[00:18:04] Denis: Inside this, processor and, then you can do like the metrics, uh, metrics, some multiplication, the inference, and this is application. Very is very suitable for AI specifically because like you need to, for inference, you need to just do a lot of huge number of, uh, Similar, almost identical, uh, calculations.

[00:18:28] Denis: And, that's why, I think that's. But at least this is the technology that we think might be viable. it's still very early, like the, they have the prototype, but, it's still like a few years, before, the first, the first commercial applications, but the upside is huge. It's the total rest of all markets.

[00:18:49] Denis: Again, it's three, 400 billion. Dollars and it will be growing because, a lot of tech companies, but also, like data center operators [00:19:00] will need to process and train, uh, a lot of data for AI for LLMs. And, this is, of course, it's not a component that will see it somewhere in the, in your PC, but it will be the essential part of most of the data centers, which are using a huge amount of data for, training AI. So I think that we think that this, market is huge and the application of, uh, for these optical processors is just enormous.

[00:19:32] Rahul: Okay, um, just to confirm, Lumi is still using the traditional transistors, but with photonics

[00:19:41] Denis: they're using, they don't use chips. They use, it looks like a box, like a shoe, shoe box. and it's called, Oh, you can, you can call it optical processor.

[00:19:52] Rahul: Okay. It's a, so

[00:19:54] Denis: Just, yeah,

[00:19:55] Rahul: to do with the silicon process.

[00:19:57] Denis: silicon photonic chips and kind [00:20:00] of just. put it aside and do a device that will sit on the rack. It will be still an essential part of the architecture to be connected maybe to other GPUs, through the classic GPUs and data center, with the all the modulators and, uh, for the detectors, like transceivers, um, it will, but it.

[00:20:22] Denis: It will perform like it will be focused only on, AI training.

[00:20:27] Rahul: Yeah. And, uh, you guys have also invested in this company, Acatonix, right? Aca,

[00:20:33] Denis: Ecotronics,

[00:20:33] Denis: is exactly the company that, develops the digital, optical computing. yeah, the one that potentially may replace, um, uh, the core of PC, in PCs if they, succeed to, reach the commercial scale.

[00:20:48] Rahul: a processor, sort of optical

[00:20:50] Denis: With, with logical gates and all the, all the digital, capabilities that will not be confined within a [00:21:00] single use case. Like, Lumi, for example, does for AI, on the AI, but Iketonix will potentially be able to just compute any kind of, any kind of calculations, just way faster, just same as, CPUs do for our PCs.

[00:21:16] Rahul: yeah, yeah, yeah. So why is, uh, like optical computing important now? I kind of feel that because of the way AI is progressing, the traditional sort of silicon based computing is going to really hit a wall and we need

[00:21:39] Denis: Yeah, basically there are two major opportunities, which may result in like, there, there are two major challenges, which may, uh, be turned into very huge opportunities, like trillion dollar opportunities. the first challenge is that the amount of data is [00:22:00] growing. I think one of the forecasts is that, By 2050, the amount of data that we consume every year will grow by 500 times. It's, if you, if you may think of it, it's, it's insane, actually, it's similar as the amount of data that we're consuming now. And then you take like the Stone Age. that's how much data we consume back then. So in, in 25, like 20, 27 years, we will consume so much data that it will be just another, it will be another level of, um, how we may imagine like the whole world functions.

[00:22:43] Denis: And of course, this. of data cannot be processed, cannot be stored, cannot be transmitted to the current infrastructure like, well, I, we've even, uh, did, um, uh, like a thought experiment of trying to, [00:23:00] um, It's to imagine how many data centers you would need to construct. And it's basically, you, if you just build data centers for this amount of data, you will basically cover the whole or the whole Switzerland will be like in data centers only.

[00:23:17] Denis: And then comes another problem that data centers are actually quite, um, they quite power consuming.

[00:23:26] Rahul: and, again, different forecasts, some forecasts, uh, Say that, data centers consume like 4 percent of the, global power. Uh, some say even close to 10 or even 12%. And, global emissions, that's like pretty much similar numbers.

[00:23:47] Denis: And you may imagine how much more emissions will be created because of this. And, uh, that's another huge, huge problem that, maybe, here and, [00:24:00] um, this is not on one hand and on on the other hand, Moores law is kind of breaking. So, um. Uh, the size of transistors is approaching atomic scale, and then you have to deal with quantum effects.

[00:24:20] Denis: You can either try to, somehow, fight with them, uh, suppress them, uh, you may, or you may think of a different architecture, because So far, what we see on the market is that the only solution is just, we just increase the number of transistors, that's all. And that would work to a certain extent, but we already can see that the, like the graph of, number of transistors, on a single chip is kind of, was linear, but then, but now it actually is accelerating and it's a huge problem.

[00:24:56] Denis: It's a huge problem because the supply of compute power will [00:25:00] not grow as fast as. Demand. there was, there was a paper by, um, OpenAI, uh, forecasting that the demand only for AI training will grow. the, the, the de the demand for data used for AI training

[00:25:16] Rahul: we'll double every three to four months. which is, there was another article then demystifying this, saying that maybe not three to four, but it's still, it's very significant difference. And, then apart from AI, you also have streaming. Which is another thing and then edge devices, which may be, are still not that big of a segment, but it will grow. And,self driving cars.

[00:25:44] Denis: Autonomous, autonomous vehicles, one of them. And again, if we, if, if we take the horizon of, 20, 30 years, that would be, yeah, that would be a huge challenge, with such amount of data and, uh. That's [00:26:00] basically the whole rationale. another thing right now is that now we really see the progress in optics and materials that enabled, um, um, company startups to, uh, solve this problem on the engineering level.

[00:26:17] Denis: And just because of the demand, because of the problem, we see a much stronger willingness of, corporates, for example, of large tech companies to invest more money into this, to, support this research. We see a lot of government funding into this space, a lot of grants, which kind of, it's a combination, but I think that pretty much answers the question right now.

[00:26:39] Rahul: Yeah. Yeah. And, um, so the, the major sort of impact that such a breakthrough in optical computing would have is obviously on AI. And then you also mentioned about things like, self driving cars, anything in particular. I think if you can. [00:27:00] Have a completely different compute sort of platform. It's going to have an effect on everything, but the Majority of it would be in a specific industry like if yes, what would be those industries?

[00:27:14] Denis: Um,

[00:27:15] Rahul: Or use cases.

[00:27:18] Denis: that's the thing that, for photonics that, that's probably because, because we haven't discussed quantum, uh, I, I, we usually tend to, uh, discuss quantum together with photonics. Uh, but, um, I think probably just kind of, um, getting a bit ahead, um, in con uh, in contrast to, uh, quantum where you have like specific industries.

[00:27:41] Denis: Like a multiple of them, but they still will have a limited number of industries that would benefit from quantum. In photonics, like everything that will, that users compute will benefit because yeah, you can, every industry uses [00:28:00] computers, every industry will now use AI and, essentially photonics is. It's an infrastructure, a next generation infrastructure that will solve the bottlenecks for faster compute, for faster training, and for, again, for energy efficiency. So I would say the beauty of photonics is that it's kind of universal. It's not like it will help. companies in, pharmaceutical space and pharmacy, but it will not help in fintech, it will help everyone because it's a universal infrastructure.

[00:28:35] Rahul: I Guess another way of looking at it is like this technology will be pushed To, uh, like make it real, the push to make it real would be like a lot harder because of the breakthrough in AI.

[00:28:50] Denis: Uh, it's one of the factors that drives the demand. but, there are all major other factors like, [00:29:00] streaming, for example, remote work, which are on the demand side and on the supply side is again, the new, combination of materials. Um, the new technology of lasers, also several factors that enabled, like, for example, companies like, and why try to, manipulate light in such a sophisticated form and create those frequency comps that, enable to basically sleep light and manipulate it.

[00:29:30] Denis: So. It's a, it's a combination as always.

[00:29:34] Rahul: Okay. And in terms of talent, uh, and startups like operating, trying to build, in the space, where are you seeing those coming from? we are both in Singapore.

[00:29:45] Denis: I would love to say yes, but, in photonics, surprisingly, I've even before, before this, podcast, I did a, I did a quick research. Um, and, uh, uh, obviously there are, uh, research centers like in [00:30:00] us, a star, which, do a lot in, for example, silicon photonics, um, but. I wouldn't say that there are a lot of, first of all, photonic startups in Singapore, at least I haven't seen many of them in comparison to Europe or the U S and in Asia in general, uh, excluding China, because China is a different universe.

[00:30:27] Denis: They have. Quite a lot of, interesting startups there, in, in, even in optics, like their quantum computer from, um, uh, university, uh, is, is using for optical, uh, technology, but, um, Obviously, Japan is also quite strong with NTT and Sony and other, players, like University of Tokyo, for example, is, quite prominent in this space.

[00:30:54] Denis: But it's still a fraction of what we see in Europe [00:31:00] and the US. Especially Europe, surprisingly, um, there is even, uh, a rating of, the most prominent... researchers in photonics, um, research researchers, which have like the, some commercial, applications, like, uh, uh, including like a startup co founders.

[00:31:20] Denis: and, like for example, one of the co founder of, Lumai, was recently, included into this list. he's from Oxford. and we see that I think 30 out of 130 people were from the U S. One was from Japan, one was from Australia, and, rest is Europe. So, it's being Germany, France, UK, and then, like, across the continental Europe.

[00:31:49] Denis: So, it's still, it's still Europe, um, that's why we... Although we are very interested in APAC, but in photonics, specifically, Europe is still [00:32:00] a very interesting space in terms of talent, both on the scientific and engineering side of things. Um, but yeah, still, I think, in Asia, in APAC, in countries like, Singapore, Japan, from time to time, we see some teams which do something similar, but maybe not that revolutionary.

[00:32:20] Denis: Like for example, for tonic chips, I've seen several startups. I even talked recently to a startup from Indonesia, which does put on a technology from, related to photonic chips, but, It's still a fraction of what we see in Europe.

[00:32:36] Rahul: In terms of the background of a lot of founders, right? I, I find, feel like, they have semiconductor sort of background or they are physicists. Would you agree?

[00:32:47] Denis: not necessarily like, for example, um, the founders of LUMAI, like, since you go, the co founder of LUMAI, who studied in Oxford, he has the background in quantum [00:33:00] physics and optics. so it's more on optics side of things rather than semiconductors. Sometimes if it's, uh, Related to, um, some applications and CMOS architecture, then of course, there are also people with semiconductor background, but I would say it's both.

[00:33:19] Rahul: Not necessarily, depending on the, uh, like for data centers, for example, like optical conceivers, they do not have like direct correlation with, uh, semiconductors. Okay. And, uh, Yeah, let's talk about, quantum computing in comparison, to optical computing, right? So you mentioned the use cases for quantum computing is specific, but scaling is a problem so far, and also maintaining the coherence of the quantum qubits is, is, has been a problem for a long time.

[00:33:53] Rahul: But if there is a breakthrough, then it can really solve complex problems like cryptography And, and yeah, a [00:34:00] lot of these complex issues, right? So, uh, but in comparison to that optical computing, it increases the speed and also the power efficiency and the applications is more, uh, widespread. So, I was listening to this, uh, interview with this light matter CEO, right?

[00:34:17] Rahul: he, he was saying that. Maybe we don't need quantum computing at all because deep learning has been able to kind of mimic very complex problem solving. So let's say we can achieve the speed and the efficiency and combine that with deep learning, the existing deep learning sort of breakthroughs that we already have.

[00:34:38] Rahul: Maybe we don't need quantum computing. Would you agree with such a statement?

[00:34:42] Denis: would probably not agree with this. because, we can have both. Actually, it doesn't hurt having both of it. Um, and, uh, uh, there is, uh, something that you can do with, deep learning, but then there's just way [00:35:00] much more than you can do with quantum sensing, for example, in, I think one of the implications that comes to my mind is in mining, I think Saudi Aramco, is one of the companies. I think someone who is working with them on the, by helping Saudi Aramco to, um, better analyze the mineral reserves, using, deep learning, um, by basically projecting like what they found the data they already have. And trying to extrapolate what they would found if they drilled, like, a few, a few miles, a few meters, a few hundred meters, deeper.

[00:35:40] Denis: This is simplified, but that's basically the, the principle. While quantum sensors, for example, which is, again, it's not quantum computers. It's just, it's sensors using quantum mechanics. they can. Solve the problem that [00:36:00] they can create the whole 3d map of the reserves with, extremely, way higher precision and get the data like the a hundred times more or a hundred times more precise data that Saudi Aramco would have.

[00:36:18] Denis: With the current sensors and the current technology exploration technology, and then they can apply AI on top of this and even improve this data. So, I don't think it's, first of all, it's not, it's not irreplaceable. It's basically quite complimentary. And there are different, there are different applications, as I said, quantum sensors is one of them. And, um, there are a lot of applications for quantum sensors, but then you have quantum computers, quantum computers, uh, so, uh, helping to solve extremely sophisticated optimization problems, which is a different [00:37:00] thing, you may think about. Um, for example, optimizing the load and, micro grids, which are distributed using, solar panels, windmills, EV and charging stations, different like, uh, different connectors.

[00:37:18] Denis: and the quantum computer computing may solve this because, the biggest problem, like in a, in a grid. In power grids is that, the sources of energy and, uh, sources of demand are becoming increasingly unpredictable. And unstable, like, 50 years ago, you would have like a single power station working on coal.

[00:37:47] Denis: And then you have like a, a high voltage line, connecting this power station with a big city where you have all the major consuming, substations. And that's it, that [00:38:00] you don't need a quantum computer for this, but if you have a distributed network, with, all the different renewable sources, which are, uh, changing according to the weather, which is, Quite hard to predict unless you have quantum computer, by the way.

[00:38:16] Denis: That's another use case that I could talk about a bit later, but That's where you would need quantum computer because it's really hard to simulate What's going on inside the system. Same for drug discovery, for example simulating how a single molecule would behave in different environments with different effects Um, with different, in different conditions, um, it's just now it's taking years for drug discovery companies to simulate it and to do the R& D research, to do testing.

[00:38:55] Denis: but quantum computers, those companies can do testing way faster. or [00:39:00] material science, basically very similar, um, use case to drug discovery. One of our quantum computing companies, Pascal, uh, which is a French, um, quantum computing, developer. They use called atoms as a, as a tech stack. They have quite a few.

[00:39:22] Denis: different use cases with corporates already, and they, they have revenue. So it's a, it's one of the, uh, one of the few is quantum computing company that already has some commercial traction. because it, and this case actually shows to a higher extent how many corporates are really trying to understand the, all the use cases and trying to experiment.

[00:39:44] Denis: with all these use cases and they're ready even to buy those quantum computers or, connect to quantum computers and pay for, um, load. Of using them, and, I think this area is still, it's a huge area. And they're [00:40:00] still like, when I said, there are specific use cases, there's still a lot of them quantum can solve many of these problems.

[00:40:08] Denis: so I think, in this case, I wouldn't probably agree that, um, AI. Can make a quantum computer computing redundant. No, it will basically be complimentary. And there was just way much more problems that. AI will not solve, will not solve as good as quantum computer.

[00:40:31] Rahul: Okay. Okay. I think one thing that we did not talk about is the challenges of optical computing. Is that, uh, got to do with just manipulating and controlling light?

[00:40:43] Denis: there are different challenges. There are challenges, um, in terms of, uh, also integrating it with, traditional CMOS architecture. That's for silicon photonics, for example. Um, like optical transceivers, [00:41:00] optical processors also have their own challenges. Um, I think it's too technical to kind of mention all of them, but in general, kind of saying, that, it's still quite early stage in terms of the, like, how would you integrate such processor, for example, in your data, center infrastructure, like, how would you integrate it with your classical, architecture?

[00:41:31] Denis: That's something that is more like an engineering problem rather than a scientific problem, but it needs to be done. It just takes time. It just takes time to, test it, to experiment, to find the right configurations. But, I think this is one of the major, and then another challenge that I think also mentioned that the architectures are just extremely sophisticated for photonics just because we've been dealing with semos and semiconductor [00:42:00] like silicon based architectures for many, many years before it's become like a state of the art technology, for photonics.

[00:42:06] Denis: It will probably also take, it's already taken quite a few times, as you mentioned, like the first, the first silicon photonics were in. Started to develop like in the, in the eighties, and we've seen the first applications like in the two thousands, but, uh, still it will take some time, but I think, we think that now actually it's, um, very close to implementation and to commercial fraction, like for optical processors, for AI, it will take like several years actually.

[00:42:40] Denis: It's way, it will, it will happen way sooner than quantum, than the real quantum revolution. So, we think that that's one of the attractive aspects of, photonics in general.

[00:42:53] Rahul: And uh, from an investment perspective, uh, are you seeing enough opportunities? That's one. [00:43:00] Second thing is, where are you seeing those opportunities? In the specific optical computing

[00:43:06] Denis: In, well, first of all, Also, small disclaimer, I'm not, sourcing, most of our companies in, uh, Europe. I'm looking at what we have in Asia, which is, uh, so far, not that many. Um, but in Europe, from what I see from, um, my teammates, we actually see quite a lot of that. Photonics is still is a niche, uh, segment, um, because, oh, because there are just not that many, um, specialists, not that many scientists in photonics, and you cannot scale them as fast as. developers, for example, that's why, although the demand, is growing really fast, but the supply of, uh, talent is quite limited [00:44:00] and quite niche. And that's why, the steel, the number of. Like maybe 10 to 20 products, really good products that we've seen in Europe. Um, and probably the same number in the U S and maybe, I may be mistaken again, because, we have, special people who are looking at this space in Europe and us.

[00:44:21] Denis: but it's like, it's not, it's not huge number of, opportunities just because, uh, they're not that many teams who work on this.

[00:44:31] Rahul: Yeah. And, in terms of, so what would be like the, the business model for some of these startups? Like, it's just like a lot of them, for example, like matter, they are, um, building for cloud infrastructure. Right. So,

[00:44:53] Denis: Um, It's hard to say because it's very early, at least [00:45:00] for the companies that we invest in, in photonics, um, they trying to figure out. Um, so they, they pretty much probably what they figure out already is the profile of customers, but they're trying to figure out how they would fit it into their infrastructure.

[00:45:22] Denis: And who would be like, what would be the go to market strategy as well? So this is something that is still under development. because for like, if we don't talk about just platonic chips, which. As you mentioned, it's already quite established, um, area, although very limited, they are, I think business model are very simple.

[00:45:44] Denis: Just you supply chips to, um, other semiconductor providers or to, um, cloud providers, like, AWS, et cetera, to, manufacturers of, uh, optical equipment. Or equipment just for data centers and try [00:46:00] to integrate into the solutions. I may probably just guess or kind of forecast that, the new wave of optical computing startups will have similar approach.

[00:46:12] Denis: at least for AI, like for analog computing, for digital computing is, again, it's very early. It's just, um, too early to say, what would be the exact. like what would be the exact, sales channels, how they would like, would it be, I don't know, subscription or would they integrate the solution into like, existing supply chains?

[00:46:35] Denis: But, I think it would be quite similar to, it may be projected to the, to what we have seen in the class computing world.

[00:46:44] Rahul: yeah, but, but surely if it works, then it would, be highly profitable. Scalable solutions that fit right into the VC sort of investments

[00:46:55] Denis: That's what, yeah, that's what we think [00:47:00] it will be. hopefully it will, um, fulfill its promise, uh, because the, um, uh, because the size of the problem is just very huge. And essentially, um, for most of the such companies. exit strategies, not really like creating a sustainable business, reaching profitability, like shipping, uh, millions of chips.

[00:47:27] Denis: Um, at least not for early stage VC investors. Like for us, we, we invest in companies that like. At seed stage, when they valued at like slightly less than 10 million, uh, and for us having substantial return would be already enough if the company, reaches like a billion or 2 billion valuation. Um, and for many, uh, for many, uh, companies, The [00:48:00] strategic acquisition would be a more probable scenario, because, they are developing technologies. That are kind of threatening the, not, okay, not existent,

[00:48:16] Rahul: Existing

[00:48:18] Denis: don't want to, I

[00:48:19] Denis: don't want to say existence, but they're threatening or disrupting the areas that are, that Traditional, the current leading players are, working on like NVIDIA, for example, with, with GPUs for AI, like Lumi, developing new generation technology for, for the same use case, or like, our first quantum investment that we did in 2013, they, Basically disrupted, like they, they did quantum communication devices and, they just started to disrupt the telecommunication, uh, industry and got acquired by, South Korea telecom, because [00:49:00] it was just a threat for the company. Um, and this is, I think the most probable scenarios and some of those exits, you may think. Of course of Microsoft acquiring, uh, open AI for the $10 billion. Uh, but also another interesting, uh, example of, Xilinx, which, was doing, basically high performance computing, infrastructure was acquired by AMD for, I think 1440 $7 billion.

[00:49:34] Denis: So. Depending on the size of the problem and on the maturity of the technology, you, the exits, the acquisitions can be very, very large. And, um, you don't, you don't necessarily need to wait till the company kind of really becomes [00:50:00] profitable and builds like a traditional business, like traditional SaaS business.

[00:50:04] Denis: So that's. I think that one of the most probable scenarios for most of such deep tech, companies, startups, for example, in optical computing and quantum computing spaces.

[00:50:16] Rahul: Yeah. Yeah. I mean, I think this is true across a lot of other other tech as well. Like self driving cruise, which sold to GM and then that was the exit.

[00:50:28] Denis: Yeah.

[00:50:29] Rahul: He took like another. Seven years or something for them to commercialize.

[00:50:35] Denis: Some technologies, already pose a threat as soon as you realize that they may work or if they work, like if they prove to be working, then you just extrapolate what will happen next. And, if you're smart enough, then you may probably think that, okay, now it's better to stop it [00:51:00] now and be a part of this process or just to...

[00:51:03] Denis: Um, yeah, try to integrate it into, into our, infrastructure, rather than creating a way till the competitor is created, because in this case, your competitors may invest in this company and they can benefit from it. So yeah, in deep tech, that's the most probable scenario, usually.

[00:51:26] Rahul: Yeah, this was great, Dennis. Thank you so much for taking the time

[00:51:30] Denis: Thank you. Thank you so much for having me and yeah, hopefully it was, helpful and insightful, happy, looking forward to a new podcast as well.