Conversation with Ayar Labs CEO Mark Wade on Silicon Photonics, Co-Packaged Optics, and Next-Gen AI Infrastructure

0:08Ben Bajarin: Hey, welcome to another episode of The Circuit. I am Ben Bajarin.

0:13Jay Goldberg: Greetings, programs. I'm Jay Goldberg.

0:15Ben Bajarin: Well, we've got some very special guests. We are joined by Mark Wade, CEO of IR Labs, and Jesse Leiter, who is, you can give your title, I think, Head of Finance, IR. Why don't you guys say hi real quick, and then we'll jump into it. Yeah, yeah. Hey, everyone. My name is Mark Wade. I'm the co-founder and CEO of IR Labs. Ben and Jay, thanks for having us on. Yeah, thanks, Jay and Ben. And my name is Jesse Leiter, and VP of Capital Strategy and Investor Relations. The role has changed. It had a strategic finance element to it before, Ben, so you're not completely off on that. But, yeah, I'm really excited to be here.

0:55And I think Jay and I have been speaking for maybe a little bit less than a year now. He mentioned the podcast. Ben, I know you've been talking to Mark more recently. And so, yeah, excited to be here speaking to your audience and on your platform.

1:08Ben Bajarin: Yeah, it's great to have you guys. This is a very hot topic. I know that you guys, not being a public company, there's always people asking us, like, hey, what's your thoughts on IR Labs? Like, we're trying to get more information. So I think this will be super informative for our audience. And, you know, my context is, because you guys haven't heard me say this, but Jay and I joke about this for a long time, like the circuit is slowly becoming more of a networking show, which is the irony of this is, you know, back in the day, right, we all covered networking, but it was relative to telco and it was super boring.

1:48Ben Bajarin: Like, great. You've got more switches. Sure, Fiber's great. This is all interesting. And when we started this podcast, it was largely compute because what was going on in Compute Trends. And then network became this whole thing, right? Interconnect and whatnot. And so now we've spent so much more time also talking. But it's relevant because interconnect is a wall. It's a problem to solve. It's one of those big hurdles to get over. But GPUs and even memory is scaling faster than the interconnect wall is, right? And so I think that's super interesting. So at a high level, like as we start, you know, Mark, why don't you just give our audience in case they aren't familiar with UIs, just the high level pitch for IR Labs?

2:31Yeah, so let me try to start with a one or two sentence framing, which is that large scale AI data centers are limited in their revenue scaling by the performance of the interconnect in their compute systems. And so that's a claim that says, hey, if we want to go beyond a path to continuing to scale AI data center revenues, to continue to grow the supply of token throughput, to improve the tokens per second per dollar per watt on these kind of tokenomic, and while simultaneously improving the performance of these frontier leading-edge AI models and their architectures and things like the reasoning models, we have to improve the interconnect.

3:19So that's the kind of high-level motivation, which is that we view solving this interconnect bottleneck as one of the core major hardware level bottlenecks that needs to be solved to keep the AI roadmaps running. Now, if we double-click into that a little bit, you know, the bottlenecks are coming from these electrical interconnects that exist in the so-called scale-up fabric. Scale-up fabric connects a number of GPUs and AI accelerators together. and you've got this issue in electrical interconnects, which is that the faster you want to go, so the more data you want to send between these GPUs and accelerators, the shorter distance you can go.

4:03So we call that the bandwidth distance trade-off. And the only thing that we know how to solve this, or the only way that we know how to solve this from first principles from the grounds up is to swap over to optical connectivity. So that's where we say, hey, we've got to get a new generation of deeply integrated, you know, CMOS scale optical connectivity into the scale up fabric. Solve this bottleneck from the ground up, you know, where the real physics problem is happening. And then we enter the promised land beyond. So that's the quick snapshot of what IR Labs is trying to do.

4:44Jay Goldberg: So I've covered optical for a long time, all through the telecom ups and downs and ups and more downs. And the one word I if you ask me, like to do word association with the idea of optical networking, what's the one word that comes to mind first? It's pain. Yeah, this stuff is hard to do. It's tripped up many people over the years again and again. Why is optical so hard? Yeah. You know, I do think if you want to go try to build things in this area, you have to, you could call it masochistic. You have to enjoy the process of pain and suffering and of solving hard problems and solving complex problems.

5:29I'd say, you know, optical devices, optical solutions, part of the challenge is that to really get a full end-to-end electronic and photonic solution to work. You're talking about the combination of multiple domains of technology and expertise. And, you know, in previous paradigms in the industry, like in telecommunications, you had specialization of those different domains. So, you know, you'd have entire companies that just provided, you know, a single laser. Or you had entire companies that provided the DSP engines or parts of the electrical circuits that connect to the optical devices. And then you had entire companies that just build the optical devices.

6:18So I think part of the challenge is that each of these domains is difficult, and you've got to put them all together in a way that can actually build a product and work end-to-end. And, you know, there's material science involved. There's device design. There's manufacturing back-end challenges. And so, you know, it's kind of layers upon layers of deep, technical, difficult problems. and optics kind of layers in, you know, all the, anything that's difficult in CMOS and high-performance electronics, you know, you then layer optics on top of that with different supply chains and a different kind of materials and device requirements, then there's just a lot of difficult problems there.

7:11And so you'll find, you know, you'll find, if you come to a company like ours or other photonics and optics companies, yeah, it's a very high density of PhDs, of people with physics and electrical engineering and mechanical engineering types of backgrounds. So I think that's just the truth of it.

7:33Jay Goldberg: So why now? What's changed that makes this buildable today when it wasn't, say, five years ago or when you started? Yeah, well, let me draw from a little bit of what I just talked about, which is that one thing we recognized early on, And I didn't touch on it in the intro, but myself and my co-founders have been working on this for 15 years, first starting mostly at MIT and Berkeley. And we started IR Labs coming out of MIT in 2015 while we were still students. So I also like to say all I've done in my adult life is try to get this technology to work. So we're fully committed and all in on it.

8:16but let's talk about the why now and also what had to happen to to enable things and i want to go back to saying hey well how did it used to look and i just described these kind of you know telecommunications products and solutions you have to look at it from a and i'd include data communications in that you have to look at it from a what problem were they solving in the marketplace what was the value of that problem and what was the right configuration of of companies and and supply chains and value stacks to solve that problem. And the reality was the volumes in Datacom, actually the volumes in Telecom were fairly small, so you had kind of bespoke relatively small and niche supply chains that were able to deliver to them.

9:03Datacom was supposedly this huge market that was much bigger than the optics community was ever exposed to. And now that we're talking about going from what was traditional datacom all the way into optical connectivity coming out of GPUs, AI accelerators, optically connected memory, scale-up switches, this is a 10 to 100x larger volume than the datacom market was serving. And so the why now is about saying we needed a market application whose revenue and whose high value revenue was actually gated by connectivity and chip to chip connectivity, rack scale connectivity and rack to rack connectivity.

9:52And the reality is that before we had large-scale AI explode onto the scene, there just wasn't a market application that was large enough and had the high enough value revenue gated by connectivity. So now we have that, and that really started – the awareness of that started a few years ago. For a lot of people, ChatGPT getting released in our industry was the wake-up call. And the reason it was a wake-up call is because everyone very quickly understood how the core computation of these applications scales. and so a lot of the work that that we and others had been doing you know showing how connectivity was going to gate and be a critical limiter to building large-scale distributed computing systems and you know we've been working on that since 2010 and before so we had all the kind of core technological ideas the the kind of grounds up ideas all that was in place and we'd made a lot of progress.

11:01But, you know, we needed the spark of a market application that made large-scale distributed computing highly valuable. And that's what ChatGPT releasing did. Now, we could talk more about it, you know, saying, hey, well, ChatGPT came out a few years ago. So, you know, why isn't iHerlabs IPO'd and why isn't CPO everywhere? You know, I'm sure we'll dig into that next. But I would say the tier one volume ecosystem finally got fully interested, you know, leading up to the chat GPT release and the year after chat GPT released. And then the reality is, you know, the tier one high volume advanced semiconductor ecosystem, you know, can't pivot on a dime.

11:48Ben Bajarin: And maybe before we get into some of the timing of these things, like, maybe just talk a little bit about kind of the the manufacturing challenges you know we hear this a lot I know again you got to be deep in the weeds you got to be in foundries to do this but like everybody keeps telling me like it's just really hard it's just really hard to make these the wafer side is hard like all of it so I just love just brief color help us understand like why the manufacturing side of this is also really hard yeah and I think we are seeing people kind of go through their own educational process, trying to understand what are really all the complexities around building photonics, you know, the way that we're trying to build them now for co-packaged optics and even other form factors.

12:33And let me try to come at it from a few different angles, because people, what I find is people typically, you know, that everyone's coming from their own understanding and you have to map it to a few things. But one of the first things to realize is we say that we're building an optical engine. And our optical engine is a combination of a leading-edge advanced node electrical IC. So right now we build in the TSMC coupe process, which the EIC is a three-nanometer EIC. And on that EIC are hundreds of millions of transistors. So don't think of, even though now the word photonics, silicon photonics, optics, But you should think of this as, you know, photonics adding a new high-performing leading-edge advanced capability to leading-edge CMOS nodes and leading-edge advanced packages.

13:30So the first thing to realize is that, you know, people in general kind of go, okay, I think leading-edge CMOS is hard. I think advanced packaging is hard. People know about co-host, people know about the supply chain shortages and challenges on getting these kind of large package multi-die AI accelerators manufactured and produced. So you should just bootstrap into that entire understanding of, and I've got some eye candy to show here, but a large multi-die package AI accelerator is a difficult thing to manufacture and build without optics. okay so now we're going to say i'm going to take this large leading edge large substrate multi reticle uh you know coos like advanced package and add an optical engine and the optical engine is a combination of leading edge eic's and a new photonics wafer manufacturing process flow and the new thing i have to do there is also attach an optical connector to this package So I think that's the easiest way that I, at least people who are coming from the understanding that multiradical, large package size compute accelerators are difficult to build.

14:51And now you're adding a new advanced capability into that package. So, you know, difficult on difficult makes it more difficult. Right. So now you could say, well, why is everyone doing this? Why is everyone so motivated? Well, it's because, you know, we've really hit the physical limits that are possible to escape the amount of bandwidth and escape the radix of connectivity out of these large packages. And, you know, you've got multiple technologies within that package that are just, you know, fully pushed up against their physical bounds. And so you need a new feature that can come in and open up a new degree of freedom of scaling.

15:32So, and that's just on the CMOS and kind of large package size. And we should dive into it next, but there's things to talk about on the laser manufacturing. There's things to talk about on the production test architecture. How do we get photonics bootstrapped into CMOS scale end-to-end wafer testing, assembly test? You know, that's really where, you know, the current crunch points and, you know, cycles of learning are being accumulated and focused on right now. Yeah. Yeah.

16:08Ben Bajarin: Yeah. Let's get back to that in a second. I want to talk a little bit about some of the timing and kind of different, you know, solutions out there. And I don't think, you know, we don't necessarily need a like, this is where each one from, you know, LPO to NPO or CPO fits. I think, you know, CPO is interesting. But I am curious, like there's this, well, one, figuring out the timing of this is really hard. We also understand that everybody building infrastructure is kind of taking their own approach to this, right? And so if they mix and match interconnect, it could be copper, it could be PCIe, it could be micro-LED, it could be optic.

16:46Ben Bajarin: All these things may exist in different implementations to get over this. But when you guys kind of look at this landscape, how do you see interconnect playing out? And I ask that within this context. Do you think kind of one solution maybe wins at the end and becomes dominant? Or do you think, you know, this, it just, again, it might be a big mix and match of all of these things as guys try to figure out their rack designs. Yeah. So I'll tell you how I think about this, which is that, you know, in the last couple of years from our largest customers and even not our largest customers, you know, we spend time with the, you know, well-funded startup accelerator communities and things.

17:31the demand for a solution that makes optical connectivity exiting the package and makes that look like something that can be high volume scaled through a CMOS-like manufacturing flow so that customers can start to think of optical I.O. within the context of their foundry OSAT model of the world. The demand for that is just unquestioned. And part of it is because once we get to that world, it simplifies a lot of the product design, the form factors, the kinds of configurations and choices you have to make. So I do think there's a trajectory as we get past this kind of crunch in the next three, four, five years.

18:24And when we land in the world of optical connectivity is now just viewed as an advanced feature coming from a foundry. That simplifies the world a lot for people because you can get out of this chaotic land of, hey, is it a pluggable? Is it a board integrated component? If it's a board integrated component, What's the form factor? You know, what's the physical technology inside of that? What supply chain is delivering to that? You know, there's a lot of kind of permutations that you have to manage the complexity of to simply overcome the fact that I can't get this from my CMOS foundry. So I do think the longer term view, and you know, when I think when I say longer term, let's say the 2030 and beyond view starts to simplify and get a bit easier.

19:15But the shorter term view when, you know, the reality is that these things are not just fully available in high volume production today. You know, it forces everyone to scramble to find, you know, well, what's my 2027, 2028, you know, kind of stopgap solution going to be. And then that triggers a full exploration of the solution space, which is a combination of, you know, which technologies are close enough to try to give me the performance that actually solves my problem, but also what supply chains are supporting and delivering those technologies. Because I can have the greatest technology in the world, but if you can't ramp true volume production, then it's not a solution for customers really trying to build high volume Rackscale architecture and deploy them on time.

20:07because everyone's optimizing to, hey, what's my time to first token that generates revenue? So yeah, I think in the short term, we're going to live through a chaotic, turbulent set of technologies, form factors. But I think in the longer term, we're going to land this new generation of silicon photonics into a TSMC-like foundry plus OSAT flow. And companies like ours are focusing our efforts on building the highest performance chips in that environment. And we think once we prove that it can be delivered at high volume scale, it simplifies a lot of the customer problem.

21:03Jay Goldberg: let me let me sort of step aside for a second what what is what's the nerdiest detail of your technology uh that you just get sort of excited about that may be obscure or super esoteric to other people well yeah you know i i was let me think about that for a second and i may touch on a couple of things. And, you know, my own background is my academic training was in physics and electrical engineering. And I spent my graduate years, you know, working on the silicon photonics device technologies that are now underneath everything that's happening in co-packaged optics and even things like NPO. So, you know, I can get really excited still about the core device technology.

21:49And I designed a lot of those core devices in my previous years. But I'll say, you know, what I've learned about myself is that, and a lot of people coming from physics and, you know, that background think like this. I tend to get most excited about trying to understand things from the grounds up. And then, you know, whenever I run into something that I can't quite fully wrap my head around yet, I spend most of my energy and most of my excitement kind of in those areas. So what that looks like for me today, and I'm having so much fun with all these coding agent models that are out now, Claude Code, GPT-6 Astra, some of the open source models.

22:35And so I'm spending time building a full end-to-end digital twin manufacturing model of what this new world of co-packaged optics, laser supply chain, TSMC packaging, all the test insertions. and so you know i won't i won't show it to you but i how do you build the right model at the core of that it's how do you implement it as a directed acyclic graph so i can do you know lazy runtime evaluation to keep track of all the details that need to be kept track of on the manufacturing model and you know i'm actually getting very excited about the manufacturing is very complex, but if you have visibility and see how it's all coming together, there's phenomenal opportunity to optimize details of that manufacturing that we think will be critical to actually getting to a volume production solution.

23:38And what do I mean by details? Hey, depending on my business model and exactly how the materials are yielding and performing coming out of all the different core manufacturing steps, what's my optimal test policy across five test insertions? And which one makes sense for me versus which one makes sense for my customer? There's actually a very rich set of interesting questions that get asked and have non-obvious optimal solutions in that landscape. So strangely enough, that's what I'm spending my nights and my weekends on right now is burning a lot of coding agent tokens to build a digital twin model of all these things.

24:27Ben Bajarin: Nice. Nice. All right, so you guys have made kind of a very particular bet in a very specific solution. So I always like this question, right? And it's kind of broadly not just the one area you guys are going full bore in, but also just like some of the things that I get there. So what's the technical bets, I guess, like the things that you guys are betting on as a company? Yeah, you know, and I'd actually say the way I view it is we have not, okay, you could say we've made a very specific bet, but that specific bet, you know, to me is very broad and obvious, which is that you're going to have optical connectivity coming out of large scale, leading edge advanced packages.

25:17and to do that you have to have optical chips come out of CMOS-like leading-edge foundries and supply chains. That's really the core bet that we've made as a company and that was really the founding bet that we made as well to say there's going to be a future where optics moves inside of the compute package and optical connectivity moves inside of the computing fabric. We used to call it the compute fabric, the memory semantic fabric. You could go look up our kind of hallmark academic papers at the time. We built the first ever processor to DRAM optical interface and published that in Nature back in 2015.

26:04So the foundational bet was optics is going to go into computing fabrics. If you want to deliver something at that scale, compute products in that space are built in leading-edge CMOS-like supply chains. And to us, that means leading-edge foundry like a TSMC. And that leads us into saying, well, why did we go fully focus on TSMC Coupe? But also not just the foundry, foundry plus OSAT. How do you build this full advanced package, which has to have a lot of these pieces coming together? So to me, that's an obvious bet. And then you back yourself into everything that you kind of go drive down into in the technology weeds starts from that insight.

26:53So we could get into devices. We can talk about micro ring resonators and micro ring mux, demux filters and, you know, dense wavelength division multiplexing on the laser side and external light sources and why we chose all of that. But, you know, to some degree, those are the details that came from a higher level bet. And the higher level bet was we have to make photonics look like CMOS leading edge scale manufacturing.

27:24Ben Bajarin: Yeah. No, that's a good point. And, you know, I would love some call. I hear from people, maybe not everybody, but I hear whispers of like folks concerned because I'm not going to say that TSMC is not immature, but you'll hear Tower and you hear Global Family, and it makes it sound like others that they're farther along, right? That TSMC is just not. So you're making a big bet on this. You guys are working with TSMC. You're in-house. Everything's under one roof, plus with your customers. So just give us some color, like your confidence level, that TSMC is the right bet for this and that they're moving forward on a path where there's confidence, right?

28:04Ben Bajarin: Confidence that TSMC is the right play here. Yeah. Well, one, let me start by saying I want to make a couple of general statements, and then I'll get specific on TSMC Coupe to the degree that I can because we've got lots of wafers flowing out of the fab right now, so I think we're in a good position to make some general comments on this. But one thing that always annoys me a little bit is that it depends on your mental model of how you think hard things make progress. And hard things make progress by you always going through this phase of, if you want to say like, hey, are you guys having challenges on getting things working?

28:42Yes. And, you know, I don't know that there's any other model that says, you know, what I tell our company, we have to assume that we're going to go from very difficult to figure it all out to fully working and fully in production. And that's not just on the CMOS Foundry side. That's on almost every single thing. So, you know, and I know where some of this commentary comes from because, you know, again, people, a lot of people in the finance world oftentimes are trying to figure out how to move positions on a daily, weekly, monthly basis. But I think in the technology world, you know, we make progress, you know, monthly, quarterly, yearly, right?

29:22So there's kind of a mismatch of the language that we use to describe these things. But let me kind of go to TSMC Coop. And one, IR Labs is a fabulous company. So we hope that we're living in a world, I don't know, 10 years from now, 20 years from now, where we've got a rich set of interesting foundry solutions that we can develop products in. But one of the first insights to realize is that a lot of times people make these comments trying to compare one foundry's photonics wafer process technology or photonics chip to another foundry's photonics wafer. But the thing to realize, a photonics chip is nowhere near the final product form factor.

30:16So the final product form factor looks like this full advanced leading edge package. And so the way we think about it is, you know, my optical engine, it needs a very high performing leading edge advanced CMOS EIC. It needs to be 3D integrated and advanced packaged with a photonics chip. You know, that needs to become a basically a wafer of this, you know, TSMC coupe wafer. it then needs to go through what's becoming a scalable post-foundry OSAT ecosystem, and you need to integrate it into the most advanced packages in the world. And so only after that can you say, oh, I've got a real path to a differentiated product in a real market that can consume it at volume scale.

31:05And that is where we've chosen to make our bet, because you could say, well, hey, can't you use a photonics chip from another foundry to go build a pluggable transceiver or maybe some NPO modules. Yeah, you know, and actually, I don't have a lot of commentary on those spaces. I think, you know, plenty of photonics foundries are going to successfully supply into those kinds of form factors. But when it comes to saying the opportunity and scale up and to really attach into a compute fabric, into a GPU, AI accelerator, optically connected memory, you know, we're members of the Envy Fusion, NVIDIA ecosystem.

31:45So if we want to attach to a real NVLink, NVSwitch, that takes a different kind of set of requirements. And we define our success metric in how do you build a product, how do you build a supply chain, what ecosystem is going to successfully deliver to that intercept. And so on TSMC Coupe, one, TSMC has been a phenomenal partner. We've been working in TSMC Coupe for, you know, say approaching four years now. And we felt confident that the value that we could help bring to TSMC, and we're like this with all of our manufacturing partners, is that we know how to build photonics. And we know how to build electronic photonic solutions.

32:34And TSMC is, you know, the number one foundry in the world. And they know how to build high volume, high quality, high performing, you know, wafer scale technologies. And silicon photonics at the end of the day, from a wafer foundry process perspective, is just not that hard to do. You know, you need high quality silicon waveguides. You need some materials like silicon nitride. You need a good photo detector material. And, you know, we and almost everyone else are using germanium in, you know, for our detectors. You need a good way to get fibers connected, you know, you need to get the light onto and off of the chip, which comes through the fiber connection.

33:23That ends up being probably the most challenging part. But then the rest of what you need looks like advanced node CMOS and advanced packaging. So I think we just view, yes, photonics is a critical, critical piece of what we're building. And it's necessary to have, you know, phenomenal performance in your photonics chip. But it's one of those things that's necessary but nowhere near sufficient for the full solution.

33:50Jay Goldberg: what what are the what are the biggest hurdles to adoption that you're hearing from customers and here i i know i know there's obviously technical challenges you're facing i'm really looking at the customer side of this um what what are customers saying that is sort of preventing them from adopting this yeah you know so the i tend to break the customer sets into a few different classes. And it also shows you why, you know, a lot of people comment on the strategic investors and strategic relationships that we've made over the years. And we were very intentional in building our strategic investor, strategic development partner relationships, because it was reflective of how we thought we needed to work with the industry to bring these technologies forward.

34:40And, you know, that really comes from kind of two different directions. One is that you have to work very closely with the high volume supply chain that's going to scale this. So we have multiple rounds of equity investment from TSMC, from some of our other go-to-market partners, and we do the thought experiment all the way out to the Rackscale infrastructure. We made an announcement a couple of weeks ago that WeWin is one of our strategic investors. But we also worked very closely with who we thought was going to experience, or maybe where we say where the highest value proposition of co-packaged optics would overlay with the highest value products in the market.

35:20So, you know, who's building GPUs? Who's building AI accelerators? And so, you know, on our cap table, we've got, you know, multiple rounds of investment from NVIDIA. We have multiple rounds of investment from AMD. We have multiple rounds of investment from Intel. And so that kind of anchors that collection of leading-edge GPU customers and leading-edge foundry and supply chain partners. That's the first group of people who have to solve this at volume scale. And we see very high conviction and a lot of progress, let's say, amongst that set of people. It's outside of that circle that faces a more challenging problem, which is that they need to wait until they don't have a lot of ability to burn down volume supply chain new technology risk.

36:16So the leading guys burn down the new technology risk. they need to successfully deploy. And then beyond that, everyone who is building their SOCs through ASIC service partnerships, well, they need to view it as something that has nearly zero risk coming through one of their ASIC service partners. And by the way, that's why we also built very close strategic relationships with MediaTek, with ALCHIP, with GUC. And, you know, that builds the pathway to not only have leading-edge GPU guys adopting this stuff, but also build the go-to-market partnerships such that, you know, places like hyperscaler accelerator products and even, you know, even startups can start to say, well, you know, I can now get access to buying and leveraging the IR Labs optical engine through an ASIC service integration partnership with one of the ASIC service partners that Our Labs works closely with.

37:22So I think of it as 2028, 2029, I think we're going to see a lot of exciting things from the leading-edge GPU guys. And then 29, 2030, 2031, you're going to see it proliferate to the rest of the accelerator community. But you have to realize that by the time they're bringing it in, And it needs to essentially be zero risk and just viewed as a, you know, I've got enough capacity secured and I've got an ASIC service partner that knows how to integrate it. And, you know, I can treat it just like how I integrate HBM. Right. So you've got to get to where these optical engines are thought about like HBM.

38:08Ben Bajarin: That's a good that's a good point. All right. So Jesse, I want to throw this question to you. Mark just mentioned you guys have a lot of strategic investors. You also have, obviously, you've taken more rounds. So you and I were talking about this, just the way you guys have thought about being strategic about your cap table, which obviously matters for existing inventors. But maybe just talk a little bit about that and then, you know, the hopeful question, like, any thoughts on IPO? Yep, sure, Ben, thanks. So a couple things there. I think, you know, early stage deep tech startup company, you know, like Jay and I, I think we spoke about this at dinner.

38:50Silicon Valley VCs, for example, have not been active in this space for many years. And so early on, you typically try to convert early stage deep tech funds who get the business model and are willing to take that risk on a pre-revenue asset. And simultaneously, to Mark's point, building those ecosystem development partners that are strategic to commercial paths over time and figuring out how you get to that volume scale production story. It takes many, many years of deep integration there. And so, you know, you get the finance up front, you start developing, you know, new technologies and iterating on those.

39:25And so I think, you know, we were shipping a two terabits chiplet, four terabits chiplet, eventually an eight to many customers. But getting into their, like, you know, R &D labs and obviously seeing kind of an end-to-end solution that works and developing off of that, built really, you know, kind of great ecosystem partners and understanding, cycle of learning for us, but ability for us to also show proof points to institutional investors, which is really going to fund and underwrite kind of the path forward. And so, you know, in our series D, to Mark's point, I think, you know, it's very unique in that you had a strategic participation from, you know, NVIDIA, AMD, Intel, and TSMC, but we also brought in, you know, two really high quality institutional investors that led the round in that round.

40:04And then from there, you know, set a path with a valuation that we thought was very sensible. You know, we didn't want to have to grow into a very high valuation. We had, you know, what's called execution risk ahead of us. And so we wanted to be kind of mindful of building kind of the right cap table ahead of, you know, the challenging path ahead. And then fast forward, you know, 12 months, you have a CPO switch roadmap kind of laid out by Jensen at GTC. The cell side analysts start writing actively on CPO for the first time after, you know, us trying to get them writing about it maybe a year earlier.

40:38But when they start seeing this getting deployed or at least talked about into even scale out roadmaps from the tier ones, you start seeing a lot of activity and questions being asked as to this technology, what the path ahead looks like and how we might be able to scale. And so we were able to obviously get in front of, I'd say, high-quality investors and talk about the next 12 months and convert an exciting round in our Series E earlier this year, which was led by a big long-only fund, Neuberger-Berman. We also had participation from Sovereign Wealth Funds, QIA in the Middle East, GIC in Singapore, and a number of crossover investors that, again, are high-quality folks that could be valuable for us on the IPO path.

41:19So folks like Sequoia, Inside Partners, ARC, 1789, all kind of high-quality names there alongside, to Mark's point earlier, the GPU ecosystem, the NVIDIAs and AMDs, and MediaTek and Alchip. And so to Mark's point again earlier on the hyperscaler front, MediaTek winning some of those projects out in the TSMC process and flow for the TPUs on the inference side, inference chip side, and then Alchip obviously with AWS. And so us being integrated into that, you know, that back-end channel and being able to deliver CPO at scale to multiple players, that's kind of set us up, I think, well with the investor base to say, you know, this is where we are, this is our goal.

42:00And to your question, Ben, on kind of the path forward, of course, we want to be strategic about thinking about who's going to be there for us at the IPO. And so, you know, we're in front of, I think, the right investors now. We put out that press release last week that Mark mentioned on the WeWin strategic investment on the primary side, which topped our capital on balance sheet, added this year-to-date$650 million. And there was also some press that came out with Reuters about a secondary transaction as well, which allowed us to also bring in what we would call more high-quality institutional investors that could be valuable for us on the path forward.

42:40The only thing I'd add to that is, you know, if you do the spreadsheet math, the market opportunity and the volume from a number of units and just even some, you know, I won't give you our exact ASP targets, but, you know, you do some reasonable ASP targeting. It's you see that that optics going into this scale of opportunity is, you know, tens of billion, probably approaching a hundred billion dollar market size. And so it's not hard to motivate the kind of spreadsheet financial math. And we view our job and also where we're uniquely at is to get the end-to-end supply chain delivering to that volume.

43:22So that's where we're spending most of our time. That's where I spend all my time. And, you know, the opportunity size is incredible. the intercept timeline being in the kind of 28, 29 type of timeline, you know, we think creates a very exciting path for how we would think about an IPO. Yeah.

43:43Jay Goldberg: So I've worked on 30, I don't know, a few dozen IPOs over the years. And I see this phenomenon a lot where very technical longtime founders, CEOs encounter the street for the first time. And you've touched this a couple of bits here. And the questions that they ask aren't necessarily the questions that are top of mind for you. So what's a question you wish investors ask you but never do? You know, it's interesting, Jay. I think you're right. And one thing I've been doing, I have to give credit to Jesse for helping do this. As much as my time is busy and I never get to eat any food at these nice investor dinners that they bring me to.

44:28because I'm talking the whole time. But over the last couple of years, I have... It's interesting because the Silicon Valley world kind of abandoned this space of investment for a long time. So strangely enough, it was actually the Wall Street side who were growth investors, public investors, crossover investors. If you look at our Series E, it was a majority of that capital other than the strategic partnerships with MediaTek, NVIDIA, AMD. but the financial capital came from public investors. And so I've had to kind of, you know, cut my teeth on just building my skill set on, on, you know, seeing how they think, what they're really measuring, how they're thinking about the business.

45:12And I'll say this, you know, we had to find enough of those investors in the series D and series E that were willing to dive deep enough on the technology. However, you know, if we execute, my IPO roadshow should be much simpler because we're going to be shifting from having to explain lots of details on our technology and go to market. You need to explain enough of that to prove to people that you have a robust, defensible core that's not easy to just catch up to quickly. But also, we'll be speaking the language of financial metrics in terms of high visibility into revenue ramp. And the nice thing is it kind of makes my job a little bit easier.

45:53Well, there's hard things and there's nice things about this, But, you know, the relationship is so close with your leading customers and you have such high visibility and supply commitments into what the ramp really looks like. You know, from that perspective, it's got a nice attribute of saying, you know, it makes it a bit easier to predict once you're actually in that manufacturing ramp, what things are going to look like from a financial metric point of view. But, you know, if we do our job well at IR Labs, you know, by the time we're out on an IPO roadshow, we'll talk about technology enough to get people to realize, you know, this is not just some easy thing other people can rotate into.

46:38But we'll also be speaking the language of, you know, how do you actually value this in the public markets? What do the financial metrics look like? You know, what's the longer-term strategy? um um and but to go back to your question of what do i wish they ask look at this dinner that jesse took me to earlier this week with with jeffrey's um for the first time in the u.s in a dinner like that i'm getting asked about uh you know test equipment vendors that that i wasn't even aware you know wall street was was tracking uh you know probers insertion one insertion two insertion three and people arguing with me a little bit on that.

47:15But it's been interesting. Over the last six to nine months, I've seen the public investor guys really get into the weeds in the world that I live in. And it makes the dinners more fun for me because they're actually trying to dig in and figure things out, which is a more interesting conversation for me, even though I don't get to eat any of the lobsters that are in front of me and stuff. I would also add to Mark's comments on that, that, you know, some of these investors that we've built relationships with, they were very, very transparent on kind of their ability to not or their inability to kind of fund today, but also that, you know, they're in it for the long game and they want to hear what we're saying, what we're doing, what we are able to execute on and like kind of how that aligns with what we're saying, obviously, over time.

48:04And that builds the trust with the company. It builds a relationship with obviously the manager team and kind of our timelines and our commitments and so that they are at the IPO to support us. And so it's not just about kind of converting, you know, something right now. It's about building a relationship that is understood over time and is well aligned to kind of that IPO timeframe.

48:26Ben Bajarin: Perfect. Yeah. No, this is fascinating, you know, guys. It's interesting times because, you know again the the kind of high level view i keep coming back to which i think just so interesting for for the timing that you guys thought about in terms of deployments here is you know the race is really to build compute domains right so that you've got these giant clusters that function as one computer network together and i just like we're so far away from that really happening which boggles my mind because the models necessitate that like to serve a giant model at scale, you need a big compute domain.

49:05Ben Bajarin: And everybody's making trade-offs right now because we don't have them. And so I say that to say just again, how early we are, like how early we are in this process is still like the observation I keep coming back to. But you're right, it's a big opportunity. We try to size this market as is others. I think most have the space that you guys play in between 30 and$50 billion by 2030, which is a great number, right? You're coming from a small base, right? This is going to scale quickly. So super, super interesting times. We really appreciate the conversation here. I know you guys are going to have more to show at OCP, so hope everybody checks that out.

49:44Ben Bajarin: And we're happy to keep following along your journey, and we really appreciate your time coming on the circuit. Yeah, thanks. This was a great discussion. By the way, we didn't even talk about lasers, so that could be a future episode, because that's the other big part of this story. We will talk about lasers at a future time. For sure. We'll do a whole segment on lasers. Okay, good. I'll look forward to it. Thank you, guys. Yeah. Well, thanks for listening, everybody. We appreciate your time, and we'll catch you next time on The Circuit.

50:13Jay Goldberg: Thanks for listening, everybody.

50:31Thank you.