Kopin Corporation (KOPN) Earnings Call Transcript & Summary

October 20, 2026

NASDAQ US Information Technology Semiconductors and Semiconductor Equipment special 38 min

Earnings Call Speaker Segments

Cortney Harding

attendee
#1

All right. Hi, everyone. Thank you so much for joining us today. We are here to discuss the coming Trillion Dollar AI Interconnect Opportunity. So my name is Cortney Harding. I'm the Founder and CEO of Friends With Holograms. We are a spatial computing and artificial intelligence agency. I'm joined today by 2 incredible luminaries, and I will let them go ahead and introduce themselves.

Michael Murray

executive
#2

Sure. Thank you, Cortney. I'm Michael Murray, Chief Executive Officer, Chairman and Director of Kopin Corporation. Pleased to be here today. James?

James Altucher

attendee
#3

I'm James Altucher, Strategy Adviser at Fabric.AI. Also, I write financial newsletters about this intersection between AI and finance. Glad to be here.

Cortney Harding

attendee
#4

Well, thank you both so much for coming. So if you have questions for us, you can go ahead and drop them in the YouTube chat stream. We have somebody monitoring that or you can send an e-mail to events@kopin.com. So please feel free to ask questions, and we're going to keep this conversation going.

Cortney Harding

attendee
#5

So let's dive in. So for both of you, how confident are you that Neural I/O is going to become a standard building block of next-generation AI data center infrastructure rather than one computing option among many. Michael, do you want to kick off?

Michael Murray

executive
#6

Yes, sure. I'll kick off. So firstly, light is free, number one. Two, it's much lower power than copper. But thirdly, there's issues that we have to traverse in terms of the technical viability of microLEDs. This has been discussed at length in many, many papers, including ours that we just announced yesterday. I think from a confidence level, as we sit here today, I'm 100% confident that microLEDs will have their day in AI data centers as well micro-VCSEL as an example. So the technology is adapting to the growth demand that we see in AI. But bigger than that, let's just put AI away for a second. Just think about computing. When you think about copper traces that I used to use in electronic school several decades ago, they haven't changed folks. Now when we think about quantum computing, AI data centers, they're just electronics themselves, we have to be able to interconnect and talk data much quicker, higher bandwidth and lower power than we currently are today. It is the problem with AI. The interconnects are the problem. I'm very confident that not only will we be able to use microLEDs and micro-VCSELs in co-packaged optics and near-packaged optics in the very near future. When I say in very near future, we're talking in the next 18 months.

Cortney Harding

attendee
#7

James?

James Altucher

attendee
#8

Yes. Just at a 50,000-foot level, people often wonder, oh, is this AI -- and I remember this back in 1999, is this AI thing a fad? Is it -- everyone is getting all hyped up, but then they're going to move on to the next thing. Let me just tell you, I sort of think the world is divided into 2 types of people. So there is the category of people who have used AI to the point where they say, "Oh, I can compress 3 months' worth of my work down to 10 minutes. " And then there's the second category of people who will eventually be in the first category. So that's it. So AI is just -- and this category here is only 1% of AI users. So the demand to build data centers, it's not going to be $1 trillion. It's not going to be $5 trillion. We see $10 trillion of infrastructure build-out happening between now and let's say, 2031, 2032. 10% of that is going to be spent on these thing called interconnects, which connect the GPUs. Every GPU is going to have 5 or so interconnects hooked up to it. So if you have 1 million GPUs or 10 million GPUs, you're going to have up to 500 million; if you have 100 million GPUs, which is what the estimate is, you're going to have 500 million interconnects being built. Right now, they all use copper, but the laws of physics say copper can't get any faster. They're at their limit. Then you're going to have laser, but power is a big issue. Lasers, if you watch any science-fiction movie, it seems like lasers take a lot of power. And like Mike was saying, microLED is just -- it's like 1,000 flashlights is what a microLED is. And that doesn't require -- it requires 1/20 the power. And by the way, it's just as fast or faster than laser. So this -- everyone knows this is the future. And then it's just a matter of how do you get the right technology in that interconnect shape. So it's going to be microLEDs. Everyone knows that. And then we feel we have the right IP, the right technology and have a lot of evidence backing this up. Kopin has been doing this for a long time on the manufacturing side. So we're very confident that this is going to scale and that the AI world requires this to grow.

Cortney Harding

attendee
#9

Yes. I mean that was incredible. Michael, do you have anything to add about how microLED will have a genuine advantage over copper and conventional laser optics? I mean James spelled out that was it.

Michael Murray

executive
#10

Yes. I think -- so let's just do some simple math. Currently, in the chip that we have a microLED, it has over 16 million individual LEDs or 16 million individual lights. If you think about that going slow and wide, you're able to transmit a tremendous amount of data having 16 million lights as an example. So you don't need a fast LED. You need a slower LED, but you have lots of lanes, lots of lanes of traffic. So Kopin builds that today. Moreover, we're now building them in color. But most importantly, we're building them here in the United States. And that's critical, and I'll talk about why that's critical in a moment. But from a microLED standpoint, you're able to have this 1.6 to 3.2 terabit transmit and receive capability in a chip that is a 1 millimeter square type of die. That is absolutely revolutionary for an interconnect as an example. And this microLED technology has been worked on in this specific instance in this specific application for many, many years by many companies actually. The problem that they've faced, a problem that we've already solved is the culmination of those little lights as an example, because light doesn't travel in a direct line. It actually starts to [ buoy out ] just like a flashlight does. You want to be able to control that light. Kopin has been able to do this for many years because we build microdisplays. What makes us specifically different than a lot of the other companies that are trying to do what we do is they're optics companies. They're used to one fiber optic going one way, the other fiber optic coming the other way, and those 2 shall meet in the middle. What Kopin does better than anybody else is actually build these displays with 16 million LEDs and something the size of my thumbnail. We already have figured out the light culmination and the light budget and link budgets of this technology, and that's where other companies have stumbled for the last several years.

Cortney Harding

attendee
#11

And talk to me a little bit about what you mentioned a second ago with manufacturing in the United States.

Michael Murray

executive
#12

Yes.

Cortney Harding

attendee
#13

I think that's also really critical. There's a really interesting article in the New York that came out yesterday about semiconductor manufacturing and all the issues between Taiwan and the U.S. and other places. So like spin out a little bit about how it's a benefit to manufacture in the U.S.

Michael Murray

executive
#14

Yes. So I think one of the really key milestones for investors is Kopin just received another IBAS award. So I'm going to pause on that for a second and double-click on it. So an IBAS award is basically an Industrial Base Analysis and Sustainment Act award that the U.S. government will award companies that are bringing technology out of China or out of Asia and putting it here in North America for critical technology acquisition areas, of which microLEDs are. So Kopin just received another one of these awards. The reason being is we have to have that supply chain here in the Continental United States for things like HUDs going in aircraft, things like Soldier Borne Mission Command, which is a day and night vision technology for the U.S. Army as well as other applications like thermal weapon sights and what have you. But this is really important. United States came out with an [ EDEP ] through the NDAA and FTC organizations that have basically said that no copper interconnects or optical interconnects can be made in China that touch U.S. data centers or U.S. GPUs. Therefore, now Kopin just became very viable and important to not only the United States through the Department of War, but also integration systems like DISA, 4-letter agencies, 3-letter agencies who are, by the way, the second largest consumer of AI data centers.

Cortney Harding

attendee
#15

What about the talent pipeline in the U.S., though? Because I know that's been an issue for other like semiconductor and other skilled manufacturing work here. Are you finding issues with that? Or do you think there's enough of a talent pool in the U.S.?

Michael Murray

executive
#16

Definitely, there's enough talent in the United States. But we have to go to the talent. So we actually just opened up a design center in Dallas, Texas for that reason. There's a tremendous amount of wealth of talent in the Dallas area in optics. And we're working with our partners, Fabric.AI that are based here in New York and Kopin is headquartered in -- just outside of Boston in Massachusetts. So we're pulling that talent in the locations that we find them. But there is a gap, and we're using AI to develop some of this technology, and we're using the design resources of our partners as well.

James Altucher

attendee
#17

And let me just add to that. This is why we're so -- at Fabric, we're so excited to be working with the manufacturing at Kopin. As Michael said, you need this manufacturing and this part of the supply chain in the U.S. It's not like AI data centers are holding kind of the collective intelligence of all American IP right now. We're not outsourcing that to China. That's not happening ever. And so such -- their display technology, their microLEDs are in the helmets of fighter pilots, soldiers, all over the world, American soldiers and American use cases. So -- but it's very important with this rise in AI, with the fact that it's very important to realize GPUs, everyone's focused on GPUs, NVIDIA, AMD and so on. Yes, GPUs have increased in speed and bandwidth, maybe 20x over the past few years. But the actual connection between the GPUs, which is where almost all of the data and calculation is happening is how these GPUs communicate. That's barely improved at all. Because again, the laws of physics clamp down on copper. So Credo, Astera Labs, they build all the interconnects now. They're using copper-based solutions. So that's why NVIDIA wants to use laser-based or microLED based. We're at the leading edge of this microLED revolution that has to happen. The communication between these GPUs has to increase during this $10 trillion build-out. And again, 10% of that build-out is going to go towards these interconnects. That's really where the focus is going to be. And that $10 trillion is going to be spent on U.S. manufacturing, which is why we're so excited to work with Kopin. In terms of the talent pool, again, there's a head start. There's the AI revolution has been happening since whatever, the launch of ChatGPT past 4, 5, 6 years. But manufacturing microLEDs is not brand new. This has been happening for a long time. So again, we're standing on the shoulders of giants. We're building this for IP. We're sharing IP on the whole AI universe. So very excited at this point.

Michael Murray

executive
#18

And actually, I'll add one thing to that. A lot of investors ask me, well, why now, right? That's one of the biggest questions I hear all the time is why interconnects now? And I gave that a lot of thought over the course of the last few months that we've been talking to some of the biggest companies in the world about this technology issue. And one of the things that I realized is that Moore's Law is completely dead, right? I mean Jensen Huang just -- it was a speed bump for NVIDIA. So -- but what happened was, I think a lot of the interconnect companies looked at it and said, well, we don't need to be as fast as a terabit or terabit 6 or 3.2 terabits because Moore's Law is dead. We're not going to get that speed, and we're not going to see this AI trajectory that we're now seeing. Well, they were all wrong, right? So Jensen -- that was a speed bump for Jensen and AMD and Intel. They now have technologies that are well past Moore's Law. The problem that, that created, though, is they're driving a Ferrari. The interconnect is a bicycle and then they connect to another Ferrari. The problem is the middle bandwidth. And it was about the investment and the belief that Moore's Law was going to gate us from that technology development, and it did in copper. And that's the thing is it actually did create copper development because it can't achieve the rates that certainly light-based optics can do.

James Altucher

attendee
#19

Yes. And just to add to that, Moore's Law always talked about basically the speed and potential of one CPU, one chip. But compute now, you have to think about it in terms of what's the speed of your data center. And the data center is not one CPU. It's lots of CPUs managing lots more GPUs, managing lots more interconnects with thousands of wires in between all of them. It's a gigantic mess. People are untangling that mess. This is how you untangle that mess, and that's the future that we're building for.

Cortney Harding

attendee
#20

Yes. I mean the issue here is really the timing of everything, right? Like I mean, ChatGPT was really only a few years ago, right? And everything has grown so exponentially. And so I think right now, we're just in this really interesting period where we're kind of racing against time and there's always time constraints. And like volume production is expected around 2028 rather than earlier. So is there a risk that competing interconnect technologies mature first and make Neural I/O [indiscernible] ramps up or like how are you thinking about this in terms of kind of this race against the clock that we all seem to be in?

Michael Murray

executive
#21

Sure. No, it is a race. And one of the things that I've been a big proponent of, and I think we're doing it the right way, is we're walking before we run. I would love to go straight to a monolithic IC today. However, the way that we're going about our development is really thinking about the technical issues that we've seen other companies fail at, culmination. We've seen them fail at that. We've seen them fail at being able to place LEDs in a continuous manner that isn't random, the ability to connect and have a link budget to those LEDs. We looked at what was being done in the marketplace and took our places to say, these are the issues and connectivity issues and technical issues that we need to solve. We're doing that with our friends at Fabric.AI. We're going to show that as a demonstration at CES in 2027. So the demonstration itself will focus on what are the technical issues that we're going to solve to get to a monolithic IC design. Once we move to an IC design, those in the community that are electrical engineers and optical engineers will be able to see the lines of trajectory to get to 1.6 terabits to 3.2 terabits per second, and that will be software definable, by the way, in our chipset called Neural I/O with our friends over at Fabric.AI. So...

James Altucher

attendee
#22

Yes. And let me just add that you say volume production is going to start and start to boom in 2028. Right now, we're in 2026. Guess what? 2028 was yesterday. Like that's already -- that -- 2028, that chip has sailed. So we're talking -- there's, let's say, something like 100 companies involved in the ecosystem of building a data center. We're talking to many of them. We're seeing what they're talking about, where their gaps are, where their hurdles are. And they're designing data centers now for 2030. And the standards are going to change, the speeds are going to change, the chips are going to change, but they know this. They're very -- trillions of dollars are being spent. So they're not messing around. And so we are building for those times. Like we're not -- we're kind of in the discussions right where we need to be with this technology. If we had started talking to people a year ago, 3.2 terabits a second wouldn't have made sense to anybody. Now it's -- now we'll be right at the right point.

Michael Murray

executive
#23

Yes. And I think to answer the question a little bit more crisply, it's a race. Absolutely. But I think we're all confident now that microLEDs will have a lane in near packaged optics as well as co-packaged optics. So for us, a year ago, sitting here, we had a lot of design questions. We had a lot of concerns about the manufacturability of this technology, so did the market. Now we're all convinced. Jensen Huang put $6 billion to work in light-based optics for interconnects for a reason. He knows they know that they have to move in this way. So yes, there are other competing technologies like micro-VCSELs as an example, but you're not going to put a micro-VCSEL inside of a GPU. It's just not going to work. And if it does work, hey, hands up, great feet of engineering, but I don't think so. So I think certainly starting at the CPO level, microLEDs are going to have a home. I think Kopin and Fabric.AI, building this IP together, Kopin building this technology in the United States for likely the U.S. government and the U.S. government data centers will have a very key role in what the future looks like in AI data centers for sure.

Cortney Harding

attendee
#24

Yes.

James Altucher

attendee
#25

I also just want to say that all of these -- there's a lot of different layers of this stack. So like Google, Amazon, Microsoft, let's say, they're the ones building these enormous data centers for the OpenAIs and Anthropics of the world. But you really have to look at what are the picks and shovels of the picks and shovels of the picks and shovels. So when Google says, okay, build me a data center; NVIDIA says, okay, well, we have the GPUs. But then, okay, NVIDIA wants to know, well, who's going to connect our GPUs together. And maybe Marvell or Broadcom or someone raised their hand and say, we could do it, and we have a solution. And then NVIDIA says, "I love your solution, but what's your backup solution. Oh, I love your backup solution. What's the backup of that backup solution?" There's a lot -- this is a big category. And again, nobody is messing around. Like nobody is going to -- nobody -- you have to have 99.99% uptime and you have to have -- everything has to be serviceable in the field. Like this is not a theoretical thing. This has to work for a multitrillion dollar spend. So that is years for many years. So that's what we're building towards. And it's complicated, but that's the hurdles we're currently getting IP on right now. And look, it's going well.

Cortney Harding

attendee
#26

Yes. I mean that's why this is, again, the $1 trillion opportunity is digging these levels deeper beyond what everyone is talking about these days. So just to sort of continue the topic of Neural I/O, do you see this being licensed to or adopted by competitors? Or are you going to keep it proprietary and on differentiation? Like how is the -- what's the thinking on that?

Michael Murray

executive
#27

Yes. So I'll start and then have James kick in from Fabric.AI perspective. So from a Kopin perspective, we look and feel very much like a foundry. And we're developing the capability of Neural I/O along with Fabric.AI and other customers as well, including the U.S. government because it's a critical technology area for them. But again, we've already built microLEDs that are flying in planes that are serviceable for over 10 years of service. I always ask a lot of our potential customers, what are you going to do with a micro-VCSEL when you have to tune it or replace it. MicroLEDs, we have a different way of potentially using software to create uptimes and guard against any sort of pixel out. So that sort of question, I think there's a broader market for us. And one of the things you'll see in our investor deck is we now have a vertical called Foundry Services. And that's building microLEDs for other companies that want to use them in other ways, whether it be quantum computing as an example, or we're working with other semiconductor companies that want to use this technology for co-packaged optics. But from a Neural I/O perspective, do you want to talk about that?

James Altucher

attendee
#28

Yes. Let's say, Neural I/O or any interconnect. And again, interconnect to remind people, connects 2 GPUs, 2 CPUs, GPUs to memory, 2 TPUs, whatever.

Michael Murray

executive
#29

Rack to rack.

James Altucher

attendee
#30

Yes, rack to rack, the whole thing. So there's -- in any interconnect, there's maybe 9 different components. You have the transmitter, the receiver, maybe it's fiber in the middle. There's lenses to couple to the fiber. There's lots of different pieces of an interconnect. At Neural I/O, at FABC, Kopin, we're going to have IP, and we are already having IP at every level of every point within this 9 component interconnect. And that gives us a lot of revenue opportunities, whether it's selling the IP, partnering with suppliers, partnering with manufacturers, partnering with the hyperscalers. So all of this is happening as we're building.

Michael Murray

executive
#31

And I think one of the -- exactly right. So I think one of the interesting things that we're seeing now from NVIDIA is NVLink is moving forward with a new standard. And that's going to be something that we're going to enable. That IP can be developed by Fabric or by Kopin or both. And we'll offer that IP as a block to any customer that wants to develop their own transceiver. And Fabric.AI will get paid a royalty for that. So very much so Kopin is focused on the government side of things because we're based in the United States, and we have some areas of entry point with the U.S. government today. And then Fabric is going to move forward with developing the IP for more of the hyperscaler market. But that IP is transferable, and it's a way that I think Fabric is going to enjoy a very large revenue stream with not much overhead. They don't have to build a $100 million fab like I do. So it's a great business model. If you remember Arm back in the day, that was the Arm model. They licensed their IP technology to other companies that use it as competitors as well as vendors.

Cortney Harding

attendee
#32

Nice. So I mean, you mentioned a few customers, right, the government and some of these hyperscalers. Can you talk about other types of customers that you're engaging with right now?

Michael Murray

executive
#33

Yes. So there's the picks and shovels customers as an example. I was very shocked by our first white paper download list. There were companies that downloaded. I can't mention them here today, but that I never heard of before that are many, many billions of dollars, but they are the companies putting the racks together. There are companies putting wires together and the copper wires and fiber optic wires together that are all very interested in working with Fabric.AI and Kopin to actually make those interconnections, the physical interconnections as well as the optical interconnections between the chips and the boards and racks. So we've learned more about this market, I'd say, in the last 3 months that I think we learned in the last year. So very surprising to see the number of potential customers with new applications. We're actually seeing demand from things like vehicles. There are some applications in vehicles that might need some of this technology. Certainly, the Department of War and Department of the Interior is very interested. And then certainly, we're also seeing the quantum computer space, be very interested in this technology as well.

James Altucher

attendee
#34

Yes, absolutely. And let's say, between SpaceX, Google, Amazon and all the hyperscalers, they're going to -- I mean, even this next year, it's probably going to be close to $1 trillion that we'll spend. But again, over the next couple of years, $10 trillion. That's like rain that's going to pour down on, again, first, the NVIDIA, then the Marvells and the Broadcoms, then the Celesticas and the Fabrinets. It's almost like I wish I can make an ETF of everyone who's downloaded our white papers in the past few weeks. So Michael and I, we've written 2 white papers now. One was just released yesterday. That's sort of a road map to the future of connectivity in the data center and even AI. And so if you could -- basically, that entire supply chain is a customer at different levels. So -- and we're seeing it more and more who are the components of every single part, tiniest, tiniest components of this giant ecosystem.

Cortney Harding

attendee
#35

Yes. So I mean you mentioned data centers, and it seems like a good time to bring up the fact that we're heading into the elections in the U.S. in a couple of weeks, the midterms are coming. Data center construction has been a huge issue on both sides of the aisle, and there's a lot of people that are concerned about it, upset about it and believe that AI data centers are going to have a negative impact on the environment. So when you look at what you're doing in your solutions, talk me through how much water and power could actually be saved, right, as we're kind of coming up with these issues, like what does this look like in terms of adopting your technology to make this better?

Michael Murray

executive
#36

Sure. I'll kick it up a level just because it's fun to talk about. But look, I'm a certain age where I remember when the Internet was going to end us all, right? I remember all the Internet is going to make it. Y2K, the world is going to fall apart, AI is not going to end humanity because it doesn't have one of these -- it doesn't have opposable thumbs, right? So it's going to be useful for a number of years. That's one. Two, the amount of people that still don't have the Internet is quite a lot, right? Therefore, the amount of people that don't have access to AI is even larger than that. So we're actually at the beginning of AI data centers, not in the middle, not at the beginning or at the end, we're at the start of AI data centers. So it is a topical issue for the United States government and certainly local municipalities. So from a standpoint of electricity savings, we did some math early on that says that we can save around 27,000 watts of power per rack, and there's 1,000 racks per kind of, I'd say, mid-level data center. So if you do all that math, that's about $25 million per year of energy savings, just in electricity, switching from copper interconnects to microLED interconnects. Now there's lots of data that supports that. So from a cooling perspective, you don't see these copper lines lighting up like a toaster oven. You'll see light, right, and waveguides or fiber optic cables in the middle. And that's going to save tremendous amounts of power and therefore, tremendous amounts of water. Would you agree?

James Altucher

attendee
#37

Absolutely. So again, just to highlight what Michael just said, microLED-based interconnects are about 1/10 to 1/20 the power usage of what's in the data center today. So all the arguments about powering the [indiscernible], we're trying to solve that problem. That's one of the main problems we're trying to solve. The other thing about -- I get it, people are concerned about AI, and there's a lot of headlines. But I'll just tell a quick story. A friend of mine the other day had some issue with his -- her thyroid. And she put the -- her doctors are saying one thing, she put the scans into ChatGPT, just in AI. And he said, No, no, no, you need to ask your doctor about this, this, this, this. He did -- he responded to those questions. She ended up having surgery, removing the thyroid, curing the cancer. So AI, again, is just at the beginning. It's -- we could talk all we want about, oh my gosh, terminator robots killing humanity. That's not going to happen. It's saving lives. It's helping people. It's increasing efficiency, and we're trying to make it as power efficient as possible.

Michael Murray

executive
#38

And remember what I said earlier, we have a Ferrari, bicycle, Ferrari issue, right? You remove the bicycle in the middle. Now your bandwidth is 1:1. And I think that's what NVIDIA, AMD and all the big GPU, GPU players are wanting to see is that bandwidth delay and issue in the middle needs to go away. Therefore, you get better bandwidth efficiency, lower power and you're transmitting light. Light is free. Copper is not free. It's very expensive. Moreover, I think from a standpoint of where we're going, 3.2 terabits is just a bump in the road friends. I think the market has some sort of idea like there's going to be 3.2 terabits and that's it. Our road map goes well beyond that, probably 3 to 5x that speed in bandwidth.

Cortney Harding

attendee
#39

Yes. I mean I think the other thing that's not discussed as much is the conditions under which copper is currently mined, right? And how terrible that is both for the environment, for the people involved in the process. I mean a very small handful of people get very wealthy and everyone else is in these terrible conditions, whereas light is just Light.

Michael Murray

executive
#40

It's generated by silicon.

Cortney Harding

attendee
#41

Generated, yes. And so like talk me through a little bit of that and sort of are you making that point when you talk to other people about this?

Michael Murray

executive
#42

Yes, it's a great point. We haven't really. We've been very much focused on the semiconductor development, those 5 technical key problems that we're going to show at CES to say, these are the 5 technology issues that we see with microLED adoption. This is how we've solved those and how we're going to solve them in the future. However, having said that, I've been very active on Capitol Hill, working with the Senate, working with congressional folks on this problem. And part of the education that needs to go with the policy as well as the technology development is education. And we're writing a lot of white papers for that reason. A lot of people think we're just writing white papers for the thrill of it and the downloads of it. Yes. And it's education to solve some of these issues so that the congressional people and folks in the Senate, when they vote, they can vote with confidence that there is a solution in the future that not only will allow data centers to thrive and AI to thrive, but it will also be safe and safe from intrusion. One of the things I talk a lot about is when you have an interconnect in a data flow, that interconnect can educate up and educate down. That's a bad place to be if you're an adversary or a good place to be if you're an adversary to get into a network. So I think encryption and interconnects is also a topical area that folks just haven't been talking about. And now with the NDAA and FTC coming out and saying, no more Chinese silicon in those areas, they're starting to figure out where these chips can call home.

Cortney Harding

attendee
#43

Got it. I mean what else does the AI industry misunderstand most about the cost of moving data? I mean we've sort of circled around this question, but are there any things that we see it's like a fundamental misunderstanding of this topic?

James Altucher

attendee
#44

Yes. I think again, I think the entire market has forgotten how important communication between GPUs are. So Jensen Huang, he has an amazing story. Basically, he kind of sketched out what a GPU is on an Appian in a Denny's in 1992. And was he planning that his -- the way he would turn into a $5 trillion company was AI? No. He thought it was just only this will be good for games. Later on, it was good for crypto mining. Guess what? All AI needs matrix multiplication, which require GPUs, not CPUs, but GPUs. But guess what, you need millions of GPUs to actually do this at the level that society demands. And you need to do it efficiently, you need to do it fast. And that efficiency and speed have not arrived yet. But that's where, okay, we're going to go from copper to lasers to microLEDs. That's where you get the speed. Michael mentioned potentially 16 million lanes instead of 1 lane per LED on the microLED or you need efficiency where, again, we're going to be 10% to 5% of the energy cost. So this is a huge area that people have kind of forgotten about because almost there's like a glamor around the GPU. The GPU needs a team. It's just the coach and the interconnects is everybody on that team now.

Cortney Harding

attendee
#45

I will say this about the Jensen Huang Denny's. It's still open. It's in the suburbs of Portland and how young kids are going there because they're very excited about it. They're very excited about the technology, and they want to see where Jensen got his ideas, and I presume you eat some pancakes.

Michael Murray

executive
#46

If I worked at that Denny's.

Cortney Harding

attendee
#47

He did work at the Denny's. He has...

Michael Murray

executive
#48

$5 trillion grand slam.

Cortney Harding

attendee
#49

$5 trillion of Denny's. His 2 jobs are Denny's and NVIDIA. Best ones...

James Altucher

attendee
#50

Good career.

Cortney Harding

attendee
#51

Yes. So why is microLED potentially suited to solve all these problems rather than just being sort of another optical interconnect technology? Like what sets it apart?

Michael Murray

executive
#52

Yes. I think the price per watt, one; two, the ability to put thousands of microLEDs in a 1-millimeter square package makes it very unique over micro-VCSEL, as an example, or lasers and certainly over copper. That's why it's so unique. At Kopin, like I said earlier, we have a chip right now that has 16 million LEDs, and it's about the size of my thumbnail. So if you do the math around that, how many LEDs can you put in a 1-millimeter squared? Well, we're doing that math. And we're not ready to tell the world about how many we can put there and our switching speed just yet. But we're -- in our paper that we released yesterday, we're talking about 3.2 terabits now, today, now. And our road map is well beyond 10 terabits after that. So I think there's a tremendous amount of capability in microLEDs, but there are 5 technical issues that we have to traverse. And I'm happy to say, as we sit here today, all 5 of them, we have figured out. And the biggest one being culmination of light and the ability for the interconnect to work with fiber or things like waveguides or over the air. So we're going to demonstrate that with our friends at Fabric.AI in January. There'll be iterations very quickly thereafter until we get to a monolithic design because we see the market and we see the government, we see the Department of War wanting this technology now. So we're very confident where we're at.

James Altucher

attendee
#53

And I think you have to look at not Oh, how the financing is happening with NVIDIA doing -- or is AI going to kill humanity, all these headlines, you kind of have to move past that and just realize, hey, this is beneficial to society. So let's solve these issues about power efficiency and the environment and all these guardrails you have to put on AI. We're going to solve these issues. But you have -- I would encourage people to download the white paper that Michael and I just wrote, you can get it at the Kopin website, for instance, or the Fabric website. And we discuss what happens when you have a microLED-based data center, whether that's 5 years, 10 years in the future, who knows, but we're going to be there every step of the way. And it's a very different future than what people can possibly imagine. And we outline what that looks like.

Michael Murray

executive
#54

Yes. And I think exactly right. And one of the things that I think James and I come to the market with is somewhat of a grown-up mentality of it's an and, it's not an or, right? Copper will still be used. We're not saying that. Fiberoptics and micro-VCSELs will still be used, but so will microLEDs. And when you think about Kopin and Fabric.AI as companies and as investors, that is a $1 trillion market that Kopin and Fabric.AI are clearly marked in. And so I think we're punching well above our weight. I think we have great technology that's proven that is proven by the U.S. government and Department of War, and we're building our own production line here in the United States to build this technology. It's an and, not an or.

Cortney Harding

attendee
#55

Yes. Exactly. Well, we could keep going all day. I'm sure we've just scratched the surface, but unfortunately, we do have to wrap it up. So I'll throw out a final question. So what milestones should shareholders follow over the next several deployment stages? And that's, again, without relying on specific product revenue forecast, but what are kind of the big things that people should be paying attention to?

Michael Murray

executive
#56

Sure. So I'll take that one if you want to add. So one of the big ones just happened. We just received another IBAS award from the U.S. government. So the government is funding Kopin to build more microLEDs in the United States. That same production line that's building color microLEDs for things like thermal weapon sights, night vision goggles, like Soldier Borne Mission Command and HUDs that go into rotary wing and fixed wing aircraft, that same production line will build our microLED Neural I/O chip and capability for NPO, CPO and foundry services. So we're building that fab right now. We're putting in the pieces as we speak. And so that was a big milestone for us. A lot of people, I think, were waiting to see that. Secondly, we have our date in January, where we're going to be showing a number of different devices at CES. So CES isn't only about Neural I/O, although that's the belle of the ball, but we'll have other technologies there as well, like neural display, the world's first bidirectional microdisplay and a few other little tricks that we've got up our sleeves. And then we have a number of roadshows thereafter like Photonics East and West that we'll be showing off the demonstrator. But we'll also see some milestones in the next little while of hopefully, government coming to Kopin to develop this technology in several different ways. We've been working with a number of agencies in the government on this technology, and there's a lot of interest there. So one of the milestones will be what we'll be doing with the U.S. government, how that will translate to Neural I/O and our friends at Fabric.AI. And we're actively negotiating several new customers on the Fabric.AI side as well as the Kopin side that I think are early adopters of this technology. So customer announcements, sample announcements are all going to be in 2027. Anything that I missed?

James Altucher

attendee
#57

Yes. Look at our IP. So as we file IP, you'll see where we are in the development cycle. And look at our partnerships because that also tells you where we are at in the development cycle. And by the way, look at our hires because that also tells you -- but those are 3 factors. I would add, by the way, Michael's factors are very important as well.

Michael Murray

executive
#58

They were great.

Cortney Harding

attendee
#59

All right. Well, thank you so much, everyone, for joining us today. You can download the white paper at the Kopin website or at the Fabric.AI website and it's a fantastic read. Have follow-up questions, you can feel free again to submit them to events@kopin.com and make sure to follow us on -- you should follow everyone on socials. You can follow me on socials@cortneyharding. It's just my name. And if you guys want to share your socials.

Michael Murray

executive
#60

Just follow me on LinkedIn. My socials are a mess. Yours is better.

James Altucher

attendee
#61

@jaltucher.

Cortney Harding

attendee
#62

All right. Thank you so much, everyone.

James Altucher

attendee
#63

Thank you.

Michael Murray

executive
#64

Thank you.

Cortney Harding

attendee
#65

Thanks.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Kopin Corporation transcript — plus 261,000+ transcripts from 13,000+ companies, speaker segments, AI summaries and full-text search — through the EarningsCalls.dev API.

Get the API View API docs →

This call discussed

For developers and AI pipelines

Programmatic access to Kopin Corporation earnings transcripts and 261,000+ others is available through the EarningsCalls.dev REST API. Plans from $24.99/month — full transcripts, speaker segments, full-text search, and the recently-added /api/v1/transcripts/recent polling endpoint for ETL pipelines.

More tools for stock research

OptionsBell: an unusual options activity scanner that tracks options flow across all US stocks and emails you when large trades appear in Kopin Corporation or any ticker on your watchlist.

StockMarketScan: a stock screener with alerts: 24 pre-built screens refreshed every 5 minutes, watchlists and email notifications. Both offer API and MCP access and fair monthly pricing.