The Next Winners of the AI Data Center Boom

with annual revenue growing from 1 billion this year to 24.8 billion next year and then 71 billion by 2028. So this is a massive massive scale. >> What's up everybody? It's LGDE here and welcome to Milk Road AI, the daily AI show that's always looking for bargains on stocks, even if the bargain might be even bigger next week. Today's July 21st, 2026, recording on July 20th. AI stocks topped a few weeks ago. But does that mean that they're completely cooked? Our lead AI analyst Melvin definitely does not think so. And today he's going to explain a very overlooked sector of the trade, photonix, the light that moves the data to the chips and what part of the rack is needed to make that happen. He's also going to share two companies that he likes in the sector and how their revenue is set to take off as the data center buildout winds up for the next leg. If you want to see his moves on these companies and how he's played Neoclouds, Robotics, and also made huge gains on Micron, Nebius, and AMD this year, you got to go Milk Road Pro at the link below. It's just a dollar for a 7-day trial. And a reminder that our podcast today is free and it wouldn't be possible without our partners at Securitize the Regulated Rails for tokenization and BitGet stocks 2.0 with Real Liquidity, Real Dividends. Keep an ear out later in the show for a message from them. All right, Melvin, what do you got for me today? >> Well, thanks for having me back on LG. Um last week you know we talked about Neoclouds and people seems to have really enjoyed that show and I'm back and today I wanted to talk about a sector that is you know bit of a beatd down at the moment and that's fatonix and optical networking um basically how uh data actually moves from these AI data centers um between each other between the GPUs. Um, and I know the second I say fatonics, half of you are going to be like, "Okay, this sounds crazy. I don't know what that is." But stick with me here. Uh, it's actually not that complicated because once you break it down, there's And once you break it down, it's really easy. And there's so much money to be made in this trade still. Um, so yeah, >> I don't even know what phonics is, man. But you said there's money to be made, so I'm all ears. >> Exactly. Perfect. So before I tell you what fatonix is, let me give you some, you know, context first. Um, if you can just put up that first chart, LG. Perfect. Um the big four hyperscalers um Amazon, Google, Meta, Microsoft pushed their 2026 capex plans uh past 700 billion combined. And that's after they all raised uh guidance again following their um quarter earnings, their March quarter earnings. And to put that in context, go back, if you go back to 2018, um these four companies combined were spending something like 30 to 40 billion a year in total. And by 2024, that number was roughly 261 billion. And this year, as I mentioned, it's already past 700 billion. And it's not slowing anytime soon because last week, Morgan Stanley raises 2027 and 2028 capex by 9 and 10%. And so now they're essentially calling for 1.23 trillion in capex in 2027 and 1.4 4 trillion in 2028. So there's a massive amount of capital that will be directed towards the AI infrastructure buildout. Now inside that spend um networking like the wiring um and the switches connecting all these GPUs together is already eating roughly 15 to 18% of total clust uh total cluster capital cost. Um so in a 200,000 GPU cluster today um optical transceivers which are basically the components that convert electric signals into data uh into light sorry um so data can move between GPUs at extreme speeds alone are consuming about 17 megawatts and 435 megawatts total power. Um and that's a real chunk of this you know of the build. So reason why I bring all this up is everybody's been focused on GPUs, the memory trades. We talked about Micron a couple weeks ago. We talked about NeoClouds last week, but there is this this third bottleneck and they had a bit of a runup. Um, but they have gotten absolutely wrecked along with some of the semiconductor trades over the last month or so. And I think that's where the real opportunity is. And we're currently going through a massive massive buildout. Um, and the key question is how do you connect um, tens of thousands of GPUs together so they can act like one giant brain instead of just bunch of chips, you know, sitting next to each other, not talking to each other uh, fast enough. And that's what photonix is. And let me explain that in actually more details using that image. So easiest way to understand photonics is this. Electronics moves data through copper using electricity while photonics moves data through uh glass fiber using light. Um and this image actually explains it really well. The data if you look data starts as an electrical signal gets converted into light by the optical transceiver um and then it travels through the fiber gets converted back into electricity so the GPU on the other side can read it and that's what photonix basically is and the copper like copper has been like the standard way of the way computers move data for decades um because it carries electrical signals through like the metal wires and and it extreme works extremely well uh when the distance is short because it's cheap, easy to manufacture, very reliable. Um and that is why copper is still used inside you know our data centers. Um and uh but copper starts struggling as the cables get longer and GPU and the data speeds rise and this is where light and photonics becomes useful because it can carry more information farther while using less power. But copper is still the best option for short connections while fiber becomes the better choice for higher speed links. And if you actually look over to the next chart and this image shows this perfectly um because um a large AI system can require as we know thousands of copper cables and once those cables stretch across several racks, signal quality drops while heat and power usage increases and this becomes harder to manage as AI clusters grow from one rack to system containing hundreds and thousands of GPU GPUs as we have now um and if you look at the next slide actually it actually paints the picture for the long-term direction that we have uh because as I said traditional copper connections create more heat and it loses more power as you know data travels farther but silicon fatonix and CPO basically convert the signal into light much closer to the chip allowing data to move faster across fiber with lower power use and less latency. And this is why we need both, you know, both data centers need both cable and or copper and fiber. Um, like I said, copper handles the inside the rack while fiber controls the the racks and across larger systems. Just going to pause there for a second to point out that the [clears throat] market is showing signs of something kind of different happening and our analysts at Milk Pro are all over it. They spent the last couple weeks making a lot of trades, getting out of some positions, and then getting into a lot of new ones, getting ready for the next wave of robotics, space, or even kind of picking some different AI winners. If you want to see what they have in their portfolios, what position is their opening, it's just a dollar in Milkroad Pro at the link below. Maybe I'm dumb. A fiber cable has light in it. >> Correct. >> I didn't know that. I didn't know that. And I got to admit it to you guys. I have fiber internet. the CEO of the the company that provides fiber internet lives around here. I'm just not a tech guy. So, I just didn't know that part. [laughter] An idiot. Um, okay. So, when you're talking about a fiber optic cable, the optic part of course is the light part. All right. That's correct. That's I'm >> can have a lot of different there's different components of optics which I will get into as well. Um but photonics when I say is mostly the the way in which they communicate through light >> inside the data center. Um so now that we understand how why photonix becomes more useful right like as you know AI systems get bigger the next question is simple how close can we move the light to the chip because the longer the the signal travels through copper um before turning into light the more power it uses and more signal it loses. So the goal is to move optical parts closer to the main chip. And what I mean by this, if you can go to the next slide, it it shows it perfectly. So this image shows the three main steps. So first you have pluggable optics, which is what we use mostly today like in your optical transceiver modules like the 160T um that AOI makes uh PAM PAM4 or the DSPs. These are basically called the pluggable optics. There's a bunch of them, bunch of different ones. The optical module sits at the front of the system away from the main chip. If you look at the image, you have the chip on the left hand side and um you have the optical receiver modules on the right hand side and it sits away. But it uses more power but it is easy to remove uh and uh replace. Okay. And a lot of our data centers are built this way. But the problem is this is kind of inefficient because as we get bigger and bigger into more training and more GPUs, this is a very inefficient way. So what do we do? We go into near package optics. Near optics moves the optical module closer to the chip on the board. This shortens the the copper path and lowers the power usage. And we all know how much power these data centers need. So, if we can save, you know, power, why would we not do it? We need the most efficient way to train these uh uh GPUs. >> Well, do you know how much closer they're able to get it? Like, do you know the efficiency that an NPOS can provide? >> It's dramatically higher. >> It's significant enough. >> Exactly. To, you know, because the thing is it's not easy to replace data centers and all these cables and everything, right? So, it's it's significantly enough to make a huge difference. >> Great. Okay. So, that's what an NPO does. Continue. Finally, or the third one is the co- package optics or CPO, which is basically puts the optical engine um directly besides the chip. If you look on the slide, um the the signal only travels a very short distance through copper before um it becoming light. So, it uses much much, you know, less power. Um and then there's one more step with this which is the optical I/O um where optical sits directly at the side of the chip. It's not on this slide. Um, this could even use less power, but it is still the newest and the least proven approach. U, but the road that should be coming pretty pretty in the near future as well. So, the road map is simple. Move the light closer to the chip because the closer it gets, the shorter the copper path becomes and the less power a system uses, >> right? So, it's a huge efficiency. So basically, if you can get if you can get if you can get these things really really close, then you reduce the need for this copper cable that needs tons of heat and cooling as well, right? It needs it needs the heat, it needs the energy, and it needs the cooling. So, and like you were saying, this creates like a 30 to 40% power loss to be to be even just be doing that part. So, basically, the closer you can get these things together, the more efficient it is. And long term, that's naturally how people want to build their data centers, right? any new data centers coming out, they want to do the CPO part, right? They don't want to do they don't want any of the other two. They were like, listen, we we especially since you can do it, you might as well do it that way. >> Um to really optimize that that what you're doing. Okay, let >> exactly what it is. So, now let's talk about the size of this opportunity. Um because this is where the numbers start getting crazy. Um Goldman Sachs estimate that networking content per AI compute system could rise from 315,000 today to roughly 9.4 million in Nvidia's next generation, which is like a 29x increase. The reason is simple. As I mentioned, AI systems are just getting much bigger and bigger and GPUs need to move uh far more for more data between each other and networking is just is becoming just as important as the chips themselves. And Goldman's bigger claim is that total uh addressable market could grow from 11 or 15 billion today to 154 billion by 2028, which is 9x increase in a few years. And Goldman Sachs is not alone on this. Like counting sees AI clusters uh grow from 17 billion to 100 billion by 2030. Morgan Stanley estimates scale up networking could reach 73 billion by 2030 while city interconnect optical interconnect could reach to 92 billion and you get the gist of this. So almost everyone agrees on the direction of this that AI networking is like becoming a massive market. >> So so that's what you call so that's what you call photonics. So this like optical part is AI networking is or it's a part of AI networking. >> It's part of AI networking. >> It's part of AI networking. Okay. Yeah. Okay. >> So much more components that goes into it >> of course. But yeah and you're talking about Hold on. We got the I'm just going to go back to the original slide. Um. >> Yep. because you have like copper connection copper connections circuit boards and other >> there's tons of other components but it's it's specifically this part of the not just getting these pieces close together but also the actual cable right or is it just or is it just building the module that has them close together like you're saying in your third um your third type here the CPO >> bit of both >> both okay yeah so just the overall efficiency of that transfer of info >> correct correct >> got it okay >> but the most aggressive massive you know part of Goldman Sachs forecast is actually the co- package of optics that we just talked about. Um Goldman believes CPO could go from 91 billion by 2028 uh or market by 91 billion by 2028 uh with annual revenue growing from 1 billion this year to 24.8 billion next year and then 71 billion by 2028. So this is a massive massive scale. So >> wow. Okay. So the revenue has barely come in and is that revenue coming from like the deals that have been put in place for new data centers basically like is that yeah okay okay and what what are the companies sell that do this like what what part of it are they selling like they're selling this actual the entire system right like the cop package optic system they're selling that whole thing or they're buying from other people they're making >> yeah there's different parts of it they're selling lasers they're there's they're selling optical modules they're selling there's a bunch of systems that goes into this each assemble the CPO? Do they assemble the CPO and ship it or the CPO is assembled at the data center? >> They ship it to Taiwan and they they assemble it there. >> Okay. Okay. Okay. Got it. So the TAM within the next two years is 91 billion. The estimated revenue right now the revenue this year is what 1.2 billion or 1.5 billion whatever you say. >> 1 billion basically >> 1 billion. Next year it's expected to 24x >> and in two years it's expected to 70x and that projection was Goldman Sachs. Right. >> Correct. Correct. But >> but I do want to say Goldman Sachs is uh the most like aggressive on these numbers. Um but like I I do believe in the direction that we're heading to um that you know co-ackage optic optics is becoming like a major part of AI infrastructure. But like I said, this aggressive is a little too much aggressive in my opinion uh about how quickly the industry can you know actually build it and deploy it because the problem is the supply chain still has to scale. The technology still has to be test tested at like these larger scales and the data centers are clearly not going to replace everything overnight, you know. So this may take some time. >> It could take time. A lot of things could go wrong along the way. But in the current path of like we're going to build all these data centers, we're going to need way more compute. This is this is how this part is expected to grow basically. >> Correct. And there's other part of this, right? Like the plugable optical modules that we just talked about earlier. Um that that basically what that is is they sit at the front of the switch and converts electric signals into light. You know, they're not going away either. They're still going to be used in in these data centers because if you pull if you actually look at that chart um Goldman expects that global module yep market to grow from 34 billion in 2025 to 69 billion in by 2028. The biggest change is the product mix like you have the older modules which is the 400 gawatt becomes a smaller part of the market while you know the newer modules like the 800G or the 1.6T 6T and eventually we're all moving to the 3.2 terowatt products take over. So these are like all these um you know different things like plug modules and coal package optics. These are all just going to grow uh massively. Um and it's not like a one size or one market takes all. They they can still continue to grow together. >> Mhm. Okay. >> So the next question you you know why am I telling you this? How can you like benefit from all this? Right. That's like the key question that you know you always want to answer. Um so I do have some recommendations on some of the some of the stocks you know you can buy and the first stock I want to talk about is AOI which is applied opto electronics. Um it's a very small cap trading around 102 right now. Um market cap is around 8.2 billion. So, this is a very tiny company and I got to flag this right away because this stock has traded between $18 and $233 roughly in the past 52 weeks. So, there's been massive ups and downs with this stock. This is very >> why why why has there been so much so much volatility? This is a very high beta stock because this is sentiment driven um and there's a lot of news that comes out regarding these you know um co- package optics and around this sector. So um yeah this is just a very high beta play and uh yeah that's the main point I want to make with this because if you're actually going to buy this you got to understand what you're signing up for. Um so so what do they do? What do they actually do? Easiest way to think about it is it is a company that builds fiber optics products that help move data using light. So it's the large its largest opportunity is in data centers where it produces high-speed optical products for 100 gawatt, 400 gawatt, 800 gawatt and now 1.60 gawatt connections. And if you look at the image um this image actually shows where all the markets AOI serves. You have the data center market. You have the broadband access. You have the telecom um and fiber to the home and the sensing field. And the data center business is just the part that's growing the fastest. And the main reasons main reason why the investors like me are like you know paying attention to the story because their data center um transceivers run from all the way to 100G all the way to 800G with 1.6T 6T now entering the market too. And the what's cool is most of these use AOI own in-house lasers. Now I'll I'll explain what that is in a bit. Um but their most established businesses historically has been the cable broadband business, but the mix is shifting pretty fast um to their AI data center business um that they sell all the optical modules that they sell. Um but their real edge is is that they grow their own lasers um chips in house and the lasers matters because fiber cannot carry data without a light source. So by making its own lasers a AOI controls one of the most critical and supply constraint constrained part of the entire module. They got their own manufacturing process at their fab in Sugarland, Texas. Um, then they shipped those over to Taiw Taiwan for packaging. But most competitors, either design or manufacture, but they but they don't do both. and and and a AOI actually do because this matters way more than it sounds because there's actually currently a industry-wide shortage of these high-speed lasers because a Chinese competitor um locked up reportedly locked up 70% of one of the major, you know, laser supply capacity. Um, and because AA ai makes their own, um, they're somewhat, you know, insulated from that whole bottleneck. But the cav caveat with this is they're not unique here. There's competitors like Coherent also runs their own laser fab and honestly they're like kind of further ahead. Um, so there is competition in this in this field as well. >> Their advantage for AOI AOI is that despite competition, they own most of their stack, >> right? like they they yeah they they don't they don't rely on too many others for the components or for the manufacturing. >> Well, for their their in-house lasers. Yes. But there there's lasers, right? >> Yeah. There's a whole different side of this, right? Um because >> here's what you what's actually like interesting about them right now. Their 400 gawatt ship uh shipments are up something like 10x over year over year. Their 800G RAM is accelerating after some qualification. Um and the firm like some there was some delay like last year but they're they they're well past that and um they're you know I think that makes up about 5.6% of their data center revenue but by April they disclosed that they had 124 million in cumulative 800G orders from a major hyperscaler in like a month. Um including 7 a single 71 million order in there. Um and the management is actually guiding 800G to become the dominant data center contributor this quarter and their 1.6T volume order came in in March. Um deliveries are starting in Q3 and the earnings on on the earnings call management actually said this out loud and I love this line. this they basically said the revenue is limited by your production capacity and our supply chain not by market demand and they expect that demand to keep exceeding capacity through at least mid 2027. So their ability to grow is right now depends on their buildout and they can't build out fast enough because everything is just basically getting sold out. Now there there is obviously risk to this. You know I I think AOI is like I said it's a very high beta stock. So you really have to be careful with this stock. But it is it had it big run up. Now it's kind of consolidating around the 100 um dollars. And I think this is a good time to pick up this stock right now. Another major concern is their customer concentration. Um, and it's actually getting more concentrated. Um, their top 10 customers are 96.6% of their, you know, revenue last year. Um, and and just two customers made up 82% of their total revenue last year. Um, so that could be a really problem if they Yep. if they if somebody pulls back. >> Real world assets like funds, treasuries, and private credit are still running on rails built decades ago. Gated, paperwork heavy, slow to settle. Everyone's talking about tokenizing them, but far fewer can actually do it and do it without cutting regulatory corners. Securitize can. It's the SEC regulated infrastructure bringing real world assets on chain. Nine years in, native tokenization, not wrapped, backed by Black Rockck, Morgan Stanley, and Kathy Woods Arc Invest, and chosen by the New York Stock Exchange, Vanek, BNY, and Apollo to do it at scale. It's the regulated bridge between traditional finance and crypto. Tokenized the world at milkroad.com/securitize. Everyone's tokenizing stocks these days, but almost nobody's doing it right. Thin liquidity, prices that drift from the real thing, dividends that just vanish. Bit stocks 2.0 is different. real NASDAQ and New York Stock Exchange depth through licensed brokers. Price is mapped one to one, dividends paid to your account in real time. Plus, you get the lowest fees in the market at just 0.04%. And you can trade them like any other crypto as margin in earn in grid trading. Tokenized stocks finally done right. Head to milkroad.com/bitgget to get started. Oh, isn't that the case though? Isn't like I don't know what's the term for that for like customer concentration or whatever you call that? Like isn't that most of the companies that we talk about on the show and [snorts] in the AI bottleneck buildout that there there's like five customers like [laughter] the hyperscalers are the customers and that's it. >> You know >> it's all the hyperscalers. >> Yeah. Exactly. Yeah. There's just that's the only people that are building that have the capacity to build these things and buy all these components. Right. Yeah. >> Exactly. But they're also up against Chinese vendors. Um like there's >> of course there there's so many competition because they can make these you know uh modules much cheaper in China. So that I think that's the biggest risk that you know um AOI has. So um but the second name Marll ticker number MRVL >> um this is a completely different animal um trading around $193. Market cap is around $172 billion where AO AOI makes physical product and grow lasers. Marvel's a fabulous chip designer that sits on both sides of the AI bottleneck, the compute silicon itself and the data center movement layer um connecting it all. And this is the optical networking um TAM that I described earlier. Their COO um actually put this really well. He said that the AI bottleneck has shifted from compute capacity to connectivity and Marvel's whole whole portfolio is basically built around that. um around that shift and it spans um spans about five different layers of tech. So first they have P A4 uh which are optical DSPs which is like the optical modules that I talked about earlier which are basically the brains sitting inside of optical transceivers including the modules that AOI and others built and Marvel's basically the leader in creating that and they have basically have close to a monopoly with this with their 1.60 generation and with their Nova platform. So they're doing really well in there. Um second is their coherent DSPs uh for long reach connections between data centers up to like 100,000 kilometers. Then the third is their um physical layer chip um that every chip to chip connection depends on. Um then they also have fourth they have the Ethernet cable business um which which revenue crossed 300 million this year and they're guided above 600 million next year. So this is a very ever growing business. They're about to double. And the fifth is their custom AI chips for hydroscalers. Um they already have they're already building chips for um you have the ad AWS trainium. They're building chips for Microsoft. Um they also have like 50 plus pipeline opportunities um with an estimated $75 billion dollar lifetime and they're targeting over 100 billion in custom ship revenue by fiscal 2029. And yeah, and this is and I've talked about this before. There's a growing number of hyperscalers are building out their own chips because as I mentioned before, everybody's tired of Nvidia and because Nvidia's chip is way too expensive um and they need to control their margins in house. So this is going to continue to further accelerate. Um Marvel's the second leader in this um there's Broadcom which is like the big player. Uh but I continue I I think Marvel will capture a lot of Broadcom's re um the market as well in building out these custom chips, >> right? And you were saying their revenue. Let's go back because you have a revenue chart here for them. >> Yep. That's coming up. >> So it's right now in January 2026 is annual revenue, >> correct? That they were doing. Yeah. Okay. So they had nine just just over 8 billion and for 2026 it's projected at 11.5. So a 30 40% increase. Okay. >> Correct. >> And then and then almost a doubling within two years basically after by 2029. Okay. >> Yep. But C custom chips, right, is only half of the Marvel story. The other half is moving data and memory fast enough to keep those custom chips busy. So one of the fastest growing opportunities for Marvel is something called CXL, uh, which is the computer express link. Um, basically what that is, it's a standard, uh, CXL is a standard that allows servers to share and expand memory pool. Memory instead of being limited to whatever DRAM is physically attached to one machine. How I would think of this is like turning memory into a shared pool that several CPUs and AI accelerators can access it when they need it. So we all know the biggest bottleneck in in me like biggest bottleneck right now is memory and this should alleviate like minor minor sum of it by sharing um this memory pool. So now if you could put up that chart because Morgan Stanley actually put up um recently more than doubled its forecast on the CXL memory controller market. you got to go back to >> um yep they now are now expecting to reach around 2.1 billion by 2030 um from his previous estimates of 990 million. It's also tripled its forecast for the CXL switch market from 664 million to around 1 9 billion roughly. And um Marvel is one of the clearest bene man benefacies because of its structure of product line um because it covers the exact categories that Morgan Stanley just upgraded. um like the memory like the expansion um near memory accelerator the CXL switches um and uh and to be frank CXL is still only one small part of like Marvel's memory strategy um there's like a bigger strategy at play with their you know acquisition of Celestial AI and it photonic fabric technology. So basically, Celestial is designed to solve what is called the memory wall. And basically what that is, which is the growing gap between how fast an AI can process data and how quickly memory can deliver that chip uh that data to the chip. Now AI accelerators keep getting faster, but the connections between the compute and the memory are not scaling at the same speed. So, Marvel claims photonic fabric can be more than twice as power efficient as copper. Um, it delivers like latency below 200 nano speeds. Um, basically what what I'm trying to say, this is supposed to be really fast, right? So, and and they acquired this back in March to, you know, get into the market as well. And why this matters so much is as I mentioned before Marvel already designs custom chips. The the switching silicon like connecting those chips and many of the optical components moving data through the system right now. Celestial gives Marvel the mix missing optical link that can sit much closer to the processor and the memory. So this is a very very much a strategic play by Marll. Um, sorry I keep pronouncing it wrong. It's Marll, now Marvel. Uh, but it's just [laughter] I'm a huge Marvel fan. So, >> you were watching the Doomsday trailer while we're talking, weren't you? >> I was. I was. [laughter] >> Looks terrible. Anyways, uh, that's great. Um, okay. Okay. Marll, we'll get it right. Um, and this is also, you know, another another trend I've noticed about Marll is that it's a Jensen company as well. >> It is, >> right? And that's something that it's like you've stressed, I think Vincent has stressed this as well, but you've been the main backer of this. It's like listen, Jensen Huang doesn't does not just randomly mention companies, nor does he and he's in he's an investor in them as well, right? >> That's right. Jensen invested about 2 billion back u back in like May or something um with the company through the NV link fusion which basically allows like custom AI chips to connect with Nvidia GPUs and Nvidia's broader networking system. Um what that means is that that means a hyperscaler could now use Marvel design accelerator while still connected to Nvidia infrastructure. So Nvidia is effectively supporting a company whose customers are building the chips partly to reduce their dependence on Nvidia chips GPUs like I said because Nvidia can still benefit if those chip remain tied into his networking platform. So that's the brilliance of Jensen and Jensen has called Marll the you know the next $1 trillion company >> next trillion dollar company. That's right. >> Exactly. And and I fully believe that and I fully believe that. I don't know when that timeline is. Uh but this stock stock has dropped dramatically over the last um 3 4 weeks. Um >> look, so we've got [sighs and gasps] Yeah. So it's down what is it down from uh from its alltime high? How do I get my little Here it is. >> 30% >> even from that alltime high. >> Yeah, it's down 45% from its all-time high which was on What day is that? June 18th. So, in a month, in a month, it it dropped. Well, it dropped 45% and then today it's down 40, it's down 40% from that all-time high. So, it has two green days in a row last Friday and today, Monday. >> To be fair, this stock did require some pullbacks because I think it went up 20% or 21 25% because Jensen called it the next hyperscaler. I don't think it should move like that based on those comments. So, >> oh, so this huge pump kind of going into the end of April here kind of post Iran war scares. Exactly. >> This this was this was a huge this was he said that >> he said that and it pumped like 25%. >> Right. Okay. Okay. I still believe that I still believe this this could likely hit um you know a trillion dollar company in the next you know x amount of years. Um so the financial >> which is nice that's that's like a 6x right because at 172 right now 172. So and it could be and what you're saying too is that like that's not an immediate thing. So, it's something that over time, right, you could you could even see could you see better entries coming soon as well, Melvin? Even though it had a 45% pullback. >> No, I don't I think this is going to be at the lowest. I think it dropped down all the way to 170s like couple days ago even. Um I don't know what the lowest was, but 179. Yeah. >> Yeah. I don't see this going below 150. Maybe it could drop to 150, but I don't see this going below 150. Um I >> depends on market conditions, too. There could be, you know, especially these days, any kind of headline can can scare the market. But yeah, >> exactly. Okay. But, you know, earnings are coming up. So, um, you know, they they should get a really good guidance based on all that. So, like I'm going to remain bullish on this company, >> right? And AOI despite its insane just to circle back on both picks that you're kind of giving us today. What is you're saying? Their chart has been like totally berserk this year. Let's take a quick look. Wow. Okay. This is a similar chart, man. So, these two these two would you say? Would it be fair to say that these are almost kind of bundled together as photonic companies in recent times? This chart it's just pretty similar in terms of the last like three months. >> Yeah, last three months. But like yes and no because these are kind of similar but not at the same time because AOI is like 8 billion company. Other one's like 190. Yeah, much higher. But everything runs on sentiment sentiment these days, right? Like so the sector tends to run up together. S tend to come down together. So, um, yeah, I expect all these things to climb back up because we had a bit of a weakness in the semiconductor trade that's largely seems to go away because of the Kimmy K3 news that came out that because of open- source models, we need more compute than ever, which means more data center buildout, which means all these TAM is going to continue to grow despite, you know, all the nervousnesses around this trade. But I expect Marll to continue to grow and continue to, you know, um, reach 1 trillion in the in the coming future. >> So, here's a question for you. This is our third straight episode, which comes out. They always come out on Tuesdays with Melvin and I. There's a third straight episode where you kind of give us like a sector snapshot, right? Last week we did Neoclouds. The week before we did robotics, right? And totally different totally different areas, especially robotics is deeper into the future. Um, and Neoclouds is kind of more like right now, the next five years. What like my timeline is 24 months which no my timeline is 36 months right in the middle of all that what out of those three sectors how would how would you rank them in terms of RNR or even like what's the safest what's the riskiest in that time >> yeah I would say the safest would be neoclouds because they are just growing at a very fast pace as I mentioned before Kimmy K3 is a really bullish sign for these neoclouds because we're going to need massive amounts compute more than ever before and that's going to that's going to accelerate and you know they're I was looking at some data and I think Nebius and all these other company or specifically Nebius is going to crush their earnings. I am a little bit worried for them in Q4 because of their windland uh vin sorry I'm pronouncing that wrong because of the data center but I will still still pick neoclouds at first then memory being second memory was still going to be a bottleneck there's a lot of debate going on right now when about like when the me memory bottleneck is going to go away and a lot of people are like even Morgan Stanley put out a chart saying that it's going to alleviate by end of this year which I don't see that to be the case I expect this to continue into 20 like at least the middle of 2027. I can see this going past 20 like into 2028 as well. So, um the memory trade I'm still bullish on them. As I said, SKHX just got listed on the market. Um and Micron is still, you know, we had a bit of a pump up today. Um because what's funny is Micron is expected to reach $400 billion um in almost profit by in like couple years. um and they're expected to buy um 40% of UBS actually projected that they're expected to buy 40% of their share back u you know um so that be a massive boost for their um their stock as well. So, so I would go Neoclouds memory and then um what we are currently talking about >> in terms of risk >> risk. >> So, this would be the riskiest. >> A AOI would be the specifically the riskiest. Yes. >> Okay. Okay. Okay. Makes sense. Okay. Well, you guys take that as you will. If you do, for anybody listening, if you do want to know how Melvin's going to play this [laughter] you got to go milk road pro. you actually you you you do already have a lot of the stuff that we talked about today uh and in the last couple weeks in your portfolio there. So if you guys want to see what that is um make sure you go check it out if you just if you honestly just want to look at it and then leave. It's a dollar to do that so you can do that. But I feel like we have a lot of great stuff and you want to stay. Um so make sure you check that out. Melvin, I mean this is awesome, man. You have any final points we should think about for for Photonix? >> Stay invested. buy these stocks at a at a you know very very good price and your three-year you will thank you thank yourself. So buy now buy now sell later. This is not the time to panic. This is the time to accumulate. [gasps and laughter] >> Oh awesome. Okay. Well I like that. It's good philosophy uh going into the rest of the week. Melvin, thank you as usual for a great presentation. Great uh little thesis here. I'm excited to see what you got next week. >> Thanks LG. >> Want to stay ahead of the biggest [music] technological shift in history? Subscribe now to get insights straight from the sharpest minds in tech and finance. Quickly, you'll [music] note this show is for educational purposes only. Nothing here is financial advice. Investing always carries risk. Never invest more than you can afford to [music] lose. Thanks for tuning in. See you in the next one.