0:00 [01:00:00:00 - 01:00:28:10] On March 2nd, Nvidia invested $4 billion into two companies most people have never even heard of. Not an AI model company, not a chip designer, two companies that make components out of light. In that same week, Marvell closed a $3.25 billion acquisition of a photonics startup. That's over $7 billion flowing into one corner of the supply chain in just a matter of weeks. And when that kind of capital moves that fast, something fundamental is shifting. [01:00:28:10 - 01:00:32:09] So let me try to explain what's actually happening here in plain English [01:00:32:09 - 01:01:10:03] As of right now, inside every major AI data center, thousands of chips need to talk to one another, and they do that by sending data through copper cables. Today's fastest connections run at about 800 Gbps. And copper can barely even
0:47 handle that. So the next generation doubles it to 1.6 trillion bits per second. And at that speed, copper begins to fall apart. Because there's too much heat, there's too much power. So on top of all of that, the signal itself begins to degrade even over short distances because it's an exponential problem. And in this case, that's not a software problem that you can just simply fix with a patch. It's a physics wall. [01:01:10:03 - 01:01:30:04] And the workaround is pretty simple. And this is where silicon photonics comes into play. Instead of pushing electrical signals through copper, you push light through tiny glass pathways etched directly into a silicon chip. Same kind of chip manufacturing that the semiconductor industry already knows how to do. It's just with photons instead of electrons. [01:01:30:04 - 01:01:41:09] matter. Speaking of what matters, the part that does is the electricity bill. A single AI data center today can consume as much power as
1:38 a small city. Think of 100 megawatts or more. [01:01:41:09 - 01:01:47:04] When you double the data rate, copper's power consumption doesn't just double, it actually scales exponentially. [01:01:47:04 - 01:02:06:10] And this is where photonics completely changes the entire economics. At the individual link level, optical connections are 3 to 3.5 times more power efficient than copper. Now multiply that across an entire data center with 40,000 or 50,000 links. That's not just a 5% improvement. [01:02:06:10 - 01:02:35:01] Every hyperscaler that's building out AI infrastructure, think Microsoft, Google, Amazon, Meta, they're all running into the same wall right now. That's why the money is moving so quickly. And the market reflects that urgency. Silicon Photonics is roughly a $3 billion market today and it's growing roughly 25 to 29% annually.
2:27 Conservative estimates put it around $10 billion by 2030. But the most aggressive projections say it's going to be closer to $30 billion by 2034. [01:02:35:01 - 01:03:15:06] In my mind, this is not a nice-to-have technology. It's an absolute must-have. Every data center is going to have to make this shift. And everybody already knows it. That's why Nvidia already committed $4 billion to this space in a single day. So looking at this in kind of a high level, I think about this space the way that you would think about a basic building. Of course, you're going to have your base layer. There's massive companies where Photonics is a growing piece of something much bigger. Then there's the middle layer. These are the foundries and the contract manufacturers that physically make the transition even possible. And then there's the top layer, which are the pure play names with the most direct exposure and of course the highest upside.
3:15 [01:03:15:06 - 01:05:04:08] Once again, I'm not a financial advisor and this isn't advice, and this is just my personal research framework. But as always, if you get any value from this, please consider pressing that like button. Since I already mentioned that Nvidia is investing heavily into this space, let's go ahead and start right there. Where Nvidia, they don't just sell chips anymore. They sell the entire rack scale AI supercomputers. Their latest platform, Vera Rubin, ships in the second half of 2026, and it packs 72 Rubin GPUs and 36 Vera CPUs into a single rack with 1.3 million components. And here's what matters for this video. Silicon Photonics is built directly into this system. The Spectrum 6 Ethernet switches use integrated silicon photonics, delivering five times better power efficiency than the traditional network.
4:01 The Super Nix run at 1.6 TBS per second per GPU, the exact speed where copper really begins to hit the wall. Photonics isn't an accessory in this system. It is the entire infrastructure that really makes this work. Look at it this way. When Jensen builds something into the Nvidia platform, the supply chain is going to quickly follow. It happened with CUDA. It happened with Infiniband. Now it's happening with Photonics. Of Nvidia's $215 billion in fiscal 2026 revenue, $197 billion came from Data Center alone. That's the market that these photonics companies are really going to be selling into. And it's just the beginning, and it's just accelerating. Staying within the base layer, the next company is Broadcom, where they are the current industry leader in co-packaged
4:47 optics, which means they're just integrating optical components directly onto the networking chip instead of plugging them in separately. And they happen to be on their third generation of this technology, running at 200 gigabits per lane. And a fourth generation, at double that speed, is already committed to. [01:05:04:08 - 01:05:58:21] And Silicon Photonics is still just a small piece of their total revenue today. But as every data center on the planet begins to make this transition, it becomes a meaningful growth layer on top of an already dominant business. And moving on to the next company is Cisco, where most people know them as the backbone of the enterprise networking. And that's exactly why they're going to be in the space too. They acquired a company called Luxterra back in 2019, which was one of the original leaders in Silicon Photonics. Since then, they've been building their
5:36 own photonics integrated circuits and digital signal processors in-house. They're already making 800 gigabit Silicon Photonic transceivers within today's data centers. And they showed a co-packaged optics prototype in a 25 terabit per second switch. I mean, hopefully you can see it. Like this is the direction where everything is heading. Already Cisco is on track for its strongest revenue year ever [01:05:58:21 - 01:06:15:21] Look, Silicon Photonics isn't why you're going to buy Cisco today, but it's why Cisco stays relevant, well, tomorrow. As every data center transitions from copper to optical, the company selling the switches and the optics that go within each of them is sitting in a very comfortable position. [01:06:15:21 - 01:07:46:18] AI infrastructure doesn't just stop at data centers. It extends into defense, where autonomous
6:21 systems are the next massive frontier. And that brings us to today's sponsor, Spark AI Inc. A software company building next-generation GPS-free target acquisition and autonomous navigation technology. Spark AI's proprietary technology gives drones full autonomy, the ability to locate, target, and navigate entirely on their own. That means no GPS, no radar, no external connection. A zero-signature system that works across air, land, and sea. Now here's where it gets compelling. Spark AI just appointed Matt McCran as CEO of its US subsidiary. He's the former CEO of DroneShield, a roughly $3.5 billion company. And his focus is all about turning Spark AI's geolocation technology into deployable defense solutions and forming partnerships with major defense
7:10 contractors. Then the company just announced a strategic partnership to deploy that platform in Ukraine, marking their first expansion into an active defense market. And the defense and commercial drone markets are projected to exceed $100 billion plus over the next decade. And Spark AI utilizes a scalable, high-margin licensing model. As of today, the stock is up over 122% year-to-date at a price just over $1. Where the CEO over the entire company, Anush Manzuri, has the conviction of insider ownership of 34%. Now, if you'd like to learn more about Spark AI, please check out the link down in the description. [01:07:46:18 - 01:08:52:17] Now I'll move on to the last company within the base layer, which happens to be Intel. And I know Intel is a little bit of a controversial pick, but the stock has been through quite a lot. But
7:57 here's what most people are already missing. Intel is arguably the most advanced company in the world when it comes to Silicon Photonics research. They've been at this for over 25 years, and they've shipped more than 8 million Photonic integrated circuits with over 32 million on-chip lasers. No one else is even close to that type of volume. Their optical compute interconnect chiplet, the OCI, does 4 Tb per second bi-directionally. That means it can go both directions. And it consumes just 5 picojoules per bit, compared to 15 for traditional pluggable modules. That's a 3x power efficiency advantage. And here's the kicker. Intel is the only foundry on the planet that offers an optics-based foundry option for its customers. That means that other companies can come to Intel and say, "Hey, build me a custom chip with Photonics
8:47 integrated directly into the package." To me, that happens to be a massive strategic asset. [01:08:52:17 - 01:10:37:03] They are high risk, but the upside is significant if they happen to execute. Now let's go ahead and move up to that middle layer, where silicon literally meets light at manufacturing scale. Foundries and contract manufacturers. The companies that take a design and turn it into something you can actually ship. And here's how you begin to know that this layer really matters. Think Taiwan Semiconductor. The world's largest foundry just announced a dedicated platform for mass-producing photonic components in 2026. I'm sorry, but when the TSMC builds an entire production line for a particular technology, that's the ultimate validation signal. But the companies already entrenched in this space already have a bit of a head start. And in semiconductor manufacturing,
9:36 head starts compound very quickly. So let me show you who's already there. The name I'm going to start with is Tower Semiconductor. They are a specialty foundry, and they don't design chips, they manufacture them for other companies. When someone designs a photonic integrated circuit, Tower is the one fabricating it on silicon wafers, at scale, across their fabs within the US, Israel, and Japan. They run multiple photonics platforms, including one with lasers built directly onto the silicon. And their CEO said it plainly. They are, by far, the leading foundry in silicon photonics. Revenue from that segment hit $228 million in 2025, more than doubling from the year before. They've committed $920 million to expand the capacity by more than five times,
10:22 with over 70% already reserved through 2028, and it's backed by customer prepayments. They already have a direct partnership with Nvidia to build 1.6TB optical modules, one that I didn't even mention at the top of this video. [01:10:37:03 - 01:11:20:16] And now we're going to move to the next company of Global Foundries. Where Tower is the specialist, Global Foundries is the scale player, one of the largest contract chip manufacturers in the world with fabs in the US, Germany, and Singapore. Their edge is combining silicon photonics fabrication with advanced packaging. So they can build the optical chip and the complete module all under one roof. They doubled photonics revenue to over $200 million in 2025, and they expect to nearly double it again in 2026. Management says that they are on a path to $1 billion run rate
11:13 by 2028. Doubling itself once is interesting. Three years in a row is a full-on signal. [01:11:20:16 - 01:12:33:17] But photonics is only about 3% of their total revenue, and the overall top line has been relatively flat. So in this case, patience is a little bit of a requirement. Now move on to the name that ties this entire middle layer together. And that's Fabernet. They're a precision contract manufacturer based in Thailand, and they don't design anything. Other companies design the product, and Fabernet just builds it at scale. So you've got laser modules, 800 gigabit and 1.6 terabit transceivers, optical engines for co-packaged optics, all assembled in over 2 million square feet of factory space with submicron alignment tolerances. That kind of precision is extremely difficult to replicate,
12:03 which is why customers rarely leave once they're qualified on Fabernet's lines. So here's why this name is underappreciated. Remember the two companies that received $2 billion each from Nvidia at the top of this video? Well here's a little bit of a secret. They both manufacture through Fabernet. That money flows through as increased manufacturing orders. Fabernet benefits from every single dollar that's spent in this space without bearing the technology risk. Their financial profile is extremely clean. [01:12:33:17 - 01:15:16:22] Their book value doubled in just five years. So if you want the safest entry point into silicon photonics, this one might just be it. Now let's move on to the juicy top layer. These are the pure play photonics. These companies live or die by how fast this transition
12:49 begins to happen. The first company at this layer is Coherent, and this is the world's largest vertically integrated photonics company. They make the laser chip, package it into optical engines, and they assemble the final transceiver module, all under one roof. Most competitors only play in one of those layers. They're also one of only two companies on the planet that's making both indium phosphide and silicon photonic engines, on all that scale. Their revenue roughly five times from $1.1 billion in 2018 to $6.3 billion just over the past trailing 12 months. So their most recent quarter was at $1.7 billion, and that's the best in their company history. The gross margins are recovering from 30% to 36%. And on March 2, Nvidia announced a $2 billion strategic investment with a
13:36 multi-year purchase commitment. All that, and they have a peg ratio of about a 1.1. And now I'm going to move on to Lamentum, and they make the lasers that power the entire photonics ecosystem. Every silicon photonics chip needs an external laser because silicon can't efficiently generate light on its own. Lamentum's lasers go inside other companies' products, including their competitors. So they're going to win regardless of who takes transceiver market share. In under two years, this company went from a $547 million net loss to $252 million in trailing net income. And not to be outdone, Nvidia announced that same $2 billion deal structure here, the largest laser commitment in Lamentum's history. And one of the items that I look at is their peg ratio, which is
14:23 0.61. It happens to be the cheapest in this entire photonics universe. And now we'll move on to the last name of this top layer, and that's Marvell technology. Their digital signal processing chips go inside virtually every 800-gigabit transceiver. That's on the market today. That's roughly 50% of the market share. When light comes out of a photonic chip and needs to be interpreted as data, that's Marvell's silicon doing all the work. And they also design custom AI chips for Amazon and Google. In February of 2026, they closed a $3.2 billion acquisition of Celestial AI, a company that's building optical chiplets that connect computing chips using light instead of electrical traces. It's 25 times greater bandwidth at 10 times lower latency than
15:10 current approaches. And if that happens to hold up in production, it changes how entire data centers are going to be wired together. [01:15:16:22 - 01:17:14:23] Their quarterly revenue nearly doubled over the last eight quarters, from $1.1 billion to $2.2 billion. And their adjusted earnings per share tripled from $0.24 to $0.79. The shining light is the fact that the company shows persistent gap losses, and that's entirely from that $1.3 billion a year in amortization from that prior acquisition. But if you look at their cash flow basis, the business is extremely healthy. And for those of you that are value seekers, this company is sitting at a pay ratio of $0.64. And they have an earnings report that's going to be coming up pretty soon that's going to be the first to potentially include Celestial AI, and it's going to set their fiscal guidance for 2027. So it's very high stakes, and it could be a
15:58 bit of a catalyst. When we look at all the layers of photonics, here they are. And these are my top 10 companies. But of course, now I'm going to break it down on how I would invest $100 into this space today. Now here's how I'd break that out, based on what I see as the strongest outcome when it comes to these photonics. And just keep in mind that I invest asymmetrically. I tend to put a little bit more heavily towards the mega caps and less into those smaller companies with those higher risk ratings. And of course, I'm putting it on the screen right now. Here's the bottom line. Copper is hitting a physics wall. Every major hyperscaler knows it, and over $7 plus billion moved into the space in just a matter of a couple of weeks, just to prove that point. Silicon photonics is no longer speculative. It's going to be the next foundation layer of the AI
16:46 infrastructure. The only real question is which companies are going to capture the most value as this transition truly begins to accelerate? As a reminder, my free newsletter breaks down a lot of these moves that I take every week. The link's down in the description. As always, drop a comment with your first ticker that you're going to start researching from today's video. Now, if you want to see more deep dives like this on the AI infrastructure stack, then please subscribe so you don't miss the next one. And as always, thanks so much for watching, and I'll see you next time.
