The $10T AI Buildout Has a Photonics Problem

0:00Yannick De Koninck:Welcome to Sorcery.

0:10Herwig Van Hove:The potential of photonics. The technology to interconnect these GPUs.

0:14Yannick De Koninck:Global payroll and HR platform.

0:16Herwig Van Hove:8 billion operational data points. Scale-ups over startups. You'd rather have a million of something than one of anything. Building the intelligence layer for defense.

0:24Yannick De Koninck:There isn't enough power, there's not enough memory chips.

0:27Herwig Van Hove:We're at the end of physics of what we can still do.

0:41Yannick De Koninck:Janik and Herwig, welcome to Sorcery and Strike's annual gala. We're here in Monaco and we're going to go deep into photonics and what you're building with FEMA. But before we get into that, we're going to go into a long conversation on photonics. So, Janik, I'd love to start with your background and then, Herwig, we'll get into yours. How did you get into photonics?

1:05Herwig Van Hove:So, I was one of these nerdy kids who locked himself up in his bedroom playing Lego all day. So, naturally, I went to engineering school and it was 20 years ago. And back then, Silicon Photonics was this niche upcoming branch in engineering with no prospect for any job at all. So I decided to jump on that ship. I did make my PhD in that field. And then by that time, I had the opportunity to move to San Diego to join a startup company there that was trying to commercialize Silicon Photonics. One of the first companies to do that and to successfully do that. I moved back to IMEC where I worked on laser integration for silicon photonics for a while and then five years ago I joined NVIDIA back then when I still had to explain to people what NVIDIA actually was and so we started from scratch built the silicon photonics technology there into a product ready maturity and now I'm starting at TEMA.

2:15Yannick De Koninck:Ulrich, how did you get involved?

2:18Herwig Van Hove:So I've been an investor most of my life, turned operator very late, but I've been focused on deep tech for quite a while and so got involved in the fascinating world of photonics almost a decade ago and realized three, four years ago that the train was leaving the station and there were some gaping holes in the supply chain. And that is what we're trying to solve with Thema, realizing that this visionary choice of Yannick 20 years ago to study the field of silicon photonics was something that has put Belgium on the map and has created a talent base that's broad and deep. and I've spent my life selecting, honing and nurturing talent.

3:11Herwig Van Hove:And so that's what we're going to hope and do with this.

3:13Yannick De Koninck:So let's lay out the state of the market and how chips have gotten to this point. Why copper is going out of date and why there's a focus on photonics and lasers. Can you explain a little bit of that and also why Europe? up?

3:29Herwig Van Hove:So I think the market is the part that really dragged me into it. We saw a much bigger opportunity than what most people in the field were seeing. That was because we were talking to the CTOs of data centers and realized indeed what their need was and how photonics could help solve this. And we were listening very carefully when some of the gurus of the ecosystem, like Andy Bechtelsheim, were quoting numbers that far exceeded what even the most optimistic analysts were saying. And we felt that they were, you know, very well informed. And so that's what pushed us to do this. And to answer your question about where we are today.

4:21Herwig Van Hove:So if you think about AI, you think about GPUs, right? You think about these compute engines that actually run the models, but the models have gotten so large that they no longer fit on a single GPU. So what we need to do is we need to interconnect multiple GPUs together to run these models. Traditionally, that's been copper interconnects, but as these GPUs get faster, they need more data, They need more data at a faster rate and copper is running out of steam. And that's why we're transitioning to optical interconnects, which can which bring much higher data transfer bandwidth. So, yeah, that's where.

5:02Yannick De Koninck:So I recently interviewed Tony Kim of BlackRock in Paris at the Rays AI Summit, and we were breaking down the entire industry. He says there's a 10 trillion dollar rebuild that's happening. And one of the most affected and going to be the most effective places for that spend is in these data centers because the current old infrastructure of a data center is not built for AI. It's completely inefficient. It takes a lot of water. It takes a lot of energy. And the chips that are being used in those are just they're just all outdated and they cannot handle the parameter models that are in the market.

5:39Yannick De Koninck:So I would love to get kind of an understanding from one of you, the difference in architecture of those data centers and how they're transforming with this new build.

5:52Herwig Van Hove:That's probably for you. Yeah. So if you compare a traditional hyperscale data center to a modern AI factory, one of the big differences is that there is way more traffic between different servers. If you think about your traditional Google search five years ago, it was basically your traffic is routed to one computer that finds the answer and sends that answer back. Now, if you think about modern AI infrastructure, there's, as I said before, the models no longer fit on a single GPU. So there's already GPU to GPU communication. But now in the age of agentic AI, there's a model that is doing a call to some tool, which is on a different server that's returning the results to the GPU.

6:42Herwig Van Hove:So there's a lot of traffic within the data center, which is very sensitive to latency, basically the time between when all these different calculations can run. So what that means in reality is that you need a much flatter network instead of a really deep hierarchical tree of switches. you need a much flatter network that means far more connections between uh uh direct connections between switches and gpu so basically a much denser network than what we had in the traditional hyperscale if you're asking a cto of a data center what keeps you awake at night okay the the the energy consumption is answer number one because it's existential to them they don't know how to solve it.

7:24Herwig Van Hove:Their CEO is saying, we need to grow, we need to grow. And he's saying, okay, but where? Because nobody can deliver the grid. Google negotiated for three years with Amsterdam and had to move its plans to actually now Belgium because they simply can't get the grid. And so that obviously, with the ambitious growth plans that they have, requires a much more fundamental solution that can scale to the same size as the problem, right? And so there are many data centers, there are many more being planned. Therefore, you know, upgrading the grid or building new power plans cannot happen in the time that is required.

8:07Herwig Van Hove:And that's where, you know, every deep deck paradigm shift needs a killer app for photonics, more than 20 years in the making, 30 years in the making on the science side, energy consumption by data centers and AI is that killer app. And so now we get to where we are today, which is, that's great, but nobody was prepared for the speed of change. The speed of change happened November 22, chat GPT comes out, it goes viral like it's never seen before. And with the growth of that came this requirement or this realization, okay, we need to switch to photonics. Nvidia took the lead in that. They had been talking about photonics for a long time, but they said, okay, we need to make the switch happen.

9:01Herwig Van Hove:24, they were even louder about it. 25, they really pushed the market. And 26, they showed the market what they meant with their investments in Lumentum, Coherent. They've invested 4x, the total market size, in the first six months of this year. And so now everybody realizes, oh, the supply chain is not ready. The innovation is ready. The photonic scientists have done a lot of work. The photonic designers have done a lot of work. But you need to work that all the way back to the beginning, which is, you know, all the way from the raw material over the tooling, over the infrastructure to build.

9:41Herwig Van Hove:And I think that's where the pain point sits today. And it's part of what we are trying to solve.

9:48Yannick De Koninck:So how many photonic companies are out there? I interviewed the Lumentum CEO, Michael Hurston. And even just looking at that company, they are spilling over with demand. Their stock went up a thousand times in a year. Insane. and he had like just stepped in maybe a year and a half ago so he like really came in an opportune time but i'm curious like what does the state of the market look like for the companies that are existing and how does it all work together i think the the again the the what needs to be

10:19Herwig Van Hove:understood is the the the shift in in in size of the market photonics for years was an academic endeavor with some applications that was being prepared for in the supply chain at that scale. And now all of a sudden, NVIDIA steps in and says, no, no, guys, this is much, much bigger. And so, you know, obviously, Lamentum Coherent are the two leading companies in the West doing this. They're the only ones scaled at that size. If you look at the rest of the industry, the traditional players, the TSMCs, Global foundries towers of this world have made a tremendous effort investing in the part that fits in their supply in their production process which is the the front end the silicon photonics so making the wafers but if you deliver a wafer to nvidia that's not what they ordered they ordered a final chip so there's a lot of other things required um lasers first and foremost which is what Lumentum and others are making.

11:24Herwig Van Hove:And so that's going to have to scale by a lot. And that's challenging because the throughput of the machines is not at all where it needs to be. The efficiency, the yield is not what it's supposed to be. And the substrates, the raw material, is really, really cornered. the market is i mean there's only a handful of suppliers and on top of all these problems you now have the east west divide which is getting quite problematic so in all of that we think europe is in a in a sweet spot because in europe you have industrial capacity that was available industrial infrastructure that was available that's what we bought you have photonic talent that has been trained for decades.

12:21Herwig Van Hove:You have industrial talent that was available because all the big foundries closed. And so the combination of all of this, in addition with a desire of European sovereignty, makes that we're like in an ideal spot to do this from Europe.

12:40Yannick De Koninck:So let's go to the founding of the company and how you guys came together. What was that process like? Yeah,

12:48Herwig Van Hove:we have been building the concept of a pure play photonic foundry at a much larger scale. We were looking around. We were seeing that all the big guys were doing the front end of the process, as I said, the silicon wafers. And we were seeing a lot of innovation happening, but that was mostly university spin-out scale. and so there wasn't really an industrial player um in uh in the middle and as we were making plans in in 23 24 to do this there was i guess some serendipity involved in that uh a large-scale european foundry went uh went out of business and so we we took the occasion we jumped on it we we we bought it and it delivered i think a very nice combination one 20 kilometers away from imac world leading research center lots of talent two lots of industrial talent as i said an infrastructure that was ready to go right at the time when everybody needs it because when you want to build a foundry of this size you're talking about a three to five year project It's between getting the permits and the licenses and building that whole.

14:09Herwig Van Hove:It's not just clean rooms. The infrastructure is quite complex. So having all of it ready to go in 24 was a serendipitous event, I'll call it that. And so that's where we jumped on it. And then we started talking to both sides at the same time. when I say both sides, we started talking to the industrial talent and the photonic talent and said, look, Europe has invested, successfully invested for 20, 30 years in photonic innovation. This is the time for industrialization. And, you know, Yannick and others jumped on board instantly. So that's where we are.

14:56Yannick De Koninck:This episode is brought to you by Brex, my favorite. You become what you spend on. And I refuse to spend my time on work that shouldn't exist. Expense reports, receipt chasing, and manual closes. The companies building what's next from Vercel, OpenAI, Anthropic, Granola, and Deepgram all made the same call. They all run on Brex. Brex is the intelligent finance platform that combines cards, expenses, and banking into a single stack with agentic finance built in. AI agents that handle expenses automatically enforce policy before spend happens and close your books in minutes. That's why Sorcery runs on Brex, so I can spend time on building and not busy work.

15:41Yannick De Koninck:It's time to get Brex AF. Learn more at brex.com slash sorcery. That's B-R-E-X dot com slash S-O-U-R-C-E-R-Y. Bye. Turing is training the next generation of AI with tasks that require real expertise and real world judgment. That's why companies like NVIDIA, Anthropic, Salesforce, and Gemini partner with Turing. Turing builds realistic reinforcement learning environments and data systems based on real operational traces, the kind of infrastructure Frontier Labs need to train superintelligence. Visit turing.com slash S-O-U-R-C-E-R-Y. Janik, from a technical standpoint, where are you at?

16:26Herwig Van Hove:So yeah, I was in a great spot at NVIDIA, right? we were building the next generation of chips and and I wasn't planning to go anywhere but when I started to talk to Herrick I and I learned about the ambition of this enterprise and and what he was planning to do I realized that this was a once in a lifetime opportunity for me to join a mission that's going to solve some of the core problems that are slowing down. Probably one of the biggest or the biggest technological revolution that I will witness in my lifetime. So I decided to leave the Golden Palace and jump onto the pirate ship and set sail.

17:16Yannick De Koninck:Selfish question. What is it like at NVIDIA? What have you learned there? Like operating principles. What is it actually like?

17:25Herwig Van Hove:It's an incredible company. It feels like a startup. It's very agile.

17:35Herwig Van Hove:When the idea is to fail fast, so we start building stuff. If it doesn't work, move to something new. It's a, in my experience, a very apolitical company. it's the the org chart is just a piece of paper if there's a project and you can contribute you contribute to that project doesn't matter where you are in that in the org chart um so yeah it's um it's an incredible company and then for people who don't know what imac is can you also explain imac and where it fits in the stack sure so imac is a and it's not imac by the No, IMEC, Inter-Universitary Microelectronics Center. So microelectronics center, that tells you something about how old this institution is.

18:31Herwig Van Hove:So it started over 30 years ago and has been doing, has been spearheading research initially in the area of electronics. But since sometime also much broader area, one of these is photonics. So they started, IMEC was one of the birthplaces of silicon photonics in the world and has grown to be the number one research center when it comes to advanced electronics technologies in the world.

19:07Yannick De Koninck:And where are you guys at in terms of commercial progress to date? When are you going to start rolling out your customers?

19:14Herwig Van Hove:Our biggest challenge is the timeline customers are putting in front of us. The good news is that the volumes they're asking for even surpass our wildly ambitious projections. So that's certainly very positive. The constraints for us now, the whole market, Coherent, Lumentum and others, everybody is facing the same constraints. The tool builders are not building anywhere near enough tools. And the tools themselves need to be further improved for throughput, for efficiency, for yield. And then even more problematic, the substrates, the raw material. And so here's the good news. As I said, Europe has been investing successfully in innovation.

20:09Herwig Van Hove:And so some of these problems, even though they didn't anticipate this shift, were being researched. And so throughout Europe, Greece, Germany, Spain, France, Switzerland for sure, there has been a pipeline being built on tackling individual technical challenges around this. And so to answer your question, our timeline is to start producing next year in 27 and work towards a full ramp up by 28. So our timelines are short, but Europe is wide awake and ready to go. That's not just a slogan, we feel it every day. So there's a lot of collaboration going on and we're finally starting to really work across borders.

21:17Herwig Van Hove:and I'm very optimistic about Europe playing a significant role in this.