← 返回

2026.10.09 | TB | Optical | Soitec Photonics Substrates & AI Datacenters

原标题：Soitec – Photonics Substrates & the AI Data Centre Surge

#optical #materials

Third Bridge · 2026-10-09

Analyst-led • 7 Oct 2026

Companies: SOI FR, 4063 JP, 6488 TW

David Herisson (DH) Executive, Sales & Business Development, Advance Imaging at Exosens

Saad Maqsood (SM) — Third Bridge Sector Analyst

Agenda

Key Insights

Introduction

SM [00:00:02]: Welcome to Third Bridge's Interview today entitled Soitec – Photonics Substrates & the AI Data Centre Surge. I'm Saad. Today, we have with us David, former Product Manager at Soitec.

Just before we get started on the call today, David, could you please state I agree or I disagree to the following statement: You understand the definition of material non-public information and agree not to disclose any such information, or any other information which is confidential, during this Interview.

DH [00:00:39]: I agree.

SM [00:00:39]: Could you start us off with a quick introduction to your background, please?

DH [00:00:48]: My background is more on the technical part because I started with a PhD in optoelectronics a long time ago, I would say. I joined Soitec in 2019, and I left Soitec end of 2024. I worked as a product manager in different business units, starting first with the power product and then shifted directly to a Photonics SOI product. At the end, I was working mainly on the POI, piezoelectric-on-insulator product.

Questions

00:01:41] Soitec's Photonics SOI [silicon-on-insulator] sales doubled YoY in Q1 FY27, and management now expects full-year FY27 revenue to more than double in Photonics SOI from just above USD 100m in FY26. Given how the market is looking and moving, how credible is that growth trajectory for Photonics SOI?

DH [00:02:28]: I believe that growth is coming from two different things. The increase of data centres all over the world, so increased demand of transceivers and high-end transceivers mainly. That's one of the reasons that this market is growing now. There is a second part as well, which is very important also is what we call CPO, which is still in development because for me, it's not a complete mature product, and it's not implemented in data centres so far. A lot of customers are developing a brand-new chip for CPO application, and they need a lot of wafers for that. They are using a lot of wafers for their own development. You have for me two markets in parallel, the standard transceiver market and the CPO market. The demand is increasing, of course, with AI.

00:03:34] You mentioned that CPO [co-packaged optics] are still in development. When do you think CPO demand will really ramp up?

DH [00:04:02]: In fact, technically speaking, it's a complicated integration. The complication is not on Soitec's side. It's more on the chip manufacturer side because you need to integrate all the optics directly on the standard chip substrate, which is a bit more complicated. It's a kind of heterogeneous integration, which is not straightforward. A lot of companies are working on it for more than 10 years now. It becomes more and more mature. I guess it will be ready on the market in the coming probably one or two years from now. This kind of design will help to save power consumption in the data centre, mainly, will lower the cost of the full system and will improve the rack density in the data centre. It's a real improvement, which may pop up in the next two years.

SM [00:05:19]: Soitec's management mentioned three key areas – pluggable transceivers, CPO and NPO [near-packaged optics]. Could you provide some colour on the role NPO plays for Soitec?

DH [00:05:40]: I used to work mainly on the transceiver and on the CPO and NPO, I'm not really familiar with it today.

00:05:51] If we compare the longer-term growth opportunity over the next five years for Soitec's transceivers vs CPO, will both accelerate at a similar pace, or do you think demand will be stronger for one?

DH [00:06:21]: I think both will exist at the same time. CPO chips will be more expensive at the beginning. It will bring a lot of better quality and a lot of improvement in terms of performance for all the integrators inside the data centre. Both will coexist in the coming years. Today, on the market, you can find low-cost transceivers made without SOI, in fact, because for very low cost and low performance, you do not need SOI. The transceiver market addressed by Soitec today, it's for the high-performance transceiver mainly. Low-cost transceivers still exist on the market now, even if we have more and more data to manage in data centres and so on and so on. Yes, I guess there will be enough room for CPO and transceivers in the future, of course.

00:07:29] Will Soitec eventually come in and replace all the low-cost transceivers with higher-end ones, or are there two completely separate markets?

DH [00:07:46]: No, I think it's a matter of cost because a low-cost transceiver today is 6-7x lower in cost compared to a high-end one. There are still some data centres that still do not need to manage a huge amount of data compared to other ones and some functionality also in data centres that do not require a large amount of data. I guess Soitec will only focus and is focusing mainly on the high-end part. Today, the high-end part is addressed mainly with a 12-inch, so 300mm substrate made by Soitec. The 200mm are showing a bit lower performance. It's due to the 200mm vs 300mm processes, which are a bit different. For Soitec and for the competition, it's the same.

00:08:59] Is the market ready to adopt Soitec's high-end work on the 300mm side, or is it still comfortable and more established on the 200mm side? Is it a no-brainer for a customer to go for the company's high-end offering?

DH [00:09:35]: In fact, if you have a look at the world leader on the transceiver and the world leader is Intel in terms of quality and performance. To achieve the performance they are offering today, they need very high-end substrate. They cannot do with a standard and regular technology of transceiver like the 200mm. They need very high-end processes and very high-quality supply. Yes, there is a market for it. Soitec wafers are already adopted by many different customers today for their high-end integration and high-performance integration. I'm really confident on the future that it will grow. In terms of competition, we can talk about the competition, it will be very difficult to beat what Soitec can offer today. I'm not talking only about performance because the level of performance from the competition can be equivalent to what Soitec is offering, but one of the key advantages of Soitec is they can offer high quality in very high quantity. They have this capacity to offer very high quality of substrate even in high-volume manufacturing. That makes a difference, in fact.

00:11:23] Who would you consider to be the most direct competitors or alternatives to Soitec on the market?

DH [00:11:34]: Today, the main competitor for silicon photonics is Shin-Etsu, SEH. GlobalWafers is trying to rise a bit as well. Today, the main advantage of Soitec is just what I explained before is that there is no limitation in terms of quantity. If a customer would like to have, I don't know, 200 wafers or 2,000 wafers, they can ask Soitec to do it. They can ask Shin-Etsu or they can ask GlobalWafers, but for the future order, if they want to grow in terms of manufacturing and increase the quantity up to 10,000 per year or 50,000 per year or whatever, they need to go to Soitec because the manufacturing capacity of Soitec is today the only one that can offer the same quality in volume. That's one of the main advantages for me of Soitec today is to keep the balance between the quality and the quantity.

SM [00:12:44]: Would you say Soitec is the best-positioned, if not the number one player, on the silicon photonics side?

DH [00:13:01]: Yes.

00:13:11] What growth potential do you see for Soitec? For example, if the CPO qualification phase slows down or gets delayed, does that have a direct negative impact on the company? Is there a lot tied into customers and their qualification and adoption of CPO?

DH [00:13:57]: In fact, I think the manufacturing line is well managed. The equipment that you are using in the manufacturing line to make Photonics SOI or to make RF circuitry and so on are all combined, so are all in common. If you have a decrease in the demand of Photonics SOI, for example, you can increase the manufacturing of RF and so on. You can play with it. Yes, of course, you can have an impact if the Photonics SOI demand is going down, but it can be replaced by other products as well in the Soitec portfolio.

SM [00:14:47]: Is there some flexibility there?

DH [00:14:52]: Yes, there's a lot of flexibility. By the way, they are using two different plants and facilities, one in France and one in Singapore. That helps also a lot to have flexibility on the manufacturing line.

00:15:08] Soitec has two main plants – Bernin and Singapore. Would you say that, at least for the time being, Soitec has enough manufacturing capacity to cater to the large boom in AI infrastructure and consequently, silicon photonics?

DH [00:15:52]: In fact, the capacity today is enough for the demand. They can address everything in Singapore or in France. I guess it's going to be okay. There's a lot of possibility to increase the volume in Singapore as well, which is good. In fact, I will not expect a very big boom also in the photonic market because you just mentioned before that Soitec revenue has been announced to be EUR 100m in FY27, for example, which is already very high, but keep in mind that compared to RF, it's not a consumer market. It's still a professional market. At the end, even if there is a boom, it will be limited in terms of quantity. The demand will still be there. Transceiver is a market that you have to replace the transceiver also every probably three or four years. The market will always be there, but you cannot expect a boom like we had a boom on RF a few years ago, I would say.

00:17:23] Do you see any bottlenecks today for Photonics SOI? It sounds as if Soitec is in a good position with capacity, and it's quite flexible. We discussed the customer qualification side. Is there any particular bottleneck for Photonics SOI, whether on the customer side or the company's?

DH [00:17:58]: For me, the bottleneck or what might slow down the development is CPO maturity, which is today a bit complicated to set up. There's a lot of constraints in the CPO design, in fact. Today, it's not straightforward for the chip manufacturer to make it. For me, that's a bottleneck because when I was working at Soitec even 3-4 years ago, the big players were working on it, on this development. Today, it's still not in high-volume manufacturing. I was expecting to have it in high-volume manufacturing in 2023-24. It's still not there. It's a bit delayed in time. It will come, of course, and for sure, but that's probably the main bottleneck today.

00:19:19] Soitec management has been quite open about the fact that margins should improve as Photonics SOI revenue contribution grows. Could you roughly quantify how much higher the Photonics SOI margin opportunity is vs the historical group margin? Is this a huge bump potentially coming in?

DH [00:20:00]: In terms of margin today, the Photonics SOI wafers are higher compared to the other products, I guess. I'm not sure that they share the margin per product line. I don't know, but my guess, and I'm pretty sure of it, is that the margin is higher on Photonics SOI product compared to RF and POI.

00:20:51] Do you think the Photonics SOI business can maintain high margins for the foreseeable future, or do you think it will eventually become a high-volume but slightly commoditised substrate business? Will margins eventually have to come down to the level of the group or other product lines, or do you think it will always be a bit higher than RF [radio frequency] or POI [piezo on insulator], etc?

DH [00:21:23]: I think it will always remain higher as it is today, but it also depends a lot on the competition because today, I would say the end customers are more forced and the chip manufacturer are forced to work with Soitec because of the quality and because of the quantity they can deliver. The competition cannot really follow for this quantity, but if SEH or GlobalWafers are able to rise in terms of volume, then the competition will come clearly and then the margin will go down, of course. That's more on this part that I believe that the margin could go down on this specific product if the competition is rising. The margin also of the chip manufacturer is very high on this kind of product. That's why they have a lot of interest on it because in terms of quantity compared to other standard chips like imager or whatever, it's very, very low, but the margin is very high on the transceiver and on the CPO and so on.

00:22:51] Considering how Shin-Etsu Handotai and GlobalWafers have performed historically in terms of innovation and ramping up, are you more worried about either catching up to Soitec? Does one of them have a better chance of coming out with something more credible quicker than the other?

DH [00:23:22]: SEH is mainly based in Asia. To date, they can offer very good product. They are also part of the shareholder of Soitec. It's a supplier, it's a customer and it's a shareholder of Soitec at the end. It's a bit complicated situation. They are offering a very good product in terms of quality, and it was a strategy on their side not to grow too fast on the volume, and they can change their mind. GlobalWafers, it's a bit different because some people don't like the unique product offered by Soitec today. I guess the US government is also helping GlobalWafers to rise in terms of quality and quantity to have a bit more autonomy on this kind of substrate. I think there is a lot of support provided to GlobalWafers and more coming from government organisations or public organisations on the other side of the Atlantic. Both can be dangerous, I would say, but not for the same reason, at the end.

00:24:53] Is Soitec better-positioned in silicon photonics vs in transceivers or CPO? I'm assuming that there are a lot of lower-quality offerings on the transceiver side, as we spoke about earlier. Is it the same in CPO as well, or are the low-cost competitors more on the transceiver side, such as the connectors being used?

DH [00:25:22]: I would say that Soitec is probably more positioned on the very high-end and on the CPO compared to the competition, so more on the CPO, yes, because today, the market of the high-end transceiver is not that high because it still remains very expensive and not probably bringing enough added value for the data centre. The market is there, but it's not rising probably as expected and CPO is the next step. I see more the added value of Soitec on the CPO compared to the transceiver.

00:26:36] Do you think there is something Soitec can do to demonstrate added value for its higher price on the transceiver side, or would it need to lower prices to enhance its positioning with high-end transceivers?

DH [00:27:09]: Not an easy question. I don't know, in fact. It's a bit complicated to answer this question. I know that Soitec is really active on the high-end transceiver, but the competition as well. I see more added value on CPO, and I will explain you why. It's just because on the CPO, technically speaking, the customer, the chip manufacturer will have a different technical requirement. The wafer will be a bit different compared to what you use in the transceiver. In the transceiver, as we said, it's a generic design. Everybody is more or less using always the same design. Then you might have a bit of difference in terms of performance because the wafers are a bit better than the other one. That will make the difference.

In the CPO, it's a bit different. You have a special requirement coming from the customer. It means your research and development team and your manufacturing team needs to have flexibility in offering different design and be able to ramp up very quickly on it. For me, that's the main advantage of Soitec today is that this particularity and this knowledge in R&D, how to change all the parameters of the product you're offering and still be able to ramp up very quickly in very high volume. I'm convinced that Soitec can make the difference mainly in the CPO market. I will be more cautious on the transceiver market, I would say.

00:29:39] Do you think there is a risk of chip manufacturers finding alternative solutions? Is there any scenario where they don't go for Soitec, Shin-Etsu or GlobalWafers, but for a different material completely? That would leave Soitec in a tough position. Is there any end market where that is a risk?

DH [00:30:27]: There is an alternative solution existing on the market by using a different material to make Photonics SOI chips like lithium niobate, for example. Today, it's working well, but the difficulty is that it's a heterogeneous integration. It's not 100% compliant with CMOS manufacturing. That's one of the difficulty today. It's just like you make your chip and at the end of the manufacturing, you just pick and place the Photonics SOI device, which is made with another material. Today, you cannot integrate it at wafer-level manufacturing for the main reason that it's very dangerous to use other material like this one in terms of contamination and so on. It will not be fully compliant with the standard silicon manufacturing line. Yes, it does exist, but it's still a bit complicated. I don't know if it's compliant with high-volume manufacturing today, this solution. The solution already exists.

00:31:55] What potential do manufacturers see in lithium niobate? Can it lower costs or improve performance in certain aspects? What's the appeal if lithium niobate works out and is made compliant? What potential advantages do these manufacturers see?

DH [00:32:23]: It can be lower cost and it can be multiple suppliers as well. The competition in the market will be probably higher compared to what you have today in the standard silicon photonic. That may help also to decrease the cost.

SM [00:32:52]: Could you quantify how much cheaper it could be to go for lithium niobate vs standard silicon?

DH [00:33:12]: No, I don't know. I cannot answer these questions because I have no idea of the manufacturing cost of lithium niobate chip, in fact, and all this kind of a pick-and-place integration at the end of the process. I have no idea of it. What I have read so far and what I have learned so far on it is that it was working. It was made on a low volume, and it was working well, but I don't know if it's compliant with high volume. In this case, I have no idea of the potential cost and reduction of cost by using this technology vs standard photonic.

00:33:59] Soitec mentioned that it's going for more multi-year CRAs [capacity reservation agreements] and cash deposits from customers. Historically, in this business area, are CRAs effectively firm demand and locked in, or can customers still technically change the volumes depending on how the optical architectures evolve?

DH [00:34:41]: In fact, the manufacturing of the Photonics SOI wafer is very specific. It's compared to the standard RF or power or whatever, it's a very long process in terms of manufacturing time. There's a lot of constraint compared to the other applications. I understand that Soitec don't want to put too much allocation of the manufacturing line and cannot put all the priority on the Photonics SOI wafer. You cannot have a complete dedication of your manufacturing line for photonic SOI wafer. That's why they are just probably limiting the volume in the fab. They ask customers to book in advance, some quantities and so on because it will be too risky for them to dedicate the full line to photonic wafer. It will be too risky because if they do so and then the customer in two, three months, four months say, "Okay, I don't want this wafer anymore. I cancel my order," and so on. It will be a nightmare to come back on the previous situation. The allocation is probably defined with some constraint in the fab, but just to be able to manufacture all kinds of product at the same time. They cannot have a dedicated fab for photonic SOI.

00:36:30] Industry reports suggest that Soitec is really the only player in industry-standard Photonics SOI wafers, although Shin-Etsu and GlobalWafers may be working towards them. Do you think it will be very hard for a new player to come into this field for the foreseeable future? Are there very high barriers to entry because the specifications vary so much by customer?

DH [00:37:22]: Yes, I guess there's a lot of technical barriers, in fact, to reach the same level of performance. I do not see any other competition rising in the coming 3-4 years. It required a lot of investment in terms of footprint, also all the equipment in the manufacturing line and so on. There is a know-how on the Smart Cut process, which is very, very specific. It's not straightforward. I don't think so. Apart from GlobalWafers and SEH, I will say that nobody can come up with an equivalent solution, I mean, Photonics SOI solution in the coming 3-4 years. I don't think so.

SM [00:38:27]: Would you say Soitec has good positioning then?

DH [00:38:39]: Yes, definitely.

00:39:10] Soitec secured a long-term POI supply agreement with Skyworks in March 2026, so clearly some good traction there. How attractive is the POI opportunity for Soitec in the next few years, whether from a revenue or margin perspective?

DH [00:39:50]: For me, POI market is very specific within Soitec, and it's a completely different business strategy compared to the other one. Today, two different things. With Photonics SOI, for example, or RF or whatever, Soitec supply a substrate. This is a base of a very complicated circuit that will be deposited above and manufactured above. The substrate is very important. I think it's something like 10% or 5% of the end product in terms of added value. All the added value will come from the design of the chip manufacturer. The design and the quality of the circuitry and so on. A good substrate is helping a lot, but not making all the performance of the end product. In POI, it's a bit different. The circuit which will come on the POI, it's a very basic and very simple one. Just to give you some example on photonics or on RF, a chip manufacturer will make something like 30-40 levels of photolithography on top of it. It's a lot of added value of design on top of it. With the POI, it will be something like three or four. More than 10x less. All the added value is coming from the substrate in case of the POI. The substrate become more and more important in the final product in this specific case.

It's two different things. The main objective of the POI is to make filter. RF filter, it can be GPS filter, it can be a different kind of filter. The idea of using POI is to have a better performance at a lower cost, which is completely the opposite of Photonics SOI today because Photonics SOI at Soitec, the objective is to make very high-end and high-performance wafer to reach the best level of performance on the market. For me, I see it completely different in terms of market and in terms of customer and so on. This kind of filter manufactured on POI are very, very efficient. I know there is an agreement with Skyworks, but there is also a lot of different customers for POI. One of the really, really good advantages of Soitec on this is that they can create really adaptive materials for this new kind of filter. As part of Soitec now, you have research and development teams that are able to make design of filter, which is a unique position within the company.

Soitec is able to evaluate its own substrate by going to the end product at the end, if you see what I mean, which is absolutely not possible on the other hand. It's nice added value. I think POI has a great future in the coming years. I will not say that margin will be very high because the objective of POI is to address, as I told you, low-cost market, but in terms of volume, it can be very high. It addresses today the consumer market because Skyworks is for cell phone application. In terms of application, you have a cell phone, but you have also the glasses, for example, just like the Meta glasses and so on. Every time you have a radio frequency communication between objects, you need a filter. The potential market is very huge. Yes, POI is very attractive, and there is a big potential of growth on it. I would say more on the quantity compared to the margin.

SM [00:45:14]: It seems that the margin side can be handled by photonics.

DH [00:45:24]: Yes, exactly. You need to find the balance within the company.

00:45:27] What does competition look like on the POI side? How well-placed is Soitec to get volume?

DH [00:45:49]: I do not recall exactly the name of the competition, but there's a lot of competition in Asia as well. There are some people that try to manufacture by themselves. Chip manufacturers try to manufacture by themselves the POI wafer to make it at a lower cost. I think that, again, one of the main advantages of Soitec is that every time they develop a new stack or a new wafer design or whatever, it can come very quickly in volume afterwards. That means once you have made some samples to be evaluated by a dedicated customer, very quickly after the validation of the customer, you can ramp up in volume. Same strategy, I would say, of Soitec company, always the same. It's one of the advantages of Soitec to become a leader on the POI, even if there is a lot of competition mainly in Asia today.

00:47:08] How would you rank the growth trajectories of Photonics SOI, POI and RF SOI over the next 2-3 years, just from a pure growth numbers perspective? Which do you think has the most vs least potential for Soitec?

DH [00:47:41]: It's not really obvious to say because RF and POI are really linked to the cell phone market today because this is the main market, I would say it's larger than only the cell phone. The cell phone is the main driver today in terms of quantity. This market is going down over the last few years. Probably other connected object will come in the market soon, like we see the glasses and so on. In terms of growth, in terms of quantity, I see more Photonics SOI growing in terms of quantity and in terms of revenue, but today, if I'm not mistaken, RF SOI revenue is still higher than general POI and photonic SOI. I do not see Photonics SOI revenue doubling or tripling in the coming 3-4 years. I think RF will still remain at a very high level. There will be a growth more on Photonics SOI, but I'm not sure that in terms of revenue, it will reach the revenue that you used to have on RF SOI in the past because in the past, Soitec used to do something like EUR 500m per year only on RF. It will be very difficult for POI or silicon photonics to reach this level without being in the consumer market like a cell phone market. Today, for RF, I would say, all the cell phones in the world, they need Soitec wafer inside. They need it, in fact, it's mandatory because what has been developed by Soitec is unique for the RF filter.

SM [00:49:58]: I think RF SOI has such scale right now from the past. It might be difficult for photonics to contribute more revenue than RF in the next 2-3 years, but it might be able to in the long term, depending on how the mobile market goes.

00:49:58] You mentioned that Soitec has pretty good capacity levels, but if it needed more, how easy or difficult is it to reallocate existing capacity from a lower-demand area to a higher-demand area, such as silicon photonics? Is that easy, or is it a 3-6-month process?

DH [00:51:12]: In fact, what I really appreciated when I was at Soitec is that there's a lot of flexibility in terms of manufacturing mainly. It all depends on the market demand and the market request. A few years ago, Soitec has decided to spend a bit more money on the development and on the manufacturing line to make silicon carbide wafer. We did not mention that part today, but finally, at the end, it was not a success because there's a lot of competition and the product was not mature enough and not really for the market. I was really impressed by the flexibility of Soitec to create a manufacturing line in, I would say, a few months' time. They were able to set up a new building and a new manufacturing line in at least one year or something like this with new clean room and so on because they have decided to put all the priority on it.

Finally, the market was not there. It was probably the wrong strategy and so on. We can blame them or not. Sometimes you don't know what the market will be and what the competition will be in 3-4 years' time. It was not a success, but it was amazing their capacity to create a line. If there is a very high demand in silicon photonics, I'm sure they will be able to grow their manufacturing lines, can be in Singapore, can be in France. Both are available, but they will be very, very active on it as they always are.

SM [00:53:25]: That was a great conversation, incredibly helpful. I think that goes without saying. It was a pleasure speaking with you. I hope you enjoyed it, too.

DH [00:54:13]: Yes, it was very nice, very open discussion. Thank you very much for that.

Transcription ends at 00:54:31 of the recorded material