← 返回

2026.10.05 | TB | Memory | Swissbit Industrial Memory & AI Tailwinds

原标题：Swissbit – Industrial Memory Demand, AI Tailwinds & OEM Commercial Strategy

#memory

Third Bridge · 2026-10-05

Analyst-led • 1 Oct 2026

Companies: 5289 TW, A005930 KR, MU US, A000660 KR

Steffen Allert (SA) Independent Consultant

Louis Knight (LK) — Third Bridge Sector Analyst

Agenda

Key Insights

Introduction

LK [00:00:02]: Hello, and welcome to this Third Bridge Interview entitled Swissbit – Industrial Memory Demand, AI Tailwinds & OEM Commercial Strategy. My name is Louis Knight. I will be facilitating today's Interview with Mr Steffen Allert, former Director of Sales Security Solutions at Swissbit. Steffen, thank you for joining. Welcome to today's call.

Before we begin, please state I agree or I disagree to this following sentence: 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.

SA [00:00:52]: Yes, I agree, and I will not distribute any information which is not publicly available.

LK [00:00:57]: Could you begin with 60 seconds on your relevant experience, key responsibilities at Swissbit and wider background, please?

SA [00:01:09]: I was working for Hyperstone until 2020, until the time when Hyperstone was acquired by Swissbit. I was the right VP of Global Sales for Hyperstone for the flash controller sales. When Swissbit acquired us in beginning of '21, I was heading the integration of the sales and marketing team into the Swissbit organisation until 2023. In end of 2023, I took over the global responsibility for building a newly established specific security sales team within Swissbit. In December last year, December '25, I retired.

LK [00:01:57]: Could you describe what Hyperstone is within the Swissbit portfolio?

SA [00:02:06]: Hyperstone is a flash controller manufacturer. The NAND flash devices, which are manufactured consist of a controller and of a flash and some other smaller parts. The controller is the brain of such a solution. Hyperstone developed and is developing such controllers. Before Hyperstone was acquired by Swissbit, those controllers were also sold to third parties or to other companies besides Swissbit. Hyperstone is a part of such a solution story.

Questions

00:02:46] What would you say are the top three most important strategic questions facing Swissbit in the long term?

SA [00:02:55]: The most important question today is how does the allocation period continue? Will it last for more years? Will it end soon? This is an important part because this will drive the availability of the storage devices. Of course, it's also driving the prices in the market. This is publicly available, for example, on a website like DRAMeXchange, where you can see the price development over the last 1.5-2 years, where the price is multiplied by a factor of three, four or five, depending on when you start that analysis. The most important strategic question by far is how does that continue? This is by far the most important question. All other questions besides that are less important. This is the most important thing to prepare for.

00:03:49] What is your personal opinion on how long Swissbit's allocation period could continue for?

SA [00:03:58]: There are two things which really influence that. It's first, the demand coming from the AI and data centre side, which is absorbing more or less all the volume. If that ends, this allocation ends, this boom ends. The second thing is, of course, the NAND flash suppliers like Kioxia, Micron, Samsung, SK Hynix, Sandisk, they build new fabs. When those fabs come into the game, of course, the number of storage, the flashes will increase. Therefore, the prices, the cost for those flashes might go down because the demand might more balance with the available products. These are the two factors which will influence the continuation of that trend, of that boom, which is publicly available, a build-up or an introduction of a new such NAND flash plant requires 2-3 years to bring that operation up to speed.

You see already new fabs coming into the game. Kioxia just announced that they're going to introduce a newly built fab in Japan end of the year, beginning of next year. This is the first message I heard that the new fab is coming into the game, but I'm sure others will follow. The more products are available, it will more and more balance with the request, with the market itself. If that is balanced, of course, those costs on the NAND flash side will come back. With that cost coming back, the prices will come back definitely, will go south. When does that happen? Nobody knows, but from my perspective, my personal opinion is this will last at least until '28-'29, but when the point in time where that turns is really coming, that's, of course, we would need a crystal ball on that one, but it will come. The question is when. From my perspective, not in the next three years.

00:06:37] Could you outline the various products and services that Swissbit provides?

SA [00:06:43]: Yes, actually, this is available on the website more or less, it is available on the website. It's two main product areas, it's security and it's storage. On the storage device side, it's different interfaces, which are served SD cards, USB, eMMC, PCIe. These are from the technology, more or less the same. It's different on the interface side. It differs on the speed mainly by PCIe being the fastest one. Those devices are used broadly currently in the data centre and the AI area, but it also gets traction in many other markets. This is the most modern standard, so to say. On the security side, it's also visible on the website. It's three main areas. It's storage with security. The storage devices on the memory side are upgraded, call it, with security functionality.

There is a second part, which is the TSE business, which is going into cashier systems as an anti-fraud protection. The third part on the security side is FIDO keys, so multi-factor authentication, hardware-based solution. These are the main areas of product. For more details, I would suggest to visit the website from Swissbit because there's much more detail available.

00:08:25] To what extent is it SaaS vs hardware components that drives Swissbit's sales?

SA [00:08:36]: As you can see on the website, it's hardware devices we are talking about, but each and every hardware device does not run without software.

LK [00:08:48]: Does Swissbit provide the software?

SA [00:08:53]: The software is necessary to run those systems. Therefore, it's necessary. For example, the flash devices, the controller I was talking about, the controller needs a software to manage all that stuff. It cannot only be done in hardware. There's always software involved in that in such devices, but it's built-in software more or less. It's not provided separately.

00:09:16] Does Swissbit provide maintenance services or is it only installation and software?

SA [00:09:24]: Define maintenance. It's not a robot where you need to do a lot of maintenance stuff. Of course, if a problem occurs, but this is nothing special with Swissbit, the companies need to analyse the problem where it comes from and solve it. That's more or less the basic things.

00:09:54] How would a data centre use Swissbit's products and services?

SA [00:09:57]: The data centres, you can divide them into two applications. The primary application, of course, is storing a massive amount of data. We are talking about exabytes, petabytes of data, which are stored in these data centres. These devices are usually served by the large flash suppliers because having such a huge volume of storage data and those guys manufacturing the devices itself, they have a price advantage because they can sell this amount of data coming out of their fab, while others which are not manufacturing those NAND flashes by themselves, they need to buy it from one of those big guys, but there is another application in the data centre, which is managing the whole data centre. It's managing the whole plant.

It's managing the heating system, it's managing the cooling system. It's managing the whole plant. This device does not need petabytes of data. It needs probably a couple of hundred gigabytes or some terabytes, but the main application or the main requirement is the reliability and availability of the service. If those devices fail, the whole data centre might go down. This is definitely different to that what is happening on the mass storage side because on the mass storage side, you have a thing which is called RAID. It's a redundancy solution. More or less, we explain it in easy words, it's a mirroring of data.

You don't store the data once, you store it several times, two times or three times. If a device in that application fails, nobody cares because the data is still on other drives. You simply exchange it and throw the device away. This is different on the boot device side. On the boot device side, if such a device fails, of course, it's also built redundant, but if those devices fail, it has a much higher implication on the service level of the data centre. It's two applications. One application is mass storage. Quality is not such an issue. Availability is not such an issue. On the other side, boot drives, availability is an issue, definitely.

00:12:41] Generative AI LLMs [large language models] are trained by CPUs, which are dense racks of computer chips using memory, but Swissbit chips are the NAND flash controllers, which control how the data is written and stored, more relevant to how the process operates than the CPU stacks that you see in a data centre. Is that correct?

SA [00:13:16]: The CPU is one thing and the data storage is another thing. The data storage device, they all function the same way. They receive data from a CPU and they store it. It's as simple as that. If the CPU asks for the data, the devices, the storage devices need to supply the right data to the CPU. All those devices operate the same. It's the same for the mass storage I was talking about, but also for the boot device. The CPU ask for something and the flash storage device needs to provide the data, the right data.

LK [00:13:56]: You mentioned that there is a new fab being built for these NAND flash controllers. Is that correct?

SA [00:14:06]: I haven't seen the fab yet, but it's a press release from Kioxia.

00:14:15] When Swissbit talks about the nanometres of the chip itself, is that how you would also measure the NAND flash controllers, the storage chips, or is it only for CPUs?

SA [00:14:32]: The nanometres is the manufacturing technology. If we talk 5nm, 3nm, 7nm, whatever it is, this is the usual measurement for the newer gen (inaudible 14.48). On that, you can probably make up the generation of the chips, the regular chips, I call it, I'd say regular chips, CPUs and that stuff. On the NAND flash side, it's more about the level. We talk about 3D NAND and the 3D NAND works in three dimensions. It's not a planar solution, it's a 3D solution. There, it's about the levels which are available. It's not about the nanometres at the end of the day.

LK [00:15:28]: It's a different measurement, measured in a three-dimensional way. Is that right?

SA [00:15:33]: Yes, yes. It's a different measurement, but it's there, on the nanometre side, the more advanced it is, the lower is the number. 3nm is smaller and I don't want to call it more modern, but more state-of-the-art than 7nm. If you talk about the NAND flashes, the higher the number, the better it is, but if they talk about 300 levels, this is higher density than 200 levels. That way (talking over each other 16.06) Swissbit the numbers are different.

00:16:09] Is Swissbit operating at the most state-of-the-art level of NAND flash, 3D NAND with 300 layers, or is it operating in the mid or low tier?

SA [00:16:24]: Actually, I don't want to comment on that because this is not public information.

00:16:40] What impact do you see the new NAND flash fab you mentioned having on the market? Is it a significant additional supply that could impact pricing, or is it just an example of how rapidly the market is expanding?

SA [00:17:00]: Actually, a single new fab will not change the overall picture, but if all of those participants, the large participants, I mentioned those guys, each one adds more, one or two or three fabs, then it will change the overall picture, but introducing a new fab will probably release it a little bit, but it will not have a significant impact from my perspective, my personal opinion because the demand is so high, which cannot be fulfilled today that one fab might release a little bit the pressure in the market, but it will not significantly change the overall picture. As an example, if 10 fabs come into the game, that changes the picture.

LK [00:17:52]: How many NAND flash fabs are there in the world? 15-20?

SA [00:18:02]: Actually, I don't know. Each and every number I tell you is wrong. I don't know. It's also a question of how you measure those plants, because in a single house, in a single plant, there can be one line, there can be 10 lines. I don't know. At the end of the day, you need to calculate the manufacturing lines and not the fabs itself. There can be a fab which has two lines, and there can be fabs where there are 10 lines, and that's quite a difference. I don't know.

00:18:40] What is a general market CAGR for Swissbit or for the market? Could you provide a personal estimate of a sustainable organic YoY sales growth rate in the market for, say, five years?

SA [00:18:58]: If we focus on the key markets for companies like Swissbit or ATP or Virtium, they are all operating more or less in a similar area. I call it industrial, but there is much more than only industrial automation. It's not only factory automation. When I talk about industrial, I also talk about defence, I talk about medical, I talk about Netcom base stations, mobility like trucks and trains. These markets are growing continuously over the years, and I expect them to grow also over the next years because automation is increasing, but the shift between the market itself is important. For example, defence was not an interesting market five years ago, was low volume, high expectation, was not really interesting, but now defence is getting much more traction. Not only because the traditional military equipment is increasing being built, it's especially on the drone side because if you just take, for example, a main battle tank, the main battle tank, maybe there are 1,000 sold over one year.

Even if there are 10 storage devices in it's 10,000 systems. It's not a big deal, but if you talk about drones, it's more or less a thing which you send out and then it never comes back. Those drones are sent out in the level of thousands, for example, in the Ukraine war. Each and every day, there are hundreds of drones which are sent out, which never return. It's much higher volume on that. The situation itself is not good having such a war, but in that respect, the volume on that side is increasing significantly, while the automotive area might suffer a little currently. This is where the shift in the markets is a fact, but the overall markets are growing. Some of these markets are growing.

LK [00:21:15]: Could you estimate a CAGR figure? Would you say double-digit YoY sales growth?

SA [00:21:21]: My personal assumption. It's nothing which I have reviewed in the web or which is based on a thorough analysis. I would say it goes between 5% and 15% per year, but this is a rough estimate. It goes up. The trend is upwards, but there are cycles, so up cycles and down cycles, of course.

00:21:46] Swissbit's key growth drivers include defence autonomy, with kamikaze drones and main battle tanks, which have a lot more memory and digitalisation, etc, and much higher volumes, given NATO spending priorities. The company also has data centres because the CPU chips need the memory and storage to operate. It also has slower-growth areas, but defence autonomy and data centres are the two most important. Is that correct?

SA [00:22:22]: I would say, yes. Data centre definitely. On the defence side, on the industrial side, I would say currently, the drone market is definitely a very interesting market because it's growing. Every day, there are new drone manufacturers, new drone developments announced coming up, coming into the game. There is no end of that story from my perspective because just seeing it in Germany. The German military, the Bundeswehr, maybe that's a little bit to stick, but they almost don't have drones. They need to invest in drones significantly. It's not only the military, it's also the police, for example. This is definitely an area which is interesting, definitely. This is, by the way, not only true for Swissbit. My market assumption in general. Which of these companies benefit from that? That's another story. This is about the strategy of each company in that area.

00:23:35] How many NAND flash chips do you need for every one CPU that will control a device such as a drone or a household item?

SA [00:23:50]: Don't call it NAND flash, let's say storage device. I'd say one at least. You need one because you need to store the operating system. You need to store some operational data. Therefore, you need at least one storage device. Now it's about the complexity of the system. We are talking about main battle tank. If we talk about main battle tank, it has a fire control. It has a motion control. It has whatever. There are definitely more than one. In a drone, there is usually only one storage device.

LK [00:24:27]: Typically, you would have one CPU and then you'd have the storage device components with it. Is that right?

SA [00:24:36]: Yes. Yes, you could say it that way because when we just talked about the tank, there is also more than one CPU. There's usually one storage attached to each CPU. There are other specifications or system designs where you have a huge central storage, which serves more than one CPU. This can also happen, but this is a matter of the system design of the system provider. It's not a matter for the storage provider.

00:25:14] EVs have many CPUs nowadays with a lot of semiconductor chips. They must need the same level of storage devices. Is this an opportunity for Swissbit?

SA [00:25:28]: The automotive market is an opportunity for everybody, but it's not new with the electric cars. The electric cars might require a little more of that, but even the cars with the combustion engine, at least the current solution, they are also equipped with many, many, many electronic devices because they also are working towards autonomous driving with all the sensors, with all the automation in that. The electric vehicles might have one or two more, but in general, each and every car has a lot of these devices in it. I know from a couple of years ago, this is combustion engine cars, usually, there are at least 10-15 of these devices in each car.

Maybe the electric vehicles do not have 10-15, but maybe 15-18 in that, but on the other hand, an electric car, for example, does not need a motor control and engine control because it's simply an electromotor, which is easier to monitor and manage than a combustion engine. This might even reduce the complexity of that, requiring less electronic equipment, but actually, I don't know the details of the electric cars, but I would say there's not a huge change because of moving from combustion to electric, the number is similar. My expectation.

00:27:10] How easily do you think a company such as Swissbit can transition from its core industrial, automotive and netcom niches into the hyper-competitive data centre and cloud storage market, with boot devices?

SA [00:27:25]: As I said, on the mass storage side, I would say this is not a market for smaller companies like a Swissbit or Innodisk or ATP or Virtium, but the boot device in such a data centre is a market because this boot device requires the quality and the availability of a storage device, which is required in the traditional markets of these companies. We talk about availability. We talk about long-term availability. We talk about reliability, which is not so much the case on the mass storage side. As I said before, if a mass storage fails, nobody cares, simply exchange it, but if a boot device fails, this might cause a real problem. Data centre is a market, but in a niche application. The niche application is boot device.

00:28:33] OpenAI, Microsoft, Google and other companies are training super intelligent LLMs right at the edge of innovation with AI. Is this application too complex for Swissbit to capitalise on?

SA [00:28:57]: At the end of the day, an AI data centre, and I would call it the traditional data centre is not differing very much. There's mass storage, petabytes, exabytes of data and there is a boot device. Just summarise it under data centre. If it runs a large language model or if it just stores your and my data, this is not much of an issue. The issue on the AI data centre is that you need higher speed so that the response time is high, but in that respect, we do not talk about the NAND flash. We more or less talk about the HBM memory, the high-bandwidth memory. This is completely different technology. High-bandwidth memory is volatile. If the power goes away, the data is gone. It's like DRAM. It's more or less a fast DRAM.

It cannot be compared with the NAND. In the NAND device, you store data continuously, non-volatile. If you store some data there, it stays there. If you store data in the HBM, if the power goes away, it's gone. Therefore, to speed up the response time of such a data centre and AI data centre, HBM is more of an issue than the NAND.

LK [00:30:22]: Does Swissbit provide the HBM [high-bandwidth memory]?

SA [00:30:25]: No. HBM is like a DRAM. It is supplied by guys, for example, like SK Hynix. I would say SK Hynix is the market leader in HBM.

00:30:39] Would Swissbit provide the same technology to hyperscalers and LLMs – which are expanding the most rapidly today – that it would supply to a very simple data centre, such as for colocation and storing memory for, say, consumer photos?

SA [00:31:01]: In general words, yes, but always the devil lies in the detail. There might be some different requirements. Let's just make one example. If you supply a data storage device into a real industrial application, you talk about a temperature range of minus-40 to plus-85 or even 105 degrees. You can operate at any of those temperatures. In the data centre, everything is cooled, heated, so you have a constant temperature more or less. The requirement in that respect is less than in the real industrial application because there is simply not that temperature variation because all runs at room temperature, 20-25 degrees. There's not that variation between minus-40 and plus-85, for example. In that respect, it makes this application easier, but in other applications, it's similar. This is general technology. This is nothing to do with any of the suppliers there. It's simply the environment.

00:32:17] Can Swissbit scale its cybersecurity and fiscalisation products – such as FIDO2 [Fast IDentity Online 2] authentication keys, microSD, security tokens and encrypted storage – into broader global enterprise and public sector markets?

SA [00:32:34]: The answer is always yes. You have to take the right decisions. I'm not taking the decisions for Swissbit. Let's phrase it in a way, the products are definitely interesting. Otherwise, Swissbit would not offer them. I think this is nothing special about what I say there. The anti-fraud protection for the cashier systems is getting more and more traction in Europe. In Germany, we have the strictest rules for that. Therefore, the highest need or the need for higher security in that area. Other countries in Europe were not that strict in the implementation of the EU law, but they are catching up with Germany. They are going towards the German requirements. This means the standard anti-fraud protection measures might no longer be sufficient, which opens up potential new markets, of course, but this is true for Swissbit, but also for the competitors. What trend is also obvious, and this can be also seen and read in publications, it's going away from hardware-based TSE to cloud TSE. The security measures are going from a local hardware device into the cloud. This is a trend. This makes the rollout easier, but it also opens up more competition on that side because building a hardware device is more complex than realise these applications or these heavy application in software.

00:34:31] How important is competitive differentiation in Swissbit's markets, given that it is largely supply-driven and supply-constrained? Does competition matter at all, if everybody is sold out because of the capacity shortage?

SA [00:34:52]: Competition is always an issue. There is no market where there is no competition. I would say on the memory device, currently, it's not so much about competition. It's more about if you have enough basic material to supply the number of products your customers are asking for. This has nothing to do with Swissbit. This is the overall general situation in the market. Currently, it's not about competition in a classic way. It's more about, can you deliver? Are you able to deliver the products? This is the most important question today for everybody in the market. This is not specific for somebody. It's specific for everybody. Do you get enough flashes to supply your products? That's the key question.

LK [00:35:57]: Would you say all of the flash capacity is sold out?

SA [00:36:04]: Yes.

00:36:05] What is the competitive differentiation for Swissbit? Can you go to a customer and say you have capacity and the customer will then choose you, or is there any other competitive differentiation?

SA [00:36:24]: Of course, in the traditional markets, in these high-end markets, you still need to supply those products with the quality, which is expected by the companies in that area. If we talk about consumer, the quality expectation is much different to an industrial application. If you have a USB stick and that USB stick fails, and your pictures from your last vacation are gone, you might not be happy, but you will not sue the supplier of the USB stick. You will just not be happy and maybe buy another USB stick from another supplier, but on the industrial arena, let's take a manufacturing plant of a car OEM. If a device fails there and the whole production plant goes down for one day, that's quite a cost. They will not say, "Yes, it just happened, I don't care."

No, they, first of all, want to know what happened. Why did it happen? What did you do to avoid that? What are you planning to implement that this does not happen again. This has a completely different implication at the customer compared to the consumer side. You still need to supply the products in the quality, which they expect before the allocation, but they also expect, of course, the quality now during allocation. The standard requirements, high quality, availability, reliability is still there, but what comes additionally into the game is, can you deliver? Do you have the product available? Can you really send out the product to your customer in the volume they need it? The standard requirements are they still there, but there is a new requirement, can you deliver?

00:38:21] How does one player within the market perform better than another competitor in securing more supply?

SA [00:38:35]: It depends on the relationship you have with the flash suppliers. Of course, it depends on the contracts you have to place with the guy. At the end of the day, this is a question of purchasing. Are you an established and reliable partner of one of these flash suppliers? Do they really honour your reliability, your long-term relationship with them or do they simply say, "I send it to those guys who pay the highest price"? This differs from one supplier to the other. I don't want to mention names here, but there are suppliers which really sell just to the guys paying the highest price. There are other companies, other NAND flash suppliers, which think a little bit ahead, because this allocation will come to an end. Then this seller's market will go back to a buyer's market. At that point in time, somebody might remember that they were not treated well by one of those guys. Therefore, it depends on that. It depends on the set-up on the NAND flash supplier side, but it also depends on the relationship, which have been established over years by the NAND flash storage manufacturers like Swissbit, Virtium, Innodisk, ATP.

LK [00:40:12]: Who would you say are Swissbit's closest three competitors?

SA [00:40:16]: I just mentioned those.

00:40:20] How do you think Swissbit differentiates itself from its competitors?

SA [00:40:29]: First of all, they all make great products. Technically, I would say they do not differ that much because at the end of the day, they more or less use all the same NAND flashes because there are only 4-5 NAND flash manufacturers available in the world. You cannot differentiate very much on that side. The controllers, they also need to cope with specifications provided from the host side. You cannot deviate on a PCIe specification because it has to function in each and every PCIe environment. Therefore, there are some possibilities to adopt some parameters in the firmware of such a device for specific customers, but this is very limited, I would say. The real differentiation is the customer access. Do you really have access to the decision-makers and the customers? Are you in the discussion early? At the end of the day, the decisive factor is that you know early enough about new developments because if you know early enough about new developments, you might be able to adopt your product offering according to that.

Of course, you can't do that for a customer buying 10 devices. I'm talking about the large guys buying 100,000 or one million pieces per year. Those guys will drive anyhow the market itself. All the smaller manufacturers will follow these big ones because they cannot differentiate and implement a new technology. Therefore, if you are close to a customer and see to one of these big customers, then you know early about new developments, about new CPUs they're going to use. The CPU at the end defines what will be used on the storage side. Just an example, 10 years ago, SATA was the interface of choice. All CPUs, AMD, Intel, they all provided CPUs with the SATA interface. This changed. You will find no more Intel CPU or AMD CPU with the SATA interface currently today. They all switched to PCIe, NVMe.

Therefore, you needed to know that your customer is switching from the old-generation SATA CPU to the new-generation PCIe CPU early enough that you are able to cope with that change so that you are able to develop your product, which fits to that new decision of your customer. Therefore, from my perspective, the most important thing is being close to your customer, being a reliable supplier. There's always a point where something goes wrong, where a device fails, where a customer of your customer is not happy. This is where the customer relationship is defined. If you answer fast, you answer efficient, you provide the right information on the right time and do the countermeasures to avoid these failures in the future, this is where you differentiate.

00:44:09] How strong is Swissbit's strategy to secure supply and maintain supplier relationships better than its peers?

SA [00:44:23]: I don't know details on that one. I would say all of those market participants are aware of that. How they do that in detail, I simply don't know, but everybody knows in the market what I just said. It's nothing new. Some might be able to do it better. This is a regional thing. Swissbit being a European company, of course, is best-positioned in Europe, while an Innodisk or ATP is best positioned in Asia and a Virtium is best positioned in the US, simply being local.

LK [00:45:05]: Swissbit's Berlin manufacturing facility serves as a moat against non-European competitors in an era of supply chain regionalisation and regulation such as the European Chips Act. Is that right?

SA [00:45:51]: The EU Chips Act does not push the non-European suppliers out, but of course, it prefers European solutions. Digital sovereignty is an issue, definitely. Of course, the EU Chips Act is not limiting competition, I would say, because this would not be legal. The World Trade Organization would not allow that, but what it says is we need to take care that we are not completely independent from non-European suppliers. In that respect, each and every company working in Europe, having plants, having development teams, R&D teams in Europe have an advantage. Therefore, of course, companies like a Swissbit or Infineon or STMicro, NXP, they have an edge over non-European companies, because simply being European, having R&D teams here, having the competence here, manufacturing these products in Europe, this gives them an advantage in the European market, definitely, yes.

00:47:05] How well can Swissbit defend its premium pricing and high margins when NAND flash memory prices suffer severe global down cycles?

SA [00:47:20]: That's a tough question. You have to somehow to cope with that. You have to be close with your customer. As I said before, it's very important that you are close to your customer, that your customer is aware of what you are supplying. They will not switch suppliers without a very important reason. An important reason could be that your products don't work as expected or your prices are much too high than compared to the competition, but when that happens at which level, that's a question. That differs from one company to the other, from one market to the other. There is no easy answer to that, but the solution is being close to your customer to see and to hear what's going on at the customer. Is new competition coming up? Are they about to qualify a competitor of you? You need to be close to your customer. This is what helps you to stay into the game into a significant business with any customer when the competition increases. Again, it's customer relationship from my perspective.

00:48:44] Is it true that, when selling NAND flash memory, the typical unit is a 30.72TB solid-state drive?

SA [00:49:13]: No. Usually in former times, it was always a multiple of two. It was 2GB, 4GB, 8GB, 16GB, 32GB. Now, this is no longer two because there are things in such devices which reduce the available memory for you as a user, just to provide additional quality of service, having a higher speed because you have more space left to manage the data flow on the device. This is called overprovisioning. For example, you have a 64GB device and you say it's a 60GB device. Then you as a customer, only get access to 60GB and the device is 4GB still left in addition to that. With this 4GB in addition, you can increase speed, you can increase availability, you can increase reliability. The more you provide that overprovisioning, the better it is for your management, but the less data you can store in it as a user. Therefore, you cannot say 30TB or 30GB is better than 60GB or 64GB. There, you really need to go into the details. You need to see how they manage that. What is the technology behind it?

LK [00:50:41]: I read that for data centres and enterprise IT, 30.72TB is a standard commercial capacity size for high-density SSDs [solid-state drives], such as from Samsung, Solidigm or Kioxia. Would you not agree?

SA [00:50:58]: I don't know why it should be 30.7TB and not 30TB or 30.8TB or 31TB. If this is a standard, it's a standard. It's not really a standard. It's just a number.

00:51:09] What do you think about NAND pricing? What is the unit of measurement that we should be thinking about – petabytes or terabytes?

SA [00:51:21]: It's usually dollars per terabyte, per gigabyte user area. If you have USD 1 per 1TB, this would be a measure. This would be an indication. This is not a realistic figure, by the way. It's just an example to make it easy.

LK [00:51:50]: What is a realistic NAND flash price range, not specific to Swissbit, but generally for the industry?

SA [00:51:58]: Actually, I do not know because the prices are changing so much and they went up significantly. Actually, it's pretty easy to follow up. It's also on DRAMeXchange. This is a very good source of information. I'd say, two years ago, USD 0.07 per gigabyte was a rough estimate, but this is years ago. Today, it might be USD 40 per gigabyte. I don't know, you can look that up in DRAMeXchange because it's just a number. To have the realistic number, you can just go to DRAMeXchange and see how much a, let's say, 32GB drive costs today. Then you see approximately because there is still the controller in it and the manufacturing, but you also get raw NAND prices there. It's a spot market at the end of the day. There, you can calculate that actually. This is really a spot market. It changes every second. Significantly sometimes.

00:53:10] How sustainable is Swissbit's roughly EUR 250m in annual EBITDA through memory market fluctuations over the next five years?

SA [00:53:28]: No comment. Sorry.

00:53:34] What do you expect from NAND flash pricing over the next 2-3 years?

SA [00:53:41]: From my perspective, it will stay as high as it is today because I don't see the end of the allocation. The price will only change when the allocation ends and the allocation ends when there is more supply or less demand, but from my perspective, over the next three years, there will not be a huge change on that one. It is still a seller's market, and it will stay a seller's market for the next three years, my assumption. Maybe it's only 2.5 years, maybe it's five, but from my perspective, I would say three years is a safe assumption.

00:54:17] What happens if there's geopolitical de-escalation in Ukraine and the Middle East so defence expenditure pulls back, particularly across Germany, France and the UK? At the same time, what if data centre demand sees a drop off, so all the CAPEX from Meta, Alphabet, Google, OpenAI, Anthropic, etc, dies down because data centres become less of a huge global phenomenon? What could happen to Swissbit in five years?

SA [00:54:57]: Not only Swissbit. This will be a bloody mess. Of course. What happens to each and every company, don't know, but of course, from my perspective, and this is not only true for Swissbit, this is true for everybody in the market. This is even true for the big guys like Micron or Kioxia or Samsung. Maybe Samsung is not that much depending on it because they are a conglomerate, though they offer much more than NAND flashes, but for the simple NAND suppliers and HBM suppliers like SK Hynix or Kioxia or Micron or Sandisk, those guys will also suffer significantly. Everybody in that market will suffer significantly when that comes to an end. You need to prepare for that. That means, for example, the big manufacturing guys like Micron or Kioxia, if they now invest, let's say, and each of those fabs costs, I don't know, USD 10bn-15bn, to build such a fab. If they build 10 of those fabs, everybody, everybody invests USD 150bn in these fabs. The allocation comes to an end. These fabs are not loaded anymore. This will be a tough situation for everybody in the market. Maybe this could be even a chance for the NAND flash manufacturers, the NAND flash storage device manufacturers like ATP or Swissbit or others because then they need to get somehow their devices sold to somebody at any price. They might even benefit from that, but that's the crystal ball.

00:56:45] If Swissbit enters into a crisis mode over the next five years, but for company-specific reasons rather than macro ones, what would be the key reason? Would it be a product fault with major clients? Would it be the disadvantage of being in Europe?

SA [00:57:16]: Again, this is not only Swissbit. I think the key of each and every company is the people. You need to have the right people and you make them and then you have to make them stay. There is a war for talent is the expression for that at the end of the day. With the baby boomer generation like myself going to retire, that might increase. The problem for Europe here might be that our universities and education system is not really prepared for these new challenges. An example, TSMC is investing in Dresden, huge fab, great investment. Great for Germany, great for Dresden, super, but they bring Taiwanese experts to Dresden to make the system run. This is a risk. This is not a risk for Swissbit. This is a risk for Europe. You need to have the right people with the right skillset. I'm not sure that we are prepared for that. There's nothing to do with Swissbit. This is my personal opinion. We need to invest into state-of-the-art education. I'm not sure that we are doing that. That's from my perspective, the biggest risk, having the right people at the right place.

LK [00:58:51]: Are you seeing the talent in Europe move abroad or is it just that the US and east Asia are performing a lot better with regards to developing their own talent?

SA [00:59:09]: I don't think that many talents are moving away. There are some moving away, but there are talents coming, but I think especially in Taiwan and South Korea, Japan and the US are much better positioned in the education of these needed competencies, semiconductor, for example, AI, so expertise in large language models, expertise in generative AI. I think we are, in Europe, not up to speed with that. We need to invest in that. This is my personal comment, sorry. I think the working behaviour of some of the younger generations, sorry to say that, is not that they can cope with a very, very powerful work relationship that they have, for example, in Asia. They work 8-10 hours per day, five days per week. In Germany, the guys coming from university, they talk about work-life balance.

LK [01:00:34]: Let me close by saying thank you so much. Really appreciate it, very insightful Interview. I appreciate that and enjoy the rest of your day.

SA [01:01:06]: Thank you, Louis. It was nice talking to you.

Transcription ends at 01:01:15 of the recorded material