Hi, everyone, and good afternoon. Thank you for joining us for the 3rd day of Needham's 26th Annual Growth Conference. My name is Nick Doyle. I'm part of the semiconductor team here at Needham. It's my pleasure, pleasure to introduce Tower Semiconductor. Tower is a leading foundry for analog semiconductor solutions across many markets, including consumer, industrial, automotive, mobile, infrastructure, medical, and aerospace, and defense. The company owns and co-owns fabs in Israel, U.S., Japan, and Italy. In the summer of last year, Tower and Intel mutually agreed to terminate their $5.4 billion merger after passing the August 15th outside date. Tower received a reverse termination fee of $353 million, which the company is mostly reinvesting into the business. Joining me from the company is Dr. Marco Racanelli, President. Marco is gonna go through a 30-minute company presentation.
We'll use the remaining time to answer as many questions as we can. Marco, thank you for joining us. Please take it away.
Great. Thank you very much, Nick, for the introduction, and thank you, everyone, for the interest in the company. My name is Marco Racanelli, President of Tower. What we'll talk about today is the next big step for Tower, and I'll kind of spoil the surprise. The next big step is really scale. It's scaling up the business and creating an improved and better business model. We'll show our long-term business model for the company and for our investors. So before I start, standard safe harbor provision regarding forward-looking statements that we may make during the presentation containing some level of risk. Let me begin by simply introducing the company. For those of you that are not familiar with Tower Semi, we are a pure-play foundry, meaning that we manufacture semiconductor components for our customers.
We have over 300 customers globally, but we don't market products ourselves. We manufacture those for our customers. We focus entirely on analog technology. The focus is on RF, and you can see the breakout of our Q3 revenue split, Q3 2023, the last quarter that we reported publicly. The RF sector is split into two sectors. RF Mobile, primarily handset-driven business. RF SOI is 26% of our corporate revenue. RF infrastructure, we'll talk in more detail what that means, 10% of our revenue. The other major technology we invest and focus on, power management. Power ICs were 27% of our revenue, and power discretes were about 17–19% of discrete and other revenue you see on the pie chart.
The last area of focus is sensors and displays, and that's about 18% of our revenue based on the Q3 run rate. We serve a wide range with these technologies, serve a wide range of end customers and markets, as you can see here, and we have multiple manufacturing sites across three major geographical areas: the U.S., Japan, and Israel, so Western-focused areas, but cross-qualify a lot of our technology for assurance of supply to our customers. So speaking of our fabs and our scale, our increase in scale, that's represented on this slide here. We're adding significant 300-millimeter capacity to our overall capacity fleet, 300-millimeter being the more advanced wafer size that we manufacture today. We're largely in 200-millimeter or 8-inch fabs, with one 300-millimeter factory, and we're adding scale and capacity to our 300-millimeter fleet.
First, the existing factories, before we talk about the expansion, we have two factories in Israel, two factories in the U.S., and two factories in Japan, and only one of those is 300 millimeter, the one factory in Japan, 12-inch or 300 millimeter, meaning that that is the most advanced factory that we have in the company. Focus going forward, add capacity, to sustain growth in those more advanced nodes, and we are now adding capacity in two additional factories, one in Agrate, Italy. This is a deal that we have with STMicro, where we have about a third rights over about a third of a capacity of a new factory in Agrate, and ST has the remaining two-thirds of the factory in Agrate, Italy.
And then the latest agreement that we announced after the Intel, the breakup of the Intel agreement, is capacity in one of the Intel factories, and that's in Albuquerque, New Mexico, where, again, we have 300 millimeter capacity. Currently, we're transferring technology and being brought online, and we'll be in manufacturing in 2025 and ramping 2026 and beyond. In Italy, we're actually beginning manufacturing presently, so that's a little bit of an older agreement and is starting to ramp today, will become more significant in our revenue profile towards the end of this year and ramping from there. So we're bringing new 12-inch capacity, overall, to address some of the most advanced technologies and the, the demand for the most advanced technologies from our customers, and this will lead to better scale for the business and better efficiency. We're a manufacturing company.
Scale will bring better efficiency. So on this slide, we show what this means to the bottom line, to our financial model, by showing our Q3 2023 annualized run rate. That's where we are today in revenue and representative profit lines at the gross, gross profit, operating, and net profit levels. Then you can see how that translates as we fill the new capacity that we're putting in place today. So on the revenue side, the additional capacity, once at 85% utilization, and 85% utilization is roughly the model of all of our factories. We try not to exceed 85%, because lead times then get stretched out, so 85% is really the model of the company. So once we fill all of the existing capacity...
and the capacity of the new factories you saw on the prior slide, we can increase our revenue profile by 1.9x to $2.66 billion. So $2.6 billion is the maximum revenue of the company with all of the current factories that we have on our plate filled to 85%. And you can see that then the returns are multiplying at a higher rate, as you would expect from efficiency of scale. So gross profit by 2.1x, operating profit 2.9x, and net profit reaching $500 million per year with these factories in play. So what do we need to do to get to this point? One is finish off the investment in those factories, and we've talked about for the Intel capacity, we have a $300 million CapEx investment.
So we're buying tools for that factory over the next 2 years, and that is committed CapEx that we will spend over the next 2-2.5 years. And on the Agrate side, we have an additional $250 million to spend to build out the rest of the capacity that is in this model. So to achieve this model is finishing off the spending that we've already communicated to the street, roughly an additional $550 million in CapEx over the next 2-2.5 years. And then the other portion of that is necessary to reach this business model, of course, is to fill the capacity.
And so what I'd like to turn to in the next several slides is sort of giving you a little bit more color on the markets and how we address the demand side, or how we create the appropriate demand to fill those factories. And so let me go around the pie chart and give a little bit more color. Start with one of the markets that we believe for us is one of the more exciting markets, the RF infrastructure market. What do we mean by RF infrastructure? What we mean is that we are building today components that go into optical fiber transceivers, optical transceivers, primarily for data center, so datacom market, but also telecom and other markets. But data center and artificial intelligence now is beginning to dominate the volume picture for us in this space.
And these are built using our high-performance silicon germanium and silicon photonic technologies. And that market is expected to grow significantly. This is some third-party data from LightCounting, showing from where we are in 2023, so where we reported last quarter, for example, to 2027, just the number, the sheer number of transceivers will roughly double. And so the market is expected to double in the next few years, and that's very good. But in addition to this, today, most of that 10% of revenue that we're showing, as only relates to the first bullet here. We're making silicon germanium components for those transceivers, and that's the electronic piece of the transceiver. Going forward and starting to ramp right now is the second bullet, which is the silicon photonic component. This is yet another chip that we will make for those same transceivers.
And over time, silicon photonics has the opportunity to grow and attach to more and more of these transceivers. Silicon photonics is replacing an existing, more discrete technology, and at the lower data rates, that attach rate is relatively small, and therefore, the volumes today for us in that technology are relatively small. Most of the revenue that you saw is silicon germanium components. But at the higher data rates, the expectation is that silicon photonic will have a much greater attach rate. And over the long term, therefore, if we're selling two chips into every transceiver instead of only one chip, it has the potential to double our market, our available market. So a doubling of the units and a doubling of available market has the opportunity for us to quadruple our presence in this space over time.
Speaking of silicon photonics, we are a leader today as a foundry in the space, working with over 50 active customers. We are in production and announced a partnership with InnoLight. InnoLight is number one in the space of building optical transceivers for hyperscalers, and the other customer and partner that we've announced is Marvell. Marvell, of course, a tier one IC provider into that same space. So two leading customers with over 50 active customers in silicon photonics to fill kind of that promise of that quadrupling of this total market. And this is just some backup data on the silicon photonic market, what Yole thinks about it, and they believe a 36% CAGR is reasonable in the time frame that we're in.
What's interesting about the Yole report, and so certainly reflected also in the business that we see and the products that are being introduced in our foundry, is that so far we've been talking about that green bubble, the datacom bubble. But there are many other applications that can make use of silicon photonics that are represented here by Yole, and also represented by the tape outs and the products that we see from our customers, such as sensors, optical computing, lidar for automotive. So there are many other applications, and you can see all of those bubbles are growing significantly based on Yole data as well. On the left-hand side of the chart, I show a bar graph, and that's actually new products being introduced into Tower Foundries on our silicon photonic platform.
So you see we've been active for a few years, but you can see that shape. And there's a little bit of an inflection point in products being introduced between 2022, 2023. Typically, the shape of new products being taped in within a couple of years turns into some level of revenue. So the expectation is the level of the revenue slope will look something like the number of products that are being taped in. So very strong activity right now, a little bit of production starting with InnoLight, Marvell, as we've announced, and expected to grow significantly. So this is part of the growth story. The next couple of slides, just very quickly, about why is it that silicon photonics is gonna have a greater attach rate at the higher data rates?
And by the way, those higher data rates are most, if not all, of the growth in the optical transceiver market, as particularly going forward, the 800 gigabit node. Well, part of it is cost, right? You're assembling a bunch of discrete components into a single die, so that's across all the data rates. But part of it is also that it can—Once you have all of the components in a single die, you can move that die closer and closer to the ASIC that you're trying to communicate with, ultimately, eventually, even co-packaging the ASIC with that transceiver. And that reduces the losses in the traces between where the ASIC sits in a box in a data center and where that optical connection, that pluggable optical connection is.
And so that becomes more and more important at higher data rates because you're improving the bandwidth, you're reducing the losses. So ultimately, eventually, there will be really a performance reason for why silicon photonics will take over in this market. Today, it's primarily cost. Still a very good reason. Now, the other die that we make for these optical transceiver is a silicon germanium die, and here there's the potential for some content growth for us because of the trend for linear pluggable optics. Today, for the higher data rates, each transceiver has a DSP at either end that cleans up the signal as it's being transmitted. And what people are realizing is that functionality of the DSP typically also resides in the host. So in the ASIC that's actually communicating, that processor can do some of that cleanup itself.
And so by adding some analog content to the silicon germanium transimpedance amplifier, the receiver, or the laser driver, the transmitter, by adding some content to that silicon germanium die, which is what we produce today, we don't produce the DSP. You can actually eliminate the DSP in many of these connections. And this is something that many of our customers are building and taping out in our fab, and we are starting to see some production volume related to this, and this is another very good trend for us in increasing the content in this high-growth market for us. And by the way, the reason, yes, you can lower cost, you get rid of the DSP, you can reduce power consumption, the DSP is very power-hungry, and very importantly also, for the data center guys, you can reduce latency.
Those are the three biggest care abouts in an optical link. If you ask a hyperscaler, what do they care about? They care about those three things. The link also has to work. That's the last thing. So it will replace some fraction of the links, and that is very good, a very good story for our silicon germanium market going forward. The next big chunk of the pie is our mobile market. This is a bit larger business, a little bit more stable in the sense that it has seen significant growth. Now, where do we go from here? What we first of all produce in this market are wireless front-end components, as you can see listed here, RF switches, antenna tuners, low noise amplifiers, PAs, and our RF SOI, and also RF silicon germanium technologies.
Now, the number of handsets going forward is not growing tremendously. So where is the growth gonna come from for this technology? It's really the continuation of the 5G, 4G to 5G transition. And in developed countries, maybe in the U.S., the expectation is, okay, that's already perhaps reaching some plateau, but around the globe, around the world, certainly that's not the case, and that will continue. If you look at India, look at other developing nations, there are many years of transition still to come. That's represented in the blue part of that graph. So as that blue part of the graph grows, each 5G handset has significantly more RF content than a 4G handset. So that's where the growth in this business for us will come, is in that blue part of the curve.
I'll talk about millimeter wave next, which is. We had a press announcement with Renesas a couple of days ago, regarding millimeter wave. And here, millimeter wave, we're not talking about millimeter wave in a handset, but we are seeing a significant design activity, design wins, as we announced here, and some production also. Looking at millimeter wave, using silicon germanium for phased array receivers or terrestrial receivers for satellite-based internet services. This is a relatively small market today, so as we said in the PR, we are in production, relatively small.
However, it is expected, the number of users is expected to significantly increase over the next several years, and if those projections come to be, and certainly it will increase at some rate, using that $80 million number, it drives about a $400 million of revenue opportunity for silicon germanium, or the TAM would be about $400 million from a very low level today. So another market that can grow significantly for us, and we're partnered with a tier one in the space, in Renesas. So the next major segment, I'll talk about is our power management, section. About 27% is power IC, and this is really the focus. The discrete market, although it's about 17% of our revenue, really, we're not focusing on the discrete market any longer. We're really focused on the power IC market.
This is by far the largest analog market in the industry. So about $24 billion in IC market, serving all major semiconductor segments. So every electronic device has to have some sort of power management, whether power is coming from a battery or coming from the wall socket. So significant market. And then if we break up that market, across different voltage ranges, you can see from five volts to 120 volts, you can see that the highest units or the largest part of the market is at lower voltages. And that makes sense because these are low voltage, battery-operated or handheld or handsets and other mobile devices. At the higher voltage end, it's more automotive, it's industrial, so very good segments to be in, but lower volume. Traditionally, that is the market that we've been playing in.
The higher voltage market, automotive in particular, drives a lot of our volume today in power management. That's been very good and very sticky business, but it's lower volume. What we're doing now, particularly with the Intel facility, where we have a lot of 300 mm capacity, we are transferring now our 65 nm BCD technology, which is a very attractive technology to attack that lower voltage part of the market. We announced a couple of major design wins from customers or customer wins, really, platform wins from customers. One, a major IDM and another, a major fabless, that is playing in that lower voltage part of the market. We will utilize the Intel facility that we've announced to ramp those customers. This gives us access to really the biggest part of the market that we have not traditionally participated in.
So a significant growth opportunity for us in power management, and again, it's not because the market is growing necessarily at a huge rate, but because we're poised for very strong market share gains, given the technology and the capacity and the customers that we have now. The last area that I'll go through is our sensor and display area, which is 18% of our revenue based on Q3 run rate. And here we have two areas to talk about. One is our sensor area, CMOS image sensors, and we really address high-value markets. So it's not really cell phone cameras, but it's medical, industrial, automotive, dental markets for X-ray where we can provide value and extract value in a significant way.
So although it's a small part of the total pie that you see here, it contributes disproportionately from an ASP perspective, from a bottom-line perspective. And then the second area is micro-LED or micro-OLED displays. We have partners in both, looking at, eventually, emerging applications in AR/VR and high-resolution displays. The growth in the space, particularly in the areas, is really. It depends on, in CMOS image sensors, it depends on which specific area, the focus is on. You can see some areas are, growing faster than others, and in green are the particular focus areas for us. And you can see that those green areas are growing faster than the overall market. And to get specific about, the three, markets that we today, our revenues derive from, most significantly, is medical and dental X-ray.
Here we own a very significant market share. Our wafers are very high valued because of unique stitch field technology that allows us to create very large die. This is important for large die X-ray systems. High-end photography, where we have very high-value wafers due to unique pixel IP. This is where performance is key, and the pixel is the light-sensitive element. And so having a combination of stitch field and a backside imaging technology allows us to have best resolution and performance overall for the most demanding fields in cinematography, broadcasting, and high-end photography. And then industrial machine vision. Now, here we extract high value from our wafers or provide high value, mainly because of a global shutter technology. This allows you to take very rapid images, which is important in a lot of industrial applications.
So very high-value wafers, even though from an overall revenue contribution, smaller than some of the other segments. And then these are the areas of potential growth for us: 3D imaging and biometrics. This is an area that we're very active in today with time-of-flight technology. And as I mentioned, potentially an explosive area over time is that of displays. And eventually, displays on glass will be perceived to run out of steam at some point from a resolution standpoint, and silicon can provide a solution to that. We are partnering and have partnered with a couple of key customers in this space, and certainly, as that takes off, we're poised to participate in that market. So come back to the overall model. Again, the strategy of the company is scale.
We are putting the scale in place with the two announced partnerships, with one with ST and one with Intel, both creating 300 mm capacity, which traditionally has been, kind of holding back our ability to gain market share from customers. Now that is wide open, we are engaging with new customers, and we're bringing those to those two factories. As we fill the capacity in those factories with some of the technologies that we've discussed, we're then able to exercise a better, business model, as you can see here, and improve our financials accordingly. So with that, I want to thank you for your interest, your attention, and ready to take any questions that you might have. I really, yeah, it's a very good question. I don't have an idea.
I can see that there is some activity there, clearly, from our standpoint. From my perspective, it looks very attractive, right? I think the key is, you know, hyperscalers have gotten used to the DSP, the programmability, and so, the fact that what I presented is true, that it's lower power, it's lower cost, it's lower latency, so those are all things that they care about. But there can be some, you know, barriers just because it's not the incumbent technology at the higher data rates. And so, you know, there will be some places where it makes sense and some places where kind of the DSP technology will remain. How that balance will play out, I think we'll wait and see.
I think there's nothing negative, it's just additional growth, because even in the other solution, we still have drivers and TIAs that we supply. So it's just additional to that for us. Right. Right. Yes, a very good question. The question, by the way, sorry, I should repeat the question, is what do we need to do to compete with other foundries since our business model is not to compete on technology node, right? It's not to go to 3 nanometer. It's really to stay in more mature nodes, and what do we do to compete in those nodes? And what we do to compete is really focus on those analog parameters that customers care about. So if we're talking about RF, RF-SOI, it's reducing the losses in a switch.
That can lead to better battery life in a handset. It can lead to better reception. How do you reduce losses in the switch? You reduce RON, COFF in the transistors. How do you do that? That is the IP that we bring to market, as an example. In power management, it's similar. Reduce the losses in the transistor at higher voltages. How do you do that? Well, you have to create transistors that can operate at high voltages and still not have a lot of resistance, not have a lot of losses. And we do that. Right now, our 65-nanometer technology has some of the best figures of merit relative to any other foundry in that respect. That leads to better efficiency, a lower power consumption in operating those chips. Image sensor, it's similar.
It's, you know, the image sensor is sensing an image. How well, how efficiently are you collecting the light? That's the pixel IP. And, how broad is your wavelength that you can absorb, for example? Can you see at night? Can you absorb some of the IR spectrum where, you know, that allows you to see better at night versus not? What is the resolution of the pixel? How small is it, or how large? In some of these cases, you wanna collect a lot of light, and still not leak or not have a lot of noise, right? And when larger pixels... So it's basically engineering, to answer your question. You have to engineer and understand the market, get very close to the customers, and that's what we've done in silicon photonics, for example.
Sorry for the long answer, but silicon photonics is a very good example. Got very close to the leaders in the space, InnoLight. Yeah, it, it, you know, InnoLight, in that case, and Marvell, those are the two leaders in the space. We got close to them, understand what they need, and then work with them to create a solution that they need. But now that solution is a foundry solution, so any customer can come, and we've built up IP that is very relevant in that market, and now a customer wants to come and use it for a Lidar application. They're coming and using it for a Lidar application. And we developed kind of a barrier of entry because of that, right?
Because it's an engineering know-how knowledge that another foundry would have to somehow surmount, and as they invest in that, we are already ahead. So that's in most of our markets, it's really engineering in the wafer. We do not. In fact, if you look at the model, actually, the new capacity starts out at lower margin, right? Because filling our current factory, our current factory is already, basically, the fixed cost is already paid for by the wafers that are there now, even though it's not full. So when we fill that additional capacity, the drop-down to our profit line is about 50%. So each $1 in revenue returns 50% in margin. With a new factory, that's not the case, because you haven't covered the fixed cost.
So overall, those wafers would be at lower margin, drop-down per, dollar of revenue increase. If at that point, so at that point, a small perturbation may be very similar to what we have today. So we haven't assumed that those prices somehow are magically going to be better, let's say, from an ASP perspective, but we have to cover the fixed cost before we get to that same model we have today. So we, what we are showing is, okay, what is the end result, right? When they're all full, that's where we're at. Yeah. No, we had no ability to complete the deal. And, and it was really driven by Intel, but there was no ability to complete the deal. Because, yes.
So there was no deal on the table, that, "Hey, if you just do this, you can complete it." There was nothing like that. Yeah. 50. Yeah. Yeah, yeah, we'll look at that. Like, right now, the focus is on growing the scale, and so as you say, we have the $550 million plus about $180-$200 million a year of maintenance, what we call maintenance CapEx, that we use for everything else. So that, that is where we are right now. We'll continue to look. Right now, we want. In addition to that, we want some flexibility, so that if opportunities come up. You know, we've, we've had very good opportunities in the past with TPSCo, the purchase of the Panasonic factories, good opportunity with the ST deal, good opportunity in the past, we purchased a factory from Maxim.
So you never know when those opportunities come about, so we want to leave a little bit of flexibility. As we get through this phase, you know, we'll look at you know other uses of cash, but at least for now, that's the thought. Right. Yeah, it's a good point. Right, it's a good point. Right now, we're seeing so much opportunity that the instinct is to just focus on those investments, and it's a good place to be in that respect. Yep. Yeah. So first, the depreciation. We use a 15-year depreciation, linear depreciation. And then the second question was? Yeah, we haven't really published. We just showed the full 85% level. We haven't really done anything in between.
Just to kind of bookend where we are today and where we will be in the ultimate model, where this capacity is online and filled. But we haven't really shared the rest, other than to talk about, for example, like we explained earlier, when we fill revenue in our existing fabs, typically it's a 50% drop-down from revenue to the gross margin. It's not, it's not going to be hockey stick, you mean in re-- the, that suddenly it goes from zero to 85%? If there's good... It's not-- I mean, it's gonna be, again, not a, you know, it's gonna get better and better as you fill, because as you fill, you first absorb the fixed cost, right?
When we start production, start depreciating, and then you have to absorb that fixed cost, and then the margin gets better and better after that. Yes. Yeah. Yeah. Yes, it's a good question, actually. So the question is, the LPO, the linear pluggable optics, where silicon germanium can take some of the content currently in the DSP, what links specifically are likely to move in that direction versus remaining with DSP? I would say the shorter links, they're easier, right? But in a data center, you tend to have a lot of shorter links. And then, you know, where you have control of both ends of the connection. In a data center, typically, that's what happens. Yes, yes. But again, that's for the hyperscalers to decide. This is just an observation, looking at what is likely to happen. Yeah. Yeah.
Yeah, we will be rational in how we ramp up the CapEx spend-
Right.
so that it's not way ahead of when it gets filled. To that point exactly.
Yes.
Yes.
So-
Yeah. Yeah, I can say it's a very long-term agreement in both cases. We—I don't think we've published the length exactly, but it's a long-term agreement. We have to make our customers comfortable, and so it, it's not a few years. It's a long, long term.
Long term.
15-year. Yeah. Yeah, I guess that's for the auditor. I'm not a CPA, so let them think. Yeah, that's a good point, actually. Those are. Yeah. Yeah, so yeah, the answer for those that may not have heard, yeah, those are our tools, so it can depreciate 15 years, even if they're in a different facility, 'cause we moved them. Yes. Yeah. I think, you know, obviously, the market we can't control, and I think that's the biggest risk always, is that the market timing doesn't cooperate. But in our view, I mean, that's just a matter of timing, right? The up cycle will come, and then the fabs will get back to a good utilization rate.
So just the timing of when we need to fill them, because, you know, depreciation is starting to hit, and now suddenly the market is soft for a little bit, and that could leave us exposed for a quarter or two. I mean, that would be the risk, because I think from the other side, we have a line of sight over the demand, we have the customers. Execution is certainly important. In Agrate, we've done a very good job of getting equipment into the factory, qualifying our technology, qualifying products, and now we're starting to ramp, pretty much on the original plan that we had published. So we're looking now to replicate that at Intel and then ramp accordingly. Okay. And, was it-- Yeah.
No, I think. Well, there's always risks, but today, the factory is operated without interruption, and it's in the north of Israel, so it's very far away from where the conflict is taking place. So at this point, there's, you know, nothing unusual happening. Factory continues to operate as normal. Some employees were deployed, but we covered that with overtime, and it's a small percentage of the total. Okay. Very good. Thank you very much for your interest and attention.