Okay.
Okay. We'll go ahead and get started. Good afternoon and welcome, everybody, to the second day of Needham's 28th Annual Growth Conference. My name is Quinn Bolton. I'm the semiconductor analyst for Needham & Company. It's my pleasure to host this fireside chat with Navitas Semiconductor, founded in 2014 and headquartered in Torrance, California. Navitas is a pure-play, next-generation power semiconductor company and industry leader in gallium nitride, ICs, and high-voltage silicon carbide technology. Joining me for the company is Chris Allexandre, President and CEO. Chris, thank you for joining us at the Needham Conference.
Thank you for having me.
Chris, you joined Navitas on September 1, so you've got about four months on the job now. Can you first just give us a little bit about your background?
Yes. So thank you, Quinn. Thank you, everybody, to be here. I've been in semiconductor pretty much most of my career, 25 years. Started in TI for 17 years. I worked in Europe for TI. I worked in China as well, part of the leadership team. Then I moved to the Bay Area in 2013, worked at Fairchild, and then joined Renesas for the IDT acquisition. And my last job was to run the power division of Renesas, where I quite transformed a GaN company. So that's how I got exposed to GaN.
Excellent. Since joining the company, you've had an opportunity to meet with the team and with customers. Can you share your thoughts on just the team's technical capabilities and what messages have you heard from the customers?
Okay. So when I took the job in September, the first thing I did is to go out and meet all the customers, current customers and potential customers. So I met hyperscaler companies, I met GPU companies, I met computing companies, I met grid companies, as well as industrial companies. They all said the same, which I think we all see now in the SAM and TAM and potential is there is a big pivot with the AI catalyst. Everybody has to re-architect their power conversion. They have to get more density, higher efficiency, more power level into the same footprint. And this cannot be done with silicon. So clearly, the first message is, you guys have the technology we need. You have high-voltage SiC. We'll come to that in a second. You have GaN. We need GaN. We need SiC. So that's the first thing they said.
Number two, they said, this is a risky, challenging transition we need to do, and we need to do it with speed. So please bring the expertise that you have. We didn't start GaN yesterday. We've been pioneering GaN from the beginning, as you mentioned. In the GaN space, we shipped about 300 million units of GaN, mostly in 650-volt, which is the high-voltage GaN. We pioneered that in the mobile market, which was really the market that got GaN at scale. And we got experience. We know where the skeletons are. We know how things work. And that expertise that customers say, please bring that to us. On the SiC side, it came through an acquisition of a company called GeneSiC that we acquired a couple of years ago that basically pioneered the high-voltage SiC.
So SiC is kind of well known in the lower voltage used for EV, but there was a niche market for high-voltage, super high-voltage SiC like 1.2 and 1.7 and 3 kV. And that technology is also needed for the grid. So that's what they said. And number three, they said, do it yesterday. So that's why I pivoted the company very quickly away from consumer mobile, not just for financial reasons, but because I wanted to have the entire cycles of the company going after those high-power markets.
Navitas, when you joined, had already started to pivot away from the mobile and the consumer markets. Your Navitas 2.0 strategy, I think, accelerates that transformation to the high-power markets. Just spend a minute elaborating on your Navitas 2.0 strategy and what the focus markets under Navitas 2.0 are.
So you said it, right? 1.0 had started this pivot away from consumer. I just accelerated it. And the benefit of that is it was really about resource reallocation. So today, the entire company, all cycles are on the high-power markets. The high-power markets are AI data center, which is very famous and front and center of everybody. Number two is grid infrastructure, which is related to the AI data center, because if you want to do what you're trying to do in AI data center, you have to change the grid. Number three is computing, which is going through the same transition. We'll talk about it later. And last is industrial, right? So that's kind of the big pivot. We focus on two technologies. We are not confused. We do two technologies. One is GaN. We have 650 volt. We pioneered it.
We've been in that business now for 10 years. We released mid-voltage GaN a couple of months ago, 100-volt GaN, and we've been very clear that as the market expands, we'll go lower voltage on GaN as well as higher voltage. On the SiC side, we have the stack to support 650-volt all the way to multiple-thousand volts, but we only focus on 1.2 kV and above. The reason we do that is our technology brings differentiation when it comes to reliability. Grid and PSU need reliable technology, so we want to focus where we can make a difference, and we think the EV market is a bit of a crowded market, so no need to. As a fabless company, we don't want to play there.
Okay. So de-emphasizing auto for.
Completely. We support the customers that adopted early GaN. We have a few customers that started GaN in OBC, mostly application, but we're going to support that, but we have basically no roadmap, no R&D, and no focus long term on the EV space.
On the focus for the super high voltage silicon carbide, talk about some of the end markets where you see the need for the 1.2 kilovolt and above silicon carbide and describe the competitive landscape in super high voltage SiC.
So SiC is a huge market. It's between $5-$6 billion right now. It's going to move to $10 billion in 2030. But if you look at it, it's actually mostly low voltage SiC, right? So it's EV, industrial, 650 volt, 750 volt. If you remove all this, the market is down to $100-plus million. As I said, we have the technology to do all of the above, but we focus on 1.2 kV, which is called high voltage SiC, and ultra high voltage SiC, which is 1.7 kV and above. Why? The 1.2 kV is used today in AC/DC conversion inside data center, so-called PSUs. So here we compete with the traditional SiC vendors, the onsemi, the STs, the Infineons, the Wolfspeed. Anybody that basically supports EV with low voltage actually competes in that 1.2 kV. We also focus on 1.7 kV and above.
We announced a couple of months ago a 2.3 kV, 3.3 kV module-based SiC, right? That was announced around the October time frame. That's tailored to the grid infrastructure. So if you want to think about grid, if you think about the revolution that's happening in the grid, right, where you really have to change the way you transport and convert to enable AI data center, you have new applications like battery energy systems, solid-state transformers. All those need super high or ultra high voltage SiC, 2 kV and above, and that's what we are focusing on. And here, I would say we compete with a lot less people.
There are not many Infineon as the technology or well Wolfspeed as the technology, but we don't see many of the traditional SiC vendors focusing more on the EV and the PSU, having these super high voltage, ultra high voltage SiC modules.
Excellent. Maybe spend a minute just talking about the company's total SAM opportunity under the Navitas 2.0 strategy. What's that SAM today and where do you see the SAM growing by, say, the 2030 time frame?
I'll start by saying we just posted on our investor relations website a few days ago some slides that actually give detail on SAM, TAM, evolution, and definition of high voltage, ultra high voltage, because there's sometimes a lot of confusion, right? If you look at the SAM that Navitas 2.0 is going after, which has just detail, right, it's about $3.5 billion in 2030. That comes from a few hundred million now. It's about 60%-75% kicker. The split between SiC, high voltage, ultra high voltage, and GaN is about 50/50. So $1.7 billion, $1.7 billion, basically between SiC and GaN. Most of that $3.5 billion is coming from AI data center, $1.4 billion, and grid at $1 billion. So $2.4 billion of that $3.5 billion is actually coming from those two drivers.
Then you have computing, which is about $4-$500 million, and then you have industrial. We focus on those two because this is where we expect the technology disruption to happen. So think about GaN as really an inside data center play. Think about SiC as outside of data center play with a little bit in PSU or AC-DC before you convert to the new architecture of NVIDIA, which is the 800-volt HVDC, right?
Okay. One of the questions I've had from investors is GaN and silicon carbide have been around for a while. They have advantages over silicon, yet adoption has been slower in many cases than people expected. What do you think some of the biggest challenges to adoption of GaN and silicon carbide have been?
It's a very good question. I would say SiC is much easier to implement. So SiC, if you think about the ChatGPT moment of AI, right, if you think about the ChatGPT moment of SiC was really when Tesla adopted SiC in the first EV train, right, that they wanted. So really, SiC has been accelerated since then. But SiC is kind of very traditional like silicon, easier to drive. GaN is a lot more complex. GaN switch at a much higher frequency, driving a lot of issues like EMI and thermal loss, and also the transformer design has to be different. It's very unforgiving. You can't just implement GaN in a casual way. It's not going to work. It's not going to switch. So the system-level expertise and the system-level change has been a bit holding customers to move from silicon to GaN.
As a matter of fact, when we drove the adoption of GaN into the mobile business back a few years ago, we really had to do more than just delivering the FETs and the ICs. We really had to step up and show customers how to implement this in the system level to give them confidence. Today, I think this confidence is there. First, there is necessity. You cannot do what you're trying to do from an AI data center point of view with silicon. So there's necessity. Number two is GaN has now multiple years behind. I think the expertise at the system level between 100-120-watt chargers carries to the other high-power application. I think the concern being less, there's still concern, right? We are basically educating customers to basically have less concern.
Excellent. Lastly, before talking about the various focus markets, Navitas is smaller than many of the companies you compete against. How do you differentiate yourself? How do you win against larger customers?
I get that all the time. So I tell you, this is not kind of me, of course, selling the company, but I actually think size is an advantage. This is about speed. When we talk to the computing guys trying to adopt GaN faster to basically reduce the size of their power bricks and get it higher power. If we talk about the grid that I have to deliver this new grid by 2027 to enable the new data centers architecture, when we talk to the hyperscalers, they want this yesterday. So speed and experience makes a difference. We don't have a fab. We have a strong partnership with the lead foundries. We own some of the technology. We move super fast. And I think that has to be an advantage so far.
We have enough cash on the balance sheet to not worry any customers that we're going to be around, and the focus is executing and getting things on the board.
Excellent. Turning to data centers, which is, I think, your largest part of the SAM today. This is obviously lots of different power conversion applications in the data center. Can you talk about which applications within the data center you're targeting? And we'll get into the 800-volt rack in a second.
It's very good you asked me that question because there's a lot of confusion because there's a lot of change in the architecture of a data center. I'll start with where we are today. I'm not going to rewind the tape and how we got here, right? Today, most use 48-volt as what we call IBC, which is intermediate bus converter, right? The first application is converting AC to DC. As you grow up in power level, you have those PSUs, power supply units, AC to DC brick that convert AC to DC. And that's a SiC play mostly. That's a 1.2 kV SiC. And that grows. We are moving from 10 kW, 10-kilowatt PSUs to 20-kilowatt PSUs to 30-kilowatt, stacking more power inside the same footprint. That's number one. Then you have the 48-volt, which was traditionally silicon.
With the adoption of what some hyperscalers call the high voltage buck architecture, GaN is getting into 48 volt. So you have some 48 volt designs converting 48 volt down that is now starting to use GaN for efficiency reasons. That is small, but that's the first time GaN gets into a data center. And that's 26. Then you have the big pivot, which is the 800 volt HVDC that NVIDIA and now the entire market is adopting, right? This is really kind of skipping the AC-DC and moving from 800 volt DC all the way down to 48 volt, 12 volt, or even 6 volt. That PDB, as they call it, which is a power distribution board, that PDB has a huge content when it comes to GaN.
On the primary side with 650, on the secondary side with mid-voltage GaN right now and maybe going to low voltage in the future. So that's kind of where we focus. The other application like power shelves and things like that, there's also circuit protection. You're going to hear from some of our competitors, the JFET SiC is a small content for SiC in circuit protection hot swap, but the bulk of it is really AC-DC as a transition and, of course, HVDC.
Okay. So on the 800-volt DC, we do AC to DC conversion. Maybe outside of the data center, we bring 800 in. We've got 800 up and down the busbar, each rack. You come onto the compute board, you have an 800 to either 48, 800 to 12, or 800 to 6. One of the questions we get is, is that a silicon carbide opportunity? Is it a GaN opportunity? Because I think there's a lot of confusion and I think a lot of the silicon carbide companies are saying, hey, 800-volt, we've got that for autos. This is right up our alley.
So what I would tell you is we've seen only a few exotic implementations of the 800-volt HVDC with SiC. We supply 1.2 kV SiC, so we sit at those tables. But we've seen, to my knowledge, only one real implementation. Most of the implementation is GaN-based because SiC doesn't get the same level of density than GaN does. So if you think about the primary side of the 800-volt HVDC, it will be mostly entirely a GaN play. Now, where some of the confusion comes is not everything is going to pivot on day one to HVDC. So you're still going to have a lot of AC to DC inside data centers for the next few years. And that is still PSU, which is using SiC.
That's why when we talk about SAM, and again, I refer the investors in the room or online to look at the website, we talk about this $3.5 billion, right? And that $3.5 billion SAM in 2030 is based on $1.4 billion for data center and $1 billion for grid. We assume about 220 gigawatts of AI deployment by then, 50% of that deployment being HVDC. What this means is another 50% will be still traditional architecture, which is AC to DC and 48 volt down. Now, there are high models. One of our German competitors is putting a model where 75% is going to move to HVDC, which is going to reduce the SiC SAM and increase the GaN SAM. But overall, to your question, 800-volt DC is a pure GaN play, in my opinion.
NVIDIA, at least for their ecosystem, are they specifying a material or a technology, or are they just saying, hey, we want an 800 to 12, 6, 48 step, let each supplier develop itself?
So what we have seen as a big change is the hyperscalers are a lot more involved in the system architecture definition of the boards than they were historically. We've seen the same, by the way, in the old days of mobile, where the mobile used to subcontract to the ODM and they basically took control, right? So this is really a collaboration with, you mentioned NVIDIA, but the others about how to make this PDB of this 800-volt DC brick the best way, how to use how many of the GaN FETs you're going to put on the primary side, how many 100-volt GaN you're going to put on the secondary side for the 48-volt, what transformer you're going to use, how you're going to lay out the board.
We participate in this discussion with our application engineering team, with our system teams, and we basically collaborate with customers. That's why I say you can't improvise GaN. You can't come in and say, I just did a press release. I do a tape-out. I'm going to have GaN next week. I'm going to sit at the table. This is very complex. Silicon has been around for 40, 50 years. This is very different. It requires a bit of system expertise that we carry over from the 1.0.
As investors think about the high voltage rack architecture, can you share any thoughts on what's the total power, sorry, total power conversion content per rack or per megawatt, if there's an easy way to think about it, just sort of sizing this opportunity?
So from a GaN perspective, and I put some slides together, right, but I have the numbers here. From a GaN perspective, we think it's about $10,000-$15,000 of GaN content per megawatt. Again, that's based on some level of redundancy, the level of deployment I talk about, and as well as the percentage of HVDC being deployed. Because, of course, more HVDC means more GaN content. Some of our competitors have a higher model, but we think 10 to 15 is reasonable. On the SiC side, it's 20 to 25. So it's actually slightly higher. And the reason it's slightly higher is you have the inside data center with AC-DC, and then you have the grid, which is outside of the center that basically supplies the 800-volt DC. So the content in SiC is actually slightly higher because of that.
But that includes the grid opportunity, which I guess in your data center TAM of $1.4 billion is excluded.
It's excluded. It's the $1 billion of SAM that is grid, right? So if you think about GaN, think about 10 to 15. I think NVIDIA refers to $20,000-$25,000. $10,000-$15,000 per megawatt, $20,000-$25,000 for SiC, both AC to DC outside of data centers and all the way down to the GPU.
Okay. So we just talked about the high voltage, 800-volt, high voltage rack opportunity. Will Navitas continue to develop SiC for the traditional power supply part of the market?
Definitely. I mean, as I said, the next five years, we're going to see an increase as a percentage of the 800-volt HVDC architecture. But even if the most aggressive models you see have 25% by then, that is using the classic architecture, which is where you need AC-DC, so 1.2 kV SiC. So we continue, we're going to release our Gen 5 in a couple of weeks, which is an improvement of figure of merits to drive costs down and be able to help the data center guys to basically combine more power in the same brick. It's really about density, not just about architecture initially.
Okay. So Navitas can go from sort of grid to core. A lot of other companies claim they have similar capabilities. I'm curious for your perspective, who do you think really has that grid to core capability in addition to Navitas?
So first of all, we go all the way from AC to the secondary side of the 800-volt DC, right? Because what we believe is we serve the first stage with SiC, and I talk about that super high voltage SiC. We cover the second stage with high voltage GaN, 650-volt, mid-voltage GaN, 100-volt. We believe GaN will continue to go down in voltage. And as you go on the secondary side, which is mostly silicon today, it's going to go down to 20-volt GaN. And so we'll be able to go there. And then as GaN continues to get closer to the GPU, I think we're going to be able to serve the grid to the GPU. I think the only competitor I can think of that has really all the technology is Infineon.
They really have, like us, the high voltage SiC with 1.2 kV, the super high voltage, ultra high voltage SiC with 2 kV and above. They have GaN high voltage. They have mid-voltage GaN. They're working on low voltage GaN. And they have silicon. So they're working on the whole thing.
Okay. Great. And then of the $1.4 billion TAM for data center, do you have a rough sense how much of that is GaN versus silicon carbide?
Yeah. The $1.4 billion is about $400 million SiC and $1 billion GaN. As I said, it's mostly really driven by the HVDC change of architecture.
Perfect. Perfect. Grid and energy infrastructure is sort of your next biggest part of the SAM. What are the key applications that you're targeting in this market?
I think, Quinn, I would tell everybody that this is probably the most underestimated part of the SAM. Everybody focused on inside the data center, and I get it, right? This is the edge and this is the big change and this is a fast disruption. To achieve that, we talked about the fact that you need to deliver on the back of a data center an 800 volt DC. You can't transport DC. So that right there changed the way the grid is working. The grid is not the most efficient. So you have applications like solid-state transformers. Transformers are big. They are heavy in metal. They are not the most reliable. And they weigh a lot of weight. And they are, as I said, fairly big, right? So moving to solid-state transformers is really kind of electronification of the transformers.
They get the efficiency to a much higher level and they get the size to a much lower level, right? So that's number one. Battery energy storage system is also an important one. So any application that is grid-tied, that transports energy, that converts AC to DC is actually booming and really driven by the AI catalyst. So for me, this market is AI-driven. Without the AI data center, we would not have the debate we have today about how the grid is going to support all this and why we have to change from transformers that have been around for the last 50 plus years, 100 years to today. I think if Edison would come back today, he would see that it's the same grid as a long time ago.
You're talking about the solid-state transformer opportunity. What is the opportunity for your high- or super-high-voltage SiC in a solid-state transformer?
So of course, it depends on the type of SST, the level of density. I talked about $20,000-$25,000 of SiC content overall, right, per megawatt. Two-thirds of that is in the SST space. So SST is really kind of the HVDC disruption, if you think about it, right, outside of data center. And this is really kind of where we see most of the SiC content. Now, this is, again, this is 2,000-volt, 3,000-volt and above. This is a module play. There's a lot of complexity. It's a new field. It's new customers on top of the usual suspect. And I refer to you to look at the blog from NVIDIA that talks about all those SST companies. There is a lot of money being spent right now in this, and I think it's going to be a big change.
And it's going to carry over beyond AI. Once we start to replace those huge transformers to solid-state transformers, once we start to put battery storage and as a system around the big cities to kind of stabilize the grid, I think it's going to be used across the grid, across the board.
Have these already started to be put into deployment? Is it still on the come and really more tied to some of these 800-volt high-voltage DC data centers?
I think things started with battery energy system. This has started a while back. I think some of the big names like Tesla and a few others have worked on this for a while. This is now starting to deploy in China. They've been deploying that already for a while. Then any utility grade, utility level type of solar or renewable energy has been in the works already to kind of supplement the power creation from the classic way. I think the SST is really in the works right now. There is necessity to deploy that by 2027 because it will enable the 800-volt DC, again, on the back of the data center. So there's a lot of activity going on in 2026. I think it's all-time high.
Excellent. Okay. And then I believe your company had said on the last earnings call that you're on track to ramp the solar microinverter with a lead customer in 2026. Can you talk about what capability does Navitas bring in that solar inverter opportunity and how large is the micro opportunity over time?
So solar has been evolved over the last few years with the reduction of subsidies and so forth. It has been less of a big market expected. If you think about solar in the context of utility grade and large solar farms, I include that in the SiC SAM for grid infrastructure. But when it comes to commercial or consumer solar, I think this market is not going to be that big. And this is really something we are defocusing on to really put all our cycles on the high power market.
Got it. Okay. So that part, the consumer residential sort of gets de-emphasized. You'll continue to invest in sort of the infrastructure side of solar where it requires a much higher voltage sector.
Exactly.
Okay. Are there any other grid or energy infrastructure opportunities we should discuss before moving on?
I think, as I said, the most important anything that will help us to get the grid more stable, more efficient to deliver this 200 gigawatt, probably even to 300 gigawatt by 2030. AI deployment is key.
Okay. Great. Moving on to performance computing, the company has de-emphasized mobile and consumer applications, but you will continue to target the higher end of the computing market. Maybe just discuss what segments do you mean by high performance computing or performance computing?
So computing is a $220 million unit market, roughly. We are talking about above $1,000, above $2,000. So we're talking about the portable workstation, the super high-end notebooks, so-called AI notebooks. As those guys get more powerful, they need more power. So that's where your traditional 150 watt, 200 watt charger is going to have to be highly more deployed. And those bricks are fairly large today. So GaN, for the same reason than in mobile, when mobile moved to 100 watt, is needed in terms of adoption to reduce the size of those chargers and to get to a different level. We see on the horizon and not very far away from now, high-end notebooks deploying 250 watt, 350 watt, even beyond that in terms of watt level for chargers. Those cannot be done in silicon. They would be large. They would be too big.
You're going to have to have a backpack to carry them. So what we see is the adoption of GaN will follow the same acceleration as AI comes into clients, as the customers on the high-end side need to basically deploy inbox or out of box 250, 350 watt chargers. The content just to get level set everybody is on a 250 watt charger, the content for GaN is about $5. If you go to 300 watt, it's $6. If you go higher, it gets to a much higher single digit type of content, right? So think about 220 million unit notebook market. There's about 40 to 50 on the super high-end. This is going to grow with AI PC.
Multiply that by $5, $6, $7. You get to a $400 million market that will not get commoditized for the next future because right now it's a race to a higher power, and we've seen the same mobile. When it's a race to a higher power, it's a race for technology. Once you get to plateau where it's about you've reached the smallest footprint at the same level of power, then it gets commoditized, so I think it's going to take a few years before it gets commoditized.
What's the competitive landscape at that performance computing? Is it the same folks that played in the mobile and the consumer chargers or folks like Innoscience? Somebody you worry about moving up into that level of performance or will it be more of the Infineons and some of the other GaN?
I think we see the same competition landscape. We see Innoscience in China. We see Infineon and Transphorm/Renesas on the outside of the U.S. What I think has changed is, first of all, this super high-end market is a lot more U.S., Taiwan-centric than mobile, which was very China-centric. Number two is you have a very large customer in Cupertino that represents a significant part of those super high-end computers, right, above $1,000, $2,000 for the price of a Mac, right? So I think the landscape has changed. I think the IP play in GaN is becoming real. You've seen InnoScience being in disputes with Infineon and a few others. You've seen EPC and InnoScience being in dispute as well. So we see outside of China, less of the commoditization and the competition from Chinese inland vendors than we saw in mobile.
Okay. Great. Wanted to move to manufacturing. TSMC last year announced it would be ending the GaN production by mid-2027. This led to Navitas first introducing a partnership with Powerchip in Taiwan and more recently with GlobalFoundries that has licensed the TSMC process technology. Can you discuss sort of the relationship with both Powerchip and GlobalFoundries? Which products will you manufacture at which foundry? And then I've got a couple of follow-ups.
Okay. So we announced a partnership with Powerchip about a year, year and a half ago. We actually announced the release of our first 100-volt GaN back in October, which is based on Powerchip. And as I said earlier, the first application in data center for GaN is actually 48-volt. You need 100-volt GaN for that. So that was the idea behind this. And we have a good partnership with Powerchip. We're going to basically release and supply coming from them on the 100-volt. Directionally, the partnership with Global is a lot bigger and a lot more holistic. First of all, it's not a foundry partnership. I mean, if you refer to the press release we made, Tim Breen, their CEO, myself, we spoke about it to the press and during the release. It's a technology partnership on top of a manufacturing partnership.
What we see is we use their baseline, which we enrich. We add capability and the heritage and the 300 million units of device and system knowledge that we have to enrich their 650-volt. We clearly have intention to bring mid-voltage to low-voltage GaN with Global. I think Global is kind of the long-term partner on Navitas. We use X-FAB on the SiC side, which is manufactured in the U.S. We wanted a strong U.S. foundry partner for GaN for national security as well as to make sure that we are producing in the United States for data centers and grid, which is obvious for me.
Multi-year partnership with, so the best way to think about it is Powerchip is a good partner in the transition, but down the road, we're going to pivot really from TSMC to Global, first on 650-volt, then on 100-volt, and then going down to even lower voltage GaN.
Okay. Perfect. And then as you make that transition, are you confident you'll have access to enough supply from TSMC until you have ramped GlobalFoundries mostly, but Powerchip as well to high volume?
TSMC has announced phasing out their GaN fab in Taiwan by mid-2027. First of all, we are still supplying from TSMC today. We're going to be in production with TSMC until mid-2027, right? The strategies I've made, we've made is as we accelerate the ramp of GlobalFoundries, which is going to be in production by end of 2026, really mainstream in 2027. We basically are buying an extra year of buffer from TSMC. The idea here is to be able to give customers, as we ramp into 2027 GF base, we are basically still producing from TSMC in 2027, able to ship until late 2028, sometime in 2029 with a buffer and give customers a smooth transition.
And it's like every transition, what you want is to ensure that you get the rescue in case things get delayed, but mostly to not pressure customers to qualify this new foundry right away, right? So the strategic customers that I mentioned will get the option to keep transitioning for another year or two out of TSMC and then transition to Global. But we're going to ramp any new NPI starting today, is taped out at Global. And anything we're going to release is going to be Global-based.
Okay. Perfect. I think we're getting close to the end of the session, but I just wanted to see if there are any questions from the audience. No? Okay. Well, Chris, thank you very much for joining us at the Needham Conference.
Thank you, Quinn, and thank you, Needham. Thank you, guys.