And then also, you know, double-edged sword was hit by, you know, the falloff in silicon carbide with EVs. I think one of the things that many of you have maybe seen the announcement with Navitas and NVIDIA, you're starting to see gallium nitride and other compounds that are using wafer-level burn-in for AI and data centers. Hopefully, Gayn, you're going to talk a little bit about that. With that, I'll turn the mic over to you.
I'll try and work that in.
Oh, yeah. And Chris, our CFO. Just in case the question's too hard or so. They always ask me to remind you of this. This presentation's out on our website or will be shortly right afterwards. I'm actually going to go through it very quickly. There's twice as many slides as normal, but I just want to skim through that so you guys get a feel and understanding. We have a breakout group for Q&A afterwards as well. If you have more questions, figure out how to get in touch with us, and we can do that as well. Okay? Aehr Test Systems has been in the semiconductor test business for a long time. We specialize in electrical test systems for testing good from bad and reliability testing to basically do stresses and strains on devices to weed out infant mortality that would otherwise fail in a field.
There are more and more semiconductors that need this type of testing and other trends that are really driving our business. There is a thing called a bathtub curve if you get into this, but basically all devices, once they're good, have a certain failure rate. Then over time, if they have not failed, they are less likely to fail until they wear out at their end. This is the old, things tend to fail right away, and if not, they last for a long time. There is all kinds of physics involved in that. We specialize in identifying what this curve looks like, and then in production burn-in, actually weeding out all of those infant mortality for applications that an early failure would be a problem for. We have a very large customer base. I will just call people's attention to this.
We have added a number of customers to this versus the one that's been out there is we've added in the packaged part burn-in from the Incal Technology acquisition customer list here. It is a very vast representation of the market leaders around the world that we're engaged in, in both packaged part and wafer-level burn-in. On the wafer-level side, again, the whole point of this would be by testing at a wafer -level versus when it's in a different later on in the process, when many devices are put together in a single package, there's huge advantages in cost, yield, other implications. We have proprietary technology that actually enables this. We were the first and are still, you know, the key supplier in this in that it required a new tester, it required new contacting to the wafers, new handling equipment. None of that existed.
Little old Aehr Test Systems actually defined and built these things and engaged with some of the market leaders to actually implement that. We've successfully done that, and we're now starting to see the diversification of that into many different applications. We have a lot of patents around the world from the U.S. to Europe, China, and in Asia that actually are key to the differentiation up here. In addition to that, we've just got a lot of years of head start and a very large install base doing these products. Sonoma is the product name of our new packaged part burn-in system. People are like, "Wait, you've been spending all your time only focused on wafer-level." There are actually applications where the packaged part burn-in really makes sense. This was super critical with the AI change.
Prior to three, four years ago, none of the AI processor guys, including NVIDIA, were doing a production burn-in. They did qualifications, but they didn't do production burn-in. All of a sudden, these devices started going in these complex coax packages and having high bandwidth memory and other things. They've had to do these stresses to weed out the infant mortality before they go. It's a bit of a wild west out there because there's just not infrastructure and the dollars to be able to do this. Folks are out trying to compete for this space. We acquired this company last year, a very small company that was near us in Fremont, California. They had quietly been doing qualifications, meaning the actual validation of silicon on a very large number of AI processor companies.
We were fortunate enough to actually step in right as customers were asking them to go to production, and we were able to help them to ramp to production to start meeting production needs. We shipped more last quarter than they shipped in the previous three years. The customers were specifically engaged. Okay, Aehr, you have the infrastructure, the capability, the capacity, and the people around the world to do that. That was a critical thing for us to be able to do. Market drivers, you can walk through this, but kind of the key things around artificial intelligence, electrification of the world are things that are really driving our products. Generally, all semiconductors where most semiconductors are seeing the same trends that we are. The semiconductor market itself is actually going through a huge acceleration that people talk about.
As we go from $500 billion a couple of years ago to a trillion dollars, there's a huge amount of infrastructure that's being put in place. When we break that down and look at what's going on, you can, we kind of spell it out as semiconductors are getting less reliable, but going into applications where quality, long-term reliability, safety, and even security are critical. We walk through this. One slide that's new here too is just people don't understand how critically important it is to actually weed out infant mortality on AI processors. It's a little bit more intuitive when you put it into a car and you think, well, it's kind of doing inference real time as it's driving you around with full self-driving.
But it turns out you put them into the clusters and it's a real problem if one of the actual nodes goes down. There's impact on the training and inference models, increased costs. People are actually trying to put redundancy in place that's unnecessary. They just haven't figured out how to get out all of these challenges. It's a huge opportunity for us at Aehr Test Systems. We talk about the key trends driving the wafer-level burn-in, which are the same trends at wafer and package of decreasing semiconductor reliability, but increasing needs. They're putting them into applications where it's more important. Then because of KGDs, or Known Good Die going into processors with multiple chips, it's actually driving the need for wafer-level burn-in. At the same time, if you look at package, it's also driving other requirements that is a huge opportunity for us in the semiconductor space.
There's all kinds of devices where they were putting, we talked about some of the EVs with the devices going, multiple devices going into one package in a module. That is true of Intel co-packaged optics demos they've talked about, NAND flash memory, the DRAM stacks, obviously Blackwell, and those are very easy to interpret. There are many, many more of these coming because the Moore's Law, if you will, stopped. You just can't make twice as much of a semiconductor in the same area. What you're actually doing is you're putting two semiconductors into it, into one package and calling it one device. You look at the Blackwell roadmap, you know, multiple Blackwells, a Grace processor, 12 different high bandwidth memory stacks, and that's where it's headed.
This is driving the requirements, including on the packaged part burn-in side of things and driving our business. Just to try and put it in perspective about sizes, there are lots of different ways of looking at TAMs and SAMs. Rather than get into all of it, including some of the proprietary information we hold with respect to test times with these customers and test modes, people have said, "Oh, silicon carbide, it's great." It was actually the original killer app, if you will, that drove this and deployed it in a large scale to many companies around the world. Silicon photonics and silicon carbide are much smaller and gallium nitride than the flash memory, DRAM, and AI processor markets going forward. I'll just briefly touch on each one of those.
Silicon photonics, which is fiber optic communication, was both in data center, but also on a chip-to-chip. This optical I/O of chip-to-chip is something we've been talking about for about 18 months now, leading indicators with purchases of our equipment towards this for both characterization, qualification, but also production. We've just been tipping our hand to everybody about what's publicly out there, which are, as you talk through, you know, what Jensen and Lisa are saying. I think I used to have a picture of the Intel guy and I decided to remove him. There have been a lot of discussions going on there. This is where to watch. It is not the publicly available data. It's what these guys are doing. I've been saying for 18 months, they're going to surprise you and they'll announce something.
That's, you know, we're kind of the canary as well because we're seeing a ramp coming up right now for this with our products where we have a highly differentiated product testing 300 millimeter wafers with 3,000 watts-3,500 watts of devices in one insertion in a fully automated production test cell. Next one would be silicon carbide. Again, big thing was about the inverter starting with Tesla's driving through everything. You know, Jed would support this. I think we're always pretty much looking at the same type of data. A couple of few years ago, it was like we saw all the disruption and the opportunity for silicon carbide to go after these electric vehicles, displacing more IGBT.
What's going to happen in China, the reality is the Chinese vehicles are accelerating their silicon carbide adoption even more than originally was the plan for IGBT to a point where the NIOs and BYDs of the world are going completely silicon carbide. That's also true of all the guys in, you know, the Toyota and Honda side of things. The Korea, the European suppliers are all going to silicon carbide. I don't spend that much time talking about it because it's a tough discussion related to what's going on with EVs. When we first were talking about this, we were projecting this seemingly crazy idea that EVs were going to be 30% of the purchases by 2030.
At that time, it's like, "Come on, it's never going to happen." Now you're like, "Are we almost there already?" Even with all of the craziness, EVs are still growing battery electric vehicles, and it's still going to be something that will drive our business because we've taken a really good position and we have preferential opinions by the OEMs who are driving our equipment and the test times downstream. Okay? This gets into the yield implication. If you have, like, you know, up to 32 die in one of these with a 1% failure rate, you have 32% failure rate at the module. There's no choice but to move it to wafer -level. When we first did that, it's like, "You can do it at wafer -level." We've proven it on all these benchmarks. Again, this is a good business for us.
A gallium nitride, a new compound semiconductor coming along. There was a discussion around the data center. It's actually a combo compound semiconductor. The workhorse of it, it's going to be the silicon carbide arm of that. These devices are really good at switching and converting high power at high voltages where silicon is not so good at it. More and more things like the data centers, power infrastructure, optical or solid state switches, they call it, are driving a lot of what we're doing. We're doing some really fun and exciting things that are new design wins and new what we call wafer pack, the contactors to enable these devices that we think are going to grow. This is going to be a good business for us. Just as an FYI, we just won the very first production guy for this.
Happens to be the biggest guy. And we've got other engagements. Gallium nitride is going to do production burn-in, and there's lots of reasons to move it to wafer-level burn-in on us. Okay? Next one is memory. First flash, then DRAM. Big markets, they, by the way, all DRAM is burnt in today. All the flash that goes into the enterprise is burnt in today. People go, "Oh, doesn't burn-in go away?" Nope, doesn't. The actual, the devices have too high of a failure rate for their application. So they're doing stress and strain on them for so many minutes. Flash and DRAM measured in many hours, most of a day. So it's very capital intensive and requires a lot of capability. Things that are driving there, what we've given is just a heads up on here. Flash memory, wafer- level.
We engage with a customer who's asked us to prove that our system will work in their application. There's things going on in flash memory. I won't get that much into it right now that are very disruptive and changing the way they're going to be making flash memory going forward. That disruption is creating a discontinuity technically that is advantageous to our system. We've partnered with a couple of folks there to do that. We've told people we're working on the benchmark. We actually said, and this is sort of news, by the way, not some big announcement, but we said our goal was to try and complete this benchmark by the end of our fiscal year, which was Friday. This is actually a photo of the new wafer pack contactor with a fine pitch MEMS. You kind of have to be in our space.
I look at that. It's like one of the prettiest things I've ever seen. But, you know, you wouldn't, but that is an amazing piece of technology to be able to do it. This gives us the pitch to be able to do the next generation NAND devices, but also for DRAM. This was one of the critical things to prove through. We've actually been able to get the prototypes in. The benchmark is not complete, but a lot of it's coming down right now. We had some delays with some of the testers that were actually related to getting some stuff out of Japan, as it turned out. This is looking good and on track. Pretty excited about it. We'll give an update at our earnings call in a month. Okay? Last one is AI processors.
Probably the most critical thing that we did in the last year was both the work in the package part and AI wafer -level burn-in. This is an example that happens to be an AMD one I'm holding here, NVIDIA one. If you haven't seen one of these come around, I always carry. I'm like, you know, like Santa Claus carrying around a candy cane all the time. I usually have one of these in my pocket. This is so much driving our roadmap at both the package and the wafer -level because the devices, all of those devices are all being burnt in, in many cases in this form. In other cases, in a rack, there is a huge amount of burn-in going on of NVIDIA processors on the entire rack with megawatts of burn-in for days and days. That's crazy. Okay?
There's no other way of doing it. Being able to move that all the way back to do it at the die level is an enormous value proposition that we've now proven through our first production shipments that are now shipping products and everybody's leaning in to find out more about it. We talk about why it's important. Meta put out some stuff about the failure rates. We have our first production win. Candidly, everybody, including us, we were hedging our bets. You know, we're not a, "Hey, you know, it's going to happen." We were very encouraged by the benchmark. I said the customer, no one was cheering us on more than the customer wanting to make this happen.
We were able to prove it on our new high power system, testing up to nine wafers at a time, or in this case, nine wafers at a time to 3,500 watts a piece, which is unheard of in our space at even one wafer. We're doing nine at a time. This allows them to get the cost effectiveness and the throughput with a fully integrated automated system to be able to meet their capacity needs. In this case, the first order was $10 million. There's a company that we're doing, you know, 60-ish million revenue. You can kind of see how this thing can deflect. We expect them and candidly all of the markets that I've talked about so far to grow next year. Okay?
The interesting thing with this is this showed up, is installed at one of the world's largest OSATs, who is also selling to the bulk of the West Coast AI suppliers. They are parading the AI guys through and say, "Look what we have. Can we get your assembly business?" We are getting all kinds of sales leads, if you will, from the AI guys, "Can you test my part?" Exactly what we were hoping for, kind of kept it under wraps until we got here. Now this is working for us. Stay tuned for more news about the AI wafer-level burn-in side of things. Okay? Kind of a theme we have and we are really serious about it. Our solutions have a lot of technology, a lot of capabilities. We design things in a way to absolutely positively validate that you are getting a proper burn-in.
These devices, if they get shipped into an application, are extremely expensive and with the failure rates, very likely to fail. One of the big threads I had, I've spent many and done a lot of keynotes and things around the world in the automotive side on the silicon carbide side is we were running into people talking about the mission profile of an automotive car. What that is in that world is how long the device is expected to last. If you go through an ICE engine, your car lasts 200,000 miles instead of 100 before. You drive it for 30 miles an hour, it has this much, et cetera. You go through it, it's like 8,000 hours it's running. The semiconductor supply has got really good at figuring out how to make things that would last 8,000 hours. Okay?
Silicon carbide is still failing in its infant mortality at 8,000 hours. We had customers saying, "Oh, no, no, that's going to be good enough." I'm like, "Good enough?" Like, it's not good enough. Like, how long do you think an EV is going to run? What if it's autonomous? Some of the big EV players like Tesla understood this. They're like, "Don't look at the mission profile. These things, I'm looking at the bathtub curve. You have to weed those out." The people that took the commitment to do that were able to successfully demonstrate that because the reality is, isn't an automotive, you have a warranty, whatever it is. Most countries around the world, U.S. and the European ones, you have 20 cars fail 10 years after a warranty, you have a recall.
There is not enough capacity in the world to handle a recall of all the inverters in the world if they start to fail. There is more and more of that going on. People are being very attuned to the fact you need to basically test these and do the proper stress and strain conditions to weed that out. That is the area that Aehr Test Systems really differentiates itself on. I will leave you at that. I think we have 12 minutes for some questions. I will try and take some. Jed, if you, Jed usually gives me the hard ones, but if someone has some easy ones first, but any questions?
I'll kick it off, Gayn, and then maybe somebody from the audience for either Chris or you. I guess the first question is, all of these things sound great. Maybe just update on, you know, that you have two businesses now in terms of, you know, the capital equipment versus the recurring revenue or the, and from a backlog perspective, you've kind of gone through sort of the trough of disillusionment. Where are you relative to some of these opportunities? How should these folks think about, you know, timing associated with that?
Okay. I actually interpreted a couple of things. Yeah, it didn't really come out in this model. In order to actually, one of the key differentiations in wafer -level is the ability to take a wafer and pack it into this portable cartridge we call a wafer pack. The tester doesn't work without it. In fact, this doesn't work with any tester in the world, nor does it work with any handling equipment in the world. You need it all, which was a high risk profile. People were like, "Wait a minute, like who should I trust you? Can we do it?" It's sort of an all or nothing deal. When you get there, you then need to buy those wafer packs every time there's a new design. This year where we had a huge diversification, last year we were 90% was on silicon carbide EVs.
This year, which our year ended Friday, it's like 40%. Almost all of that was the consumable because they were all flat. The business was flat and we actually had, you know, a really good business in the consumable, which has very good margins. And it's important, you know, customers, of course, want you to, you know, be frugal and all, but that's a key part of it. We were able to continue to do all our development, et cetera, and extend our roadmap by just the consumable side of it. It's a business model that works for both. In general, we probably will see, you know, 30%-50% of our revenues every year be a consumable and over time even exceed 50% of our revenue. That's a good thing. That's true of all of the markets we talked about.
Just in general, where the markets are, we're literally, you know, I don't want to use the initial innings, but the hard disk drive, we have a new hard disk drive application for something. A lot of people figured it out. I never use the words. I'll let someone else do it. This customer is ramping this new technology for a read write head and it's driving wafer-level burn-in. They're building up a backlog from us and they wanted us to ship. We're going to have a very good year this year and it'll continue to grow, we believe. Okay? That's an application where there's only a couple of players that really matter in there. It's going to be lumpy, but a good business for us. It could even be 10% or more of our business next year.
You look at gallium nitride, that has a really wide amount of application space. This is something that we think will grow in a number of customers and the diversity of that space is not dependent upon automotive. We will see a growth coming up there. On the AI side of things, people are like, "Oh, did you miss it?" or, "What's going on?" In some ways, AI is just getting started, especially if you look at the applications where people will be putting them into, you know, vehicles, autonomous vehicles, robotics, other applications. The need for the reliability of those is critical. There may only be one processor in that car or in that robot. You cannot have it like freaking out on you. There is plenty of opportunity to grow on the package and on the wafer -level.
We think that this is going to be a really good year for us coming up and along that. Flash memory is something that, based on this benchmark, we are hoping to engage with on a development JDA over the next year, probably a revenue for a year from now, but a huge opportunity for us.
Just on diversification, and I'll mention this, I know you can't, but you know, previously ON Semiconductor was your largest customer and you had significant exposure to one, you know, hard disk drive, Seagate's kind of going in this direction. I'm curious, you know, how should we think about diversification? Do we think about, you know, is it going to be one leading customer in each one of the new applications? Or do you think that as these others ramp, AI, Flash, that you'll see multiple customers ramp at the same time? Because that seems to be a.
I think it's safer to think that you start with one. It's kind of even our strategy. So it's self-induced. You know, I didn't want to go tell 10 people I could do AI wafer-level burn-in until I could do one. And even in our strategy of getting it out there, it's very impactful. Like we work through what things we needed to do. We don't talk about all the technology tricks that we did to do that. And now, you know, our competitors are like, "You're doing what?" Like it's just nobody's doing anything like that. That gives us some advantages. And in general, if we don't have a lead customer, you know, I'm pretty nervous and my knees wobble about, "Let's go build something because it's going to be the greatest thing and maybe somebody will want it." We're not like that.
We tend to figure out, and so far we've been fortunate enough to literally have the leaders. Like no messing around. This is not mom and pops. The lead two customers on the FOX wafer-level burn-in platform were Apple and Intel, like from zero. It's like, "Okay." They deployed that across multiple applications and continue to use our products. ON Semiconductor, who was actually, you know, a distant third or fourth in the silicon carbide market, clearly has taken the greatest position both financially and in their strategies. We still cheer them on all the time. We haven't talked about who the GaN guys are. One of the things that's changed a little bit, SEC used to require us that if it was a 10% customer, you have to name the name.
Customers were like, "You can't name our name." It's like, "Well, we have to because of, you know, SEC rules." They changed rules. We can't really use that excuse anymore. We have a lot of people that say, "Do not say who it is." It's a good rule. We actually have code names for everybody. We're pretty sophisticated with cameras and lockdown boxes. We're completely qualified for all the Apple black projects and stuff. It's important to be able to keep your secrets.
I guess I'll keep going here, but you know, visibility was impinged by tariffs. You know, have you seen any change with respect to visibility or confidence as you're working with the customers where, you know, the ability to kind of set an annual goal or something like that has been?
Yeah, that's a really, that's on our mind right now. I mean, we're doing our strat planning. We're putting our plan together for next year. We will have the quotas and bonuses and things tied to it. We pulled guidance last quarter. It was, you know, we were fortunate or unfortunate enough to come out with our earnings three days after April Fools' Day or what is it? Liberation Day? Anyhow, and it was like, "Oh my gosh, what is this going to mean?" We had customers calling us, "Well, I'm not going to ship this or, you know, what does it mean?" We went through it. We actually spent a lot of time. I encourage you to go back and look. We spent, put a lot of detail in to show that we're really not impacted materially long term at all.
Short term, it was like, "Well, we had this prober coming from Japan for the hard disk drive. I think that's going to get delayed." We were going to ship it in May. I think it's going to be June, you know, those kind of things. In general, people have calmed down. We've immediately took action though. Like we're actually doing sub-assembly builds at sites outside of the U.S. so that things don't even come in and have to be exposed to, "Can you recover tariffs that then you ship back out again?" We're already doing wafer packs outside of the U.S. right now. We were quick. It wasn't so much to save money per se because it may not even be material. I just don't want it to get stuck in the borders. The customers have been very positive with us about that.
Like, you know, they don't have to worry about us because then they go focus on some other things. But it's, you know, there's some orders that we're expecting to get in May that didn't come in. What does that mean? They go, "Oh, nothing's happened," but there's still some of that. I think long term it'll work through. We have to decide, do we reinstate guidance in July? If not, do we wait a month or something like that? That's something we're kind of working through right now. We have pretty good visibility, but it's always hard. You don't want to miss your numbers. If I tell you how big it is, that can seem like hype too, so.
Any other questions? We have time for one.
Yeah. So kind of the easy one is on the packaged part, they typically are like 20 weeks or more. You could say the company we acquired had better discipline than we have. On the other hand, we were with production people come up to last minute and they're like, "I need something quickly." On our wafer-level systems, our lead times from the time a customer says, "I'm going to be, you know, even inside of three months," unfortunately or fortunately, we often take an order and ship it within a week. There's already sort of some stuff going on behind the scenes, but we don't announce it until we actually have an order. We've been very flexible. That's been a positive for people. Our consumables, a new consumable with a full test program application correlation might take, you know, 10 weeks.
If you have a follow-on order for quantities, I can do them in six. So, you know, it's kind of the way you might think about it is, do you have to have everything at the first day of the quarter for shipments that quarter? No. Is that good or bad? Sometimes as a public company, that's not good. It's hard because it's hard to tell what's going to happen and customers can wait till the last second I can still ship to them. It's a positive to them. It's hard to be a public company and quarter to quarter kind of stuff. Thank you, everybody.
We'll be upstairs in the Jenny B breakout.
Thank you.