video_OCP2026_和碩_Rubin液冷與L13部署_20261015

Good afternoon, everybody. Uh my name is Christopher Leung and I come in from Pegatron. Okay, today we're going to cover the AI factory involvement and also the migration. And also we talk about the Nvidia Rubin transition from the previous generation and how the Pegatron going to be adapt, you know, from the new year. Uh couple of the cover slide that we're going to be quickly introduce who is Pegatron and also what's our adoption for the AI factory. And then after that will be some of the advanced, you know, cooling solution and also couple of the smart manufacturing plus the how to do the end-to-end from L10, L11, L12, L1 3. All right. So, who are here that hear about Pegatron? Okay, we have few folks. Great, great. So, you know, we actually started back in 2008 and we are probably one of the top five, top six EMS globally, okay? And we specialize on OEM, ODM and JDM, okay? And for the server division, we actually started back in 2024, okay? And as you can see our revenue grows in the past few years is actually trending very squarely. And then also in the forecast of 2027 is actually enormous, okay? In turn of the product offering, so in the server division that we actually have liquid cooling, air cooling on the full rack scale, okay? Nvidia solution, the latest Vera Rubin for sure. And then also the HGX Rubin as well. Plus the AMD instant MI355 and the also the future, you know, 400 500 series. And then on the air cool side on the rack scale, we also have the air-cooled B300. And then to complete the AI infrastructure, you cannot just live with the AI server, right? So, we also have our own developed you know, the high-density server and also the storage and also the multi-node solution as well. With all the server that we have, we cannot miss the networking, right? The networking is extremely important to complete the overall network infrastructure. So, we also have our white label 400 gig 800 gig and also Max Founder as well. So, this is a pretty much can cover majority of the AI needs today with our client so far. So, what our expertise? As you can see from the slide, starting from the left, so we start from you know, manufacturing design on the liquid cooling cold plate. Okay, and we also do full system simulation and full rack simulation and then followed by the data center data host the water cooling manifold design and also deployment. Okay. So, we actually so far in 2026, we already do full scale L11 L12 deployment so far. And in turn of capacity, here you can see that in globally we have Taiwan, Mexico and also Austin, Texas in US. And then here is our capacity as you can see in the worldwide and as you can also be able to see on the chart. Unfortunately, I don't have a pointer. Maybe I could. Give me one sec. Are you able to see? Yeah, a little bit. Okay, so we focusing on both 415 V and also 480 V. And then moving forward on our new factory in Taiwan and also in Texas in Mexico will be focusing mainly on the 480 V, okay, to get ready for the deployment. And the other thing that we also want to highlight here is about the testing station. So, a lot of you probably aware that, you know, on the rack scale deployment, the most challenging part is actually on the full rack scale testing in the manufacturing. Okay, so I will have some of the slide in the later slide to cover some of the video you know demonstration. All right. So, this is how we do the transition. So, back in 2024 we do single L10 solution, okay, and then migrating to 2025 we are able to do full rack scale in the liquid cooling and air cooling deployment. And 2026 we actually do cluster level L12. We helping our customer in their data center to do all the full scale deployment. As we are speaking, we actually have a few team globally that actually are doing the deployment right now with our client today, okay? And then we are foresee on the 2027 we going to migrating into the L13. So, you probably wondering that how we going to do that, you know, in the coming year. Basically, as you can see, L12 is focusing on the data hall, but the on the L13 basically is from the facilities cycle including power, liquid, and also air cooling, and also the floor plan design and how to interconnect all the power whip and everything. And that require totally different team. So, we actually are building our own L13 design team to be able to capable to to handle that. And on top of that, we also provide end-to-end communication. So, all the way from the beginning of the inquiry, putting up the design, and also communicating with customer, make sure the facility we do the site survey with a customer together to make sure that the the floor plan, the deployment mechanism are working, and then then we starting to do our deployment, and then after service. Okay? So, this is how we actually trending looking forward to the 2027. In turn of AI factory side, as you can see that we actually implement a lot of robotic. Okay, back in GB300 stage, we actually already implementing the GB300 automatic production line as you can see on the video. So, they will be able to using the gauge to placing and positioning the main board, also the switching board, you know, also integrating the fan connector, and also the precision screw to make sure everything is not, you know, touched by the human being. Okay, and this is actually increase our hourly output. So, usually in the men using the human production line, roughly about six to seven unit per hour. Okay, and then after we implemented into the automation production line, we actually increased increased to almost 18 to 20 unit per hour on the L10 deployment. Sorry, L10 production. Okay? So, we expecting that even go higher on the value being later would. Okay? And also in addition, here is our Oops. Let me see if the video. Okay. So, uh order to support large-scale deployment that we also need to have the uh um very, you know, large number of the testing station. So, currently we're designing this uh production line in Taiwan to get ready for the VR Rubin today. Okay. And as you can see, so we have to have a large number of the footprint and also the uh facility need to have all the liquid and also power to get ready uh for all the deployment and get ready. Technically speaking, we are looking at about uh one each rack uh on the validation lab need to be about 1.27 hour Sorry, 1.27 day per rack. Okay. So, in order to consistent output, you know, this kind of scale, we have to make sure that we have the placement to hold those uh testing bed. And also secondly is we need to have a manpower to get ready to have all the facility keeping uh test wisely. Okay. And then um as I mentioned, moving forward that uh from the GB300, we only see about 50% plus of the automation line implementation. And then for the VR Rubin, we're actually aiming for over 90% automation. Okay. That means less and less uh human involvement. And so that increase the quality and also the quantity and also the uh output uh ratio, okay, in the overall. Uh for the uh GB300, our 10 pass rate, we're aiming for uh 75%, which is what we have today. And then for the L11, we're uh over already over 90%. And then for the VR Rubin uh pass through rate, we're aiming for about 95%. Okay. That's uh that's what we're anticipating right now. Uh so, overall, for both the VR Rubin and also the Rubin ATX for Pegatron, we are going to be ready in production in Q4 2026 and full production in 2027 Q1. Okay, so that's what our anticipation is. And talk about the scale. So, as you can see back in 2026 today so far I know the year is not finished yet. So, we already surpassed you know 10,000 XPU. So, what XPU mean is GPU CPU all together. Okay, because we do deploy liquid you know scale in both CPU and GPU together. Okay, so in 2027 we are anticipating actually we already receiving confirmed order over 200k in turn of the CPU and GPU rack scale. Okay, so definitely if you are interested to see how Pegatron able to help you in turn of your get your data center ready and then get the general available time up as soon as possible, come to our booth. Our booth is at the E2 here is a show floor and we're definitely interesting to speaking of you and see how we can grow together. Any question? No. Great. Thank you. Thank you for your Oh, yes. >> You didn't talk about the 800 volt architecture. When will you see the 800 volt architecture >> Okay, so the 800 volt. So, currently we are actually working with few power manufacturer. For example, the well-known light on you know and also Delta and also some of the partnership like a Mac Meet. Okay, so we are actually currently evaluating the 800 V today. Uh the challenge today is actually the facility side need to be getting ready. So, uh by middle of the 2027, we should be already able to deploy the 800 V if the customer uh are capable to handle that in the data center. Okay, and we have few choices for the customer um to pick and then we can adapt that. And most likely I would say the VR Ruby will be using the side car option. And then also the next generation VR Ruby Ultra is also going to be the side car option as well. Yes. >> Will you be testing 800 V in your lab? >> Yes. >> Setting up >> Yes. >> that part of your >> Correct. So, that's uh where I covered in the Let me see if I can go back to the manufacturing slide. Okay, here we go. So, as you can see here, um for the 480 is actually ready to adapt uh the 800 V. Okay, so as you can see all the new plan that we do in Taiwan, in Mexico, in Texas, they are all 480 ready. Okay, for the new side and they are 800 V adaptable. Yeah. Okay. All right. No more question? Thank you and uh appreciate your coming and uh you know, hopefully we can see you at the our booth uh in tomorrow. Thank you. >> [applause]