NVIDIA GTC Live at COMPUTEX Taipei 2025 Pregame

[Music] live from the Taipei Music center in Taiwan. This is Nvidia GTC live at Computex. The eyes of the tech world are on Taipei as thousands gather at Computex, one of the world's largest showcases in the industry with a focus on AI and robotics, nextgen tech, and future mobility. This year, Nvidia's founder and CEO, Jensen Huang, will open the conference with his signatures keynote address. Until then, we've gathered some of the brightest minds and biggest names in tech to discuss the company's past, present, and future. The stage is set, the lights are bright. It's time for Nvidia GTC Live at Computex. [Music] [Applause] [Music] Hi, I'm Ben Ritz. And I'm Brady Wong. Brady, I'm really excited to be here because as an equity analyst who closely covers Nvidia, Melius Research in New York, I know that the tech community along with investors are looking forward to hear how Jensen and his team are driving the AI revolution forward. You know, the products and initiatives announced here set the course for the AI market, not only in terms of chips, but they also draw attention to an entire ecosystem that drives the creation of real world applications that can use the technology. You know, you can sense here, Brady, a lot of excitement around autonomous AI agents, physical AI, which is led by robotics, and advancements in amazing reasoning models, which really are just still moving at a breakneck speed. So, with that, I want to turn to my co-host, Brady. You're from Taipei. What does this mean, Computex mean to the community here? Yeah. Hi everyone. I'm Brady Wong from Counterpoint and I'm excited to have you here in my hometown, Taipei, Taiwan, Taipei, Taiwan. We are at Computex, a huge tech event that started back to the 1981. Over the years, it has become one of the most important stages for showcasing new technologies. Now, it's pressing stronger focus on AI and how it's shaping the future. It's a gathering point for the global tech communities where ideas are shared and innovation takes off. This year, GTC Taipei kick off as one of the first key event at Computex, setting the tone for the week ahead. Thank you for joining us and let's dive to the exciting world of tech GTC Taipei at Computex. Well, as you can see, guests are filing into the venue and you can see that behind us. But uh the as the anticipation builds for Jensen's big keynote which is coming up soon. But today in our pregame show, we're going to explore three chapters of Nvidia's story that showcases how it remains at the center of critical partnerships across Taiwan. To that end, some of Taiwan's and the world's most important companies are joining us to set the table for Jensen's big keynote, sharing their unique perspectives and histories from the front lines of AI innovation. So right now, let's take a look at how Nvidia's humble beginnings grew into something that is changing the world at a remarkable pace. [Music] This is now a quality day here. To help us kick off today's conversations is someone I've known for over 20 years, Jeff Clark, the COO and vice chairman of Dell, a company I actually cover as an equity analyst as well. So, you're near and dear to my heart. And joining Jeff is the Dr. Y executive vice president and the courtship chip operating office at TSMC. Welcome. Thank you. Thanks for having us. Great to see you. So Jeff, I was hoping you could tell us quickly why you're here in Taipei. Well, real simple. Jensen asked me, which is no small task given we have Dell Technologies World kicking off in a few hours in the United States. But how could you say no? Yeah. He's got a way with people, right? He does. Yeah. Well, you know, maybe Jeff, uh, you can walk us through your take on the evolution of Dell's partnership with Nvidia, starting from the early days, the gaming card integration, all the way to these complex full stack AI systems where you're generating billions of revenue together. Sure. I was thinking about when the fir the first time I met Jensen, which best of my recollection is early 1996 back in the day when 3D graphics was really making its inroads inside the PC industry and we started our inter engagement with Nvidia. If I recall correctly, it was our dimension desktop. It was the Reva 128. We put it on the map and the rest is sort of history. We were on the back cover of PC Mag which back in those days meant something where that was a real piece of marketing collateral that drove demand uh performance leadership and from that very point we've had a very close partnership with Nvidia. We've had many first to markets many performance leadership products breaking into new markets like the professional workstation market uh that we entered in 1997. We entered with Jensen and his uh platforms and products shortly thereafter with what was became the Quadro product line the professional graphics. We've been doing work uh I think deep level collaboration for again many many years that led our way to GPGPUs led our way to ultimately the things that we're building today and ultimately to the partnership that I'm trying to show showcase here today. the Dell AAI factory with Nvidia, which is ultimately a deeply integrated system and taking the pursuit of accelerated computing and the architectural transformation that's underway today, working together and doing things that were near impossible just a handful of years ago. I love that shirt. I do too. Yeah, you got the Texas logo on the arm. It's good. Hi, I YJ I believe TSMC and Nvidia has been working for a long time and both share the same engineing background background and also have very high commitment to the qualities. So what do you remember about the early days of working together with Nvidia and what made the partnership so special? Yeah, I think the uh unique partnership between Nvidia and TSMC began in 1997 when our founder Dr. Morris Chan uh personally reached out to Jensen Huang in response to his request for Foundry service and after initial connection we have worked together cultivate a very deep and enduring uh collaboration and Nvidia's first product as uh Jeff also mentioned the river 128 and also river TNT a star production in TSMC at that time in TSMC's is a newly uh I established age feep fab 12 actually at that time I was uh feed 12's engineering manager responsible for the quality and yield of that particular product so since that until two 1997 to 2017 yeah I think TSNC has delivered over six million 12 in equipment wafer to Nvidia I think the number grows and but we won't be able to relieve the the new number because it's not the public information. However, it's obviously that TSMC and video will grow together and we are thrilled by the remarkable achievement that DVA has achieved become the global $3 trillion industry leader. Wonderful. Yeah. 1997. Yeah. Last year I I joined TSMC but I lived five years after five years. Oh, okay. Yeah. Well, YJ, really good to see you. Well, it's a legendary partnership that you have with Nvidia. It's it's amazing. Um, how's the technical collaboration changed over the years? Was there ever a time where in you had to tell Nvidia to change course or vice versa, they had to tell you? Yeah, I think uh my my major engagements with Nvidia start from uh when I engaged with Joe Greco, senior vice president of Nvidia advanced technology team and uh starting when we are doing the development of 28 nanometer technology and since that we have driven the uh the limit of innovation today to three or two nanometer TSMC's most advanced technology as as well as our most advanced packaging coas uh the SOIC 3D dice stacking technology. I think Nvidia is always has the most challenging product with largest die size very difficult to manufacture require the highest standard of quality and yield and uh but uh however we we work together very well and closely and also we are always uh media is amazing customer of TSMC and constant deliver revolutional product to the world through TSM. MC's uh uh foundry manufacturing service and uh at the beginning we are just a partner on product design and uh man manufacturing service but now we work together very deep in technology collaboration and development beyond the what we call the design design for manufacturing or design technological optimization or even today to system level integration. So Jeff, now we are in Taiwan and we know know Taiwan is a center of the chip production al and also the system production, right? So could you talk about the Taiwanese Taiwan chip and the system pipeline that enable those global executions for decades? Sure. But maybe to build on a little bit about what YJ was talking about, I remember entering this industry. My first design was an 8086. Less than 50,000 transistors, 3,000 nanometers as we would cast it today. 286, 150,000 transistors, 1500 nmters. And now you're putting a 100 billion transistors on a die at two nanometers. It's a 3 million times improvement in density, 750 times smaller. That's what's enabling technology. That is why we're able to do and build the things that we're building today. And then you couple that I was reflecting on uh my time here in Taipei. I think I first came in 1990. So I've been coming for 35 years. Whoa. I grew up in this industry with many of the tech giants in this on this island. And I think about the innovation they've brought, the manufacturing capability they've brought, the engineering capacity and capability. And it's able it's enabling us to take a desktop market to become a notebook market, become a server market, network market, storage market, building out the fundamental infrastructure that our company needs, our customers desire to win in the marketplace. And I think about what's happening today with the computational intensity that we're getting out of the tremendous chips that are being manufactured. Putting that together with high throughput, low latency fabric, taking that with some of the high-speed disagregated storage architecture that we have done. You bundle that together and you build these enormous computers that are capable of things again that we didn't think of two years ago, three years ago that were possible to do and we're solving huge problems and we've only scratched the surface. Yeah. Well, YJ, uh, sounds like you make the world go round. Uh, Nvidia's got a lot of great products coming. We got Blackwell, GB200's, and then the GB300. Reuben, how are you collaborating about what's next? How you shaping the future? We don't want to steal any thunder from Jensen, but how are you helping them make it happen? Well, as uh uh for TSMC, it's important for us to put the right technology at the right time, of course, at a cost that makes sense. and Nvidia one of the Nvidia's greatest strengths is their ability and agility to adapt strategy to keep pace with the ever uh continuous changing market trend. So throughout this uh years of collaboration, we have been able to uh uh develop a reason to work together, allow us to response quickly and to cope with the uh fastm moving market trend. How about you Jeff? How are you adapting to move fast on Jensen time? Speed of light. You mean you've heard that, haven't you? The speed of light. Um, look, I I think what's happened is for us, we've had to rebuild our engineering model. The time between new pieces of technology and its adoption has forced us to rethink, reinvent our engineering flows, how we produce the type of energy density, the GPU density, how we power and cool these things in record time. It's changed the way we've innovated. We're quite excited in our ability to respond. And it's the deep collaboration that we have with Nvidia that's enabled us to do that. Uh the yield was we were the first to ship a GB200 with NVL72. That was deep collaboration done in a very short period of time meeting the market needs and the resp to respond to what customers wanted and with Jensen going we need to get this to the marketplace. That's what's underway. It's changed the fundamentals of how we develop and deliver systems and solutions to the marketplace. I've seen you work. This is the fastest I've ever seen you guys work. So YJ, what are you excited to see today? What do you want to hear from Jensen today? Oh, definitely. What's his outlook on the AI the new era everybody is looking for. Yeah, that's critical. Yeah. And we understand what the future what the future application future opportunities. Yes. Yes. Definitely. How about yourself? What are you looking forward to hearing from Jensen today? Uh how the flywheel is going to continue, how it's going to speed up, the types of capabilities that will build, and ultimately the problems that we will be able to solve and improve mankind, improve uh human progress. We're building incredible machines to solve problems that have never been solved before. That's pretty powerful. Yeah, I I mean, I agree. I'm I'm really looking forward to what the applications are that he decides to highlight today. I think the whole industry is looking for it. Well, in summary, I think we heard, you know, a lot of great things here. Um, I think that, uh, you know, these are two of the deepest partnerships that Nvidia has, longest standing partnerships. Uh, Dell obviously your your history goes way back into the 80s, meeting Jensen in the 90s and now adapting throughout, you know, the whole e the whole GPU journey is unbelievable and I've seen it. Uh I'm lucky to have a front row seat and YJ your company without you the world doesn't turn on its axis. Uh it's really clear that everybody what we see today from Jens wouldn't be possible without you guys. So you know we look forward to hearing more about you. Uh also I say as an American welcome to the US. I know you already been there a while but we're looking forward to what you're doing in Arizona. Yeah. And uh you know I can come visit anytime. Sure. So uh thanks a lot gentlemen. just uh you know great great uh to speak to you. Thanks for having us. Thank you. Thank you for having us. Now we'd like to fast forward to present day as enduring innovation and partnerships from the early moments are now being realized in the AI [Music] factory. Heat. Heat. [Music] Let's welcome our next three guests. IA Chen Jang and Mike Young. I chen the co-founders and the CEO of Tren Michael the chairman of Asus and Michael Young EVP of Quanka Computer and the president of Ka Cloud Technologies. Welcome. Thank you. Great to see you. Great to see you. And hi Brades. Thank you so much. Nice to see you. Nice to meet you here. Yeah, I'm really excited to speak to you all. Um, Eva, we'll start with you. No, thank you. Maybe first. Yeah. Well, I think you're the only security uh leader um that we're that we're speaking to this morning. So, you know, securing AI is a huge deal. We get asked about it a lot. I want to learn from you. How are you doing it? And how are you working with Nvidia to secure AI? It sounds like a big job. Wow, that's a big job because you know if you think organization is like a person security before we were just securing your leg your your arms right but securing AI is you need to go into the brain you you securing the brain and the whole nerve system where the AI is sending the information into all this nerve system and um if you think about IT industry it's all about the computing power transfer the data into information and generate intelligent for people to use right so this whole stack is hardware all these people muscle to to build all those hardware and then the I call it the uh infrastructure software like bios OS network drivers and then on top of it is application software HR software finance software, R&D software, each of them then manage the data, analyze the data and generate the functional intelligence before that was before after AI actually all the data are coming from the top and go into the AI model pre-trained integrated with the whole computing power and then the application is just util utilizing all this intelligent generated from AI to present the data. Right? Cyber security is a very special layer. We inserted in between infrastructure software and the application software. We protect the whole information flow between all this software and east west traffic information flow. So we require a huge computing power to analyze all the data. That's why starting uh T micro actually starting 2002 we already try to explore and utilizing the DPU from Nvidia and later on also using GPA GPU from Nvidia to do all this security data analysis. But now with AI is even more collaboration. For instance, we are using Nemo finding the Nemo. We are using Nemo to train our LM specialized cyber security LM called Cybertron. And then we also using uh kud is the data analy pipeline, right? We use includive to make sure we can better analyze the data faster and then we can detect the anomaly using morphy's uh uh system and then finally we package all of this including the ARM and our agent the agentic AI or the agent in NIM in the NIM fashion and therefore we can faster deploy all this security into customers AI infrastructure. So that I think a very deep collaboration with Nvidia. Thanks Eva. Yeah, there are a lot of insight and you know in the morning I talked to Eva my my brain is full of the hardware. Now thanks for explanations now I I understand the critical of the software. Yes in the hardware. Yeah, you are the legend of the Taiwan technology industry and how is the ASUS approach to qualities and reliability and co-engineering with Nvidia critical to could scale the AI factory. Jensen's vision on the AI factory is indeed the next industrial revolution to push forward such an incredible revolution at live speed like how Jensen always wants things you know the importance of quality and reliability. Today's AI servers packed with massive numbers of GPUs to maximize model flops are pushing state-of-the-art. High bandwidth, high frequency and high power design to the extreme making signal integrity, power integrity and thermal reliability absolutely critical. Quality and reliability have been the cornerstone of ASUS starting from our day one making motherboards. We have always encouraged our engineers to master signal integrity, power integrity, thermal simulation, and computer architecture. For years, we focused on sot state-of-the-art. It's incredibly, you know, painful. But ASUS engineers love the challenge. Overcoming pain and bugs is a part of our DNA. So we love this kind of challenge and we really enjoy the core engineering with Nvidia which is indeed critical to scaling the AI factory. So, do your engineer talk to you, tell you they like the challenge, right? Oh, always always. They always love the challenge. Yeah. Well, Mike, you've been waiting. Me, too. Mike, you've been waiting patiently over there. Uh, tell us about Quana and QCT's approach to AI factories. I mean, it's amazing what you guys are doing. I remember thinking about you as ODM for PCs and now you've evolved so much. Yeah. Yeah. I think uh AI factory is a very interesting topic and I think uh from a ODM point of view this is really something because of uh you know what is uh AI factory you know I've been you know also trying my best you know try to understand more about this from Jensen and also how to you know collaborate with my internal team not only just the hardware but also the software then I figured it out that AI I factory is a you know especially QCT AI factory is designed to turn your vision into reality and uh you know we actually also transform your data your token ying into a practical and AIdriven intelligence output that is the token out. So in short AI factory you can treat it as a token in token out but I think AI track AI factory the most important thing the core thing is AI infrastructure the infrastructure is including the server storage network management tool and even the facility but not only just that but also you know we need to work with the open de facto kind of partnership you know the ecosystem so companies like uh you know like Nvidia ML common and also hugging face and etc you know to support enterprise with ML ops dev ops and also multi-modelings and the AI factory is not only just these two but most important thing is from the business point of view you need to fulfill customers requirement. So from the uh vertical industry how you fulfill their use cases with the AI embedded that's another thing. So we try our best to integrate the whole thing and speak the same language with our customers and provide the best fit solution for them. So this is I think is what we are calling the AI factory and from uh you know quant and QCT's point of view we we've been facing in this into our server factory our manufacturing and then we find out because of omniverse because of the digital twins you know it help us improve the labor cost cut down 50% And uh in terms of productivity increase 30%. And the most important thing is when we change the line do the production planning it also help us to reduce the time cut in half. So I think basically this is the target that we're working on and which is very much in line with Jensen's idea. Yep. Right. What I learned from media is that AI factory actually is token creations, token factories, right? Yes. And take TSMC for example, we all expect TSMC to increase the capacity, right? So we expect the AFA to create more token and create more money. That's how I said it. Business model. Yes. Yes. Yeah. Yeah. And yeah, so again the question for Mike is like do can you discuss you know how you work with Nvidia partners to develop the advanced server platform? Right. Okay. I think this is a wonderful journey you know I think I still recall that at the beginning stage we start you know my engineering start working with uh Nvidia uh from the DGX1. I think everybody knows that's the first GPU server or Nvidia in 2014 15. So it's been almost 10 years and since then up to 2021 uh we've been working with uh Nvidia for generations of HGX server machine for QCT OTS of the shell product and not only just that but also tier one CSB customers AI server. Okay. So with this so we develop ourself become a good bridge between Nvidia and CSPs Nvidia and the enterprises. So this helps. So with these all these years so we've been working with uh HGX. Let's talk a little bit about MGX. MGX is a very creative modular design architecture project which allow all the customers can catch up keep up with Nvidia's product strategy one year reason because with this you know so now you can see that Nvidia expedite pull in the schedule of their plan in direct level scale up to uh GP200 MVL 72. So with both parties collaboration and within this project quant QCT you know we also provide lot of system design system boards design and also like a mechanical and also thermal design. So I think that this is a wonderful contribution and that we love the journey with Nvidia and challenge you right. Yeah don't even mention about that you know always you know challenges work with Nvidia because Jensen I think he's always very you know how yeah creative so always we need to keep up with you know his thinking. Yeah. digital question here. You mentioned the one year RI, right? Yeah. Yeah. The first I heard that I was very surprised because that's a very fast space, right? So there so you and also maybe the TSMC and the others companies vendors must also face the challenges. Yeah. Try to keep up with the video. It's a it's a big challenge for this industry because usually you know well you know for x86 you know this kind of uh you know uh product lined up usually takes a longer time you know so mature you know but comparing to speed of light okay amidia this is different so originally when Jensen mentioned about you know one year reason you know everybody especially you know from my engineering just say are you sure you want to do this with all you know all the new advanced technology that is more important it's not mature technology but we take the challenges but you did it you did yeah we did it you know I think we challenge just preparing this pregame show thing the change speed of Nvidia's staff worker is amazing every minute Nvidia just launched a new chip. Well, that's great. Johnny, let's turn to you. Uh, I remember listening to you 20 years ago as a as a player in PCs, but how did you evolve? How did how did you get Asus to evolve to be an AI infrastructure company and now partner with Nvidia for for these nextgen systems? Okay. Our strategy is to go all in on AI. Okay. At ASUS, we have a comprehensive AI strategy called ubiquous AI, incredible possibilities which aims to integrate artificial intelligence across all facets of business from the cloud to the edge to PCs and end devices like phones and robots making AI accessible to everyone everywhere. We also invest heavily in software development from the infrastructure layer to the system software platform layer including our vital ASUS AFS AI fun service, ASUS AI hub, AIDC, ASUS infrastructure deploy center and ACC ASUS control center and the application layer like smart medical, smart manufacturing, smart retail and smart city and so on. Our goal is to provide an end to end AI factory service that includes design, manufacturing and service. Yeah. Thanks to the national level Taiwania supercomputer project which we took the major responsibility for we have establish a full spectrum of capabilities from hardware to software to support Nvidia's magnificent AI factory vision. Beyond that, we've established global server manufacturing facilities at major global locations to further shorten the supply chain for our UI customers. One of our unique strengths is that we have over 2,200 locations globally to serve customers, including tailormade SLA endto end solution services and work closely with major vendors and partners worldwide. Our successful cases cover Taiwan, Japan, Korea, India, US, Europe, Middle East, and many countries in Southeast Asia. Thanks, Johnny. Thank you. Yeah, again, a lot of insight here. Yeah, thank you. And IA the next questions is important as we are now in GTC Taipei at Computex Taiwan right so I believe Tren Michael is a Taiwan based company and could you share us what's a unique advantage come from being a Taiwan based company well actually uh Tren Micro is the only global software company that is originated from Asia. We are the largest one that is originated from Asia and um so software is the part that we are very unique in Taiwan. We are so famous with all this hardware manufacturer TSMC and all these muscle guys and but last year's GTC uh actually last year's computer text when Jensen shows that the map of uh the supply chain and the little Taiwan in there. Try micro was the logo underneath there and we were the only software company. When I saw that map, I my my tear almost come down, you know. It's so I understand heartbeating. So I think this time the AI security AI factory is actually originated from the software factory. You know, some last time uh Jensen was uh talking about this AI factory and he he say uh he call all the software engineers as the software factory worker. I I was saying oh thank you very much for calling me like I thought I'm the white collar worker now I'm the all blackheads right but that's I think is what AI factory means right if you transform from software factory to AI factory this whole ecosystem is going to change I believe uh Taiwan should play an even bigger role in the whole IT industry not just the hardware computing power um manufacturing but the integration of software hardware together to solve customers AI uh question and problem infrastructure new factory concept right that's the new factory concept before hardware company like Asus they pay software they they pay Microsoft a lots of money to to bundle this operation rating system, right? The margin from uh I I've been joking but truthfully I think if if Jensen can convince Trump that all the software worker is actually manufacturing worker then Trump's dream of like moving manufacturing back to us you already done right software worker are factory manufacturing workers right so so overall I just want to say I believe this uh AI factory concept will change the whole ecosystem of it and then working together with the computing power and the software application. Yeah. Then we can provide the best IT infrastructure AI infrastructure for the whole world. How are you? Nice to see you. Hey, you guys you guys look good on TV. Yeah, you guys are you're you're like broadcasted to this giant movie theater. You guys are gigantic. Yeah, this is a very professional studio. It is a professional studio for professionals. We have partner here. You are going to sing solo in the in the show. No, I I was I was honored to watch out for you. So So this is what happened last time. This is what happened last night. So, so it takes two hours at least, maybe three hours to seat all the people, right? Because they've been waiting in line for hours. Yes. And it takes a long time this. And so we came up with a good idea. Let's do a pregame show. Yeah. And a pregame show of industry experts just like NFL football. They're all, you know, they're all previous, they're all coaches and Right. Thank you. Our honor. Yeah. And so it's a it's a pregame show of experts. And this is what happened. So after GTC, they asked all the audience, they enjoy the pregame show more than the keynote. Come on. I don't know about you. Enjoy your Dennis delivery. They want to know your inside. So anyhow, this year. Anyhow, I'm so happy you guys are here to do this. Thank you. All right. Finish your job. Thank you. Thank you very much. I'll do I'll do my job. See you in a bit. You got somewhere to go. Yes. Wow. What a big surprise. Yeah. What a big surprise. How can he have so much time bumping here? Super. What do you think about that, Mike? Oh. Oh. You know, Jensen is always a big surprise to us. Not only just uh from the technology you know product but also the business model as you he just mentioned about you know AI factory you know it turn the technology become kind of monetization I think this is quite impressive yeah always very creative innovative that's how I know him yeah a true leader in every way what what are you looking forward to seeing today Johnny in summary. Okay. I think one one thing I would also like to take this opportunity to highlight is that recently people may have misinterpreted you know the impact of deepsek. Yeah. My belief is that you know it also it actually echoes Jensen's vision of AI factory. Yeah. It further popularizes intelligence manufacturing. I jokingly called it, you know, the poor man's Chinese kung fu. We have been watching them for more than one year. Uh, since the the publication of their first MA paper, which we were very impressed by and then the following, you know, uh, R1, yeah, and distillation are all very impressive and solid, which also enables, you know, small models. to blossom like wild flowers. Yes. Imagine one day the world will be full of AI brains. Yes. Yeah. That come from that come in different forms and you know uh and sizes including super big, medium, small and even tiny from the cloud to the edge. Edge. Yes. To PCs and ant devices. Yep. like phones and robots. Yeah. And I we are all very very excited, you know. What did Jensen inspire you in closing to say? Um well I actually will say last time he talked a lot about sovereign AI and when uh Tanker actually is very big in Europe and in AMIA all the countries the government sector and that is the strong demand that I'm seeing that sovereign AI and just like Johnny say uh every country will want their own brand right and then every enterprise will have their domain knowledge their IP in their own brand. So, right, that's what what we uh usually say. So, I think what uh Jensen inspiring me and a lot of the software people is thinking about all this AI is not just one big open AI, right? Is every every enterprise, every country will have their own AI new universe. Yes. Well, uh thanks so much. Thanks. A lot of great comments out of you, out of you all from the security side all the way to the AI factories, these great partners. Thanks a lot for for joining us. Thank you. Thank you. Look forward to talking a lot more in the future. Thank you very much. So, as a reminder to our viewers, Jensen's keynote is coming up at the top of the hour, but before that, let's take a look at how Taiwanese companies and Nvidia together are advancing physical AI to the next frontier. Heat. [Music] Heat. Heat. Heat. [Music] We're so excited to discuss the world of robotics with Scott Wong, COO of Techman Robot, a division of Quant, who helps companies develop robots. We also have Dr. Peter Pang, director of Delta Electronics Robotics and Smart Manufacturing Divisions. And finally, Bran Chang, chairman and CEO of Foxcon Industrial Internet. Thanks, gentlemen. Thank you. Good morning. Thank you. Morning. Hi, Scott. Welcome. Yeah. Um, I have a question. We know the robots in your Tech Men has a collaborated robot we call cobot, right? So, it's vision based robot, right? Could you help tell us about what the role did you see the Tekken man playing in the next wave of intelligence manufacturing especially in Taiwan and um any practical example use case you can share with us? Yes and actually we how say we start cobalt development around 13 years ago and actually the cobalt we have a vision embedded. So in the beginning we think the vision capability is must to have for the cor robot. So we not not just provide the cover the motion only. So for example the the vision here we begin and it's native. So the capability for a robot is must to have. So in that timing we provide the house smart robot robot cor robot for our customer we the differentiation compared with industrial robot is a gap. So because traa what they have the sensing only the motor angle they have in chova because just they do the some movement one point to another point. So the computing power for the robot right industrial robot just motion only system of degree of freedom. So the computing is very easy but once the the intelligent come the vision is most important for the loa like a human right we can uh we can understand the environment we can reasoning we can planning what we would like to do that's why we have this capability most important is the vision so 13 years ago we think is must have if compare the for a while I think we are very few company is how say vision is building in our uh robot so in that timing so we uh that's a long few few years we cope with Taiwan I say manufacturer company yep yeah for example semiconductor mhm and electronics even CNC machinery those manufacturer and what they need is a intelligent solution for them. Why they needed a intelligent? Because if we just take one uh object from one point to another point like the brain work to do that things but we we have the vision. So in their for example for for for CNC in their production their object come their the object don't need to fix in one precise position because our vision we can do positioning compensate the variation and the pick and after how say process and we also can check the the quality maybe some scratch And the AI function we are wrong for sure. We also few house uh corrupt robot or industrial robot the control bus we embedded how it GPU inside. So we can do uh AI inference in our control bus. So like age computing. I see. So that means the robot can do the calvation by itself and also can recognize the image and also can make some intelligence movement by itself right and with the box beside below embedded with Nvidia GPU. Yes. It can do some high level. Yeah. Language model something right. Yes. Yeah. So today you bring the robot. Is that the real robot or just the mini one? Sure. It's it's a it's a toy but Okay. Okay. our the technobizer. Yeah, I think it's easy for I understand. Yeah, you you understand what's the difference because you can see there the the vision here. Yeah. Yeah. Well, thanks. Uh, hey Peter, can you talk about your collaboration between Delta and Nvidia? Maybe focus on the integration of your Debot series of robots with Nvidia's omniverse. Sure, of course. Uh actually we are producing a lot of different kind of products for data center related of course included ambidia and uh we are building the power modules for the data centers related but our production line didn't really can be optimized or remote control quality so easily yet we have the chance to cooperate with Nvidia and based on the only and combine with our we called machine and robot twin that can based on our age computing connecting our machine and robot immediately real timely return the necessary critical parameters and back to remote for example our production line in Thailand in somewhere else in the world but we can remote monitoring measuring and even calibrate like Scott just mentioned the robotics while the machine have some strange things happened we can remote calibrate based on the omniverse thanks to the omniverse and uh we have the beginning from the real to scene we have the real production line but together with universe we becomes better and that is not enough people need we have much more we need to have much more production productivity Like Jensen just shared with us the day before yesterday mentioned about people used to consider cost but now more more and more productivities but from our side we need to be balanced. So for the new production line can we do from seem to real or not? Can we minimize the unnecessary waste as much as possible? So we also cooperate to based on the oniverse to do the virtual design even from the very beginning of the product design and then later on for the related machine robotics uh together with the necessary process uh technology to make the products correct and make the customer satisfied. So based on onus again we do from sim to real that means we don't need to really build the production line so soon but we can make sure everything that almost like the reality and minimize the trial and error after. So we cooperate with a media so closely from real to seam and seam to real. Okay. So this is what we have done and we will continuous do not only for the robots but machine process the whole production line and actually this afternoon we will show the whole factories simulation as well. Thanks. Well hey Brand great to see you. Uh I mean you're you're a company that we look at in our work very closely from your work with you know iPhones to smartphones to everything but this robotics initiative is really amazing and I think it needs more coverage. I was wondering you know how did you how did FII and Nvidia first start working together and collaborating? Okay so probably people know we are running lots of manufacturing for the electronics. So we have a more than the entire Fosskan group have more than 200 chemists over the world. So just like what uh Jensen mentioned the evolution trend of uh uh AI from perceive AI to generative AI to agentic AI even up to physical AI we gradually follow on this. So back to 2016 we built the HPC with Nvidia solution for medical research. You know we do not only make it as smart manufacturing but also like me medical research to help something like a medical car simulation like 3D rendering in terms of tumor research. You know the phase two we engage with media to build a cluster of a V 100 in Wisconsin. Probably some of people knows that we have a very big campus in Wisconsin to to produce the electronics product over there which we leverage the the machine learning and start to use a lot of tools but back to 2023 when GI is booming was booming so we start to look at that uh how to leverage the you know the rapids under which under the the isac you know So you know the complexity of material moving on the workshop is very complicated. So how you going to use AI? So we do train the AI model to move the material and send a signal to the material tower also correlated you know to the to the AGV to the AMR to the robotics. So actually we make a ecosystem about this you know to to drive the material handling and at the end the impact is uh uh like we improve the unit per hour per person about 73%. With this AI models with AGV AMR together we reduce the delivery cycle time about 25%. Whoa. Yeah. And uh more than that at the worst we are thinking how to raise up the the you know the automation level because in a lot of manufacturing process if you look at that some you really can do the automation and put the low bodies inside but for some material maybe very soft very flexible then you need more something more beyond than lower arm or AMR AGV which maybe you need humanoid robots. So we have studied on this you know to put into the production line into simulate process and uh toward like implement like a physical AI eventually. So we can res resolve all these issues and really make entire production fully automated. So with lease as example you know I think I believe a lots of company especially Taiwanese Asian company like to invest in Vietnam in this era right in this decade lot of people think about that in Vietnam maybe the labor costs are cheaper they have many people actually she's wrong Vietnam probably their population lesser than 100 million compared to China is about nothing so actually the automation level we have in Vienna is very high you know why because they don't have so many technician which can repair the machine can optimize the parameters right can you know coordinate the layout and so on so we implement AI model we implement technologies to make it easier then uh how you going to enlarge this how you going to enlarge this so we use omiverse so back to Vietnam And the factory we have in Taiwan. You can see that in 2024 both our factory in in Taiwan and we may got certified by World Economic Forum as a lighthouse factory. So Golash factory is not just automation but also come with a lot of agentic AI application already you know and uh the one the server factory we have in Taiwan is a is the first one in the world for the server you know as a lighthouse factory so going forward just like Dr. P said that we leverage Nvidia work very closely on obverse. How you going to reduce your cycle time? How you going to drive high accuracy, high efficiency, high consistency for your operation? When we have more than 200 campuses for the same product on reverse with reverse with the isac plform and video have we work on robotics you know then also lots of uh uh m uh uh uh uh training then we can keep moving forward to duplicate then we can do simulation we can do validation we can do decision making at the end we measure everything before we deployment you're pretty Good omniverse salesman by the way. Oh really? Because I'm doing that. I have a whole team whole bunch of team doing this because got it. The thing is come with we face challenge before now we see the solution. So we need to adopt invented solution to resolve the all issue we have right. So we can overcome. Yeah. So we can overcome the challenges coming along the way. Yeah. Thanks a lot of good insight and I always like to hear about the use cases and you do bring a lot of use use cases. Yeah. And next question is to Peter. I think we have uh some discussion yesterday or a few days ago, right? So could you tell us how is the partnership with Nvidia simulation accuracy and also help to accelerate the development of robust solutions? Of course. Sure. just right after the brand just mentioned. Uh we are so pleased and thanks to the Nvidia related uh USD and uh that so wonderful universal scenery descript all different kind of tools that we are using for the old times but nowadays all connect together and just like we just discussed how Isaxim can help us. How is ice can help us? For example, we do the kinematics of the robots for a long times but that is not enough. It must come comp uh integrate with the necessary critical process who really makes the product come true together and again together with Omniverse not only is SIM but also Isac gym that we can pair processing much speeder because pair processing and the speed can be double and double and also one very critical important secret is can we do the machine for producing the product at the beginning not wait until the product prototype almost complete that could be too late already. So we are appreciating the oniverse surprise award foundation model and that the cucurus can support multi-user at the same time in the same space and can duplicate to the other. For example, when you are design product, maybe you are too busy, no time, but we can see what you are doing and then we can consider ah okay we know what Brady is thinking right now. So we have to match synchronized cooperate concurrent engineering at the same time. So mention about again what uh John Jensen just mentioned the cost is very important the productivity is important and also the cycle time is very important and we are wondering if we can secure the quality that means what do the rising right at the first time so if we can do the job well at the beginning together synchronized it's idea before but nowadays thanks to only we have the chance to do it together at beginning then we can synchronize from time to time to make the unnecessary waste down to the minimum. So how about the speed must be very fast right? Yeah. Yeah. The world is changing efficiency serations though so we need to be be flexible and quick. So about for example our company's robot I start uh my first job 40 years ago for robots very very old time and I start from industrial robot corab robot our depot cor robot is not only a robot but much more important it's a solutions that can really solve the end customers problem not only motions but process and thanks again to Omniverse the simulations becomes much more real. You mentioned the accuracy. How can we do it? Not only kinematics but the dynamics. How about using the skin, the material, the all kinds of physical that we cannot identify so clear before. And also at the same time we used to do a lot of process simulation for example uh by using uh anis. Nowadays that all can be connected with oliverse or even just do it in oliverse. So a lot of challenge coming but still I believe together with omniverse we can conquer so many things not only the machine not not only the the production line but the whole factories. Just uh quickly Scott wanted to ask you one more question. Um okay you know robots used to follow scripts. Yes. Now they adapt and learn in the field, go around and learn. Um, how did you deal with that in work in industrial robots? How did you how did your team adapt and and work with that in the industrial setting? Yes, I think the how say in the for the robar right actually is uh some challenge we need to solve and for example the uh for sure the smart robot is needed for the and customer but how easy for the uh engineer they can how say imp implement the solution in their path 9 currently in automation uh industry actually is still a gap. So we uh try to provide a very easy to use solution for our uh customer and our partners. So uh for example in the tradition right just mentioned is only uh the angle of the motor we have they have but now the vision vision image is come and the force sense feedback it also have and also AI power is come and the next step is so the very uh powerful function how engineer can what they can Writing the call I think is another good thing for them. So it's take a lot of time. So what we are doing is the first thing is we first hardware hardware hardware is how say easy ready to go. What what I mean ready to go. So it mean once the robot right mean they have the the robot and we also provide the autofocus camera in our system. So they can uh how say for position and one is a sensor inside four sensor inside. So then all of so they what they have is how say for hardware really one platform and the second and the second sense is the software. Yeah. Well we're almost out of time but Brand let's start with you wanted wanted to see what you're excited that maybe Jensen could say today. what are you excited about? We'll go down the line. Um we really appreciate you being here. Yeah, I think uh this is AI era very important. So back to us when we are running this such scale of global operation we do need to leverage the new technology Peter mentioned verse Scott mentioned about robots. So you look at that the gen AI essentially require data require computing power require algorithm or AI model. So as uh the leading company for smart manufacturing network I think for group is driving uh we collect all kind of manufacturing data you know that right? So second thing is we built gradually a lots of computing power from Kong you know now we have H100 H 200 all implement there then move to Wisconsin we have a V 100 which was built in 2020 and also we have the commuting power in China with H20 then we are going to build we already built the the GB200 you know forex in Kawong for the R&D purpose So going forward we got to even build more like uh you know I think the chairman Yang Leo Foskan group mentioned that we're going to build 64 racks in Ken as the initial one and moving forward probably you heard about that first group invest a lot in Houston we're going to build more than more more the computing power with GB 200 300 in Houston to help us out so with lease then we start to run the AI model algorithm how we going to make it again omniverse Got it. On reverse. And any quick words, Peter? Oh, yes. I I really want to share with everybody uh omniverse perception sensing capability really makes omniverse not just people thinking virtual side but connect to a real site very closely and from the uh what we try to challenge uh overcome for example the computation powers that we require from the the tra traditional robot to corrupt robot cognitive robot with sensing like Scott mentioned and also humanoid robot like why because the automation last mile still using a lot of human operations with high skill that still needs a lot of help but thanks to again there is a very high degree of freedom human models that can really helpful for this so again thanks to uh Nvidia's computation powers we can do not only the cloud but also the but also embodied that not only agentic but also physical AI and thanks to everybody again hope we can fight for a better future. Thank you. Great. Thanks a lot. Thank yeah go ahead. So actually you know the next generation in the future for the coba right house how say reasoning and planning is very important now it's the house so currently we use the oniverse the cosmos the se so the foundation model like the k motion like that that's the foundation model and we are use and we also generate the genetic data SMG now we are trying and we uh modified the AI model got it currently. So we hope in near future we will have the more capability for for example reasoning and planning for the co. I think you will. I think you will. Thanks a lot. Thanks to all of you for joining us. That was great. Well, what a great session. So uh Brady, you know, you're a local uh AI expert in Foundry. What did you think about what was going on uh what what you heard today? Yeah, I think it's a critical point especially in Taiwan because we have we are the center of the chip productions and also assistant productions and we a lot of robot vendors are working with Nvidia. So I think it's a good opportunity for Taiwanese companies to start in the robotics industry. Yeah. And Ben, how about you being the respected AI research analyst in the world of finance? What stood out to you? I think, you know, Brady, I I I'm really impressed by how fast people are moving in Taiwan, all these partners, and what it takes from Nvidia's side to integrate all these partners, manage all these relationships and move at this speed and then bring along all these partners who are moving really fast as well. Um, you can see it with Dell and Asus and Quanta and uh, all all the folks we spoke to are moving so fast. I I've never seen it like this in order to keep up with Jensen's speed of light and it's it's amazing and I'm really looking forward to what he says today and I think this community here in Taiwan is I I can feel it after being here a few days. It's so strong in terms of, you know, the pride in what Nvidia is doing and the the desire to work with this company. Um it's very impressive and uh I think that you know today um what is really clear is that uh he's a very strong ecosystem here in Taiwan and both globally and everybody's looking forward to what he has to say to drive the industry forward. We were lucky enough to hear a lot about robotics today which is one of the next frontiers but uh there's a lot of other things I think he's going to say and it's going to pull uh this whole uh this whole island forward and it's really exciting to see. Yes. So, I wanted to also just say thank you all for joining us here at NVIDIA GTC Live at Computex. The countdown is on to Jensen's keynote. It's coming up really soon. And as you see, people are still filing in. So, we got to get to our seats and uh Brady and I got to get moving. Thanks for watching. I can't wait to see the keynote from Jensen. [Music] Hey, hey, hey.