NVIDIA GTC Live Keynote Pregame | Live in Taipei

NPU? Love? Love? Listen? Listen? HoloHub? Love. Love. Me? Me? Love. Love. Me? Me? HoloHub? Me? Me? Love. Love. Love. Love. Me. Me. Me? Me? Love. Love. You, you. You, you, you. Kion? Love. Love. Love. Love. Here is Taiwan. Taipei Pop Center is developed. The scene of the 2026 GTC Convention Warm-up Show. In. During the fact years. Computex in Taipei attracted from Participants from all over the world (LAUGHS) Only the conference has become an industry. The world's largest and One of the most influential AI events in the industry. Today. The week-long event has officially kicked off. Roughly per capita The keynotes will be delivered in person. Take a sneak peek around the world. The next step in the ever-changing world of AI. Before the keynote. Our warm-up talk was about A series of heavyweight ideas from Taiwan Leaders and core architects Discussing how to turn the tables Taiwan's major industries. Promoting a more advanced agency business AI and its robots. and how to promote A whole new infrastructure. NVIDIAGTCLiveTaipei For the next two hours. The most important stage in the house is right here. Events Now It's official. Cohere. Greetings from all over the world! I'm Bruce Lu. I'm Bruce Lu. I'm True. Welcome to Computex 2000-2006. NVIDIAGTC Taipei. We're here. At the Taipei Pop Music Center. Outside the plaza. We've seen a lot of people gathering today. Here. Everybody. We're looking forward to waiting for Jetson's Gino. And also for NVIDIA GTC Taipei 2016 kicks off the event. We're on it now. AI in Taiwan The key to making the AI era possible The place. The World's Most Important Tech Event Computex is just around the corner. And Jetson today Jetson will be presenting today through this keynote for everyone. Kick it off. Yes, sir. We're all looking forward to it. The real talk. It's also the first GTC. Spark and Show. All in. I'm Bruce, a Taiwan semiconductor analyst at Goldman Sachs. Lu Quinlan. I was just analyzing Tracy Tsai. Tracy Tsai, courtesy of Cohere. In-depth IT research and consulting Consulting services. Assisting global corporations and governments Make informed and Business decisions. I'm very lucky to be in Taipei. And Taipei is where AI is being built. The future. At the center of global technology At the heart of the global technology ecosystem. For the next two hours. We'll be talking to Taiwan Twelve heavyweight tech leaders. Total Semiconductor AI Server. All the way to edge devices, Robot AI factories. And how AI can really be used in the workplace. At last, everyone. We'll take a look at the infrastructure behind this. And the seeds of the future. These last conversations will lead us to To Jetson's keynote. Let's get started. All industrial revolutions start with Infrastructure begins to change. It's the same in the AI age. It takes a whole industry to build an AI factory. Ecosystems. And Taiwan is realizing this. And Taiwan is the best place to make it happen. Barry Lam is now infrastructure just liketheinternet Just like AI2. AndL3Harrisanewkindof Arproduceanewkind Ofoutputtokens thebuildingblocksofanTELUS thelargestinfrastructurebuildoutin humanityishappen. Listen right now with Listen right now with Taiwan at the Listen right now with Taiwan at the center Listen right now with Taiwan at the center. Benefit its economy and GeForce. PersistentinTaiwanfoundationcallto ThefutureofglobalAI2. Andinthisnewparadigm. Themoreyoubythemore Youmake. The server wave has been officially triggered. From the world's top semiconductor process To the cabinets that hold them. System solutions. It's all about computing servers, Network Architecture to the Most Critical Power Supplies Management and cooling technology. Today we have invited The three drivers of this wave. First of all, the first m Dr. Yuji TSMC Executive Vice President and Co-Operation Second Dr. Barry Lin Barry Lam Founder and Chairman of Quanta Computer, Simon Xie Simon Lin Wei-Tong-Tong Director and Chief Strategy Officer Mr. Xie Ming Simon Lin, Director and Chief Strategy Officer of WeiDirect. High. That we start first. Dr. Lin the Jetson used to Dr. Lin, this Jetson, once said something very impressive. He said TSMC's most important asset. The most important asset of TSMC is Trust. This is also NVIDIA. NVIDIA is willing to put the entire future of the company on the line. That's why NVIDIA is willing to put the entire future of the company on TSMC. And that trust extends to Surprise Chain, extended to Taiwan. It has sustained the entire AI revolution. But in the trust On the other side, it's a very difficult trade-off. Woe to the explosive growth of demand. No matter the advanced process, Packaging capacity is seriously in short supply. TSMC is also in Taiwan, the U.S., and the U.S. at the same time, Japan. All over the world, TSMC is massively expanding production. Behind all this is Some astronomical investments. So I'd like to ask you. How are you doing? In full satisfaction of the customer. Increase its trust and B future Risk of overcapacity Find the best balance between the two? I want that TSMC. The three most important ones. We think our competitive advantage is Technology leadership. Our production advantage. And trust in our customers. That's why TSMC Since its founding We have always trusted our customers. Very important. I want to do it now. I think in the current wave of AI. How to win the trust of customers. I How do I balance this? Capacity. We've always been a company. The most concerned thing. I think now I think now, under the extreme demand of this AI. The most important thing is still we are The most important thing is that we are expanding our production capacity vigorously, so you can see that. This 2 years we no matter in the domestic, Taiwan. In the United States, in Japan, in Europe. We've been expanding our production capacity vigorously. It's almost into all our strength. And then try to produce as much as possible. And of course after the production, then, we will also make one ourselves. We also have a very good market research department. It will also Do one here. Market analysis, not just from our customers. Our customers' customers' customers' downstream How much demand will there be to do an analysis. But At this stage we're just going full steam ahead Power of expansion, capacity. Because the demand for AI is too Exuberant. I think. Here we should still be able to always There could be a brute A good balance. Understand. Thank you. - Thank you. I think I've also often told investors. We also trust TSMC very much. TSMC helped us do all the DELL. So for the NVIDIA AI part of AI. We seem to be relatively relaxed. I'll follow up quickly. That's because now The capacity is still seriously insufficient. That means every customer every customer who wants it may want capacity. I'll follow up quickly, that is, because there is still a serious shortage of capacity, that is, There are still some shortfalls. So how do we do it in How do we find ourselves in the middle of this? How can we find ourselves in the middle of this, because we don't always want to We don't want to favor one customer over another. Or in We don't want to favor one customer over another, or one customer over another, or one customer over another. How do we find that balance? This is a judgment that requires wisdom. I think this is our chairman. I think this is a very important thing for our chairman to do with the president. He also has a very wise judgment. Of course, it is very important for us to We have to keep up with all our customers. Close contact and cooperation. I hope not. But not to our most important customer, NVIDIA. We hope that not only can we fully support our most important customer, NVIDIA, but we can also fully support them. I hope that not only for our most important customers like NVIDIA can fully support, but also for some other smaller ones. For other smaller, newer customers, we hope that not only will we be able to fully support our most important customer, NVIDIA. We also have an appropriate adjustment on it. And this in our internal. Almost every week. Every month, everyone We're discussing how to do this. The most reasonable. The most reasonable, the most able to make The most customers can go The most satisfying allocation Methods. Thank you. Okay, then I'd like to ask Barry a question. That is to say, this foundry in Taiwan From the very beginning Doing PC foundry. Now Has successfully transformed into a production Server. So now The demand for servers is so strong. Gartner estimates that This year 2026. Gartner estimates that by 2026, the growth rate will be Sixty-four and a half percent, about six percent. 6,000 $4,688 billion in spending. In AI servers will continue even through 2003. Growing to $934 billion. I'd like to ask Barry. I'd like to ask Barry in this one I'd like to ask Barry, on the back of this strong demand. I'd like to ask Barry, on the back of this strong demand, how is the foundry going to build an AI system? Able to combine Chip, power management, heat dissipation. And the network. Challenges of the process. Probably. What's that? That one. Sad market. It's an explosive growth. To that Two years ago. Our vow night. To LLM That. It's even hotter. The Listening. AR Discovery said there was a Very great use instead of humans doing A lot of work. What about this wave? NVIDIA. It was caused by NVIDIA. That's right. I remember NVIDIA. DB Uh-huh. 200 was the first step uh. From chips To the rack. A new produce. self I'm sorry. It means We used to. Hydraulic coping. We used to talk about hydraulic cope, we talked about taking a GPU, we talked about taking a GPU. Now from DB 200. We're not talking about a race, we're talking about a race. One RAM, one race, that's Chow. A unit, like a shell, means something like that. That one NVIDIA A16 That's thousands. Algorithms. So that's why. So that's going to make This The standard is easier. Yeah, what happens when you make it easier? And it's then The product that came out It's already surpassed On the market. The solution that was already on the market. So what about writing our client? Tim to Lenovo It-- It's using The latest SNP to the GPU. Technology time to technology time to technology time to technology time to technology time to technology time to technology time to technology time to technology time to technology time to technology time to technology time to technology. Work Kion timetomarket. We got it. Very big improvement. That's how it works. That's how we're going to get the whole thing going. Character this uh. Oh, my God. The industry's The industry's big. The development of the industry I don't know. That's why we're going to be able to bring about a great development of the whole personality industry. That's because of this. That's why the whole character of the industry has been developed. Self. Then, the whole concept was realized. The whole concept was completely different. So the customers are They dare to invest. He's a fast investor. I'll get it. I'll get it. I'll get it. Recovery. This is also an opportunity for the customers. Very good Medium is the best, Taiwan. It's just fine. So this already feels like the whole The process of transformation. From the general past such Single product manufacturing has To a solution-oriented. Even a highly sophisticated manufacturing. The transformation of this process is great to be able to Delivered so high A precision AI server. The cabinet system you just mentioned. So this is a OEM manufacturing in Taiwan. It's also a very big shift. Containing the ecosystem. with TSMC or NVIDIA. And the cooperation with Hsia. It's also a very important shift. Of course. That Rome is not It wasn't built in a day. Yes, it was. We are. We started making PCs in the '80s. It was RP. We didn't do it. We didn't do chips at that time. The chips were Intel's. But Circuit boards, boxes. Power supply, uh. All the parts. It's made in Taiwan I'm sorry. That's what I'm talking about. At that time, at the beginning, too. That's the way it was at the beginning. So it's almost No need to ask. Foreign parts That came to AI. He is. The world. The most advanced base. I can't say no. Dictatorship is Good, that's why I say that. After Taiwan One move. We're from the assembly plant. We are responsible for the last mile. That's why we know best. The materials, but all that. In Taiwan, so yes. Our assembly Assembly. Yes. Flat. It's all very simple. Only do it this way. This AI. Water and fire can do so Big numbers. It's really not easy. The demand is so strong. And then to be able to match that speed is not easy. So thank you. I'd like to ask SIMT again. You're right about the question. Can you share that with us? It's this in industry. In this transformation process In the process of transformation. What kind of conditions do you think are necessary? In order to cope with this Very increasingly A complex manufacturing process. Because in the past the manufacturing process was relatively monolithic. Well, it's more complex now. To make solutions and so on. And you have to be very early with the customer. Cooperation. That's a very big shift for you. You can share that. How can you share that? It's a real thing. How can we really transform in the process? Successful. And then to respond to this It's a very high-precision and complex system. It's a process of very high precision and complexity. I think just now Barry already has I think Barry has already mentioned that. From PC to now The transformation of manufacturing is huge. That's just coming to push you particularly I think Barry has already mentioned that, from PC to now, the manufacturing transformation is very big. The change in complexity. This transformation of this industry in Taiwan. It's inside the process If we don't hit a bottleneck. That's a lie. There must be a bottleneck. Some big bottlenecks. And then we'll be able to transform. I think because of the past In the PC era. The technology was relatively mature. So the manufacturing in the back The degree of standardization is very high. And in response to customer changes. It's not too much. It's not too big now. Because of the iteration of his technology Faster than before, yes. And it's even more so. So we In response to this faster and more complex. Even more systematic. We're not talking about all of it until we're talking about all of it. Including heat dissipation. When all aspects have to come in at the same time. The complexity of that is It's HP. So the first one. What we have to do is So the first thing we have to do is. Talent is Fundamental. That we have to understand To the past we The first thing we have to do is that talent is fundamental. Most of the production Processes. A lot of them are on the mainland. We are many of them. We're all setting up factories. But when this wave. Because of geopolitics. After some changes. We don't care if we pull back Taiwan or To the mainland or to the United States. To Mexico. We all found talent. There is a shortage. Because there are a lot of people on the mainland. Some of the talents we trained in the past. There is no way out. No way to transfer. So we have to re Cultivate. That's what we did a few years ago. We're just about a dozen years ago. We were thinking about this. We began to cultivate More talents. From the school and also including From putting senior people. From the mainland. To train, this is the first. The second one, because the complexity has become very large. So we relied only on the old people. We can't solve these problems. You have to have new thinking. New systems. That's why we've gone from digital to digital. From the ability of AI. After adding it in. We found that We can force it to be used in the past. Less manpower. To achieve what we want to do more efficiently. The result, this is That's a very important key point. Your system. Your business model has to be a pretty good one. Significant equation. We call it The DNA shift. The DNA shift. That's been going on for the last couple of years. I think a lot of Taiwan Our relevant and friendly Competitors, including Berry. Actually, they've done a lot. That's why we're all doing it. A lot of this change. What is more important now? Let's look at the present The ASIA of the cloud is the AI itself. Will go with Barry Lam. Will help you find the answers, he's a wave. It's gonna go down. So we deeply appreciate that. So we deeply realize that the future of manufacturing must be AIenable. A new wave of this AI. That means we are now the User of AI. We are manufacturing equipment. But we're also becoming the AI of him. What about this user? He has to make good use of AI this Power. The future of manufacturing Enable through AI Enable to become more flexibility, that has this What about resilience? Then? In the anticipation domain In anticipation of various changes in the domain He has more Capability. At the same time. In this new. Whether we do Rack. When we do the back-end. Cohere's a lot of change. It's a lot different than the old PC. There's no such thing as a unified standard. You just make a box. It's not like that. A lot of firmware, a lot of whatever. It has to be readjusted. So I think. This wave of Taiwan's transformation Absolutely not. In just seeing that. Because this wave of AI is the same This Media puts this through Open Data. I think this wave of Taiwan's transformation is definitely not in the sense of just seeing that AI is the same as this Media through Open Data. If our connotation has not gone through The past transformation. That means a little bit. The FO shift. We don't have a way to catch We wouldn't have been able to catch this opportunity. So I think this is Another wave is coming. A new stage. I think this is a new stage. Let me feel that is Taiwan No matter from this semiconductor I think this is the place that makes me feel that Taiwan, no matter from this semiconductor foundry ah. This spirit In constantly Continuously to evolve Evolution, then let us At the time of accepting new challenges. It's always evolving. That here is the key to talent. Talent is a foundation. That certainly has customer-oriented. And then the spirit of customer service. The spirit of customer service, the spirit of gaining the greatest trust from our customers. It's a very important. That step in the process Luckily, this ecosystem works together. Help us Taiwan today! In this AI infrastructure Very important position. Almost irreplaceable position. Because it's hard to find this It's hard to find an ecosystem like this. Together From the very beginning of the idea Start designing. Together, partners working together. And then you can create a market opportunity by meeting the customer's needs. I think it's really remarkable. Yeah, yeah, yeah. I was talking about that. The complexity of the product. I think we need to utilize TSMC's technology. I think we need to talk about TSMC's technology. I think I'm in it for the long haul. I think I've been working in the technology field for a long time to make this discovery. I think what we've seen in the past is Every two years we have a Morse Google. But now It looks like the AI technology is really It's too fast, and in that case. How do we look at the future? In terms of semiconductor technology How are we going to match the development of AI? In the end, MoFlow will go down, how? What is the future of MoFlow? And then advanced packaging. In other processes. How are we going to look at it? TSMC's future technology blueprint. And how to work with that. AI. Development? As a long-term I've been I've been in charge of R&D at TSMC. So the R&D side I've been in charge of R&D at TSMC for a long time. I think our attitude towards technology is I think we're paying a lot of attention to technology. The most important thing about technology is of course The most important thing is advanced technology. Front-end semiconductor process technology. That everyone talks about MoFlow that Motional Actually, it's been evolving. This we do The most important work of R&D is to continue to make this magic to be able to Continue to develop. So we are now in the last few years. Whether it's seven nano, five nano, three nano. Recently launched nano. And then we're doing A IV. There are some future developments. In fact, it's always been We've been launching them one after another. We have a strength at TSMC. We have a strength, which is that we maintain a relationship with our customers. Very close cooperation. So we have a strength, which is that we maintain very close cooperation with our customers, so we have a very close cooperation with our customers. The direction of demand for this advanced technology. In fact, we are quite aware of the direction of demand for this advanced technology. Because we work with almost the whole world. All of them. Major Designers. Leidos. So we've been able to get a good understanding of the direction of the demand for this advanced technology. We've been able to get a good grasp of the technical focus, the technical focus. I'm going to spend it too. A lot of resources. In looking at our future technology development. So we think we're pretty good at that. I think we're quite confident that we can continue to push forward. MoFlow that in addition to this. I think advanced packaging has been one of TSMC's key focuses in recent years. It's a very popular technology. This is a very popular technology for TSMC in recent years. This technology is one of the most popular technologies that TSMC has developed in the last few years. The company has been investing in this technology since about 15 to 17 years ago. It's in the works. This is this It's a very early investment by the company. It's a very visionary direction. So today we see this advanced Glow and heat become AI An important technology in the AI industry In addition to this support. In fact, we've been paying attention. We've also been working with these system vendors. With our customers. We've been in close contact with our customers. We've been in close contact with these system vendors and our customers to understand what kind of development is coming in the future. So we can see now So we can see that now the theater photonics technology, the theater photonics technology, the theater photonics technology. TSMC There should be a pretty good one here. A good market. That's expected this year, It's expected to start mass production this year, next year. Because we see that this is a bottleneck for future systems. So we'll start to develop. So we, as the R&D unit. We will always What are the things that we've been paying attention to. Not just on semiconductors. Maybe for the system for There may be some limitations for the future. It's all going to start. Earlier. MICCAI we can keep these things all right. It's there when it's needed. Right? I can ask a question soon. Is that what you just said about keeping up with clients close cooperation, can you give me an example of that? What kind of Specific close cooperation? What kind of specific close cooperation? More specific. I guess. Because there are some of our programs still in progress. So of course it's not convenient to talk to you again right now Just share. In isn't that right now? The most specific fastest growing is the Department of Chip stacking technology. In stacking technology now It's still logic-on-logic stacking. We'll come, we'll come. We're going to start. We're going to start working with some of the other ones, and we're going to start working with some of the other ones. We're going to start working with some of these other semiconductors. For example, Memory stacking In logic technology. We feel for the future. Into this Inferencing. This AI enters the Inferencing stage. This is also a future development trend. That's certainly just now. Photonics. Because we found out now Silicon is good for computing. But there are more Distance communication is still used It's the most power Inferencing, it's the most power Inferencing, it's the most power Inferencing. So this part we In recent years, we have also invested quite a lot. I think the General Assembly will soon see this technology. I think the conference will soon see that this technology will also become one of the future and then of course there are some other. I think the conference will soon see that this technology will become a very important technology development in the future, Because we also slightly It's still fully mature. So it's not convenient to share with you. I think the main thing is that I think The biggest advantage in Taiwan. Actually, it lies in the upstream and downstream integration. Because we look to the absent. What I just said is that our TSMC transmission Also to do Weaver, do chips. But now It looks like there's no way to completely We must go downstream to meet the needs of our customers. To cooperate with downstream customers. No matter some Hardware's ComponentMICCAI NSCALEparts these We have to work with the downstream customers. In person The overall integrated design must be Shipment in Shenzhen, to be completed. So this thing may be Discover TSMC in Taiwan TSMC can cooperate with midstream and downstream. A very important direction. Actually it does look like ML. We also have a lot of cooperation with DELL. We try to understand In this In the system above heat dissipation ah. What's in the power supply? What might be the shortcomings. Some parts of it. It's actually something else. We're certainly happy to work with you. But some of it. We found out that no one seems to be doing this. Then TSMC will do it. We'll evaluate whether we have it or not. We have the capacity and the resources to do these things. That Taiwan I want to have global The most complete industrial supply chain in the world. So In the production of Taiwan to Our development Technology or planning. The future blueprint is also very helpful. Yes. That I'm following up a question. Barry, Barry. What do you think about this wave of this? The wave of sorrow is coming so fast. And then you You've been in the industry for so long. What do you think about this for us? The biggest change for us people. The future, whatever it is. In the future, whether it's in this job or I've been in the industry for a long time, so do you think this is the biggest change for us? Ubiquitous AI. What kind of big change do you think it's going to have? The way we work or The way of life. Ah, very much. The experts and scholars are Speculating. ACER's impact on humanity Or even that. Medium to medium relationship. On the pope's heart and lungs. That This one too? With, say, also There's going to be, there's going to be. There's a mutual only To be able to achieve good results. That certainly is now. Now, of course. LLMHawChenAlan Ni. He put AI Instead of human beings to do a lot of things. Things. That It's like this. Is it going to make Will that put a lot of people out of work? Uh, I'm I'm I'm Thinking about the '80s. PCs came. Isn't that a lot? Cloud Attention will be paid to the cause of the meeting and No higher Uh-huh. Now. AI is just starting. Actually. That, too. It's a little bit human. I think it's very good. I think this is the one. It's a very good thing. Yeah, of course. The next AI is That. Facility. That's going to have robot. Is to work with big machines. To have that? NSCALE's ability to drive. NSCALE's ability to drive. That's all very stuff. It's all going to be intelligent. It's all going to be able to intellectualize. I don't know. In that case, human beings will be able to be intelligent as early as possible. It can be changed into one. Work. The model to do it. Higher. Hey, hey, hey, hey, hey. That's that. That's that. It's a - It's a - It's a Uh-huh. Both sides of the idea. I'm good at that. It's good. But this can be done. It's good, but it can be done. It can be done. You GNN often. You know, GNN has always told us. I think you have automation at home. When you have all automation time, then you can not. Don't do the dishes. The table. The factories of the day were Automation. I don't think so. There were no people in the factories. You're not switching. More money? Of course it's a good thing. That's what I thought later. You can come later on, to. Yes. - Yes. That's why Now. Now wherever in America. In uh. Congress la, Europe la. What our country would do Concerned about this matter I don't know. Whether or not to say how Prevent this AI from doing Bad things? Yeah. Right. Yes, this is a very serious issue. It's a very serious issue. You and us too. We have to answer it seriously. So say I I also I'm not an expert, thank you. But let's see. That one. AI's Llama wow. More and more applications. More and more applications, more and more solutions to our growing More and more problems I'm not an expert, thank you. So many scientific problems can be solved. Solved. A lot of that. Problems that we don't normally solve. He can solve them very quickly. That's not it. Isn't that a good thing? So that AI will come one after another. A lot more. New models, new startups. It's all going to come up. AI finds it. New algorithms. New models. New business models. New value. That's what's going to happen. Humanity benefits. I think Barry said it so well. I think it should be about us. We should be looking at it in a positive way. AI, can make us heroes. Do more things. Make the world a better place. We should look at it in a positive way. So this is leading us All of us. In our future challenges. To do better. For Taiwan's eco-industry It's a very good place. Because of the time. Thank you very much for your sharing. Thank you. Thank you, thank you, thank you, thank you, thank you, thank you. Thank you. Thank you. Thank you, thank you, thank you, thank you. Berry thank you. Okay, thank you thank you. Good hard pass. Actually to where. We're going to share some first. Just thoughts. That's today's sharing really It's really terrific. To please sit together. Thank you. Thank you. In the course of our sharing today. The first one is trust. And then again Talent. And then customer service. So the important point is the whole The ecosystem of this kind of cooperation Collaboration, in order to build Today's such a Successful place. In fact, you see their real idea is to say that they They are in the original job. They've done the best they can in their original jobs. But In the development of AI to him more to Overseas Extended integration with upstream and downstream. The integration of the relationship with customers. And down the line, they see it again. What is the future of Visteon for AI? How it's going to change our lives. It's really It's really. It's really, really the leader of the industry. It's really, really a leader in the industry. But we're right. The future is more confident. We feel this. In the wave of AI The wave of AI. Taiwan will never be absent. And it's been going on and on and on. In successful transformation and innovation. And then let our talents also be able to upgrade. Technology is also upgraded. Good, good, good. Good, good, good! P-era is connected. People and information AI era. Until all are wise. From AI, PC Edge Computing System To robots and industrial equipment. Taiwan's technology leaders are Putting the smarts right in. To the physical world. TaiwanhelpdefinethePC Taiwan helps define the PC era. Neweverycomputer isbecominganAI2super computerdesigntooneasI Workrnativelife. I ask Wantsonlypossibleinthedata I ask Wants only possible in the data center is now center.. isnow onyoureDSX bringinGenerativeAIDRIVEAV toeveryuser witharerunninglonelyevery applicationwouldbeNVCellof NVCell of all transform in all. transformin havesoftwareasbuildand used.. inthe havinginnewexperience PhysX.. thefuture ofcomputingisasolar withGenerativeAIrealBing industriesandopening newforinnovationand The future of computing is a solar with Generative AI real Bing industries and opening new for innovation and groups andjustthisTaiwanheldandthePC todecadetogo Taiwanishavingtoreadinventory The PC to decade to go Taiwan is having to read inventory today. Wisdom City we we Tools and Equipment for Daily Life An indispensable centerpiece of our daily lives. It's going to take over. Our PC workstations, factories, Robots and edge devices. Closer to the data. Closer to the action. Closer to the data, closer to the execution, and closer to every decision. The core. Next. We're inviting just that. We're going to be joined by three of the key players in making this technology a reality. First up, the first one, Li-Hsing Tsai. Vice Chairman and President, MediaTek CEO. The second is from Mr. Tang Huashuo, Chairman of the Board. And the third, Mr. Darling. Robotics CEO. High Risk. This is your job Across the From Founder to DELL To today. State-of-the-art MediaTek ICs This is in the technology industry. Everywhere. It's very rare in the tech industry to be all-encompassing. Idea. This is the face of AI computing. The exploding demand for AI computing. Do you think in the next 10 years. In the next 10 years, from the hardware network to you What do you think will happen? Structural changes? And In this paradigm shift. MediaTek has a very special position. You're one of the few who can All the way from the cloud. The terminal provides a complete An end-to-end solution. How are you going to capitalize on that advantage? Through MediaTek's technology Blueprint to drive The whole mix of the next generation. What about the computing part? Okay, thank you. Bruce Lu. I think I do. I think I do. A couple. Different industries. Bruce Lu. This experience. Gave me a lot of bodies. Experience. I think one of the most important things is that I-- I think one of the most important things is that I learned that no matter what I'm doing. You have to be able to Identify and deliver. Your dictionary? Value. That, for example, at TSMC. Of course leading edge silicon Technology Telecom of manufacturing Telecom build deploy the best networks whereiswhatweareLILT. And then provide the best service To the general public. That to MediaTek this How are we going to do that? Through? We're From Kion Summer to Computing. To Cloud how can we provide A series of chip solutions We can provide a series of chip solutions. Let's talk about it. Cumulus value. We've done some things recently. In In terms of MacOS in the cloud. What I've learned is that you To be able to performanceproduct performance product and performance Spark and performanceSpark. TCOTotalClassof This if you must. From a system-wide perspective. We're not talking about software. Because it's a little bit farther away. This whole hardware system. This whole hardware system goes from this rack to this rack. To the board inside. To the CPU inside. To the CPU. That's now identityAI is already It's going to be one of the most important applications. So HDevices H Devices is providing this AI's AI's AI's AI's AI's AI's AI's AI's It becomes extremely important. How does this connect it all together. I think MediaTek In this regard we In the past From cell phones I think MediaTek is going to start from mobile phones. Now We cooperate with NVIDIA, for example. Computing I think MediaTek is working with NVIDIA on Computing chips. And then And then we work with big CSPs. Then we work with big CSP companies to do CPU. I think MediaTek we are very happy we I think MediaTek is very happy that we can provide these. I think we can create and provide. More on NitinMittal value to our customers. To the society. Hope. Also let us Taiwan We can do better. Everyone grows more, thank you. I think this is just right for us The next question is this one too. It's a very special thing for MediaTek. That is also a Taiwan company relatively have The aspect of flexibility. That is to say, MediaTek is not only That is, MediaTek is not only an IC design company, it is an IC design company. You can talk to your IC You can work with your IC design company, like NVIDIA works with other companies, you can work with your IC design company, like NVIDIA works with other companies. You can also do a cooperation with your customers. This advantage seems to be at least we See other Egyptian companies around the world The more absent parts, the more absent parts. How did this happen? How did all this come about? IF we're going to talk about Five-layer cake a little bit. I think this is one of the qualities of Taiwan. If you talk to GNN. Taiwan has this flexibility. But also to have this discipline. TSMC is another best characterization. It has the flexibility to serve so many different customers. TSMC is another best characterization, both flexibility to serve so many different customers, to provide so many different technologies, to provide so many different technologies. But it also has such a strong discipline. We can make exactly the same thing. MediaTek also needs this. We can make different chips. In consumer chips in Computing chips Chips in the cloud. The people we work with can make us Customers can make us. We can make our System Partner, we can make our System Partner. TSMC. Or we can make you NVIDIA is. Our customers. And also our Design partner. This one does a Gobi Beach. This ensures CPU, this ensures CPU, GPUs are state-of-the-art. They do one, we do one, and then connect it. This is what I do every time. This is what I do every time I talk to Mr. and Mr. And Jeffrey. Everyone. Very Amazfit. So complex a chip, there are two different companies. Make it out together. This is actually in our semiconductor It's almost unheard of in the history of semiconductors. I'm very proud. I'm also very happy that our colleagues have this. I'm very happy that our colleagues have this opportunity to work with us. I'm also proud, I'm also very happy that our colleagues have this opportunity to cooperate with a world-class company. I feel that Taiwan is also I think Taiwan should have more and more Internationalization, to be a world-class, First-class company. It's really not easy, it's really not easy. Thank you, thank you, thank you. What I would like to ask next is It's now. We see that Argentina AI has begun From enterprise applications. Whether it's process optimization It's starting to spread. To personal applications. That's a good point, do you hear me? OK, what I'd like to ask is to say. What do you foresee? In the future AI or in the personal life or in the personal life. In the personal life or in the personal life or in the personal life or in the personal life What kind of work will it bring? Meaningful impact. For example, the user experience will have What kind of change. Or what kind of change in user experience will there be, either with computers or What kind of interaction with computers and devices will there be? What changes can you share with us first? It's because of the two Tai The impact of the two T ai. Okay. I think so. That is to say that indeed two A now Yes, I think so. It's a very big one. Paradigm one. When the GPT came out. I think we all know that. It's a little bit like The fourth industrial revolution. I think by the time this one came out. I think it's only this time that the two really put it Released it. I think you might want to hear some more really see What will it feel like? I'll get right to it. It's going to be all over the place. Go. That said a lot of work went into Garmin. That's why. In this two The age of AI. You can imagine. It's also an opportunity for Garmin as a whole. A sea change. Because in Garmin. This NPC, everybody. I know. He's now able to think. He can think, he can learn, he can study. ModelOps. Different strategies. So that's the whole situation. For this Gamer to play up is amazing. That last words. That is, it's not a priority story or a story or a story that's not a priority or a story that's not a priority. Script. That's because of Undertaker. It's because Undertaker is the last word. These NPCs. It could be you becoming you. This, this, this. Alice is great or Alice is great or Alice is great or Alice is great or Alice is great or Alice is great or Alice is great or Alice is great or It turns out to be the scariest. Riva. But I'd like to take this opportunity to say. Because I think everyone may Android Ollama's about part time. The meaning of the part time shift. Now. A lot of people talk about it. The idea of interaction. I don't think so. There is no real address to the end of this Why is it a revolution? I'd like to emphasize it a little bit. It's now. I want to emphasize it a little bit, that is, the interaction that we are talking about now, the interaction that we are talking about. I'm going to emphasize it a little bit. Containing the two GNNs. Rick Tsai's PC. But it's got to really work. This Model. It's time to utilize the brain's ability. Not like in the past. I think he called it interaction. That is, there's actually someone. We call it handcraft. Such a Rivabased logic. So a more fixed logic. So it's more of a fixed logic, that way. I think he used to call it interaction, which is actually someone writing, what we call hand craft, a Riva based logic. Why would it be revolutionized? We have recently seen A lot of them are starting to go wrong. And go when the second volume comes in. It is because now The new one. New creatures. It's basically. It's the same as by The brain. So it's not fixed like before. We all know the biggest problem with the softwares of the past. We all know that the biggest problem with software in the past was that it was difficult to maintain. Then there's a new change, and it's over. That's why we don't pay attention to it. We're not aware of this revolution. It means that the new one must be the new one. It's a new type of software. Cohere. Cohere is talking about live coding. That's what Cohere is saying with live coding. And he's recently He's recently added the two GM N1X. After that. And he's recently added the two GM N1Xs to it, so he's telling us the whole essence of it. LILT coding. In fact, it's not just emphasizing that it's not just emphasizing that the The whole floor is going up. And that's actually what it's all about. Now. It's really making you a new type of person. These Intel Agents. And then two volumes. The second point is that the whole thing was recently released Later on. He said it's kind of like putting that Ceiling. and then punch it up to ten to a hundred times. Because now it's like this. The brain is waiting. So it can grow arms and legs. Out, out, out of the old one. The last few decades. This digitalwell digital Assets, that through Computer. Through TELUS. It's growing out at this time. In order to And then it's like this. A's army. The power of the whole thing is dozens of times greater, Hundreds of times. So I'm happy to say, very excited. That's why I'm very happy, very excited, that the fourth supply is coming. It's time. It's time. The key is when it comes back. We all know that in the past, even if At the Assistant stage. It's still very Think. Most of them are utilizing Cloud. It's AGNN that really makes This PC plays. It's the GNN that really let the PC play, that is, to do this again, that is, to do this again. The whole multi-step planning. It's also like PDCA. And then there's going to be the whole Honest. Don't let this Because AI is of course very powerful. But sometimes there's some Nation. So Security and all that. So I want all this stuff to make this The role that PCs play now. I AMPPBC finally found The best position. It's the so-called Heavy Client, yeah. So I think. You talked about a Very critical Security. Because if I'm not able to Trust, this is Riva able's 2ta1. I wouldn't dare use that, no matter how Intel you are. So instead of talking about absolute Very important, I wanttomakeaveryinternet I want to make a very internet login agent admission very most important login agent admission very mostimportant waywanttomakea Waywanttomakea-- I want to make a very internet login agent admission very most important way want to make a also fundamentally WE Trust using TomatointervalnVent. If say I don't struct, it could go wrong. I wouldn't dare use that. So I think this is probably the future. So I think this is probably the future. So I think this is probably the future on the h-end, wherever the PC is with people, wherever the PC is with people. I want IntelliGenesis a secure. We can only be in Janine. It's very important in AI. It's important to have this Barry Lam. Right. Right. Right. Right. Yeah. Great. That's great. Is the AI, that we asked under. Because you're now from JustAI? It's not Rick Tsai in one step. So you see what we're seeing especially from this side of the number. Your big name did it in just a couple years. Such a big brand of collaborative robots. But is this robot getting smarter? It's just. It's not just about the AI model. It also needs a lot of real. Data to train this thing. I want to look at it from your point of view. Say you have so many robots now I think from your point of view, say you have so many robots in factories now, doing these collaborative robots. You've collected so much data. Then how can you put it Turn it into In the next step let More and more vertical applications. You can use it. Such robots. Or that we We'll see in the next 12 years. How different? What are the faces of the robots? Thank you. Thank you, Bruce, for clarifying. A little bit about us. Damien is a start-up, after all, compared to MediaTek. ASUS. We're still small, we're still working on it. Yeah, that was actually mentioned earlier. A lot of it is digital AI. That low VR actually It's a lot of digital AI. We perceive to be To Janine Paul. Actually a lot of it is still In digitalspace. But when it comes to FICO, AR. It's actually a big difference. Digital you can have on your PC. You can do training in your PC in Server. Do validation, but scale. Because you Segmentation But scale. But to Rare Earth Riva there's some gap, but to Rare Earth Riva there's some gap. So eventually you'll probably want to In a real environment. Inside. Mental Kion. That's when we'll say it's a good idea. Than there are comparisons. A little higher. But I'm actually I'm a very frustrated student. I used to study Cohere. Matrix, dynamics. Simply put, Kion M.D. is actually math. Dynamics is fast ics. So I'm learning this stuff today like After the new API came out. We're talking about bicycle ai coming out. It's like this is fast. What I learned before didn't work. But my teacher said my own teacher. He said you're oriental. You still have to teach the old stuff. The original model based is just now. The original model based is the model based that the chairman of the board just mentioned. So you know this most basic basic, so you know this most basic basic, so you know this most basic basic. And then go to understand you and the new What's inside the Touareg DGX Platform? Gap. So this gap. So this gap, so we collect a lot of information, if we can survive, so we collect a lot of information, if we can survive. In fact, NVIDIA is really a So this gap, so we gathered a lot of information, and if we can survive, in fact NVIDIA is indeed an initial, follow me around. Barry, our president, said something. NVIDIA Initial Such an opportunity. Two years ago we didn't believe that This rollback You can rely on training, not math. We're not going to be able to do it by math. So he did use you to learn about opportunities like this. So after we gathered that information. If it can be generated. It could be in a digital tree. We can generate it in the digital tree, in the virtual world. And then the gap. I'm saying the more rare earth River's Gap. Smaller and smaller. That's when you go to these deployments. It will be easier. So this is FICOAI. It's the biggest one here. But now The whole The migration of technology. It's really fast. We're also very exciting, we're also very exciting. We're in this era. But it's really quite Otto Group. Right. Right. That this place you think we Taiwan What is the greater advantage in this place? Or how do we utilize our There are chips from the most upstream. We can do it on the chip side. On the chip. In terms of computing power can provide To the best. What is Taiwan's advantage? How do we do this upstream integration. Let Taiwan Heatsink. Inside AI. We can create A bigger space. Okay, thank you BOS I think I'm from two levels. The first is that Taiwan provides Very good infrastructure. like TSMC, MediaTek, ASUS. They're Quanta's parent company, Quanta. So the whole Infrastructure It's actually built very, very well. That also thanks NVIDIALeverage Taiwan Such a supplychain. That's a small piece of Keysight AI, actually. We based on Infor to do training. Generalist for data AI. So we're on top of it. Because we originally had our own factory. So we know the application very well. Where. But there's no way to put it in place all at once. We do so many things like people do, but we can be guilty of so many things. That's why we're talking about whether or not this world model can be put on. One at a time. In fact, he can't put it in place at once. But it's in the process. But in the process, we But some things can be done. This is what we are looking forward to, yes. I'm really looking forward to it. Because HawChen it has It's progressed to the point where it can be pushed. We can generate action. So imagine this Agents From the digital world. To the physical world. So if through Model it If we can understand the physical properties through the Model, we can understand the physical properties. The speed of gravity. After knowing the situation. And then reasoning. In fact. In a virtual environment. Inside a virtual environment, he's going to train it to say. It can then deduce that Possible outcomes. What's the best outcome. That's infinite. The business opportunities are limitless. It's in the past that we've had whether it's industrial This is the kind of business opportunities that we had in the past, whether it was industrial or commercial. IOT or not. It suffers from It needs historical data for training. But with this reasoning ability. At MONAI Janine Paul. In many models. Like whether it's in manufacturing or in healthcare. It knows how it works. It's domain data. This operation logic time. It can go and push. After the inference Find the best decision. It can then generate actions. So you think Future Business Opportunities So big and infinite. Where is Taiwan? Could be the best opportunity to enter? I've just mentioned that. Taiwan is actually In infrastructures actually I think it's the best place to get in. Such nVent It also provides all the solutions. To the world. That we in Robot this I think Taiwan In fact, because Taiwan I think that Taiwan, in fact, because Taiwan, whether it goes to the motor, whether it goes to the motor. To GitHubBox, to some integration. I think that Taiwan, in fact, because In fact, this technology is quite strong. That in addition to providing Component But in fact Taiwan can also be like our invention Robots are Sort of a solutionprovider, we provide a robot. So if there's even that Good supplychain. In fact, in addition to providing to the world. But actually also If I am also happy to see that there are quite a lot of There's quite a few solution LiDARs like this coming out together. Because Taiwan Also Taiwan has a very good field. Because of Taiwan's various From the outside. To the road to whatever is assembled. In fact, Taiwan has a very good field. Very good field. In fact, I think this risk may turn us into a very good field. Explain it. That is to say From the original AI demand, we are now IdentityAI, or even FICO, AI to the entire CHIPS Requirement. What kind of a Structural change. Or what kind of a structural change, or what kind of things we have to pay attention to in this place, or what kind of things we have to pay attention to in this place. What are our things? We can let MediaTek. Let the Taiwan side can be faster and more to To accelerate to this AI This side to promote.. Actually I The system that I've learned I think Taiwan is very. There's a lot of them. From us being bottom up, from us being bottom up. A lot of components IPi, c, Pc, etc., etc., etc., etc., etc. But on the system side. The system includes software. With the hardware. The softwares elders are the United States. And mainland LaneNet LaneNet's law-abiding hardware. And then the system side. I think Taiwan is still We need to work harder. Because you don't understand what I've learned. You don't understand the system, we did IP. We did the I, c. We did the I, c. We did the I, c, O often have some mixmatch. You'll have a lot of time to work it out later. To change. In time. It's too late It's too late. So I'm our colleague. I said you must understand. Software architecture. We We are in the Hardware Architecture. Hardware's Architecture And the system must be connected. In this regard, we are learning step by step. We work with Video. Let's have a lot of We have a lot of good and good experience. But in the end. But in the end, this whole IDC it But in the end, this whole IDC it is actually not fabricated. It's not the same thing. Basically you do robotics. I guess I'm, I'm just showing off. But this The machine. There's still mostly MICCAI. Nicole's part. You You're in the ICC part. The three stars are certainly important. And then there's Computing. Computing. I'm feeling absolutely Computing, I think it's absolutely done, it's done I think it's absolutely done. As long as we work with TSMC And with everyone and Damien. Then everyone has to talk to everyone. Do I, I am not Any doubts, we'll definitely make it. Do very well. Samsung to see how this aspect. I'd like to answer probably lighter. A little clearer than I am. But the whole system must be I think the answer may be lighter than mine, but the whole system must be understood, understood I'm sorry. The whole system is a bit like us if we don't understand it. Like a blind man feeling an elephant. Time to market. Timetomarket. We'll have timeto Time to market. This Taiwan is very flexible. But it has to be fast and easy. But we need to be fast and we need to be good. I hope we can do this together. We must work well together. Is that right? Must be much better is this me I feel that. Very deep. That's actually what I just said. I've been talking about robots. In fact. In the United States and China They're all very powerful. Very fast. But I think Taiwan must embrace this Reality. So it's still To In more such deep plowing The type of this That's what we call it. From paper to Product. So really put the whole System's Engineering is Or the PowerStack that Janine was talking about. So you know the whole so-called later. Because now there is AI later, you know the whole so-called later, because now there is AI later. The whole thing is integrated. So to be able to do it over here. If you don't pass it like that. In fact. You might have a hard time really talking to people You can compete. So like we are talking about this issue. AI this part then. Recently Because of this to the whole AIP It's finally found a very important position. But it's the same thing. How are you going to master that? So like N1X actually did Very good Architecture Architect. In the whole of this To how much computing power is needed. And then And then you have to take into account things like the Notebook, and you have to take into account things like the Notebook. You've got the power to choose. So when it comes to actually playing RTX. I think we all know one very important thing. I think we all know a very important thing, that is, you end up I think we all know a very important thing, that is, you end up using the entry Model, or you end up using the entry Model. I think we all know one important thing, that is, whether you want to use the model of the portal or the model of the Cloud. The whole HybridModel. The last couple of years, the last couple of years In fact, the whole thing is not only In the Cloud side of the whole thing, it's not just the Cloud side of the whole thing. The evolution of the Model. In fact, especially openSource. I know that we all know that. I know we all know that China's very strong, China's very strong. So that's a lot of the stateofthe I know that China is very strong, so there's a lot of state of the art stuff. And then how about instead At the time of this Janine. Because it's becoming I'm going to be able to reason in multiple ports, and I'm going to be able to reason in multiple ports. Why should it be like this? We all know that there's been a recent start to the present This hatred is starting to take a toll A lot of money. That's why now The problem of cost is very serious. That would have been local would have been a lot of important things. Like your low Latency. I'm like you with this whole thing. security more personal these things all are It's advantages. So when these things add up. The model of the logo is in fact Because in the past, on the opensource side. The whole evolution is really fast. And we all know that when it comes to logos. You don't have to look like In Cloud, you don't have to be like that. Model is equal to Encyclopedia full-time. So So the idea in ML becomes It's very important. In fact, you're focusing on In fact you're focusing on being more really in tune with this Something to do with. And then what's more important is that you Inside the logo. How do you play it? The local ones. I don't know. So throughout our So across the board we're generally across the board in terms of the math. And the memory of this ManWorlds. And Memory's Plant, and Memory's Plant. It's very important. If we can do it thoroughly, that's why I want to say. Barry Lampapertoproduct. You. If you don't get it right. If you don't make it very thorough, you end up. It could be memorybound. You've got too much. Too much. This top is wasted. That's why I'm thinking of these words. So I think these words, all must be really deep plowing. And then you have to make it. The best balance HydraHost. You've got to get to the real thing. You can still utilize clouds. Right. So I'm saying I think it's in here Both sides are more Need this very much. Rubin hover and solar This System's eyes Rubinhover and solar Fullstack that can really bring it to life. To the fullest. That's what we're excited about. So in other words. In other words, to be a PC OEM, to be a PCOEM. It's not just about It's a platform. It's got to be across the board. A layer of application Purpose Build for what people use. Is it a Game or a creator? What's he gonna do with it? Take the needs of the layer above it. And then it's integrated together. When the user brings it in for use. It's meaningful. Go support it. That's me. I'm thinking of the penetration that you just mentioned. You can't just do paper, you have to put that through that part. I think you just mentioned that penetration. Only then can it really land. Right. So I think Taiwan used to be very proud. Our Supplychain. This. But I've always been very impressed. It's from the beginning of the era. In fact, really It's this one that's really appetizing. Intel and Microsoft That Asus is More willing to say From paper to this Product is also scary. It's a lot of work. I hope there are some F5, Inc. Itron that are around now. In this day and age, what you just said. So you can't just be satisfied with Itron The best of the best. The most important thing in design is called Design Thinking, and the most important thing in design is called DesignThinking. If you don't have If you don't really think in Design thinking, you can't just be satisfied with the most important thing in design called Design Thinking. With Desire I don't know. You're equal to the most source of things. If you don't. That's often the case. The head of the bull doesn't match the mouth of the horse. Or it's all about listening to others. I'm just saying that it's something that I've heard people talk about. So this ability is particularly lacking in Taiwan. That's why I feel that Taiwan is so lacking in this ability. That's why I feel that Taiwan is so lacking in this ability. I have been talking about it at the time of the National Taiwan University. I said, "We need to go deeper. Deep plowing includes NTU recently. In a lot of papers. I want you to see it now. The nearest continent is really Too strong. So we have to be very humble. We have to embrace reality. We have to embrace reality. Then hurry up and patch this up. Taiwan still has a lot of ready-made. These favorable points. That way there is a way to deal with such a problem. So to summarize. We heard today The three guests in the audience There are some very important things. That is, Taiwan's advantages are Our infrastructure System. But at the same time we also We need to go deeper into vertical fields. To break through our deficiencies. Whether it's application scenarios or solutions, or cooperation with third parties. In the open platform we are We have to be broader. We have to embrace and challenge in a broader way in the open platform. Because of this The opportunities in the future are infinite. Especially How about grasping it at Fusco A One? This is a very important part, thank you very much. Sharing by the three guests. Thank you, thank you, thank you. Thank you, thank you, thank you, thank you. Thank you. Thank you. AI is the most powerful tool mankind has ever had. one of the most powerful tools that mankind has ever had, and it allows us to learn faster. Create more. It allows us to solve more difficult problems. More able to turn dreams into reality. And AI has started to really move. It can reason. It can use tools. It can even interact with the real world. It can plan and integrate itself. The AI Push to Adjust with Whom Factory in Lower Taiwan Machines and supply chains are getting smarter. Stronger, too. And it never stops. TheageofAIagentsis here agentsrepresentativeSoniceconomic operator Teddytheywould beintegratedintoevery industrychanginghell companiesupgradeandhowWorkr Jetson. ThenextfrontierissuesNSCALE AIsystemswideabodythat canbeBruceLureason planandMetainthephysicalworld Taiwansmanufacturers automationleadersandRobotics companiesareyousingagentsandNSCALE arethiswealthLiDAR inaneweraofRobotics wereintelligentmachinesbecomemenintegrated partofindustryand daylife. Manufacturers in Taiwan have put AI agents They're actually using them on the job. The AI agents have been actually used in the work, hitting intelligent factories, warehouses, and Industrial systems, and robots. The next three are actually bringing AI into the Three speakers from factories are joining us. Our first Keysight Yang, Head of AI software, Hon Hai Technology Group. The first woman in the group's history. Inmate. Hello, everyone! I think you all recognize Tzu-hsien. Chairman and Chief Executive Officer of PEGATRON Technology Group Chief strategist. President and Chief Operating Officer of Taiwan Electronics. Long. Let's start. A lot of this from what we just discussed Robots. That I want Keysight you stand On this production line Warehouse and global The front line of the supply chain. I think you know better than anyone. I think you know better than anyone what the factory floor looks like. I've heard that Hon Hai's biggest worry is not that AI is too expensive. It's whether you're fast enough or too hard. Can you fulfill this? AI needs. For you. Put AIGM into every Every factory, every production line. What's the hardest part? After this is done. This manufacture, this manufacture, this manufacture. It will be now. Or the whole manufacturing? It's going to be a completely different definition. What? Good! That Bruce Lu. I'll be honest. I'll be honest, before we import this AI for When it comes to smart manufacturing. Frankly, we're worried. It's probably even higher than after. Instead, we're all relieved. In fact, two years ago We've already started We've been working on this smart manufacturing. And we introduced a program called Janice. This Johannes is called Genesis. Genesis. So we have high expectations for this project. We hope it can have This is a different world of manufacturing. Usually. After this kind of expectation. Look at this GM. We took Janine ai as It's one of the main axes of introduction. But I think, first of all. Our conclusion is good. Three numbers for everyone. The first one. We are now The capacity efficiency of the rows has increased by 50%. And then our miscarriage of justice rate went down. 50%. And then. And this is what we've done. Abnormal Cause Analysis Improved accuracy to 90%. You'd be surprised. How did we do it? Actually, I'll tell you. We actually did three things right. The first thing. A lot of people would think, "Wow, I'm going to import AI. I've got to get it done. I've got to get it done. It's a big project. It's not the first thing. You have to collect the information well. So the first thing we do. This is our Chairman Liu. Very far-sighted place. What is he asking us to do? First, collect the minutes of the meeting. In the past few years, all the minutes of the meeting are first Collect them. All in a problem Any abnormalities that occurred. Processes. Collect the minutes. We realized we were doing this thing right. Looking back today. Data is the most important thing. What's the second? You don't expect AI will do it for you right away. The sky's the limit. So you have to put it in the right place. You have to let it in. What? Your Wall goes. And what's the third? Seeing the right spot, the pain spot. The right scenario. You just get in there. Don't wait. So it's not that we're actually It's not what we're worried about. How to import. What we're worried about is not fast enough. That's why In these 2 years we use Very fast speed import Very many use cases. And these use cases are now used very successfully. Let me tell you a story about a factory manager! What is the job of a factory manager? He has to keep a few things wrong. He has to deliver all the capacity every day. But it's very simple. But you know what? He comes in every morning. How many meetings does he have to have? How much supply chain noise. Purchasing has to have meetings to make sure that we We have to make sure that we can achieve that. But not now. We've imported it. We've introduced a program called Haw Chen's A-1, a program called HawChens A-1. After this one was introduced. Now I can't say it's easy. Otherwise we'd often be all right. But basically she's in the morning now. Just come in for a cup of coffee. She'll be able to read all the information. Let him make decisions. So I'm going to say A, i, so I'm going to say A, i, B is not to replace people. It's the AI that's going to make people So I'm saying that A, i, B is not to replace people, but AI is to enable people to spend their time in a more beneficial way, More valuable. So I think that's my answer. Okay, okay, okay, that's great. I do teach. Just let you do more things yeah! He could be a bigger plant. Things are good. So what I'd like to teach next? The chairman of the board. This A-one already? From the digital world. Into the digital world. The physical world of physics. That he can reason. He can generate action. Do you think that means The production process in the factory of the future. In the middle. What are the links in the production process of a future factory? Because of the use of agents. ai to take the biggest A benefit? I think that. Past factories. We all know that there are large factories. Actually Very complicated. You'd feel that just now The rebirth mentioned just now There are a lot of material to physical manufacturing Complicated things to be solved. Let's put aside the past. Let's have an expectation. What is the future? We have the God of Heaven. We have the gods painting a future world. When the factories, when the factories, when the factories. It uses HawChen. GNN Rick Tsai and NSCALE. When it combines with AI2. The factory is no longer It's just a lot of work. It's very chaotic. The factory is no longer in a state of chaos. It will become very orderly. Fewer errors, higher efficiency. Decision-making is very logical. There's a flow, it's very clear. When the unexpected flowers When it happens. What about this AI? It has the ability to help you solve it. Let's take an example I don't know. We'll take the factory as an example. It's this one. It's this one. AI software After the environment. It has Detectors. Its sensors sense something. When it senses a certain state. It will make a judgment. Decision-making. If you have a bunch of non-Nscale people in your factory, it will make a judgment and a decision. AI humanoid robots If you have a bunch of non-Nscale AI humanoid robots in your factory. It would have Replacement of human beings It's hard to move tires. Carrying heavy hands. Heavy screws. When a bunch of these When a bunch of robots walk by. The factory AI will communicate with him. Say, "This one's a couple. Pause, I've found one in the factory. I've got a transformer overheating somewhere in the factory. This time GNNtickAI of the factory. The factory GNN tick AI will coordinate. Robot. Go to inspect Follow the inspection. There may be one occasionally. You can send a drone. This drone also has AI. It's actually a robot, too. The kind of robot that can fly, it can probably fly to a certain place. Which corner to check? Faulty parts. And then notify the robot gang. It fixes it. In a scenario like this. It used to sound like science fiction. Scenarios, Science fiction isn't what it used to be. Me. Quote. Uh, We used to be Buddhist people. Talking about people. "can communicate with the world. It's because you have a tongue and a nose, Your eyes, your mind, your eyes, Your ears, your tongue, Where's your nose? They are all your senses. The physical world. The physical world. But what about that? That righteousness is your Alan Ni. It could be your model Your model. You have this tongue. Through this righteousness. To communicate with the physical world. That's how we can get rid of the last 3 years. Uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh. type of AI2. Actually, it's only the ears. It understands your natural language. It's got a brain. It's got a mind. It's got a brain. It has a mouth. When it takes the results of its thinking. When it tells you with its mouth, it's also using nature. But what it's missing. It lacks hands. And feet and body. Okay, now. Airplane AI It makes up for the heat, the tongue, the nose, the body and the mind. This one. The factory of the future. Its efficiency. It's functioning. Sequence. It can make people too. A lot of relief. Physical labor. The current definition is this. It's Robot. It's going to be a lot of robots. It will replace humans. The most laborious, The most physically demanding jobs. So an ideal world would be productive high, and people don't have to work so hard. This is an ideal world. There's another point here. That is to say, it is to make Our human skills are also Transforming. We put our skills in The most important work. Let's make our work more efficient and more productive. That robots can multitask. It can do this, it can do that. Instead of this single thing. So it's very flexible. We've been talking about it. Uh Uh, Rick Tsai. Some people associate it directly. To robots. Robots are, in fact, robots. The so-called hundred. Stripped AI. We're talking about unmanned. It's actually a loaded Intelligence on wheels. Robots. Da Vinci. The next generation of open-heart surgery. What about this Da Vinci? It's actually a-- it's a-- it's a-- it's a-- it's a-- A robot A robot. Boston, Boston. It's a four-legged machine. Dog. When you're going to disaster relief. When you're going into a When you're going into a dangerous construction site. You can start with the robot. The dog goes it's actually a dog type. Robot. That is to say the example just given. Unmanned taxi. With a robot dog. And Da Vinci. They're actually just like robot humanoid robots. They're all airplanes. This airplane AI works with. Like the last six months. Very popular lobster The lobster. Lobster is it. It's a soft body. But you'll find it can replace you. Slide the keyboard. With automatic ordering. Stocks Stocks. You take just this FICO, AI and Lobster Generative If you combine this FICO, AI and Generative AI like Lobster. The future of human production It won't stay in the past. Can only make coffee for you. Yes, it is. I think we're going back to a The fundamental question. That is, to support such I think it goes back to the fundamental question, which is to support the development of so many AIs. I think it's a very important issue. I think it's a very important issue, and the only way to do that is to say now. It looks like one of the big challenges. How to put a little bit of electricity It's not a waste. and Stabilized to the GPU. It's a very important problem. Meta The special thing about Meta is that it's from the power grid. It's a very special thing that Meta is doing here. To the chips' power supply. Module. From GREE to GPU. It's another form of end-to-end. That means this Very few companies in the world can do this. How do you want to utilize it? This advantage. In the whole AI energy I want to say how do you utilize this advantage to play a role in the infrastructure of the entire AI energy era? A different role. I I want Delta. I think Delta is doing this. power supply product line. We started planning early. From the power grid To Chips. So there's a lot of now. The products we see now are not today. The end of the line. It's been done quietly for 5 years. 10 years, like now talking about 850. We've been working on it for a long time. Down. Going solid state. Transformer SST instead of Traditional CD holes. Chip's been going this way. Should be the slowest in 12 years. It's going to explode next year. This time. The whole power system it It's going to be ready. - It's going to be ready. Controllable. In the past the traditional AC It's not adjustable, it's not controllable. That's changed to SST. From medium voltage The high voltage keeps changing down and it is adjustable, Controllable. It can be adapted to This AI Bora SIMT. Its routing is Open is very powerful. So the traditional power system. It is to be used. A lot of capacitors. Batteries to do the instantaneous UP. That's no way. Traditional transformers have no way to make adjustments. But SST can. That's what the genes do. That's what the genes do. The whole The whole end. The efficiency of the whole end of the system is efficiently adjusted. Provide. And we can see. From the center all the way to zero. A few. It's a lot of shifts in the middle, and it's a lot of shifts in the middle. Every time it changes. Transformation. It's carnivorous. It's going to increase a lot. So if you import SST In the center. There's a big piece of the present. It's a product of the present. It'll be gone. So the number of transformations. Less, more efficient. That's a lot of people. It's a lot of people talking about you like this. Not all of their products. Kill it? I don't think so. Because this is the trend. If you don't stand If you don't stand on the wave of the trend Such industry up, such industry up. Then you wait for someone else to do it before you spend old age. That's not what you're throwing away. Established products. It's the whole game of the future. The whole market is gone. That's why Delta. In the power supply system. We really have a lot of units. We're doing it. In addition to this power supply. We also provide A lot of this. Heat dissipation. Because as long as there are VOCs. It will have heat. I also have a power supply. I also have the power system, plus the heat dissipation system, and I also have the power system, plus the heat dissipation system. You can realize the overall benefits. I think Delta's going to do it. Such a wide range of products. We already have. We're not jumping off the traditional bandwagon. The experience of product design. We put A one We put A-one into the design. Design There's one inside. Big program. It's called SmartDesign. That's From AI With. The beginning of the big model Review Market Information. Then customer specifications. And then safety requirements. AI will help with the regulatory requirements. You do the summary. Otherwise, the engineer would have to spend money. A lot of time to read that. This one reads the specs, no. Not his area of expertise. Then next. We'll take all of it. Electronic circuits Or the whole machine. into digital models. Digitized models. That's it. It's not done yet. Before the hardware. That's what we do. In a MITRE environment. Iterating. We're looking for the best. We're just going to start making them. Samples. When the samples came out. You're You'll be able to validate it very quickly. Your GMLinkedIn What's the difference between your GM LinkedIn and your hardware? And finally The final testing is also The final testing is also possible. The environment is complete in virtual. That's why we are now We can talk about the future. We'll use it. So we can now say that in the future, only companies that can use AI will have AI. We can say that in the future, only companies that can use AI will have the opportunity to use AI to companies. They may have to go to market. Because of our prediction. The whole Design Code It can be reduced by at least 50 percent. Design codes will be less, Design codes will be less. It's also fifty percent, really. Because all your NVCells are Inside the computer. It's as long as the token pays. Okay, nothing else. No money. And then when it comes out. Right away it's This. The specifications or the production conditions. It will also generate it. So it's the same as our smartMONAI. IOlink together. So let the product We call it NeMo's Introduction. The NPI process will be significantly shortened. It's going to get our production line up and running very quickly. Connect to that. The overall effect will be dramatically improved. So In Delta's entire organization Inside the activities. We have several major themes. One is Smart Design Smart myself. Very smartCallLILT. And then Try to put the whole Delta's manufacturing Complete process I'm very smart Call LILT, and then try to make the whole process of Delta's manufacturing all... all... I'm very smart Call LILT. AI to assist. To help us. Let's improve the efficiency. Look at these in Taiwan. Manufacturing, really great. They've all been integrated into the manufacturing of the four clubs. I think the factories of these three companies are all There are a lot of collaborative robots. But Keysight I'd like to ask you guys more as if I'd like to ask you a question about the robots that you seem to have put in the factories. To the new vertical Kion. For example, I heard you're doing it in the medical area as well. A lot of collaborative robots. Do you want to tell us more about that? How about that? From the factory you most originally The best factory management. The production robots go to all kinds of Different applications. What about the different applications of the factory management robots? Okay, I think we When we talk about robots. Don't forget one thing first. A very important solution. That is. Anniversary. This Jonney Shih I'm talking about. Amazing. What problem is he specializing in, do we all know? He's solving the jack. It's funny. That's because. Before there was AI. You had to do costs. It's actually very high. So if we have a lot of anomalous work. Engineer. He's- If there's no way to get there early. Know the problem, solve the problem. But he often tells himself. I Would it have been better if I had known? But now there's anniversary after that. He's not just a simulation tool. This anniversary is now helping us. In the factory. All the hardware, Software Workflow. Containing the entire environment. Including the Server that we're doing the AI on. Rick we all know the hot one. Need a heat convection simulation. This Universe can do it all for us. So we can. In the virtual environment. We can keep doing simulations, keep doing training. Constantly. Learn and then optimize. When he reaches an optimal efficiency. We put it into the physical world. So do our machines. What about our robots? We don't just throw it into the physical world without training. Inside. We must put it in the virtual world first. A lot of training. So we're using MONAIVera. We are We're working on this new factory. With the new design. And this deployment. According to our current The latest planning. And the actual validation. We're about there now. By more than 60%. 60% or more, that's a very, very big improvement. Our current robot. We have robots that are actually on the ground in the U.S. What can it do? He can now do pick Pack. And then he can do simple assembly. All of these things can be done. But, ladies and gentlemen. This is not a robot. We think it's the most valuable. What do we think is the greatest value of a robot? Flexible manufacturing. In the past, when we changed production lines. Everyone knows us for the same kind of product. When we change to a new model. We're probably going to replace all of our equipment. Right? But now we have this machine. What can we do with this? Catch and replace. The end of the line. We'll change the software again. Do some upgrades. We can start using it. This is the biggest flex we've seen. Manufacturing help. What about us now? We're not just applying it in the factory. We're also working with Mr. Rong. Put our robot just like this Bruce We're also working with Mr. Rong to import our robots, as Bruce said, into Inside the physical hospital. To help the paramedics. Because the paramedics used to shave. Too much time. Carrying it back and forth. This is the sample. Get the paperwork. Now, now, now, now, now. We don't need to be taught everything. The neural carebot can help him out. That's the problem. So we think with AI. With robots. We think one of the most important things they can do is What changes human life. That's what we're all about. That's what we're seeing, thank you. Thank you, thank you, thank you. I think I'll follow Just follow the question up, from being a regular ah. That's you. In this implementation of AI. At the time of the factory manufacturing introduction. Did you run into any of this? Resistance comes, employee resistance. And then how to eliminate Their resistance, embrace. And then willing to use it? AI does this. Because actually people are a Success or not. Can you share that part a little bit? Uh, at the moment we're We're going to talk about this. The world will be a better place. Your job. It's gonna be higher. Your output. It's going to be higher Uh-huh. At the same time. You're not using this. Bleeding at your peril. Or pay. A lot of physical effort. We have to go back and say this. The invention didn't start with the invention didn't start with the invention didn't start with the invention did. Replaced? The position of the housewife The housewife's position. It was the invention of the microwave oven. Neither did the invention of the microwave oven. Mrs. I'm kidding. The washing machine and the microwave oven give the wife more time to take care of the house. Her husband's business. Employees should be hugged too. I'm just saying, the washing machine and the microwave gave the wife more time to worry about her husband. It's AI. With airplanes, AI. It helps you spread out a lot. A physically demanding job. For example, the robot. It can work in factories. Heavy work. Like lobsters. This tick AI. What can it replace? Replacing menial work. Because some people's jobs are boring. Just looking at web pages. And then they're just pushing buttons. In the future, all this stuff will work. This will replace AI. The relative human. When society is more productive. And You will be able to develop more More interesting, Richer careers. I don't know. Uh-huh. Uh, time. The 200-year population will have to re-farm. Inside the farm. Now through chemical fertilizers. Through the plow. The productivity of everyone's agriculture. Liberated. Fifty years ago clothes were still a Expensive thing. Now- Revealed through After a lot of factories. Clothes are the same as just now. The food we just talked about. We have enough to feed ourselves. We can use this. Easily. The future is in the factories Productivity inside the factory Or productivity in the office. Hugging Face AI and HawChen love. Make our work more fun. Higher productivity. Just like the farmer. Worried about losing his job. What are you tied to the farm all day for? A housewife doesn't have to worry about not having a job. You're tied up in laundry all day, What's a washboard for? Future AI. AI. Whether it's physical AR, FICO, AI or virtual AI. It can help your productivity. So employees can rest assured that they won't lose their jobs. Your job will be more fun. I only have one washing machine to refill. It does help me a lot. We help my wife is. But retirement misses me. In fact we In importing all this smart manufacturing. In the beginning, we actually had a lot of factory managers in mind. Very worried. But it's really come to this. Our factory manager is fully embraced Totally embraced. I think. Because of the benefits. It's a process. That's it. That's it. The soles came from a lack of understanding. I'm afraid. But you're beginning to understand the staff. But you're starting to understand what it means to let the employees know what's good for them, what's good for them. They will start to use it once they have experienced it. Yes, the process is a process. And they've already That. It's the best. Because I'm able to hug. To use it. To be successful. Yes, yes. For me. Thinking about this robot? With this Generalist AI? To functional. That we came three years ago. Look at vacation, 2 years ago, One year ago you could see a very noticeable change. That's when we were looking at it 3 years ago. It's Generative AI to last year. This year to this year. It looks like the dawn of AI. It looks like the dawn of AI has been seen. So this is something that I think is It's a very obvious change in the last 3 years. We've gone from morning We've gone from the morning speaker to the presenter to the presenter. We've gone from the morning speakers to the real you guys. In using these. I think it's just Just for Mac looks very, very amazing. Yeah. - Yeah. It's So. We'll just do a little Let's summarize. OK, this summarizes it. The future of APP. will be first I'm sorry. We'd like to thank the three guests. Thank you very much for sharing today. Thank you. Thank you. Thank you, thank you, thank you, thank you. Thank you, thank you, thank you, thank you. The future of AI depends on various things The acquisition of resources The future of AI depends on access to resources. Including arithmetic power, cloud, Infrastructure, Research platforms as well as the inner letting enterprises and developers really Hands-on development tools. That's why Taiwan is going all out. This is why Taiwan is fully investing in the construction of an AI ecosystem. Full support from every step of the way Next-generation innovation. AIDevelopmentdependentonthefullstack comingtogethercomputeplatform istalentand Local infrastructure. InTaiwanthatyouKWsystem istakingshape.. connection researchchurchStartups andenterpriseswithaand platformtheyneedtobuildinthe AI.. andforthe The world. StartupsExcelorrating innovationresearch Inferencingdiscoveryand NIMPrizeascaling in.. the endtireEaglesystemisbeing enablethisisdrivinganewwave ofanimationfrom AI2factory to. to agentstofind AI2buildonthecommonarchitecture that spencefromcloudsto edge. AI is rapidly becoming the norm in every country. AI is rapidly becoming a priority that every country must invest in and support. The ability to access computing power has already increased. To national strategy. AI is rapidly becoming a priority that every country must invest in and support. Taiwan is now at the forefront. Building for Taiwan and for the world AI foundation. This is a demonstration of access to computing power resources. How important. The next three guests. This is an important part of this work. Pushers. Let's welcome our first guest, Mr. Majestic. Mr. Lie. The Majestic General Machine Intelligence Cloud. Founder and CEO Founder and CEO. Second Li Hong. Professor, National Taiwan University. Third, Rumor Central. Founder and CEO of Supercomputing SuperSense. Foxconn's AR foundation. The department. And before we get to the questions. I'd like to share a little something with you. I'd like to share a little story with you, about a decade or so ago. I had a chance to visit Jetson. At that time, I asked Jetson. Jetson, if you Can you give a Taiwan's start-ups say they want to do it in the future A start-up company. Or young students. He wants to enter the AI industry in the future. What kind of advice would you have? He gave me three answers. More than two decades ago, I still have a very deep impression. I think they are still applicable now. I think it's still applicable today. He said the first one was you You have to be full of it. You have to be enthusiastic, you have to be interested. Because if you're interested, you're committed. No matter how difficult. So the second thing is this. The second thing is that this thing has to bring out the best in you. Higher meaning and value. For the industry Or impact on business. It has to have a rapid impact. Because the value of the output is the value of the output. Continuity. And the third is that you You have to pick so many things. Which one is the most difficult. The one with the highest threshold. Because there's only the highest threshold. The most difficult. It's only the most difficult, the most difficult, that can sustain its competitive advantage. Go on. OK, so I think by today. No matter what Whether it's in a start-up or a personal job Occupations are applicable. So I'd like to start with the first question. Professor Li. It's you who has been with Taiwan for a long time. Whether it's a start-up company or a young person who thinks Since they want to step into this CAD department. What do you think? Cultivation in the future What is the difference between this part and the past? And how to make these talents to be able to In this example Identify a real challenge It's the biggest challenge, yet it's possible. Change the direction of the industry. I'd like to ask if you could share. If you're talking about now. This point in time. Of course we need to be able to. Develop. AI talent, there's talent that can deploy. AI. Talent. Talent that can apply. I'd like to mention a couple of industry professionals who have been working in the industry for a while. What kind of talent do they need now? But If we look a little bit farther. Let's say we I'm actually looking at it from five years later. The question to ask is us. Do we still need to train them? Talent. Especially our so-called Traditional AI talent. If our so-called traditional AI talents are People who can train models, for example. The future training model Is it still going to be done by humans? Will it be done by AI will do it. We are now We're already seeing too many signs that it's coming. It's going in this direction. We've seen a lot of signs that it's going in this direction. We may have just seen Open AI a few days ago. They said their model solved it. What, hard math problems. You see it all the time today. Similar news. Even in our lab, we see a lot of that. You see a lot of examples of that. Let's say there's a student who has a PhD. He's very good. He found out that there used to be a great man. A commonly used algorithm that had problems. He just throws the algorithm away. "Do you think there's a problem with this? And he said, "Well, there's a problem. Gives us a new proof. And then we published it. A very good paper. Subverting our original idea. So I would think that in terms of the future. It's very possible that those developments AI's too. The way that another AI is likely to develop AI in the future is to tell-- He said help me with this. Ampere. Or I am now. I need a speech-language model. Build me a speech language model. And then he's like an engineer. Maybe just finish the task. So I would instead think hypothetically looking back five years, 10 years, maybe what's really important to us is That one knows to optimize. The person who knows what to optimize. The one who knows. What our needs are. The person who knows what our needs are. And then he knows how to comment on a model. When the model lasts At the time of optimizer. When we keep the model going in development. When it's actually in training Continuously monitor him. Make sure it's going the way we want. Let's say I just said hypothesis. There's a person who's going to start by saying I'm going to build a speech-language model. Why are we building a speech language model? Why would any human need him? Why is it a model that we need right now. Someone needs to judge its value. And then the training process. How do I know it's getting better. Someone has to keep evaluating it. To monitor it, I'd rather think it's In five years time, that's probably the most important talent that we need. So one of the points you made. I understand that not. Just an AI technology Pursuit. But in all walks of life. You have to have a deep understanding of every field. You can only know to say this. To what degree of optimization. It is able to solve the problem. So The strengths of each field should be cultivated at the same time. So I can't just rely on knowing AI technology. I can't just rely on knowing AI technology without knowing that there's one for every industry. What's the problem in the deep business field? It's the combination that will make the future possible. AI talents are more Competitive. If you want to come up with a slogan. I think the world of the future What do you want to do than to do The point is what you want to do, and what's important is what you want to do. The focus is on what you want to do, not what you already know how to do. Want to do what you've already stepped into it has some thinking. It's had some thought. It's only able to make it that far. So that's a very good piece of advice. How to train future AI talents? Thank you. Thank you. I think of what I want to do. Young talents with mobility. We want LILT to teach Out of this is what you spent Six months. On American soil. Ask the substation. Asking for energy codes. How hard did it take to put Datacenter? This set of heart method are brought to Taiwan. That you put this Expensive and hard to get GPUs. Calculation power for Taiwan enterprises Who can use it? That's basically AllyouE, like water and electricity. Then you think this Taiwan's start-ups In terms of obtaining computing power. What is their biggest pain point? When computing power can be you like this When you lose to them smoothly. The speed of AI innovation in Taiwan as a whole How will it change? I think it's a very good question. I think it's a very good question. In the past, it might have been an enterprise. I think that in the past, perhaps enterprises have been using AI in the past. I think that in the past, when enterprises used AI, they were all talking about going to use it. These GPUs, but frankly speaking. GPUs are actually only In a few start-ups. But honestly, GPUs are only being used by a few startups, the native companies. Because it's too painful. It's GPUs are farther away from PhysX. stable. So you need to have FRA talent research. Talent, pre-screened talent, follow-on talent. And the inference talent to use it. These GPUs. There's actually a paradigm shift. Like Jetson said. This token. The concept of a factory. I'm taking these GPUs and I'm taking them and I'm taking them and I'm taking them and I'm taking them. Virtualization. Into this model service, right? Then we're going to integrate more Hundreds of models. VoiceModel. And then maybe Video Model. This one has Image Model, and this one has ImageModel. There's also a big language for languages. Model for, different Context. That would actually allow Any enterprise to quickly Using AIinstateof It's straight to the point. Go po Visteon these GPUs, go po Visteon these GPUs, go po Visteon these GPUs. Because it's very, very difficult. It's not just a Taiwan problem. It's all over the world, it's all over the world. So it's fast. So it's a quick way to get to the Like the professor just said. Is going to ask this question. And it's not like I already know what to do, it's not like I already know what to do. I don't know how to ask this question, but I can't do it. It's a very painful thing. I hope so. So I hope that we can get rid of all these difficulties, that we can get rid of all these difficulties. Simplify. I want to simplify all these difficulties and turn them into a token pattern. Let everyone use it. I think that's fine. Bright is now I think that's what we're trying to do now. So the only thing they need to think about is Bright is the only thing they need to think about is what Cart dowhatI. As long as he goes thinking out of the box. He'll be able to go fast and execute. Through token or through Model service inside use With the GPU. I'll talk about that. I'm talking about the token. I'm still going back to the Neo side. It's the Vision that's being used by the It's going to be the biggest in Taiwan. And it's the first in the world to use NVIDIA. GP300. I don't think it's just building a data center. I think it's not just building a data center. I think it's not just building a data center, it's helping Taiwan to build In the AI era, it is the first supercomputing center in the world to use NVIDIA GP 300. I'd like to ask you for your advice. I want to ask you, that is, many enterprises in Taiwan in the past, that is to say, many enterprises in Taiwan in the past. Even public sector data have to be sent Outside the country to do some arithmetic processing. And then come back that this and we have been emphasizing is to say. You have been talking about this matter. That is the key data and domain to how to stay in Taiwan. This is also the sovereignty that everyone has been mentioning The concept of AI. That in Taiwan Became a NAVERCloud. How to balance you With the traditional ones. The relationship of large CSPs. That At the same time and competition and In the case of cooperation. How are you going to find that balance? Good, thank you. Moderator I think this question has often been I think I've been asked this question a lot. Especially when I'm in the Red Sea. Group. Many of our clients are Some These Clients. And frankly. More than a year ago. When we announced that we We're going to step in. This AI, DC industry. A lot of customers did have I'm sorry, I'm sorry. I've got a lot of Kion. Today, one year later. In fact, I found that this Car has It's changing. It's changed for several reasons. The first one is GPU shortage. That's why the CSP we We've seen a lot of CSPs in the US. We've seen a lot of CSP operators in the United States. He's working with these New Group services. Provider In this cooperation. So it's already So it's not a gradual process, it's not a gradual process. Competitive relationship. It is a A cooperative relationship. The second one is CAPEX, Opus. I want to recognize Wall Street is actually interested in this piece I think Wall Street is very skeptical about this. So it's caused the stock price to be a little bit more skeptical for a while. It's not that bright. Somewhat related to CAPEX. Or. It just so happens we got it. We've got a business opportunity. The third I. I think it's this industry It's still It's in a. The division of labor is defined. Why do you say that? Let's look at the PC. With cell phones. In fact, before the earliest time. Our OEM customers are They had their own factories. So on the one hand, he has to run this So he has to run the brand on the one hand, and the factory on the other. But it's not that people don't We can do anything. So the end result is to Division of labor. Everyone is responsible for this. The specialty that they are best at. So the OEM gradually throws away the burden of the factory, and the OEM gradually throws away the burden of the factory. to Old or Master to handle. The OEM concentrates. In product development. In the marketing of the brand. On top of that. I think. The next market I think the market will gradually move towards This is a model. That is to say, our model companies. Or these cloud service providers. He just focuses. in their cloud application services. R&D. Or even model development. What's left is this. The rest of the Token or the Token or the Token or the Token or the Token or the Token or the Token or the Token The rest of the Token or the math, it's given to this willing and good at building AI, and good at building AI, The rest is left to this group of people who are willing and good at building AI and DC to do the processing, and that's it. So this division of labor. It's going to make The whole speed of innovation research and development It will actually go faster. So you're now See what you've got. The newly created guests. It's in the pioneering. New applications. Do you see him having you see a clear direction? Because we are now What we're seeing right now is Maybe one or two. There's an explosion, but we don't see a clear trend. We know Janine is going up. We know Janine's going up, we know Nscale ASUS is up, we know Nscale ASUS is up. We're seeing one or two handfuls of companies, and we're seeing one or two handfuls of companies. That you're looking at it from your perspective. Because you have some existing customers. You have to see a situation that's broader than ours. You're seeing a bigger trend. That is to say, there are no applications that will come out. And which ones might be the second wave. Which ones might be the third wave? That's a very good question. I think all of them I think all the companies. Even the modeling companies. I think all the companies, even the modeling companies, are still in a Exploration stage. That's a good question. From the application side. It's actually the CSP. Even the company knows. What he's going to do. I'm looking at it from the application side. In the whole enterprise. I'm looking at the application side of it. Because the enterprise side is the closest to the application. The enterprise side, because the enterprise side is the closest to the application side, the end part. I believe the world is now I believe a lot of enterprises around the world are still in FigureLVMH. What exactly is his Aviation. How is he going to make this AI A GM, how is he going to make this AI AGM, how is he going to make this AI AGM? I believe that many companies around the world are still figuring out what Aviation is. Apply to his Workflow. Workflow. So I think that's one. Very. Potential opportunity. Why is AI now so much potential over there. It's Because a lot of applications haven't been developed. Including swearing at themselves. We have this SmartMade in-house. InferencingSmartCity. We're actually targeting every industry applications. Think, go do this figure LILT. That's because we have the scene. So we're better at knowing that. I'm in the Mali part. What are the applications that can be used by Define it. That's why a lot of the modeling companies. That's why a lot of modeling companies, we're now We're seeing this OpenAI or even Some of the consultants that Adobe could acquire The consulting service companies. Because he knows him. He's got to get into this. The enterprise goes for each different segment. Whether it's Semiconductor Medical, defense, security, etc., etc. Segment to do this deep The band. He's going to know better. What products should be This is developed. I think we I think we are in Taiwan. In fact, in many areas are the best. I think we in Taiwan, in fact, are the leader in many fields, are the leading position, are the leading position. Including semiconductors. Including manufacturing. Even the medical part. That past Taiwan Focus is In the traditional hardware field In the past, there was nothing like us. To cover AIDataCenter in Taiwan. Put the data in Own this The local end of the local end of the local end of the local end of the local end of the local end of the local end I believe in Taiwan. I believe Taiwan has a very strong domain name. Even a lot of data. I believe Taiwan originally had a very strong domain name, and even a lot of data. To have a Evidence of computing. Capability to help him put it NVCell. And then Let the whole domain Because of the friendly math. So he can do more Value things. So I think soon I'd like to ask you a question. I can share Just in Taiwan. If that's the case From this you guys New startups With the school. Maybe with the government is how to come together to cooperate. How is the whole ecosystem going to work together? Cultivate talent. Or to establish this sovereignty. The capacity of the data center in Taiwan, the capacity of the data center in Taiwan, the capacity of the data center in Taiwan. Because it's not just a matter of math. It's not just about computing power, it's also about All the Businesses. The innovation of the business model. Or talent development. Or this. It's a collaboration between the company and the school. That's it. In terms of innovation or whatever. You can share a little bit of each one of them. Okay, I can talk about it. That's right now. There's actually a paradigm shift. It's from training to Inferencing. I can add to that. That's just some of the questions. MIMD is now two points. In Inferencing is already already insured. Everybody knows Scaling. Very fast. There's a Common dynamic, there's a Commondynamic. That is to say AI It's very, very good. In solving structureproblemset. So it's like a single source tweets, so it's like a singlesourcetweets. Like coding is based on It's basically what everybody's using. And you're also Anthropic is growing very fast. The other one is. The other thing is that the video and the image model grows Very fast, so those are the two tracks. It's already taking off. So tokens are being used again. It's all increasing. And the corresponding thing is to say Okay. Since these two industries have increased. Then ACER Cloud we'll go and serve. These things, serve these people. So it's just a few talents. There's going to be a huge increase in demand. I don't know. That I think is Taiwan is Very much can focus on cultivation. Especially Taiwan. Hardware kingdom. The firmware on top of that hardware are Very strong, strong. So it's probably DevOps in the server room. The k-8s management talent. Metal management is cluster management. There's a huge shortage of those people. There's even a shortage of people to do Networking. That's what I'm talking about. Because you're trying to scale the whole cluster. You need to do that. There's a huge need for networking talent, and there's a huge need for networking talent. That's the bottom line. The middle layer is now I'm talking about token marketing. You've heard of this stuff. It's got a company a month. It spends $500 million a month. It's just too much, it's just too much. Not every American business in America can do this. It's not every American company can play like this. So in fact it's optimization. It's about optimization. Well, it's this one. It's either that or speed up. That these things are all services to Different models. It's all about optimization, it's all about this book, it's all about speed. imageaudio. And the language model is different. Reasoning models for text length. Nonfeature models need to be optimized. The people who do the optimization. I think it's like these things at NTU. It's all very, very good to focus on. Focus. This talent. It's a talent that's very, very, very scarce in the world. That's At the upper level. warQuestionstheasset. That's what I'm talking about. All these things can make us these Cloud technology resources can figure out. We'll figure it out, right? We'll just focus on what we've done to help you. Execute. But now is how to get things all Put it together. So we're probably gonna need a new generation of these guys. In college or graduating. This is Asia Native thinking and saying. whyapplicationtobuild. So we're gonna want to say. So we'd like to say, since it's so simple. So we would like to say, since it's so simple, we can go and do this AI. Native software. So we would like to say that since it's so simple to make these AI Native software, that means one is specialized Doing this accountproduct Genius, right? You can put things together just fine. How do you execute it? We'll get it, you'll focus on it. How do you go about it? Into an application bar. So there's three main layers. It's in Application and then inference. Optimize this The bottom of the machine room, the bottom of the machine room, the bottom of the machine room. It's the talent that's needed. I think it's all Taiwan. You can go now. This part is missing. Thank you, Professor. Okay, I can cite some. Recent. Practically, we talked to the industry. With NVIDIA AI to do the case description. How about cooperation between the school and the industry? We can get a win-win situation. In fact, many times in the academic world We have empty ideas. But we are very short of computing power. So our resources are very limited. In fact, it is sometimes difficult for us to imagine the resources of the academic sector. We can't imagine what resources the academia has. We can hardly imagine how limited the resources of the academia are. I'm looking at the Internet. I saw on the net that Li Hongzhi's lab should have a few sheets. H, don't say H, don't say H, don't say H, don't say H, don't say H. A 100 are not ah, how do we calculate resources are not. This is for everyone to hear. I've heard it, everyone has heard it. The answer. So? But there we'll do it with NVIDIA AI. And then NVIDIA borrowed us. This is using TAIPEI's math. Sixteen H100s. And then we helped NTU. Our lab helped. NPUCUDA at NTU. It's a big support from NTU. Under the support. Our own speech recognition system. So from now on, our courses will be on Upload it to our course platform. NPUCUDA will be automatically generated. Subtitles, so you might think there's nothing unusual about generating subtitles. and English or a mix of Chinese and English. But our courses tend to be a mix of English and Chinese or a mix of Chinese and English or a mix of Chinese A lot of proper nouns. So we take everything that's commercially available. Systems. including OpenID's Viper don't have access to. It's very recognizable. But we're targeting us. Our own course training, Speech is a system that can do a better job than OpenID. And our speed is even better than Open AI, and our speed is even better than Open AI because we're already faster than the small, small, small AIs. So the speed can be open Five times faster than the best open model. And that brings us to It's a great benefit. This is one. It's a win-win. There's another example. This is also with NVIDIA A10. This is another example. In the project NVIDIA also provides The computing resources. And then we train a speech language. model. So the speech-language model is that it can input sound. And then he doesn't just do speech. It's not just about speech, it's about understanding everything. It knows the person's gender, it knows the person's sex, it knows the person's gender. It knows the person's gender, it knows the person's emotions, it knows everything. And then it gives an appropriate response. That's a very hot topic today. That we have a good algorithm. But we don't have the math. We don't have the resources to train the model. So we work with NVIDIA AI. 5,000 hours of data to train a model called Let's Name That Model. It's called DAS, for data. It's a 5,000-hour model, but it's a 5,000-hour model. Not a lot. But that's us. At that power. It's the maximum that we can use at that power. The amount of data. That's not a lot of data, but that's the maximum amount of data that we can use at that power, and we can't train that many models normally. The model is That's 2,000 more than what Alibaba had open-sourced at the time. The results were astronomically better than the 2,000 dollars that Alibaba had open sourced. The amount of data used. But our algorithm is better. That's why we can do it. With limited computing power. Better results. So this is a win-win situation for the industry and the academia. In fact, it's only through adversity that we can grow. So it's because the math isn't realistic enough. There are very few breakthroughs in the face of adversity. Forget it, there can be a big model competition. When we had the power, we had the power. We have to think of good ways. To that. Yeah, yeah, yeah, yeah. My talent is really with us right now. One of the biggest problems we're facing right now. So I'll break it down into two pieces. One is our internal Operation. Frankly speaking. It's very difficult to find people to go. That again. Because our first time is also Taiwan's first In Taiwan to establish such A large-scale cluster. Including a lot of precipitation ah, air conditioning ah. Even global people. In fact, all aspects of us go there Taiwan in the past did not have such a large cluster of such Talent on this side. So to some extent we Now Also recruited A lot of such related background people. You just have to have the relevant background, you're willing to learn. Then I will provide this platform I'll provide a platform for you to do this learning. And then hurry up very little the same time we also We've also started with some schools. We're talking about cooperation. Before you guys graduate. I'll give you interns before you graduate. After graduation. Ampere. Another piece. Another piece, this one is more on the supply chain side. Another piece is AI application. The other piece is the AI application side, this piece is actually the government side we talked to. We also have some cooperation with the government. We have also been thinking about how to make it work through We have an advanced force. We have an advanced force, and through advanced and smooth The government is going to do this. LK, then he LK to some new creation Or academic institutions. So go to this IonQBay these AI application talents. Will blow. You never feel like we're just like It's an extension of Team Taiwan. We do. We have the most senior seniors. We have the most senior seniors, the middle generation, and some of the new generation. Some new generation. I want this Team Taiwan. come on. Analyze, Bowl, Hope Taiwan KionTELUSIndustry. I think that should be a wrap-up for today. I think we're not the only ones here today. The bigwigs of the industry, the star pupils. I think we've all heard it. Three of you should have a lot of ideas. And then I hope to be able to put into this The important AI of the future. Good. Good. Tetra. Then again. A few more minutes. Jetson will be next door. The next venue. He's going to share with you He's going to share with you NVIDIA's vision and how it's going to build what we've just seen. So in a few minutes, Jetson will take the stage next door. The future we just discussed. We'd like to thank our guests for ending the day. Thank you for sharing the future of AI. And the insights and visions that they bring. And thank you to all of us at NVIDIA and to all of our colleagues at NVIDIA and to all of us at NVIDIA. My partner, Christy, thank you. And thank you, Bruce. Thank you, Bruce, and thank you NVIDIA for making this a great group of people. And thank you Bruce, thank you NVIDIA for bringing this group of outstanding leaders in their fields together today. And thank you, Bruce, for NVIDIA for bringing this group of outstanding leaders in a variety of fields together today. In Taipei. The streets are about to take the stage for the keynote. I think you guys can't wait as much as I can. Thank you to all of our viewers around the world who are watching. Friends, thank you! I am. I am. At the Taipei Pop Center. Tracy Tsai live. I'm Bruce Lu, thanks for watching. Welcome to the GTC. Welcome to the GTC convention. Kion. Me. Me. Rubin. Rubin. Me. I'm Rubin. Follow. You're N1X. Kion? Cohere? Cohere? Listen? Okay Okay. I'm great.