Building the AI Growth Engine

So maybe I'd like to start with you, with you, Phuong, you know, from your vantage point at FPT, you know, you're working, you're working with companies thinking about adopting AI, you're using AI yourself, but kind of what's the state of things in terms of AI in Southeast Asia from a software perspective? So first of all, thank you very much for having me here as a panelist. It's an honor to join and to talk about AI transformation. So FPT is the largest technology company in Vietnam and we have interacted with many companies in the region. I would say Southeast Asia has been focusing on AI transformation but a bit later than the rest of the world like North America and Europe. But we are going very fast. Companies in Southeast Asia are trying to go as fast as they can but sometimes we see the the stage of maturity among companies are very different companies want to utilize AI, want to make a big bet on it on AI and they have been trying to increase the productivity however AI is not just about productivity AI nowadays, you know, as an American started to say it's a super intelligence instead of the artificial intelligence. So it can be not only a powerful tool, but it can be an operating capability that an organization should have. So FPT has been trying to develop a lot of different framework and methodology to help companies. to transform with AI so most we have a five-stage framework like company would go from curiosity to AI augmented to AI standardization to AI automation and finally AI native and we found that lots of companies in Southeast Asia probably in the second first or second stage of the AI transformation it means whether you are just curious about AI Trying to test and do a lot of POC on what AI can and cannot do. You are moving to the stage that AI can become a productivity tool to help your organization, to empower your human workers to become more productive and become at the augmentation level. And we foresee in the upcoming years, in the future, Company move AI capability to a higher maturity stage which we call it AI standardization and AI automation where AI not only become a tool but become the workforce, the partners in your company and that's where AI will unlock a lot of different values. So in short, we are in a very early stage and we have a long way to go. Let's kind of move from software to hardware, and I want to bring Julian in for this conversation. So, you know, everywhere needs more AI infrastructure, whether that's connectivity or data centers or something else. But in Vietnam specifically, what more investment is needed in AI infrastructure, and how is Stavium trying to fill that gap with its projects? Okay, allow me to speak in Vietnamese. Please use the translation, please. Okay, first of all, I would like to talk about the data center in Vietnam, because actually we need the data center for development, and we have only 140 megawatt data center, pretty much smaller than Thailand or other countries in the region. The reason why data center is not developed in Vietnam is, firstly, because of lack of Because the total capacity of electricity is about 87 gigawatts only, but in order to develop data center, we need to double this number. And secondly, about the international connectivity, because in Vietnam we have about five cable lines, but with small capacity, so it does not satisfy the requirement for connectivity. And also we have limitation to import CPU from the high-tech corporations. From the U.S., as in Vietnam, we... We are now developing a centralized high-tech area in Ho Chi Minh City, and we want to bring Ho Chi Minh City to become a data center hub, similar to Cyberjaya from Malaysia. In order to serve that data center, we are developing sea cable lines named Avista, connecting I think Malaysia, Vietnam, Thailand, and Singapore We have about 16 double cables, and the capacity is about 400 terabits, which will be put into operation by the end of 2028. And secondly, we need to prepare the electricity capacity, not only the traditional electricity, But also like LNG, LGCD, or... small nuclear plant and many other renewable energy more about energy in a second but on this cables point I mean does Southeast Asia have enough cables to carry all the data that has to travel within Southeast Asia between Southeast Asia and the rest of the world are there enough cables built right now Singapore is currently the hub. Singapore is the cable line hub in Southeast Asia, and all the international cable lines, about 40 international cable lines are connected to Singapore and then to Europe, to U.S. And now we are developing the cable line and connect from Vietnam to Singapore and from Singapore to other areas in the world. In my opinion, because we are now investing a lot in building a data center, and it is also, and I think that it is a very smart FDI direction and a very high value FDI, because it will bring higher economic value, because we will approach like GPU, CPU technology I think that it is a very high-quality FDI models. Yeah, even higher than other manufacturings. And I also want to add, in Southeast Asia, we are following a model of manufacturing hub with lower labor cost. So we are now stuck in the middle income trap. So I think that it is now the time for us to think of making use of robot and AI to get out of that trap. I want to turn to you now, you know, It feels like with the AI boom, there's been a lot more attention paid to people making components, people making the stuff that goes into semiconductors, into data centers. You always hear stories about how Ajinomoto, best known for MSG, also makes the key film in semiconductor production. Or there's one factory in Japan or one mine in Inner Mongolia that makes this key component. What is the AI boom doing both for your business specifically, kind of working in connectors, but also broadly in kind of the quote-unquote picks and shovels in the components part of the industry? Yeah, actually, again, from the AI data center, we were in the AI supply chain since actually 2015, 2016, very early days. And since then... We have been growing very fast in this industry. I mean, before, I would say at the beginning, AI-related, let's say, power cables or power webs, those are only, you know, like a single digit of our revenues. But right now, I think it's more than 30% of our revenues. And like what I mentioned yesterday, actually, Actually, no. it's like before 2020, we only have two factories in Southeast Asia. And right now, we have 12 factories in Southeast Asia. And that's because the ecosystem, the development of AI data center and also the semiconductor production equipment industry. OK, so for this, I mean, it's a, again, it's a, Of course, people are talking about why don't you manufacture in the U.S. Like right now, a lot of the contract manufacturers move to the U.S. However, for us, it's like a lot of development was done here, actually in Southeast Asia, also in East Asia, so in China, in Taiwan, also in Southeast Asia. So quite a lot of things were developed here. Thank you. So that's, actually it's not, for us, it's not, of course we still have plan to manufacture more, we have also more than 10 factories in North America, but for us, it's like, mainly the AI data center and the semiconductor equipment, production equipment manufacturing will be in Asia, okay, will be in Asia, and again, we talk about not just cables, not just connectors. And it's not just modules, it's not just bus bars. We do quite a lot of things for the AI data centers. And... Actually, just recently, we just completed an acquisition to buy a company making sheet metal and also racks. Actually, some of the people here are our customers as well, because AI data center is more than just a small component. And like the bus bar, it's a huge copper rod on the back of the compute racks. And those are very important to deliver power to all the GPUs, CPUs, all those kind of things. Okay, so, and also we are talking about power width that's connected between racks. Then we have... For data transmission, high speed cables connect within racks and between racks. Then we have also for the data center, we have so-called bus way, so that to connect between racks and the whole facility infrastructure. So I mean, the demand for this stuff has surged by so much, thanks to data center construction. And I know we all think about the memory shortage, I mean, SK Hynix is talking about this going It's going to be short until 2030. Like, are there other components that people, that you're seeing shortages of? Do you think you're going to be able to keep up with all the demand? Yeah, so right now, you know, not only people, of course, people talk about memory, but you know that right now, the fiber. Fiber is a big shortage. Okay, so, of course, recently in the past couple of days the stock market is tumbling, right? However, the fiber shortage will last for a while. Okay, that's probably everybody already knows that, for example, those big fiber companies, not only in the US, but also in Europe, and also in Japan, in China, They are kind of trying to ramp up their... I want to go back to Defong and kind of dive a bit more deeply into this, you know, enterprise adoption framework that you talked about, how most companies are still in the curiosity phase. And, you know, I wonder if you could tell us a bit more about FPT's journey, thinking about AI and using AI in its own operations. And then kind of what you think is still, you know, making Asian companies a little bit more cautious, potentially, about using this stuff? Sure. We are, Asian companies are very different from the American market. First of all, I have to emphasize some things like this. So we are in the AI supply chain, we often joke with each other. So we are the top, the tip of the iceberg, where we make the software, people can see it's a so cool way AI can do a lot of stuff. We are like the gold and then now but the one who actually reap the return on investment are the hardware side. You have the search and demand and you are trying to sell the shovels for everyone that go do the gold mining, right? So for the tip of the iceberg, for the gold, right? Lots of companies are behaving very differently. From the American side, we can see that companies receive a lot of investment. They're burning billions of dollars to train large models. They are doing... they release new framework, new models every day. And it cost. There are hundreds of millions of dollars to train a large language model. But it's very different in Asia, in Southeast Asia, where companies want to spend smart money, want to do a return on investment. So whenever a CFO in the Asian country, they will ask the technology team, if I give you 10 million dollars, what do you give me back in the next five years? Or next year, even. So it's a big pressure in Asia to find a return on investment. It's not easy like that. That's why in my five stages of adoption, it's called CASAN, actually C-A-S-A-N. It stands for Curiosity Augmented Standard Automation and their native. There's a valley of death between the augmentation and the standard, where the value has a diminishing of value when you only treat AI as a productivity tool. Because you think that if you give every of your employees an AI motor, then the next day the company would be so much productive and 10% to 20% or even 50% more productive and we should have 50% more productivity and revenue, but it's not. That's a diminishing of value where people using AI and create value, but at some certain task, not the whole jobs. And how are you thinking about, again, just like how much these models cost to use? I mean, we had, again, Jens Lautner on the stage earlier today where he talked about how, you know, DeepSeek's pretty good and it's pretty cheap. I can tell you I use DeepSeek sometimes, too, and it's very cheap. But from your perspective in terms of kind of the portfolio of tools a company like FPT has available to it, how are you thinking about cost? So cost definitely one of the thing but we don't look at it too much. We are now measuring a value per token. So tokens can come with different models, different intelligence that they have and it also comes with how much tokens to finish a task. So you can now speed up a task, but sometimes you don't speed up the whole supply chain of jobs. So that's why people don't see a return immediately in the augmentation phase. You have to escape the valley of death into the standardization, where you are ready with infrastructure, you are ready with data, and you are ready with your AI re-engineering of process. So when you do that, you will see the value per token actually increase, where an AI can actually become a colleague rather than just an assistant. And that is when companies can start to see, oh, now I can do much more of revenues, and I can save this much of cost. And that's when the return on investment makes more sense. I'd like to return to you, Julien, and return to this conversation about energy. We hosted a dinner in Bangkok a few weeks ago, and there everyone was talking about energy security. I think with the conflict in Iran, people had to find new sources of fossil fuels. There's talk about nuclear here in Vietnam. There's talk about renewables here in Vietnam. But, you know, what do you see as the most useful investments that can be made in energy to help? Infrastructure as well as kind of connectivity within Southeast Asia for energy. Yeah Okay, in order to prepare for the data center to be built, Vietnam is now investing a huge amount to develop the electricity grid and electricity generator. So we are not only focused on the traditional electricity, but wind, wind power, wind power, Energy, solar energy, nuclear energy, many other types of energy, and we're now building a very big nuclear energy plant in Ninh Thuan, but nuclear modern plants will be quicker, I think, because now we are using public investment for, like, establishing the energy plant, But actually, I think it is the best option. The option is that if you are investing in a data center, you can be active in finding or building the energy plant for your data center. So it is the way that Stavian is doing right now. I want to go back to you, Felix. And you talked about how you've expanded your footprint here in Southeast Asia. And one thing that's kind of been on my mind a lot is who captured a lot of this huge growth and value in the manufacturing chain? Right now, it seems like a few economies. are doing really well. There's Korea, obviously, Japan, the Taiwanese economy. Other places in the world, like Southeast Asia, have always struggled a bit to kind of capture more of that value, to move beyond just snapping things together. But kind of what's your take on this value chain discussion and kind of whether regions like Southeast Asia can start to grab more of that value for themselves? There's probably a nicer word than grab, but you know what I mean. Yeah, so right now, you know, still a lot of data centers For example, in North America, however, we see more and more resistance from the... For example, you probably heard in the news, not news anymore, but just a couple months ago, like New York and Texas, they already slowed down their data center building, right? All right, so. And right now we are also talking about sovereign data center. So we're going to see more and more data centers built in different continents or different spaces. Okay, so I think that's actually a very good opportunity for... not only Southeast Asia but for a lot of regions and for us like I mentioned earlier you know it's why we had built up so many factories in Southeast Asia that's because the ecosystem and also you know we have you know factories in Europe in North America but it's the main purpose is to just get closer to the customers okay to where the the site the implementation will be, right? So we don't have to worry about the logistics issues like the you know hurricane or volcanoes those kind of things or you know those political issues. It's good you mentioned the sovereign AI point because that's the topic I want to bring up to We have to vote for both Phuong and Julian. I'll start with you, Phuong. You know, again, everyone talks about sovereign AI. You know, everyone says, oh, we have to have our own foundational model that's trained on our, like, own cultural ticks. We need to have our own data centers. We need to have our own infrastructure. What's your stance on the sovereign AI question? Where do you think a country like Vietnam can get the most, you know, bang for their buck for, like, in the sovereign AI space? Sure. Sovereign AI is definitely a very important topic, but it's not. Thank you very much. I agree that it should be not for everyone. Next one. We talk about the return on investment, right? So there are maybe something more than the return on investment that is data privacy, that is maybe security, national security and data intelligence. So, in those kind of... Organizations have to deal with that. They should heavily invest in sovereign AI and Other than that, it's a bit too much for you to go to sovereignty because right now it's much cheaper and faster to go with service provider and we have a bunch of service providers out there. But unless your topics really touch in data security, data privacy, for example, banking and national security, then sovereignty becomes much more important. But another thing I would like to point out in... Social control alignment AI is a totally different topic. If people are using AI that trained by foreign nations and the young people nowadays, if they use AI to ask questions, everything about their decision making. Then there's a risk of misalignment, so when I do my research on the risk of AI in Mila in Canada, people pointed out that the misaligned culture of AI can be a social damage, and that can be a very important topic at the national level to think that if we don't have an AI to align with Vietnamese It means country aligned with Thailand country, aligned with Singapore, Korean country, then Our young people will slowly shift their decisions to a very much like Western or Chinese like country. So that is a longer term of risk, but however, we should start from now in order to control it. And Julian, how do you feel about the sovereign AI question? So when we talk about sovereign AI, we need to have our own infrastructure, our own data center, our own Vietnamese operators who will exploit the data which will be compliant with the Vietnamese regulations. So in order to do that, of course, actually, Vietnamese are now using AI models from other countries, but if we want to have a sovereign AI, we need to have our own models. And when we develop the models, we need to invest on it, and it will be aligning with our culture, our economy, and our institution, and it will also ensure the national security. And it is a very important topic at the national level. So we need, from the infrastructure, from the human, from the policy, we need to have autonomy on that. And everybody needs to, yeah, I don't think that it is. Also, Vietnam, but also other countries need to focus on. And I want to stick with you, Julian, for the next question. I mean, so the big player in this region is Malaysia. Malaysia is number four in global compute after the U.S., China, and the European Union. And I think having the European Union as third is cheating because it's lots of countries. But so Malaysia is number four. You know, how can a country like Vietnam kind of catch up? To somewhere like Malaysia and kind of what would be how you would sell Vietnam to someone who wants, who's looking for a data center project? In terms of Malaysia, Malaysia went, I think, five years ahead of Vietnam in terms of data center. And they have the global framework to call upon all the big players in the world. And Ho Chi Minh City is like the data hub in the Southeast Asia. And I think that in order to catch up with Malaysia, we need to build. And Ho Chi Minh City is a good choice. And as I mentioned earlier, we need to have enough electricity, we need to have enough international connectivity, and also we need to have technical capacity, and also we need to discuss with U.S. government to have GPU and other high technology components. So if we are positive and we are active, we can catch up Malaysia and Thailand if we can You know, Felix, I want to go back to you, and I want to stick on this kind of like sovereign AI point. You know, this is a technology where there's a lot of attention being paid on it right now. You know, a lot of governments are investing in this. A lot of governments are. Quite willing to invest money in these companies, particularly hardware companies. That's what it means. There's, again, there's also lots of concerns about, like, security, you know, security. You know, you're in this ecosystem, you know, how are these sorts of questions, including kind of the sovereign aspect of it, affecting your business decisions, investment decisions, even just how you plan what to do? Yeah, actually, a lot of things are depending, as I say, I mean, close to customers is very important. And so... Our customers in many industries, some of them, they do require, they have certain restrictions on how we use their data, how, where we store their data, okay, so it's not like, you know, you can say, well, we store our data on Google, we store our data on Microsoft, that would do it, it's not like that, so actually that's one of the things I'm also, I mean, I see that Maybe we'll have a smaller data center instead of, you know, the giga, giga size, mega size data center. Maybe we'll have the smaller data centers for a special industry or, you know, some customers. You want to distribute it. Exactly, right. So some of the customers say, well, I only want, this is a very special, we want your data stored in I mean, in this, let's say, in California, over here, or in China, or somewhere. Yeah, so I think that's definitely will affect our decision. In the time we have I'm going to make a radical pivot on this question. I'm going to ask Fong for your for your help on this. FPT has AI factories. I see this term all the time. Every time I read it I don't understand what it means. Can you explain to me what in the brief time we have what an AI factory is? Sure. In order to have the AI motor you need to make You need to give it the data, feeding it, so it's getting smarter and smarter. So AI factory is the terms that... Telling the process of making an AI motor the factory is to produce various type of AI motors that you can go large so that it can be intelligent as intelligent as possible or you can go deep into some specialized for example it's very good at healthcare but it doesn't good it's not good in for example manufacturing or it can be very good at manufacturing and that is not good in for example sport. So the factory of manufacturing different type of AI And then the process of making the AI work is kind of like a factory. So nowadays, there's another term called token factory. It's where you actually make these AI models work for yourself and produce the tokens. The tokens are associated with the values of the company, gaining the efforts of the company to scale. So token factory is a new term. Next to AI Factory. Okay, I will remember those for next time. AI Factory, producing token. Producing token, exactly. So, I want to thank our speakers for kind of talking us through this key technology. It feels like it's changing week by week. So, hopefully everything we say will still be relevant in two months' time when some other new thing will change everything. But once again, I'd like to thank Julian, Fong, Felix for joining us today. Please give them a round of applause.