"If you think about the one true leverage that China has against everyone, it is rare earths. The reason they are willing to consider doing it outside of China is not economic or resource driven; it is largely geopolitical. If that is a way of constraining the US, they would do it, which means you do not supply those rare earths to the US but instead align supply in China’s favor. It is not explicit in the sense that working with one partner excludes the US, but it is incentive driven, similar to how Belt and Road projects have been structured, by making cooperation financially attractive enough that partners choose alignment. China also retains a significant edge in processing technology that is both advanced and cost competitive." - James Chai, Visiting Fellow at ISEAS

"That is especially true for commodities like rare earths, where there is no clear hero to anchor the narrative. There is no Nvidia that becomes the face of the industry, so the story is harder to grasp and harder to popularize. At the same time, that creates a niche for those who truly understand rare earth technology. It requires deep knowledge of chemistry, because the supply chain is fundamentally chemical in nature, and that technical mastery is what ultimately sets players apart." - James Chai, Visiting Fellow at ISEAS

"The discussion now is whether we have reached a point where AI is already good enough for practical use. Countries that are not competing in the LLM race, where firms constantly release new benchmarks to outdo one another, have to ask what the end goal really is. That question directly affects demand for chips. If you want to compete at the frontier, firms assume a chip lasts about three years before it must be replaced with a more powerful one. But that does not mean discarded chips are worthless. Most users are not training models; they are running inference, embedding AI capabilities into everyday products like vacuum cleaners and refrigerators. For those use cases, existing chips remain highly valuable and continue to see strong demand." - James Chai, Visiting Fellow at ISEAS

James Chai, Visiting Fellow at ISEAS and former policy advisor to Malaysia’s Ministry of Economy, joins Jeremy Au to unpack how Malaysia is repositioning itself in an era defined by AI, semiconductors, and geopolitical rivalry. They explore the country’s shift from oil, gas, and plantations toward advanced manufacturing, examine how decades of semiconductor clustering built a quiet but durable export engine, and discuss why Malaysia is now doubling down on data centers and rare earths. The conversation covers US China competition over chip supply chains, the strategic importance of fabrication and GPU ecosystems, and how rare earth processing may represent the most underappreciated leverage point in the global tech stack. James also explains why execution, not ambition, will determine whether Malaysia can capture long term value from these emerging industries.

02:30 Malaysia balances growth with redistribution: The strategy is to raise high value industries like semiconductors and rare earths while lifting the bottom 40 percent through social protection.

05:42 Semiconductor strength came from decades of compounding: Intel and other multinationals anchored early manufacturing, and local engineers accumulated expertise that later spun into globally competitive firms.

10:18 Clusters beat subsidies alone: Tight networks of engineers, spin offs, and long term continuity allowed Malaysia’s chip ecosystem to survive volatility and keep upgrading.

21:05 China uses constraint as strategy: By limiting access to high end Nvidia GPUs, Beijing forces domestic firms to innovate faster and close critical design gaps.

29:45 Chips are not oil: Frontier GPUs power model training, but most real world AI use relies on inference, meaning older chips retain value longer than markets assume.

37:22 Data centers create investment headlines but unclear spillovers: Billions flow into Malaysia, yet long term value depends on whether local firms capture supply chain and technology capabilities.

44:10 Rare earth processing is the real choke point: Deposits are global, but China controls the complex multi step processing chain, making chemistry and technology control more strategic than mining alone.

Jeremy Au: Hey James, so excited to have you on the show. I'm a big admirer of your writing across multiple media outlets—SCMP, you know, Singapore media—about geopolitics, about Malaysia, about rare earths, and so excited to have you to talk about these topics.

James Chai: No, thanks so much for having me, Jeremy.

Jeremy Au: Yeah. So could you introduce yourself?

James Chai: So, the last role that I held was a policy advisor to the Ministry of Economy. I've always been in tech, so I've worked in public listed companies in Australia, UK, covering along the whole supply chain of AI. Last time it used to be a lot of big data analytics stuff. So I was a co-founder for one startup, and then I was also involved in semiconductors. Now I'm doing a lot of research. Visiting fellow for ISEAS, and I have a couple of fellowships that are also coming up. So basically a lot of tech policy as well as research and writing.

Jeremy Au: Fantastic. And you know, I think what we wanna talk about for this topic is really about Malaysia's overall economic strategy, which I think is a little bit overlooked. I think people look at ASEAN or Southeast Asia as like Singapore, Vietnam. So we talk about Malaysia's economic strategy at a high level, but also talking about the three major pillars that we discussed, which is I think one of course is semiconductors and historical strengths there; two of course is rare earths; and then third is the data center industry, which is I think related to all of that. But, you know, I think let's kind of get started. Let's talk, what is the Malaysian economic strategy? Is there one? What is it?

James Chai: Yeah. So it is not always coherent because I think the natural state of government is that people want different things, and typically they're not necessarily in one coherent strategy. But if you think about what Malaysia is trying to do, the most classic way of thinking about it is to "raise the ceiling," they call it, as well as "lift the floor." So that means that can I create enough growth items in the economy so that I can grow out of my oil and gas dependence, right? Do less mining and more manufacturing. So anything that is high growth and high value, can I focus on those? Right? So, the three that you mentioned just now would be semiconductor, rare earths, as well as any sort of like AI related technologies, but also energy transition. Those are the core sort of growth items that the country is trying to push for. But in terms of like lifting the floor, it is a lot of like social protection stuff, creating also a country that is fair enough for what they call the bottom 40% of the people. And that requires quite significant cash transfers, making sure that they make ends meet. So how do you break out of that middle income trap without creating too much inequality are some of the thinkings that they have. But like most Southeast Asian countries, because you are export reliant and because the trade situation is very volatile now, it's about finding areas that you can still grow that is within your niche. Hopefully it is an existing strength that you traditionally has held and whether you can double down on that. So there's a very strategic way of how they look at where the future lies for them. But of course the challenge in Malaysia, it's always can you execute as well as how we plan?

Jeremy Au: Yeah. And I think what's interesting is that, you know, this also splits out into like as a result what Malaysia historically has been versus the future of what Malaysia's economy could be, right? And I think for the history lessons, like, you know, like Singapore, Malaysia used to be run under the British, right? So it is integrated in terms of—I'm just thinking about like plantations, tin, agriculture, obviously trade. So I think that was the historical colonial economy. And then I think obviously, since Singapore and Malaysia split up in 1965, Singapore has gone for the finance, free trade, high value chain. And I would say that I think maybe there's a perception, correct me if I'm wrong, as well as like Malaysia has historically focused obviously on plantation, oil and gas, semiconductor, and also I think like manufacturing as well. Yeah. So, but is that a good summary of the historical angle of Malaysia? Anything you wanna add?

James Chai: Yeah, no I think that's fair. So, oil and gas, palm oil, rubber, these things has been the foundation that a lot of industrializing economy tries to do. And then around the seventies and eighties, that's where the huge push for manufacturing actually came up. And that's where, you know, you have the birth of the semiconductor industries that's primarily driven by a lot of companies like Intel, National Semiconductor, all of those early days where what Malaysia does is really a lot of the backend stuff. Not to say super easy, but definitely lower value than the rest of the chain. But it's something that they've done pretty well for themselves. So, for instance Inari, one of the biggest backend semiconductor manufacturer in Malaysia, they serve primarily Apple products. So a lot of the iPhones, if it does very well, then they would have more orders and so on. Right? And that has been quite an important MO because the thing about it is that there are some areas here that it is important to accumulate the years. So you see countries like Japan wanting to get back to it, Vietnam wanting to go back into semiconductors, and Thailand wanting to start something, right? It is actually not that easy to compete even if you start at the lower value chain, because you do need the number of years to learn the process. And because semiconductor is so precise in what it does, it is actually very difficult to just throw money and people and expect something would happen, right? So, as a whole, you're right to summarize it, that there was this plantation and then a shift to manufacturing, although they didn't quite abandon the palm oil and rubber one, but it did go very big time into oil and gas. So the expertise that they've developed there is also very profound. It's just, it's unfortunate that oil and gas is finite and there's always this moving timeline to how much reserves they have left. But it is definitely a concern and the future won't be powered that much by oil and gas anymore. And therefore, they also need to prepare an alternative for themselves, and that requires them to get into new areas. But of course, the challenge in new areas is that to kickstart itself is not actually that easy, right? Are there areas that looks new in terms of our focus, but there are some existing base in which you can compete in, right? And I think the world that we live in now, the truth is that AI is primarily a lot of the hard stuff. Right? It is the hardware, there's the physical nature of your capital, which means your land, your water, your electricity that really contributes to it. And I think Southeast Asia really has an edge because of that, right? So I think we're probably gonna take a nice little s-curve, show all those topics.

Jeremy Au: So since we're talking about semiconductors and data centers, I think they're quite adjacent. But yeah, let's talk about these semiconductors, because I think most people are kind of surprised that Malaysia has such a strength in semiconductors. I mean, frankly, even probably say growing up, I don't think that's something I was aware of. Yeah. And I think it's become actually even much more important with the current geopolitical tensions and concerns about, you know, the security of supply from Taiwan around semiconductors. And I think it was also interesting was like the Intel CEO is Malaysian. The Broadcom CEO as well. So, one of the biggest semiconductor companies in the US is also Malaysian. Yeah. So I think that's something that has only become very prominent over the past five years, from Penang. Yeah. So how did Malaysia actually accumulate this, like you said, edge, years, experience? How did that even come about?

James Chai: So, a lot of it comes from this classic model that FDIs by all of these MNC companies are meant to make you learn something. So the very commonly assumed thing is that when they come in and naturally by observing them doing something you would learn and subsequently you'll be able to do it yourself and therefore you reduce your dependency on Intel. Right? But the truth is that process is always very imperfect because by nature of competition, they would not want you to learn entirely so that you replace them, right? There's still some degree of dependency that they wanna hold. However, because Intel and all of these big semiconductor companies like Micron has been in Malaysia for almost 50, 60 years, so surely you have learned something from it. And the truth is that there's the design element in manufacturing as well, right? So design is more higher value compared to the backend manufacturing. So the Intel Design Center was actually very prominent that they were able to eventually get to a point because of hard work as well as you know, the ability to compete and hunger at that time by a lot of these engineers. They were able to design a lot of Intel's global products, right? But of course at some point working at Intel will only get you so far because you're still an employee, right? Therefore there was a group of people that actually broke out of it and started their own thing, right? That's where it started actually building those expertise. But the other truth is that other than this local engineers that actually broke out of it, there were many engineers who worked for Intel that eventually was transferred to the US and as you know, geography is destiny. You allow yourself to be exposed to even a wider range of possibilities. And that's where Lip-Bu Tan as well as Hock Tan from Broadcom separately, they were able to combine the good fortune of being in the right place, exploited it with their business genius, as well as the diligence of constantly trying. I think that has produced an outcome whereby if you look at the largest semiconductor companies, Malaysia as a small population has an outsized representation of it, right? But the other truth is also that a lot of these Penang engineers, they are very frugal. And therefore they are your classic business person who wanna always lower cost and just heighten your profit opportunity. And I think that has served them quite well in terms of having a business model that could withstand all kinds of volatility. And I think that's why it came probably as a surprise to us that they've been quietly building this behind the scenes, but it's actually accounting for nearly 30, 40% of Malaysia's exports. So, Malaysia has always known it to be a strength, but I think manufacturing has never been appealing, right? Because it is factories that they're working in. But I think the world has shifted in such a way that we recognize, once again, the hard physical stuff as being more important now in the era of AI.

Jeremy Au: Yeah. And I think what's interesting is that there's a objective dynamic and a relative dynamic, right? I think the objective dynamic is, like you said, you know, they've been accumulating this expertise over time, agglomeration of corporations and expertise and talent to do that. I think it makes total sense. And I guess with the geopolitical tensions and the instability or perceived instability, that supply has made Malaysia more prominent as a piece. Is this that interesting? 'Cause relatively at that time there were many also other semiconductor manufacturing going on. I think there was Thailand, it was going on. Obviously Philippines in the 1950s and sixties was doing very well in terms of their manufacturing. And there was also Singapore as well. Was doing semiconductor as part of that Singapore, Malaysia nexus. And I think Singapore today still has a semiconductor GlobalFoundries presence, although obviously not to the same density I would say as Penang, right? Yeah. So I'm just kind of curious, do you have a point of view on like this relative dynamic, like how Malaysia ended up being like, I'll say the Duke or King, I guess, of Southeast Asia semiconductor?

James Chai: So if you look at the semiconductor history of a lot of how these companies have started, they're typically a few dudes—and they're typically dudes. It is a highly sort of like male dominated industry. They have always been a few people who had chemistry of working together, probably like six to 10 people, and they have a larger dream that they wanna fulfill. Right? So there's some degree of alchemy in terms of whether you have that condition, right? Governments have a role in this, but they do a very facilitative role. I can give you subsidies to attract better MNCs so you're exposed to better knowledge and better equipments. I can give you subsidies to start your company so that your offering costs isn't that high. I can do training facilities so that you can train more people to work for you, but I can never do it myself. There's some degree of needing to cluster the right group of people, but it's not entirely within the government's control, right? So the government normally starts that process. It was by the Penang chief minister at that time that he thought that we needed an area that has a high probability of winning in manufacturing, and therefore he thought that, okay, it's worth doing. It's the same story in Taiwan when it started, right? They felt like we needed something to be proud of, and therefore there's some degree of national identity that stuck to it. And if you are lucky, you would have that group of people who are diligent enough, but also with some degree of ambition that you want to make it big. And most of the time this is motivated by profit, but after that, to sustain it long enough, that's where you would need a thinking of diversification, which is like, how do I continuously deliver profits like that against the volatility that I would foresee? Because actually semiconductor is a highly volatile industry and most of the time when it hits the bottom pit, it wipes out a lot of the companies. So being able to survive is a huge skill. Right? And Malaysian semiconductor companies have been very good in diversifying their client base and trying to get more and more stable clients like Apple, right? Where you know that they would always win, right? But even among that process, it's not purely because you have like business acumen or intelligence that are greater than other people. There is some degree of the clustering effect extending as well. So for instance Hock Tan, who is the Malaysian of Broadcom at that time, was the one that actually worked together with the Malaysian company so that they actually have the Apple contract. So there are effects of clustering that could extend much further, right? So it is clusters of geographically tight knit group of people as well as clusters of people who roughly think the same and want the same thing, right? So I think it's also because that this is a continuous unbroken process that made them somewhat emerge as a winner compared to instances as, let's say Vietnam, right? As a close example. If there were periods of time where it just hollows out all of a sudden, which means that the cluster of people disintegrates or like government no longer supports it at times of maybe when it's going down and needs the extra boost, then you would lose many years and restarting, you probably need twice the amount of effort. And that's what we're seeing in Japan, right? Whereby they want to reclaim their position as one of the biggest semiconductor players, but they are spending probably a few times more than they would've spent if they had stayed in the game a bit longer. So I would say that those reasons actually contributed to it.

Jeremy Au: Yeah, I think, you know, another cohost of ours, Shiyan Koh, she's a general partner for Hustle Fund. We previously talked about how Singapore could also have built that Silicon Shield because it was very much at the forefront. But I think, you know, it's understandable why Singapore, from a government industry policy basically said, you know, this Silicon actually takes a lot of water, there's a lot of electricity, it takes a lot of space. Yeah. Oh, I wish that Singapore doesn't already have a lot of, you know, in terms of like targeting self-sufficiency. And so I think Singapore decided to focus more on like aviation engineering, focusing more on obviously kind of like water self-reliance, electricity self-reliance. Yeah. And I mean, it's still a Silicon presence, but obviously is this trying to be differentiated, right?

James Chai: That's right. So I think it's interesting. And so, you know, kind of like circling back to this semiconductor space, I think one interesting part is obviously like what does the future hold for semiconductors, right? Because you know why it seems like in the future is, we know that Taiwan, TSMC is like number one for semiconductors. They're obviously the most advanced, the most economies of scale and the largest fabs in terms of throughput. So that's a known factor. Obviously America is trying to like bring some of that talent over. I think Taiwan has agreed, yeah, 250 billion. Yeah. That's a significant sum and obviously a lot of talent to flow to, I guess, Texas to build TSMC in the US. I'll say that they say Japan is setting the bill. So, and obviously China is, I think is interesting because, you know, Biden banned chips exports to especially like the GPU chips to China. And then now Trump has reversed that, and now was letting it in, but China's not really—China don't want it. Yeah. So, but, so what's going on here? What do you think is going on with this situation?

James Chai: So, okay. There are two parts to that question. One is the US-China situation. As in maybe the question that I'm thinking about is who is likely to emerge the front runner, right? And specifically US and China, how is that gonna play out? And the second one is that, let's say if countries have semiconductors, how should they focus on where they should win, right? Your classic play of like, what should you do in your present situation in terms of industrial strategy? Now I think the US-China one, the truth is, okay, as a very simplified way of looking at this, there are probably a few different parts in semiconductor supply chain that matters the most, and what I mean matters the most, it refers to the fact that it is a choke point that is very hard to replace. Right? So there is this design portion that doesn't have a very clear winner, but Broadcom, Qualcomm are some of the biggest companies whereby they almost have a portfolio concept in terms of how they run that design firm. But a lot of countries are able to design for very little things. So for instance, you can even design chips for aircon remotes that are lower value, but you still have a certain size of market, right? So being able to design for EV chips or let's say autonomous vehicles, iPhones, laptops, those things are really top tier stuff. So a lot of people wanna get into that space, right? So design is one piece. The second piece is the fabs that you mentioned, which is TSMC. Typically think of this as after you've designed something, how do you actually manufacture the little chips out so that you can actually assemble them? This process is of course extremely difficult, right? That's why there are only a few companies in the world that's able to do it at all. The ones that are able to do it well is actually only TSMC, right? That's why the world is overly relying on them. And then there's the assembly part, which is the backend—after you've had it, you have to test it and assemble. That's the part that Malaysia does very well. And then lastly, it'll be to fit it into whatever product. So if you're gonna fit it into your data center racks, that's what you do. Fit it into your cars, that's what you do. So they're more fragmented systems level players for that. Right?

Now when it comes to the US-China situation, what the world is thinking about is whether China will be able to have a supply chain that is self-sufficient, which means from start to end, it is all Chinese companies. And for the ones that I haven't mentioned, it is easier for them to already have it, which means that different parts, different equipments, and so on, right? But there are certain things that are still very difficult, right? China can invent and manufacture almost anything around the world better, faster, cheaper now with their clustering effect all across the country. But semiconductor is the really one thing that they will still find it very difficult to do, right? So other than this TSMC reliance, they have a fab called SMIC, right, which the US has actually red listed. SMIC is nowhere near where TSMC is. The estimate is around three to five years off, right? And these things, the expertise that you have compounds over time, which means that the distance may actually grow wider if you slip one or two years, right? That one is very hard to replace, although rumors are that they are catching up faster than what people imagine.

Now, design companies—recently, there's this huge wave of IPOs in Hong Kong. Right? That's where they're really starting to do more design of GPU chips. Now, the biggest global leader here is of course, Nvidia, right? And very few companies around the world could claim that they have the full stack whereby I designed the chip and ultimately people who uses my GPU have access to this software called CUDA, right? Very few companies around the world can even get close, but the closest now currently, eyes are on the Chinese companies, right? So potentially, but they probably still need a bit of time, right? And the rest of the supply chain, it is very likely that they will be able to close the gap. So the ones that I've mentioned, like the fab, as well as the design, as well as the software that's related to the design, it's probably the hardest to do.

Then the question is, if China doesn't have it, why would they actually stop taking in Nvidia chips? Wouldn't they need it? And the truth is that they do—Alibaba, TikTok, and so on, they really need these things in order for them to still be at the frontier and compete with other LLMs like OpenAI as well as Anthropic and so on. Right? But the fact that they wanna limit the H200 chips coming in shows that the Chinese government believes that forcing your way with constraints would be a good way of actually catching up faster. Yeah. So they're now dedicating the largest ever resources in semiconductors, anywhere in history, anywhere around the world, even more than the CHIPS Act by Joe Biden. They have that one. They have unlimited talent as well. Unlimited hunger to do it. Exit options in Hong Kong are all available. So the real task for history and society to look at is: if you have unlimited resources and talent, can you defy the nature of time? Which means that semiconductors normally need a long time for you to build expertise regardless how rich you are. But if I have unlimited resources, can that change the equation or not, whereby I can defy some physics of time, right? So that's what we are looking at.

Now the second question is on what should countries that are perhaps outside US and China think about, right? Like countries like Malaysia, Vietnam, Thailand, what do they do? I think normally the strategy is twofold. You double down on what the existing strengths are to make sure that you don't lose your bread and butter, and you try to expand into new areas that you don't currently have a mold on. So for Malaysia, that means that they need to make sure that their backend is more and more competitive. Not only do you constantly get in better and better clients, you need to also improve your own technology in backend, even though it's backend and not be complacent about it, right? So, Vietnam has to do the same, which means that Vietnam is slightly lower in the chain, but they need to still continuously get clients there, right? Only then you can survive. But the next area that Malaysia and similar countries are thinking about is that they're trying to get into design, right? Which is higher value, lower CAPEX, right? Design is not factory work. It is literally laptop work. It's just, it's very complicated, right? So they're trying to get into that market so that they can have higher margin work and higher value globally. So the whole idea is that I will be able to make better profit, hire better people for better wages if I do that strategy. So every country now has to think about: how do I win at what I'm good at? And the area that I'm getting into—is it not a significant resource constraint for me that I'm able to still compete even though I'm a newcomer? Right? Only countries like Japan can still put huge amount of resources, even though they've been lacking behind for a few years and still be a formidable enough competitor. Most countries around the world don't have that kind of firepower. Yeah.

Jeremy Au: And also, for example, I don't think Singapore is gonna put the firepower behind it. So I think there's a massive like, investment push by the US to push in America, China to push for it. Obviously Taiwan continues to invest, Japan's investing, and to some extent Malaysia's investing. Yeah. So to me it almost feels like there's gonna be a glut of semiconductors, you know, in about five years time, once everybody reaches domestic sufficiency. But then these fabs are not like, you know, small fabs, right? These are, you know, large, huge economies of scale. To me at least it feels like there's gonna be a glut of semiconductors once everybody makes it click.

James Chai: So the growth model that if you look at all of the CS conferences with Nvidia and so on, what the AI folks are thinking about—and what I mean by AI, it means the LLM folks, the hardware chip folks, as well as the data center folks, right? The reason they justify these astronomical valuations is because they think the future will have tremendous demand for chips and AI that is still currently not catered for. So the imagination here is that if every household has a robot that helps you, that's a demand that doesn't currently exist yet, but you need tons of chips to be able to do that, right? And the little things, which means that things we already have, let's say, an electric vehicle—in the future, if electric vehicles get better and better, they would actually require more chips. So for instance, I dunno whether you realize if you have your handle of an EV, right? There's already two or three chips there, and these things never existed in an ICE vehicle before. Right? So that's what they're actually betting on, which is the world doesn't actually have enough chips, right? And therefore a probability of glut they feel is not actually that high yet.

Now the other equation is that semiconductor is also a very brutal industry, right? Like, let's say we assume that there will be a glut and the demand is only at a certain range and is fixed. Companies have to compete for it. And the companies that don't actually make the cut, which means that clients don't actually adopt their technology, and if they stay clientless for a while, they actually go down. So semiconductor companies are very brutal and volatile in that sense, right? So whether or not we are optimistic of that future depends on how we interpret it. But you are right in saying that no matter what happens in terms of whether the demand is fixed or the demand goes up and we haven't catered for it, there are certain bottlenecks that cannot really scale up as fast as people like. And one of it's TSMC. Can they really build that many factories to keep up with the demand that's coming in? Currently, they're not, and that's why they're overvalued, right? In the sense that people still feel like there's so much that's unexploited here. But at the same time, because they're constrained, their valuations even though are high, are not as high as Nvidia, not as high as your other AI companies because they're constrained in that physical sense.

Jeremy Au: Yeah. And I think there's also a geopolitical security risk there where people feel like TSMC, the Taiwan plants may not be a going concern as well, and there's precedent there as well. But I think what's interesting is that obviously we're talking about, you know, kind of like simple fact, which is like about supply and demand, right? And I think we all agree that the supply of chips is inelastic in a sense that people cannot switch off their supply. And with everybody investing, there's gonna be even more supply. But I can't really switch off these factories because the economies of scale are so brutal that we can't switch off the factory because it's gonna be cheaper than to let it run. Yeah. And so I think there's some similarity as well to, I think the oil industry where it makes no sense to stop production of oil. 'Cause as long as you're pumping, you just pump it out.

But I think what's interesting is the demand side is different, right? Because I think for oil, what's interesting is that once you burn the oil, it's gone. Right? And so to some extent, obviously that's up and down left and right. You know, it's like what, a hundred dollars per barrel? $1,000. Like there's all kinds of—you know, at one point it was free oil per barrel, right? You know? I know. So it was an interesting thing there. But I think what's interesting about chips is that I think obviously there's this theoretical piece of infinite compute, which we'll talk about data centers later. But I think my contrarian point of view is that, you know, chips don't disappear, which is if I built 20 data centers and for example supply is going up, constant. But the truth is if I stop demand and stop buying new chips, my data centers will still run. Right? And I think there's a little bit of an artificial—I know what we're talking about—there's a bit of fearmongering from my perspective because what they're saying is that, oh, these chips will only be valuable for five years, and the models need better and better chips and therefore these chips get outmodeled so quickly. And so, it's totally worthless after five years. And I have a little bit of like, question mark here because I'm like, I get it from a company perspective, that you wanna depreciate capital expenditure for five years or 10 years. Okay, I get it right? But the difference is like if you burn a barrel of oil, it's gone. Yeah. You switch on your aircon, the oil's gone, right? But if you say like, hey you know, there's a useful life of a chip longer than 10 years, yeah you can keep it running. And I think there's a little bit of, like, from my perspective, a little bit of false logic, which is because models must compete each other to death, therefore chips must compete each other to death. Therefore chips become outmodeled.

James Chai: No, I think that's really interesting because the debate currently has shifted a little bit simply because of what you were talking about, which is when AI first broke out, which is really in the form of ChatGPT, right? It was very important to get better and better performances because the comparison is no AI. Yeah. And therefore you really be left out without AI capabilities if you don't have chips that are strong enough to keep training for the next version. The discussion now is whether we have reached a point whereby it is decent enough for usage that countries that are not in the race—which is the LLM race, which is you constantly release different benchmarks and see who's outcompeting the other—to what end? Right? It is the question and then that affects the question of your chips. Right? Which is like, it is true that if you wanna compete on the frontier model, three to five years—three years only is how long they think they can keep the chip before they have to throw it out so that they have to put another more powerful chip in order to compete.

But it doesn't mean that the chip that you throw out is actually worthless, because if you can train actually quite a good AI usage now—and most people around the world, they're not training models. They're more like just doing inference, which is like from the model that I have, how do I have all of these AI-driven capabilities in the things that I use, right? In a vacuum cleaner, in a fridge, and so on. And therefore there is still high usage for that, right? And at some point the world needs to ask themselves whether it is important that even my fridge keeps getting better and better performance and therefore I have to throw out the chip after two years and buy a new one. Right? There are certain things that don't have that kind of lifetime. Right? And or don't have that kind of very short lifetime and therefore they can stay longer. Right? So you are right in a sense that a lot of these physical assets, especially when they're powerful enough at this stage, might actually be of great use around the world, right? For that kind of diffusion of AI. But the more frontier one will always compete for it.

And the question for the market to really decide is that I can think about semiconductors in two ways: one, the GPU level grade, that is really meant for training the most competitive frontier models; and the "good enough" grades, which is almost everything else, right? And whether that top tier one will continuously grow in the way that we've seen today. Right? Especially when the investment of that GPU has no longer a commensurate increase in performance of your frontier models. Yeah. So in other words, if your frontier model slows down, even though I keep pumping a lot of chips inside, then it defeats the purpose of me pumping that much chips. That's the whole very crude method of the scaling law, which is the scaling law essentially believes that I can get much better performance if I just stuff in more chips, more powerful chips in data centers, in more data centers. And what DeepSeek has really shown one or two years ago is that actually you don't necessarily need that method. You can innovate on your algorithm that gets you better performance while using fewer chips because China had fewer access to chips and therefore they were able to show that actually there's another side to the equation that might affect how we actually look at the trajectory of the chip supply.

Jeremy Au: And I think that's interesting because it kind of like actually touches upon I think a two-by-two from my perspective. Right? I think one axis is basically frontier model, which is about performance versus power efficiency, I think is one axis. Another axis is quite important actually, is open source versus closed source. Yeah. And it's been quite interesting to see like China perhaps push power efficiency because of the lack of chips versus obviously America focusing on frontier models. Of course what's interesting is that China, you know, advancing open source, and then I would say the US primarily focused on closed source. So I think it's just an interesting...

James Chai: Yeah, no actually for that one, right Jeremy? It's super interesting because actually it uncovers a few things. One of it, it uncovers this whole philosophy of AI question. The second one, it uncovers a very technical question as to how this would actually develop in terms of the progress of tech. In terms of philosophy of AI, I think what is interesting here is the vision of the future. So a lot of these AI companies like Llama, Gemini, ChatGPT, there was a time where they were contemplating doing open source. But now all of them has actually rolled back, which means that open source in the market are not the most frontier ones. The most frontier ones are all kept within. The fundamental logic here is that, of course, if these things are really valuable and this would actually help differentiate me from the others, I don't want other people to know, and therefore I wanna maximize profit for ourselves, right? So it's a huge revenue and profit motif.

But the philosophy on the other hand of China is that technology, especially of AI, should be actually widely dispersed as possible. Right? Because if you really think of these things as fundamental infrastructure, everyone should be able to have it. And the whole goal is to develop enough of these foundational models that are competent enough that people will be able to build companies of their own using that foundational model, right? So it represents very different visions of the future, even though ironically, a lot of this open source thinking of AI actually started from the western world, right? Now the second one, the technical question here, is that will an open source foundational model like China allow it to actually compensate for what it doesn't have? So for instance, if they're constrained in terms of chip supplies, as well as some degree of talent, right? Because the ones in US are still very strong. Using an open source model, does it help close some of their gaps? 'Cause essentially what they're doing, right, is that if I open source it, there will be more developers who will be able to improve on it, and there's this communal effect in terms of how well this model eventually becomes, right? So, it is to be seen. But it is very interesting that it's only few months ago that China's models, the open source models, has really overtaken the US, which means that most developers are using Chinese based foundation models, right? Talent is a very big piece in how far your AI would go, right? And China is really tapping into the world's resources of talents to try to get there. So I think that will be interesting to observe.

Jeremy Au: I agree with you because it feels like, you know, we can look at Linux versus Windows versus a Mac, right? And to some extent, Mac obviously was closed source but then it really succeeded with app stores and bringing the developers into the walled garden. But at least having that community—frankly the Windows phone failed because they were never able to bring in the developers or co-develop. But really the Windows, the laptop system has done well because, you know, they have enough users to make it worthwhile for developers to do. Yeah. So that's quite interesting to see that play out. And obviously Linux has open source, but you know, the people who make money are the people like running hardware servers based on Linux or servicing and maintenance, but not necessarily the core software itself. But I think what's interesting is, you know, I think it kind of goes in the next piece, which is really about data centers, right? And you know what's interesting about Malaysia has been like, I think a big push on building data centers. I think Singapore's the opposite in a moratorium is said, okay, yeah, we don't have enough space, electricity or water to run data centers. Yeah. So it was a moratorium, I think it just got released, and then I think they probably announced a very small quantum of data centers this year in Singapore, but Malaysia seems to be going like full steam ahead. Can you talk a little bit more about that data center strategy?

James Chai: So the data center strategy is at once very irresistible to a lot of countries, but at the same time, it also comes with a huge amount of uncertainty and risk that countries that eventually host them has to think about. Now, the initial sell was that a lot of these companies that invest their data centers in Malaysia, or a lot of these middle states, they're very big names, and the quantum of investment is so huge that no country would really think twice about accepting them. So you imagine all of the big tech companies—AWS, Google, Microsoft, as well as Oracle and TikTok—has actually invested in Malaysia. These quantities are all in the tens of billions or close to tens of billions, and that would be a really good headline item. There is FDI numbers that you can register, and then the real estate folks as well, the construction companies would actually benefit from it. And there is this whole AI story that they can sell, which is if you host a lot of data centers, is a very important piece of AI and that's why Malaysia has always been considered this quote-unquote "AI hub."

But the reality to that is how much of the value of that data center is captured within the country that hosts them? If, let's say, most of your data centers are actually hosting overseas workload, then in which part of the supply chain do you gain, especially in an AI sense right now? Of course, we can accept that AI has both the hardware portion, which is you start from rare earth right up to the semiconductor chips and then data centers, and then you go into the software portion, which is your LLMs, and subsequently the application that you can actually build on it. None of this has really transpired in the post-hardware portion, whereas we are seeing most of the concentration of where it is growing is really at the hardware area. Right? That's not to say that it is a bad thing for Malaysia, but it comes with a lot of new considerations as to what they mean, right? So does it mean that for Global South/middle state countries, the strategy of AI is really to play at the manufacturing and the hardware portion? Because it's almost impossible for you to develop an LLM that is globally competitive, right? Because it's monopolistic in nature. It is also to be seen whether you can build softwares on those foundational models, let's say for the legal industry, for the finance industry, for the healthcare industry that has respective use cases that will be very helpful for your country itself, right? The same kind of considerations during the startup days also apply here, which is you're essentially trying to build a software company that needs scale and countries that are small always have that kind of limitation, unless you cater to Southeast Asia as a whole, right?

So the most realistic discussion that's currently being held is that at the hardware portion, how much can I gain from having all of these data centers over, right? And I think this is an area that hasn't really been thought through so much, which is like, do you have any backward linkage companies? Which means that if I build this data center, I'm not just building the cement as well as the concrete around it, I'm also actually building the stuff inside. But how much of the stuff inside are you actually building on your own? Because if there are so many that you've built, due to the data center influx coming in, you can say that you've built some kind of expertise in being able to be part of a data center supply chain. And maybe then there will be an industry that's built from it as a result. Yeah. And a lot of the data centers also need a lot of like cooling technology and so on. Would you be able to learn and therefore develop on your own? Right? So all of these are the questions, but people can view this as an opportunity, but to realize that opportunity into some kind of capturable value I think is the next step.

Jeremy Au: Yeah. And I think for the Singapore civil servants, I think that assessment is Singapore can't do it, right? Because it doesn't have the land, it doesn't have the water. 'Cause Singapore—their focus is self-sufficiency for sure. 'Cause they can't count on the Malaysia water supply. And you know, there's only so many desalination that you can do and the water reservoirs, so can't do the water part. And electricity is so handicapped because Singapore is a net-zero kind of pledge, which, you know, fundamentally restructures the cost curve or electricity, right? Yeah. So putting that on one side here. So I think Singapore is like tapped out. Okay. So, now I think to some extent if you look at Malaysia, I think it kind of makes sense 'cause you have semiconductors and then you have data centers. Imagine you can build a chain, right? End to end, right? 'Cause you have one starting end, you have the end point. And then you build everything between like, you know, cooling technologies are being used at both semiconductor as well as data centers. You know, I was just hearing recently that data center electricians in America are making a lot of money. Yeah. Because, you know, that's a lot of cabling and they have very high standard lists. They're willing to pay if everything is correct. And you know, it's not like they're going house to house, right?

James Chai: Yes, that's right. The skill is very different.

Jeremy Au: The skill's very different—this is a higher bar. Right? And they need experience. They don't need somebody who's new. So I think there's an interest to see that supply chain both for jobs and I think some transferable skill sets. Yeah. And lastly obviously another pillar that we talked about at the highlight was also talking about rare earths, which I think was a surprise. I mean, I think one is obviously, I like to say that rare earths is not that rare. Yeah. It's the lots of pieces to do it, plus the processing of it is hard.

James Chai: That's right.

Jeremy Au: So we kind of know that, which is why China is willing to do all the processing and do all the state support needed to swallow the licensing costs, the environmental cost of that. But I think what was surprising to me was that Malaysia positioned itself in the rare earths component. Yeah. I mean, obviously America's obviously pushing harder there as well, but talk a little bit more about Malaysia.

James Chai: Yeah. So, I think since the point of doing some kind of feasibility studies, Malaysia is surprised that they have more resources that they previously haven't accounted for. And the rare earth deposits, as claimed, is one of the highest in the world, right? But of course, this is as claimed, right? And it is also like having deposits doesn't necessarily translate to having good uses of your rare earths, right? Because there are certain kinds of compounds and elements in this chemistry of rare earths that are more important than the others, right? So the whole goal is to see whether things that are naturally endowed to you, whether you can actually use it to the best use or not. And the truth is that most of the deposits and technology are still in China, which means that they have significant outsize leverage to the point that could be the breaking point as to, so US strong-arming China, right? Now for the rest of the world that has deposits—let's say Brazil, actually Greenland also has deposits; they experience one of the reasons why Trump is so keen; Vietnam has and of course Malaysia, I think some in Indonesia—but the question for all of these countries is that they will never be able to make use of it if they don't have the processing technology to separate all of these elements and subsequently the manufacturing capabilities of converting this to magnets. Right?

So the reason why rare earths is so, so, so valuable, not only is it ubiquitous in technology like EVs as well as a lot of heating technology, is which primarily is a very good heat resistor. Which means that a lot of the huge workloads that you run on something don't heat up that fast if you have rare earth elements as part of the magnet that makes it, right? So it becomes therefore a huge growth area if you're able to exploit it. But it's also an area... now when we talk about processing, right? It is actually 100 plus steps in processing. They're all tightly guarded by patents as well as trade secrets, and therefore it's something that China recognizes would be a significant leverage that they hold if they don't actually allow this technology anywhere else. Now, the reason why I wrote the article was because there was a huge sort of like thing that people overlooked: when China actually went to Malaysia, they actually privately told the Environment Minister that they are willing to share some of these technologies. Right? Now, how much of that is just talk? How much of that is just forcing alignment from the US? How much of it is genuine remains to be seen. Right? That kind of signaling tells you that if, let's say, China is willing to share it with select partners in maybe a very close-look kind of system, it would make a huge difference in terms of where the actual rare earths around the world can be exploited. So I think it's not rare in the sense of not many people around the world have deposits, but it's rare in the sense that very few countries can actually exploit it for it to be productive uses. Right? So I think that's sort of its effect on the technology supply chain. But of course the real question also is that outside of China, even if China shares with you that technology, how much can you ensure that there is no environmental effects of the wastage afterwards? And very few countries can be confident enough to say that there's none. Yeah. Right? So I think that's the other question that would arise even if China is willing to share that technology.

Jeremy Au: Yeah. And I think it is really interesting because we're talking about diversification as a national security requirement, but we're talking about the steady state supply, demand, price, or profitability of this process. Talking also about the technology licensing or control component. Right? And I think, you know, my point of view is that based on what you shared, which is news to me actually, is that yeah, I mean if China had a choice, it would never export its processing technology because it's a big source of control. Yeah. And it's very hard for people to learn. Yeah. So why export it when you have plenty of land and industry to do it? So I don't think China would want to do it, but if Malaysia is definitely gonna do it, and they have a choice of either an American technology or Western affiliated partner to do that processing versus a Chinese affiliated processing partner, now of course, the Chinese would rather be the Chinese processing partner doing that processing and having a Chinese JV and having control over that process, rather than having the, you know, the West side. So I think where Malaysia goes with that, I don't know. Yeah. But I think there's something for our futures we find out. But for me, I think what's interesting is that, you know, I was reading this article by the Economist. It was talking about how Japan had actually suffered the export controls because basically China said no rare earth for Japan. And I think Japan went through this giant nightmarish piece of like diversifying, looking for rare earth, et cetera. And what was interesting was that I was reading this whole article talking about how all the efforts they'd done to diversify and get more rare earth, and basically at the end of the whole article they were like, well after this, like, I don't know, five-to-10-year dynamic, Japan's share of rare earths is still the same. Like they're still reliant to the same extent on China. Yeah, that's right. Like I think because the rare earth absolute amount went up and so they're able to diversify to at least keep the non-China component stable, but they were never able to leave the China rare earth ecosystem/supply chain. Yeah. And I think it is right to think about it.

James Chai: We just like, if you think about the one true leverage that China has against everyone, it is rare earth, right? Yeah. Now, the reason why they're willing to even consider doing it outside of China is not really for economic reasons, or is it not for resources. It's largely geopolitical, which means that if, like what you said, if that's one way of constraining the US, they would do it, which means that you do not supply those rare earths to US, but you supply to China instead. Yeah. And of course, I don't think it's explicit in the sense that if I work with you, it's to the exclusion of the US, but it's more incentive-driven, which is how they've run BRI, all of these like Africa projects in the past, which is like: I make it so financially lucrative for you that you would rather work with me instead. And of course, China still has a huge edge in terms of their technology not only being very frontier, but also cheaper. Right? And therefore you'll be able to actually profit from all the resources that you have. Right? In Malaysia, a lot of the state level that is in charge of a lot of these resources are most keen because at the state level, they're always constrained by whatever new sources of revenue there will be. So countries like Malaysia and Vietnam are very motivated to be able to do it. But I do think that there is a geopolitical condition, either explicit or implicit, that will be tagged onto it.

Jeremy Au: Yeah. And I think that will be something that the various governments have to figure out who's willing to pay the higher price. But you know, one thing you inspired in my head is I'm pretty sure the next unicorn in America is gonna be some guy who figures out how to 10x the rare earth processing. Because you know, I'm sure obviously the current approach is very heavy and toxic, et cetera. But if someone can figure out a 10x more environmentally sustainable, which is code for efficiency... but also being able to do that, you know? Yeah. I think you imagine a lot of American Department of Defense money just go both on the, you know, funding side as well as the deployment side, right? Yeah. And I think what's interesting is that, you know, when you look at the rare earth piece as well is that it just feels like, you know, if I look at the three things we just talked about, right? Which is data centers, semiconductors, and rare earth. Actually, what's interesting to me is I think, obviously most of the conversation is about semiconductor. Yeah. But to me, at least from Jeremy's point of view, it's more like, ah, we have so many chips anyway. Worst case is there's interactions, supply for two years... exactly. But there's so much supply and people just use and keep using their MacBook, you know, M4. Oh no, you know, gotta use it for a few more years, right? There's no more upgrade to M5, M6, M7 in the example. Right? So semiconductor—I don't feel like it's there. I think that data centers obviously is getting hot, some AI models, but to some example, like, yeah, everybody can assemble data centers as long as enough space, water and electricity to do it. But to me, I think rare earths is probably like the most underappreciated or it's understated as a risk/choke point.

James Chai: Yeah. For sure I think always, but even then, right? Jeremy, would you agree that even semiconducting in the past wasn't really that much paid attention to? It was only because it became—of course the book "Chip War" came out, right? And then Nvidia rose a lot in value that everyone was kind of intrigued and forced to understand how this whole thing works. If not under ordinary circumstances, manufacturing wouldn't be such a great and big thing. I think that's especially the key for commodities like rare earth where there is firstly no real hero to the story, which means that there is no Nvidia that becomes the face for all of this, and therefore the story itself is very hard to catch on. But it also means that therefore, the niche that one would build in understanding rare earth technology... but you also have to be a bit of a chemist, right? Because it's primarily a chemistry supply chain that would really set you apart, right? But I would say in 2026 and perhaps 2027, I think the conversation would shift from GPUs to data centers first, which means that this is the year people really start to examine what are the actual spillovers as well as effects to the daily lives of people. Unlike rare earth and chips, data centers actually affect people, right? Not only in the physical sense of like producing heat as well as the environmental effects on the people around that community, but also indirectly through the water and electricity costs, right? So you see a lot of stories about tariffs going up, water being constrained. All of these things will have direct effect on the people's lives that most FDIs don't have. So I think that the conversation around there would probably develop much more in 2026. I don't think it would go to rare earths yet, but therefore it means that whoever wants to build an expertise, they have like a longer runway. But also the reason is because a lot of the most frontier know-how of rare earths are in China, and most people find it very difficult to access knowledge as well as the people who know very much about it. But that means that's even more sort of like a higher moat for whoever has the expertise of rare earths.

Jeremy Au: I think I figured out maybe a next book—I think it should be "Rare Earths King," you know, like, like... it's a great title. Oh, you asking? I'm just saying like, the kings of rare earth. Yeah, that's right. Then you're like, yeah. You know, it's like the story of the five Chinese kings that was like... so nobody rules their signs. Right? And then you have some underdogs of like some American startup trying to figure out how to do it everywhere else. I think what we saw from, you know, when China put on the export controls on rare earth, I think America folded on the kind of like tariff war. Yeah. And EU as well was, you know, basically said nobody can manufacture a plane, a car or computers without rare earth for sure. Whereas if you say the normal semiconductors, I think everyone's gonna be like, well, you know, we just use a shittier semiconductor. Some kind of replace, user experience be degraded, you know, less smart car, or you know, the air conditioning doesn't have many functionalities, but you know, like you can make... there's enough semiconductors from Malaysia, it's diversified enough globally that I think imagine a situation where a supply shock wouldn't cause an immediate manufacturing shutdown. But what we saw for rare earth, literally slowing supply was enough to stop manufacturing lines. Yeah, you know? Yeah. So it has a larger and more disproportionate effect on most products. Right? Even the ones that are not super high tech, whereas semiconductors are mostly focused on the very high tech stuff, and you can make do with probably the other technology if you don't keep producing new and more powerful chips. Yeah, right? So I think that's largely true. Yeah. Awesome. On that note, I think we keep on going, but I'd like to summarize the three big takeaways. First of all, thank you so much for sharing. I think the high-level primer of the Malaysia economy, both from the historical roots where I saw was under the British, alongside Singapore, to I think where it's going today in terms of the government policy and approach in terms of like, you know, the ceiling and the floor. Secondly, thanks so much for sharing about semiconductors. I think very much a historical view of how semiconductors came to be about, but also the agglomeration effects and regional race for building semiconductors and how Malaysia has come on top. And lastly, thanks so much for covering, I'll say, rare earths and data centers, both as I think the emerging frontiers where I wouldn't say Malaysia has won yet, obviously, but is very much pushing and investigating very deeply into that. And I think definitely a little bit in a pole position right now, but obviously, you know, I think the story has not yet been written to the same extent as it has been for semiconductors. On that note, thank you so much for sharing.

James Chai: No, thanks so much for having me, Jeremy.