(The transcript has been edited for brevity and clarity.)

CHANG-YOO: Welcome to the show, Tim. 

CULPAN: Thanks, Albert. It's great to be here. 

CHANG-YOO: Yesterday, Nvidia founder and CEO Jensen Huang broke ground on Nvidia's new headquarters in Taipei. It's a 4-hectare site that's to hold 4,000 employees. Jensen Huang calls it “Constellation”. He's aiming for it to be functional by the year 2030. Yesterday he called Taiwan the “epicenter of the AI revolution”. We know that Taiwan makes not just 90% of the world's most advanced semiconductors - so that's used for AI- but Taiwan also produces 90% of the world's hardware for AI. But when he says Taiwan is the “epicenter of the AI revolution”, what does that really mean?

CULPAN: I think a lot of people listen to Jensen's words and think, “oh, he's just saying that to appease the masses in Taiwan, to appeal to the local crowd.” And he does that to an extent, but he's actually very, very right about Taiwan's role in the AI hardware ecosystem. Unlike previous hardware booms - such as in smartphones, before that, notebook computers, and then before that, desktop computers, so that's going back 30 years - most of the AI hardware now is being made in Taiwan. 

So we start with the chips, of course. Jensen runs one of the chip design companies, Nvidia. Lisa Su, his cousin, runs AMD, another one of those chip design companies. They all go to TSMC to have it made, but that's just the start of a very long process. From there, those chips get packaged, again done by TSMC and other companies like ASE Spil, and then it goes into a module, then it goes into a printed circuit board, it goes into a server rack, into a drawer. It's all compiled and put together over many steps. Almost every one of those steps is done in Taiwan by a Taiwanese company. Now, in some cases, the actual step might be done in say, Vietnam or in Mexico, or in Houston, but every part of that puzzle can be done in Taiwan, or is done in Taiwan, and some of it then is done overseas as well. 

So, the notion that Taiwan is crucially important to the entire AI hardware ecosystem is very true. And I think it's unappreciated, not just by Taiwanese who are not even aware how important Taiwan is to the AI hardware ecosystem, but globally, I think very few people around the world actually understand that. Jensen, of course, does understand that. He's got a very tight relationship with TSMC, but he also has a very close relationship with many other partners, such as Foxconn, who puts the service together. 

And so we are now at this very unique time in Taiwan's history, where for the first time in many, many decades - really, in history - that Taiwan has control over the whole end-to-end of a major industrial development, which is changing the world, and it's Taiwanese companies doing it in Taiwan. 

CHANG-YOO: You wrote in Culpium two months ago about many of the Taiwanese companies - that are not named TSMC - that have been profiting from the AI boom. Can you walk us quickly through some of these companies we should be paying attention to? 

CULPAN: Yeah, there's a very long list. I'll plug my own work at Culpium here. 

Three months ago, I started an index called CASCI, the Culpium AI Supply Chain Index. I publish it around the 10th or the 11th of every month, and it tracks the revenue of - what I did, I identified as - 35 companies that are key suppliers to the AI hardware infrastructure. Thankfully, we're privileged in Taiwan that companies required to report their revenue every month. Most places it's once a quarter, but in Taiwan it's every month. So we have granular, almost real-time data on what's going on in the AI supply chain. 

I think the way to look at it is through four tiers, or certainly that's the way I look at it. The first tier is what I call capacity. And so we have companies in Taiwan that help TSMC make the factories, right? They're not a tech company, they're actually a construction company. Part of that we have companies that make equipment - testing equipment, manufacturing equipment. Now, obviously, a lot of these companies are overseas, such as ASML in Europe, and American and Japanese companies, but there are important Taiwanese companies. One that comes to mind is Chroma ATE, which has been around a long time, and they kind of… they didn't disappear, but they weren't hip, and now they've come back again and become hip again. 

And then we've got what I call upstream. So TSMC, obviously, is an important company there. There are many other companies that design chips. MediaTek, for example, is getting into that role now. So, there's about six to eight companies in the semiconductor area. 

Then what I call midstream, a company such as Gigabyte or Delta Electronics. Gigabyte puts together the modules. So they get the chip, they put it together onto a printed circuit board, and then they ship it off to the next stage. Delta Electronics, weirdly enough, as we speak, is Taiwan’s third-largest company by market cap. And everyone forgets who they are. The largest is TSMC, the second largest is MediaTek, and the third largest is Delta Electronics. They do the very unsexy but extremely important job of making the power supply systems that allow the electricity to get fed into an AI server. It doesn't sound sexy, but it's actually really difficult, and they are very good at doing that. 

Then we go to the downstream, which is the companies that put it all together. I think most famously is Foxconn, known uh publicly as Hon Hai (鴻海), that's the listed name on the Taiwan Stock Exchange. But then there's other companies: Pegatron, Quanta, Wiwynn, which is an offshoot of Wistron, are companies that put it all together. In many cases, these companies do more than one thing, they don't just assemble it, they might make another part of the supply chain, such as a module or something else. 

Then there's a lot of companies that really you would never have heard of if not for AI, such as Asia Vital Components, or Chenbro, or King Slide.

For example, some of these companies started out making the hardware that goes into the drawer of a cabinet, so when you pull out a drawer from your your filing cabinet, or your chest of drawers in your your bedroom, they make that metal part. Sounds boring, nothing at all techy about it, but an AI server also has a drawer. You pull it out just like you would pull out a drawer in a chest of drawers, but they're very heavy. So you need a company that can build that slide or that mechanism to hold a lot of weight to deal with the power, and all the other issues involved with making a very expensive server. 

So we've got these hardware companies in Taiwan who really are not at all connected to the tech industry, now becoming very important players in the tech industry. Which is weird, but it's great for Taiwan because it tells you that Taiwan's tech industry and the AI boom is not just TSMC and Foxconn. It's a whole lot of companies right through the island, north to south, east to west, that are benefiting from the AI boom, and I think a lot of people don't realize this.

CHANG-YOO: Thank you. I've never heard of some of those companies, so that was very helpful. 

CHANG-YOO: This AI boom - it is a tech priority for Taiwan. But does Taiwan really have the energy capacity for all this development? Electricity consumption by TSMC alone accounts for 8% of Taiwan's overall usage. It could rise to 24% by 2030. This week, Jensen Huang point-blank asked for more energy: 'Without energy, there is no economic growth,” he said. Taiwan's premier responded the next day, that using current calculations, there won't be an energy shortage before 2032. So, from what you see so far, is Taiwan's energy policy catching up with its national priority in AI development?

CULPAN: Energy is a national security issue in Taiwan. Any country that doesn't have security over its energy is in a precarious position. Taiwan is not immune to that problem, and Taiwan's energy policy has been a political and economic issue for 30 years. 

We did have nuclear energy as a key plank of our energy policy in Taiwan. Then in 2000, the first change of government, the government at that time said, "We don't like nuclear energy.” And people thought, "Yeah, fair enough, we don't like nuclear energy.” So there was definitely a move away from nuclear energy. But the alternative is essentially fossil fuel-based energy. Taiwan imports all of its energy. 

And the other problem is…there are two words that describe why Taiwanese do not want nuclear power. The first word is Fukushima. The second is Taipower. Taipower is Taiwan's state-run energy company, and it is dysfunctional. It is problematic at many levels, because the management at the company don't have control. Politicians in the legislature have control - there's a committee that has to meet to just give them the right to change power prices. And so energy policy in Taiwan is fraught on so many levels. And you do need to understand energy policy and the politics behind it to understand Taiwan's energy problems . 

Electricity is two parts - one is capacity, as you talk about, but the other is delivery. There's no point for any country to have enough capacity, without the ability to reliably deliver it to where it needs to be, and TaiPower faces both of those problems. Anyone who's lived in Taiwan for a few years will have heard about or experienced brownouts - not blackouts, so much, but like little outages from time to time. They happen a lot, even when there's plenty of energy being made, the delivery doesn't happen. So, for a minister or any government person to say we have enough energy is really not solving the whole problem. We do have an energy problem, it needs to be solved. 

CHANG-YOO: So just breaking down Taiwan's energy production from 2024. This is the latest data that we have. 83.2% of electricity is generated from fossil fuels. 11.6% from renewables, and 4.2% from nuclear. (Editor’s Note: As of February 2026, these numbers have changed - Taiwan’s energy production numbers are 81.3% fossil fuel, 12.7% renewable, and 1.2% nuclear.)

That was before May 2025, when the last operating nuclear plant was decommissioned. So Taiwan has been nuclear-free for almost a year. Taiwan also relies on imports for almost 98% of its energy. I'm kind of curious. If this is an economic priority for Taiwan, wouldn't it also be a political priority?

CULPAN: I think right now Taiwan does need to grapple with its energy issues. Nuclear is now irrelevant, but it doesn't mean it will be irrelevant in the future. As you point out, President Lai, from the ruling DPP, has shown a certain amount of pragmatism about being open to the idea. That's a huge turnaround for a party that came to power 26 years ago on a no nuclear pledge. That was a core plank of who the DPP stood for. The fact that 26 years later they're willing to come out and say, “mmm maybe I'll consider it”. That is huge, and I think the political and economic implications of just that phrase alone is really, really important in Taiwan's trajectory of energy policy. 

But it doesn't solve the fundamental problems, which as you point out, is Taiwan imports almost all of its energy. The previous administration of Tsai Ing-wen had an idea to bring in renewables - and it was a sound idea. 

One, it's more green, supposedly, right? No fossil fuels. No emissions. But renewables are also one of the few ways that Taiwan can be more energy independent, because any fossil fuel that we burn on the island has to be imported.  It was a good policy, a good strategy with very poor execution. Unfortunately, politics got in the way at the very local level, with local procurement rules, that hamstrung wind power providers. And so now global wind power providers kind of don't want to touch Taiwan. It's a big problem. 

The first thing the government needs to do is say, "Okay, let's throw that out. Let's look at the bigger picture for Taiwan.” The energy policy is more important than some local industrial need to prop up the Kaohsiung steel industry, or the Taichung wiring cable industry. All these other industries. Or the Taichung Changhua fishing industry - people don't know this, but fishing captains were getting paid by offshore wind companies because they couldn't take their boats out and fish in the waters, right? 

I'm hopeful that both parties will sit down and say, "Right, we've got to get this done.” I don't know that it'll happen. I'm not hopeful right now. I'm not optimistic, but it needs to happen. If it can happen, Taiwan could solve its energy probably in a couple of years. I personally am skeptical of nuclear, because, I'll be blunt, I don't trust Taipower as it exists now, and I think most Taiwanese do not trust Taipower. And if anyone's listening and disagrees with me, you can contact me and talk about it. I'm happy to talk about this. 

Because energy policy is technology policy. As you mentioned, they're connected. Taiwan must sort out this problem if it's going to hope to continue riding the AI wave. And so it's very correct to say that we need more energy, and if we don't have it, we don't have economic development. That is the truth. Many countries and many industries can develop economically without more power, but AI hardware manufacturing is not one of them. We need energy. It needs to be sorted out. 

CHANG-YOO: Two weeks ago, the Trump-Xi summit in Beijing was another reminder of how Taiwan is in a vulnerable position geopolitically. And on several occasions, US President Trump has said he thinks Taiwan “stole” the US chip industry. So, TSMC, it's committed 165 billion US dollars in the US since the second Trump term. Trump, in return, has canceled or added tariffs on semiconductors from Taiwan. What should we be looking for in terms of Trump, Taiwan, and tariffs? When do we find out? And what might change those numbers?

CULPAN: I think Donald Trump sees tariffs is a bargaining tool. And I want to go on record and say I'm not a Donald Trump whisperer. I'm not an expert in Donald Trump. I think most people who claim to be experts in Donald Trump, probably are not. I don’t know what Donald Trump's thinking. I don't know what he's planning, but he does seem to very much revel in being the deal maker, and everything is a bargaining chip, including Taiwan. 

The scary side of that is he sees Taiwan as a bargaining chip. But if you don't worry too much about that and recognize that he is a deal maker at heart, I think that we can take away, for example, from the Xi-Trump summit. He got on Air Force One immediately, and when he was interviewed by the press, he said, “when Xi Jinping brought up the topic of Taiwan, I heard him out. He takes it seriously. It's an important issue to him. I heard him out, but I made no commitments.” And frankly, I think that's pretty brilliant statementship. He hasn't dissed Xi Jinping. He's made no commitments. According to President Trump's own retelling, he heard him out and made no commitment. Alright, you care about it. I think that's good for Taiwan. 

Now, a few days later, he went out and repeated what he said many times before, that “Taiwan stole our chip industry.” While that's factually untrue. I've written columns about it. Anybody who knows the industry knows that it is factually untrue. I do not know whether Donald Trump believes it. Or is saying that as part of his deal making posturing. He said it so many times that I'm starting to think he does believe it, but I'm hoping there are enough people in his administration who know it's not true. 

I would actually argue that Taiwan, in fact, saved the US chip industry, because in the 1980s the US chip industry was facing a lot of competition from Japan. And into the 1990s companies like Qualcomm and Nvidia in the late 1990s popped up, and these companies couldn't get capacity from those who did have fabs like Intel or AMD. They needed someone like TSMC to make their chips for them. And so the resurgence in the US chip industry over the early 2000s, for example, Qualcomm became the global leader in mobile telecommunications. They set global standards for communications. They couldn't have done it without TSMC. And Jensen himself has said many times, “if it was not for TSMC, Nvidia wouldn't exist.” 

That is true for most chip companies today. AMD used to make their own chips, and then they sold off the manufacturing business, and AMD, which was left behind, was hamstrung and were tied to their former manufacturing partner, which held them back. They had a contract for 10 years, that meant that they had to go to GlobalFoundries. The minute that contract expired, Lisa Su came to Taiwan, got TSMC to make the chips, and revenue quadrupled in three years, right? 

TSMC saved AMD, and I do hope that the US administration understands that Taiwan is not the enemy. Taiwan is the partner. 

In terms of tariffs, again, I think tariffs are a bargaining chip. He's somewhat bullied Taiwan into investing more in [the U.S.], so TSMC Chairman and CEO, CC Wei, went to the White House, made a big pledge. I know from my sources he's going to do that again in the next few months. CC Wei will go to the US and make another big pledge. This $160 something billion dollars is not the end of it. It will rise over $200 billion with another round of investment pledges. It'll be time for when it's politically suitable to Donald Trump, so that Donald Trump has something to offer his voters. And maybe that's the point at which Donald Trump will change his mind on tariffs, or ease tariffs, or do something about tariffs. 

CHANG-YOO: Taiwan's GDP grew by 8.63% in 2025 -  growth that outperformed China, Singapore, and Hong Kong. A large part of that is powered by TSMC. So, TSMC is good for the economy, and it's essential for the global supply chain. In addition to its 18 fabs in Taiwan and 2 in China, it’s expanded its global footprint to fabs in Germany, Japan, as you talked about Arizona and the state of Washington in the US. What do you think is the long-term strategy for TSMC in trying to make everyone happy?

CULPAN: I don't think TSMC ever wanted to expand in Arizona. I don't think they were particularly excited about doing Dresden in Germany. I think they're accepting of Japan because it made more sense. It's in the same time zone, it's a few hours flight away, and, culturally, very, very similar. But TSMC recognized that they can no longer be apolitical, and they did what they needed to do. And it wasn't so much the TSMC expanded in the US because of the Biden administration or the Trump administration, but it was more because US business leaders, such as Tim Cook of Apple, Jensen Huang of Nvidia, Lisa Su of AMD, and many more were getting political pressure and wanted to be able to show the administration they were doing something for [the U.S.] - both administrations, both Biden and Trump. And so they would lean on Taiwanese companies, such as TSMC, to do a little bit more here. 

Remember, TSMC is not the only one that is expanding in the US. GlobalWafers, Sino-American Silicon GlobalWafers, it's the same company has started building silicon, growing silicon ingots in Texas. Foxconn, Quanta, Wiwynn are expanding their presence in the US. Foxconn recently said they make more AI servers on US soil than they do anywhere else in the world, which I think is significant. 

So, it's not just TSMC. All of these Taiwanese companies are expanding their footprint in the US because American clients like Apple want to be able to get up at a product launch, and say this product is made in America. “We're investing in America. We are patriotic Americans.” Now these US companies aren't actually investing. It's the Taiwanese with their money investing. The American companies are merely buying or promising to buy from the Taiwanese, but the Taiwanese, such as TSMC, are living in a new reality. They realize they can't be apolitical anymore, and they're slowly learning how to play the game of government relations, and it's been quite a learning curve for them.

CHANG-YOO: Do you think that the current trajectory is sustainable for the growth of these companies?

CULPAN: I think it'll be sustainable for TSMC to expand further in Arizona. They've only built a couple of fabs so far. They will build more. But even after all of the ones announced get built out, and all those that are rumored to be coming get built out, the US will still only account for low to mid single digit of their global capacity. It'll still 90%, be 90% Taiwan. 

And manufacturing capacity is only one part of the semiconductor puzzle. The fundamental R and D, the process of coming up with the recipes of how to make a chip, will always stay in Taiwan. It's not going to move to the US. It's just not possible. TSMC has something like 90,000 people - more than 50% of those people have a master's degree or a PhD or higher, right? An advanced degree. So this is not low-level technicians, this is advanced people doing R&D work, very high-skilled technical work that's going to stay in Taiwan for generations, because the ecosystem in Taiwan goes beyond TSMC. All the suppliers, everybody is in Taiwan, and that cannot move to Arizona. It's just not gonna happen. So, Arizona will do more manufacturing. The actual fundamental R and D, which is crucial to where TSMC is in the world, will stay in Taiwan for a long time.

People ask whether TSMC fabs in the US can be as advanced as they are in Taiwan. Theoretically, it can be there. It will always be a little bit of a lag, because once TSMC has worked out the formula for the next process node, they are naturally going to port it to the local fabs, because it's happening as it goes because it's a very fast-moving industry. There is no industry that moves faster than TSMC chip development cycle, except maybe fast fashion, right? Cars take 10 or 15 years to develop, even a smartphone is a 3 to 4 year process. We get a new node every 2 years, that's crazy fast. 

So, what we'll see is the gap between Taiwan and Arizona shrinking: where the Taiwanese will learn how to do it, will work out how to do it, and they'll port it over to Arizona soon after. But it won't be initially as much capacity or as good a yields or production efficiency. 

But Taiwan will always be ahead of the US, and that's not a bad thing for the US, because I think one thing that Americans and Taiwanese in the whole world forgets is: making a chip, that 12-inch slice of silicon, is just the first step. You could go and make the most advanced AI chip at Arizona today, and get a nice shiny 12-inch wafer of silicon with Jensen Huang's latest chip, and take it out of the fab in Arizona, and you couldn't do anything with it. It's useless to you. It has to be tested, sliced up, packaged. Those things will happen in the US in the next few years, but then you have a packaged chip. What are you going to do with it? 

Right now, what happens to those wafers? They come out of Arizona. They get put on a Boeing, and they get sent back to Taiwan. And then they get tested, sliced up, packaged, wires are put on it, and then they get flown off somewhere else to put onto a printed circuit board. And then they go somewhere else to do the next step. Those phases are not going to be in Arizona or Texas or anywhere in the US. They're going to be in Asia. 

So, the idea that TSMC, even if it had all of its global capacity in Arizona, suddenly makes the US technology independent, is just not true. It's fallacious. And so the US will always need Taiwan. And even if it gets more advanced semiconductor manufacturing in Arizona or Texas or anywhere else, the other steps after that, which are becoming more and more complicated, will still be done by Taiwanese, probably in Taiwan or somewhere else.

CHANG-YOO: Let's switch gears to talking about Computex - that's Taiwan's flagship computer trade show. It was first held in 1986, and has historically been an indicator for how the computer industry is evolving. This year's theme is called AI Together, and it has three topics: AI in computing, Robotics & Mobility, and Next-Gen Tech. What do you expect to see this year? And what do you look for in general when you attend Computex?

CULPAN: Computex, in general, consists of three parts. They overlap in terms of timeline during the week of Computex. One is the keynote speeches from all the famous CEOs come to town. So people turn up for that, they want to see the big names, you know, the heroes of tech. 

Then there's a smaller set of forums, including something called Innovex, and all these other forums that are really specialized for specific niches. And if you're into that niche, this is like, you can have a nerd-gasm over this stuff, because you get into the room and some of the smartest people in the world are in that room discussing this very specific technology, like 500G switching technology, or optical transceivers, or the latest in thermal design for hardware. And if that's your niche, Computex is just a fantasy land. 

Then there's the actual show floor, the expo. To me, that's exciting, because you can just go around and go booth by booth by booth and talk to the people that develop these products. You can talk to the product managers and the engineers who literally - you see a piece of hardware, whether it's a PCB, a printed circuit board, or a motherboard, or an end device, or a module - and you can talk to the actual person who designed and built it. 

There's no other trade show like it. That you can really interface and communicate with and swap business cards with the very person who built this incredible leading edge product at a Taiwanese company, and I love it. I mean, I love it for that reason. The keynotes are great, but you can catch it all online. It'll all be on YouTube. The forums are really cool, but very inside baseball. But just walking the show floor, pausing, and just asking questions to the people who can actually answer them. That's really what Computex is about. 

So, what I'm looking out for is, I, the questions I ask you, what are the challenges? What's the new product this year that you've been working on for the last 12 months? What are the challenges - If I talk to you a year from now, what will we be talking about? That's basically the same 3-4 questions that I ask every single person at every single booth when I work the show floor - What did you work on for the last 12 months? What is new this year? What were the biggest challenges? What will we be talking about next year? 

You ask that question of 5, 6, 7 people in a row, because they clump all the similar companies next to each other, and suddenly get a big picture of, “oh my god, thermal is the big thing this year”. And that's a real problem, because these chips that Jensen Huang is designing made at TSMC create a lot of heat, and heat is the enemy of computing. And somebody's got to deal with that, and you've just met five different engineers at five different companies whose sole job is to work out that problem. And if you don't work out that problem, the product's not going to work. And then you go to the next row, and someone's got to work out how to shovel massive amounts of data through an optical transceiver switch to the next server, and that's a really big problem, and that's what they've been sweating bullets about for the last 12 months. Trying to work it out. Trying to keep the clients happy at big American companies, right? That's what they've sacrificed their weekends for the last 12 months over. And why Taiwan's birth rate is so low, because husbands and wives never get time together because they're trying to deal with some transceiver problem or some thermal problem. 

Like, literally, that is Taiwan's economy - is solving all these problems for Western clients, and it all comes together at Computex. Like Taiwan's economy is just there at Computex. It's just one week of this is who we are as a country, in one place, in Nangang.

CHANG-YOO:  I'm curious, in your years of attending, do you remember other famous products that got made because of Computex? Or maybe cool products that were shown off in previous years.

CULPAN: I remember the very first time I saw a smartwatch. I mean, we take it for granted now. Almost everyone has a smartwatch on their wrist. But sometime around 2004, a company popped up. I don't remember the name. They popped up with a watch that had flash storage in it, and it had very low-level computing, and it was a smart watch. And I'm like, “wow, my God, this is so cool.” Like, it was just so cool. It was very rudimentary by today's standards, but I just thought, how can you get all that computing onto your wrist with us? I think it was a touchscreen display, not a very good one, but I just thought, “wow, this is amazing”. And a decade later, smartwatches were a thing. 

I think the development of notebook computers - we take notebook computers for granted because they're so powerful and so useful. But having watched notebook computers get smaller and smaller and smaller every year, but more and more powerful with longer battery life. It doesn't sound sexy, but when you watch it year after year, and you just see this kind of evolution in notebook computing, I think that's pretty cool. 

And the last few years, the massive amount of computing power that goes into what I call these big, boring black boxes, right? The Nvidia servers, and everybody else. They're huge, like they're 9, 10 foot tall. They weigh a ton, like literally a ton. They cost like $7 or $8 million dollars. They're, they're a marvel of modern engineering, because you've got to get all the electricity in there, so there's a whole lot of electrical cables. You've got to get the information shoveled around inside, so there's all these data cables. Then you've got to cool it, so these cooling tubes with liquid inside them to cool it all packed into a very tightly defined space. And the engineers who put that together had 6, 12, 18 months to manage it, because the cycle is really quick, right? The evolution of these products is super fast. 

So you can imagine that you see this thing on display at Computex, and four months ago, it probably hadn't even been finished, and the engineers were tearing their hair out going  “oh my god, how am I going to make this work? But I don't have a choice, I've got to make it work”. And they do. They pull it off. Every year, they pull it off. 

And so if people are going to Computex this week, go and have a look at the servers, and don't just kind of look at it and walk past. Go and have a look at the back of it, and all the switches and connections and tubes and pipes and plugs, and think of how they jam it all in together and make it work. It's pretty amazing. It's brilliant stuff.

CHANG-YOO:  I can really like get your love for this.

CULPAN:  I am a Computex nerd, and proud of it, like 26th year for me.

CHANG-YOO: Yeah, and that's amazing. 

CHANG-YOO: I want to take a step back from this, all of this. AMD CEO Lisa Su said on the AI industry's trajectory this week, “If AI were a nine-inning baseball game, we're probably only in the third inning.” What's next for the AI boom, and is this a bubble?

CULPAN: I think we're probably only in the first inning maybe the bottom of the first. We are very, very early days. AI has been around for decades, but I think the modern AI era is the advent of ChatGPT. I think, generally speaking, people could say that was the golden moment where it became accessible for everyone. 

The trajectory will be, I think, we will go past the current stage, which is what we call training, where massive server farms owned by Open AI (who runs ChatGPT), by Anthropic, by Google, and all these other companies - Deepseek - just hoover up. They just suck up massive amounts of data, so every post you've ever put on Reddit or Facebook has now been put into one of these models. Every “like” you've put on an Instagram post or whatever is just sucked up into these models. 

We will move past that phase over the next few years to what is known as inference, which is when you go to ChatGPT and say, 'Hey, how do I make a recipe for pancakes?” And ChatGPT knows, because it's gone through all these recipe boards on Reddit, and all the websites that talk about making pancakes. And so “inference” is when you get those models and you give the answers. And we're only at the very, very early stages of what inferencing will do, we think of it mostly in text, but with more and more seeing it in other mediums. What people call “multimodal” - so it could be text plus voice, or text plus audio, or video plus music, or still images plus text - whatever different ways it'll be. And more and more of that will be done closer to the user. 

Already, a lot of these models can be downloaded onto your computer. So you don't even need to touch the cloud. You could actually be offline and still run these models and have a chat with a chatbot. Apple is definitely going on that route of wanting to get more and more of it close to you in their iPhones. And so that will be what is known as edge, because it's close to where you are, far from the center and close to the edge, where the user is. So we'll slowly see a lot of that moving away from big cloud data centers and more and more close to the user, and that will change how hardware is made. 

The other part of that will be robotics. A lot of people are talking this year about physical AI, so the obvious thing would be a robot that can essentially do all the things that you can do on a computer, but a computer just gives you visuals or audio. But imagine a physical device that can take an instruction and physically do something. An obvious one is drive a car. A less obvious one is manufacture a product. 

And the big thing is, in factories, robotics have been a thing for factories for many, many years, decades, but when you have AI in a factory setting without any human there, then that's going to be huge. Elon Musk sees a future where robots will build robots, which will build robots. And that's what is going to be the next big thing. We will see some of that at Computex this week. It's not easy to do. It's very, very difficult. Physical AI is very, very hard. But once those challenges are solved, we’ll all realize that ChatGPT giving me a pancake recipe was like kindergarten stuff.

CHANG-YOO: I think that specific example is very interesting. On one hand seems amazing, but kind of leads me to my next question, which is, how should we look at the role of the AI boom in Taiwan from an ethical perspective? There's certainly a heightened priority here in Taiwan on the recommended use of AI in both the office and academic settings. Just this week, the Educational Minister says he wants students as elementary and junior high school to learn to use AI. But a national conversation about some of the philosophical questions about AI - that's lagging behind. 

You've been in Taiwan for the past 25 years, writing about global tech, and you've seen the rise and fall of tech trends, and had firsthand experience of how society is shaped by technology. Taiwan's current obsession with AI - is this healthy? Or are you actually very hopeful?

CULPAN:  I don't think the AI obsession is particularly unique to Taiwan. China is also experiencing it, and this AI boom was born out of the US, right? Just remember that. The US is taking it up as well. Many, many societies are grappling with the same issues. Many people worry about AI and think it's dangerous. Well, go and ask your grandparents, what was the debate in society when TV came out? What was the debate in society when rock and roll music came out? Or ask great great grandparents if they're still around, what was the discussion in society when jazz music came out? 

Every new trend that seeps into culture, because technology and culture are the same. They really are always a crossover. Television was a technology that changed our culture. Radio was a technology that changed our culture. Electricity was a technology that changed our culture. And of course, more recently, the internet was the technology that changed our culture. 

Social media was built on a lot of those technologies. I don't think social media itself is a technology. It's a use of technology that clearly has changed our culture. There is a lot to be concerned about, but it's not helpful to bury our head in the sand, and say it should go away. It's not going away. We've got to grapple with it. 

I don't necessarily believe that elementary school kids need to learn to use AI, but frankly, I don't know what the education minister means by that. “Use AI” - it's like saying use technology. What do you mean by using AI? Everyone needs to know how to chat with a chatbot and learn this newfangled thing called prompt engineering? I don't know what that means. 

I think a lot of policy makers - and I'm not talking specifically about Taiwan's education, this is not a criticism of Taiwan's education ministerial policy - but a lot of politicians know they need to kinda keep up with technology, and they need to have an opinion on it, and need to develop a policy on it. So it's good that any education minister says this is something we've got to deal with. We've got to come up with a way to think about this, but I don't know what it means to say every child needs to learn AI. 

I think that there's been enough research to show that social media has been very, very destructive to developing minds. I mean, there's a New York Times best-selling book, The Anxious Generation, which goes through this: social media itself is incredibly destructive to educational development. I'm absolutely convinced of that. I've seen it. I've seen young children, in many societies, sitting in restaurants in public and all sorts of places on their phones, not… forgetting how to communicate, not learning how to communicate. And then become addicted. 

So it's a dichotomy. We can't run away from it, but it can be very dangerous, and so I think what Taiwan and any society has to deal with is: what is the healthy use of this technology? And what are the old school skills we still need to learn? 

30 years ago, a big debate in schools was: do we still need to teach handwriting? Because who the heck hand writes anymore? Everyone types something out on a computer, or taps away on their cell phone. Or these days don't even tap away, they just speak to the phones and let the phone take dictation. But still, we teach children handwriting, and I think we should. So, the question is not so much the role of AI in schools, but maybe the question should be: which old school techniques and skills do we need to keep? I know people who think that we shouldn't even bother teaching kids spelling, because, hey, we've always got a dictionary. There was a debate 30 years ago, why do we need to teach basic maths, because everyone's got a calculator. 

I think Taiwan's not unique, just Taiwan has a lot of access to AI, because we are a very tech-savvy country. Everyone's got access to computer and a smartphone and the internet, but I don't think that it's a problem that's going to be solved quickly. I think Taiwan's bigger problem is not kids having AI in the classroom. It's - Taiwan's bigger problem is we don't have enough kids in classrooms. That's the problem we need to solve, not whether or not they're gonna have AI.

CHANG-YOO: Just within your position as a tech reporter and someone who's interested in tech, how do you have a healthy relationship with technology and AI and social media, and all of this? And how do you kind of maintain that balance?

CULPAN: Well, the first healthy thing I did when it came to social media was I closed my Facebook account seven years ago, and I've never looked back. I still use Twitter, and I will continue to call it Twitter. I have Threads. @tculpan at Twitter. @tculpan on Threads. I try to use Instagram, but I'm terrible. 

In terms of AI, I do actually use AI to help me with coding. I do know how to code. I know the fundamentals, but if I've got a coding assistant to help me remember how to use a certain library or certain syntax of writing code, I'm going to use it. But I find it still enjoyable to write the code myself. 

I don't use AI for writing, I'm a writer, and I really believe that writing is thinking, so I'm not going to outsource writing to something else, because the process of writing is how I think. If anyone reads my work, you can read my writing, my columns, and you can kind of see me thinking out loud, because that's my process. And so I would encourage anybody who's writing, just even if they're writing a journal for themselves or whatever, don't outsource writing, because writing is a really important skill for humans. We're the only species that communicates in a writing form. Many species talk to each other, right? Or dance, or use visual communications, but we're the only one that write. So, if we outsource that to a machine, we kind of lose our humanity. 

I also think that it's important to read things. I don't use AI to summarize documents, although it's tempting, because a lot of documents are long and boring, but I still wade through it, because even if AI is 100% accurate, it's often the stuff between the lines in a document. Whether it's, for me, for example, I read a lot of like IPO prospectuses, I'm a real PubMed nerd, I read a lot of medical journal pieces. And it's often the parts between the lines that an AI will miss that's really the interesting part, so I really recommend people don't use AI to write and don't use it to read. 

CHANG-YOO: I think what kind of surprised me about this conversation is just the sheer amount of diversity of technology within Taiwan. And so I really feel like that your passion kind of broke through this somewhat nuanced and jargony topic in a very understandable and a very interesting way. So, thank you so much.

CULPAN: Thank you very much. I was really happy to have this chance to discuss the topic with you. I think there are so many layers to technology and what Taiwan offers the world that people don't understand. So any chance I have to talk to anyone about the world of tech beyond TSMC and chips and Foxconn is a chance I'll take, because I think the world needs to understand Taiwan beyond the headlines and the big names. Taiwan's devil is in the details of the small companies doing hidden stuff.

CHANG-YOO: Thank you. You can read more of Tim's tech analysis on Culpium and on his podcast “Supply chained”.

CULPAN:  Thank you very much. I appreciate being here.

Episode credit: Host - Albert Chang-Yoo. Research - Albert Chang-Yoo, Zack Chiang. Audio Editing: Wayne Tsai. Transcript - Albert Chang-Yoo. Producer - Emily Y. Wu.