What Happens When The AI Bubble Pops? - Hian Goh
Hian is a founding partner of Openspace Capital, a leading multi-strategy asset management firm focused on Southeast Asia.
Most notably, he was an early investor in some of Southeast Asia's most notable tech startups such as Gojek, Love Bonito and Lucence.
Hian himself has extensive entrepreneurial experience. He founded and built Asian Food Channel, a 24-hour pay TV channel that was sold to US media giant Scripps Interactive (now Warner Brothers Discovery Inc.).
Prior to this, he was a technology investment banker at SG Warburg (now UBS) and Salomon Smith Barney (now Citi).
He holds an MBA from INSEAD and a Law Degree from Trinity College, Oxford.
In our conversation today, we cover the following as seen below-
TIMESTAMPS:
0:00 Introduction to Hien Goh and OpenSpace Capital
1:06 The Early Days of OpenSpace Capital in 2014
3:02 The Southeast Asian E-Commerce Boom
5:15 Selecting the Right Founders
9:10 The TRUST Framework for Evaluating Founders
14:29 What Makes A Great Founder for VCs
16:13 Investing in Gojek: Finding Indonesia's Unicorn
20:05 Is AI a Bubble?
25:47 What The Dot-Com Era Can Teach US
27:30 AI Economics and New Business Models
33:13 Why Focus on Cost of Distribution
35:04 Who Will Win the AI Race?
40:03 The Agentic Revolution and Hian's "Hey Gorgeous" Idea
44:05 Beyond NVIDIA: The Future of AI Chips
50:10 Jim Keller, The Famous Chip Ronin
54:18 Singapore's Talent Strategy in the AI Era
58:59 Made in Singapore: Changing Mindsets
1:03:16 Singapore Inc and Strategic Decision-Making
1:05:41 Avoiding Over-Optimisation for Success
1:08:21 Advice for Fresh Graduates
This is the 60th episode Of The Front Row Podcast
Keith 00:12
You're the Singaporean VC that funded one of Indonesia's biggest tech unicorns that came out in the last 20 years. That's Gojek, which is now Goto Group. So Nadim Makarim is the founder of Gojek. When we first met him, and I still maintain this, if you ever meet Nadim, he's got the most amount of charisma. Does he have delusions of grandeur? Absolutely. But it was very clear from day one that Nadim was exceptional.
Hian Goh is the founding partner of OpenSpace Capital and is behind some of the most iconic investments within Southeast Asia. He invested in Gojek, Love Bonito, and Lucence when they were all in their early stages. In today's conversation, we talk about what he thinks makes a great founder, the intricacies of being a venture capitalist in Southeast Asia, and where Singapore can thrive in the age of AI.
Keith 01:03
Welcome. Thank you very much. By the way, this is one I've been waiting to do. Actually, take us back to 2014 when, in the early days of OpenSpace Capital, before you were like a behemoth in the VC industry, you guys had maybe a smaller, humbler beginning. Take us through those early days.
Hian 01:21
All things start small, Keith. I think that's the first thing that I really enjoy, that you build things from scratch. A common question that has been asked of me was: you sold Asian Food Channel, it's now called Food Network Channel, so why become a venture capitalist?
In 2013, 2014, what the Southeast Asian and Singapore ecosystem was suffering from was a gap in essentially structured venture capital. Series A money, which was not abundant at that time, is the gap between seed capital, of which there was a lot of supply, and private equity-type money. It's really where the salt water and the fresh water meet at a river. You've got to be a weird animal or weird fish to survive in the environment. You have to be part entrepreneur, part product-market fit person, but you also have to be the person who can do Excel spreadsheets and so on and so forth.
Series A money in the venture capital world, which we started off as a specialisation, to me really is the most interesting part of the value chain. At that time, Singapore didn't have enough of it. When I sold Asian Food Channel and I was angel investing on the ground, I recognised that problem and I teamed up with my business partner Shane Chesson. Shane was the former head of tech investment banking for Citi and we put together OpenSpace Capital primarily to solve that problem. We got Temasek on board as an anchor investor and our first close was actually $30 million, then we parlayed that into a $90 million first fund. So that's how we got started back in 2014.
Keith 03:05
Through the late 2010s to 2020, there was a huge e-commerce boom. Everyone was investing in e-commerce players. It doesn't matter if you're a shopping platform or retail brand. As long as you're going online and you're doing shopping, people are investing in it. So by definition, you probably had a glut of opportunities. How do you filter for the right ones to invest in?
Hian 03:23
You have to look over cycles. I believe that every 10 years something structural changes. When I first started my career as a tech banker in 1999, the internet came out and that was the revolution of information that you can get anywhere in the world through a computer. But by 2010, the mobile phone revolution sort of kicked off, so now you could get information anywhere, anytime, ubiquitous. I believe now, 10 years later, 15 years later, you have knowledge and intelligence anywhere, anytime. That's AI.
In 2010, it was very clear to us that the mobile phone brought a different distribution mechanism and Southeast Asia was very unique in the sense that it never had an established base of PCs in the second and third tier cities of Indonesia or the Philippines. China actually had a gaming revolution, had a PC revolution, and then it went to the mobile phone revolution. So Southeast Asia was brand new. The joke we used to say when we talked to investors, and it was a relatively new concept, we needed a sound bite, was: 600 million people just got their mobile phones and they don't know what to do with it.
From that you segment into all the different services. You could do fintech, you can do health tech, and of course one of the biggest GMV ones was e-commerce. So that's why e-commerce became such a big thing over the past 10 years. I think it's actually inaccurate to assume that it'll be the same in the next 10 years. The past 10 years have been really good for Southeast Asia because it was about digital distribution, about the mobile phone revolution. That's how we got all these wonderful new companies in the past 10 years.
Keith 05:16
You have the opportunity at the higher level but you also have to filter out for the right founders to back. For every one Grab or Gojek, there might have been like 10 or 12 other competitors trying to take their money. So then the question is, how do you choose or how do you select the right founders for you?
Hian 05:34
Well, let's assume that you kind of figure out what the right business model is. I think as a venture capitalist, if you stay in that space and you just meet a lot of companies, you meet a lot of projects, you'll start to see the trends. It was very clear that we saw e-commerce was important. Female e-commerce turned out to be the one that has the best metrics.
Then you start seeing three or four different companies and then you've got to make a decision about which founder you back. The story I always tell is that we looked at about 32 e-commerce companies before we decided that actually female e-commerce was the most interesting, and then we brought it down to essentially two companies. One was called Love Bonito and the other one was called Pomelo Fashion. Have you heard of Pomelo Fashion?
Keith 06:21
Yeah.
Hian 06:22
Do you know the background of Pomelo Fashion? Not really. So Pomelo Fashion was started by David and another co-founder. Really smart guys. They've got an excellent track record of education. I think they come from Ivy League schools. Certainly when they came in, they had their 2x2 metrics. They used to work for either Zalora or one of those Rocket Internet companies. In a boardroom, they were very impressive.
I then met Rachel and Viola. In those days, the CEO Deon hadn't been recruited at that time. I just met people who were very good at understanding the female customer, making clothing for their Asian customer, but they had no 2x2 matrix. In that regard, Keith, I'll pose this back to you: you're in the boardroom of a venture capitalist, who do you choose?
Keith 07:15
If history teaches us anything, it might be the customer-centric founder.
Hian 07:19
Exactly. My instinct as well. Street smarts and that kind of stuff. Love Bonito by that time had already made like $6 million of revenues. It was profitable. You have to give that credit for that. But the partnership was in two minds: do we want to back people who are good at the presentation of a business versus a person who is good at the raw guts of the business?
I famously told the firm: what do you want to do, back two people who know how to do 2x2 charts but have to learn how to make female clothing, or two people who know how to do female clothing but have to learn how to do 2x2 charts? I'll be very clear about it. The situation was like there was no planning. I would ask Rachel: what is your customer acquisition cost? She'd say: what customer acquisition cost? We just post on Instagram to our thousands and thousands of users and we do it.
At scale, it becomes a challenge. What I bring that from was, to me, and this is just my personal philosophy, I look for founders who have certain qualities, and the one that had the interesting perspective was people who were emotionally authentic to the mission of selling women's clothes. I believe that over time I could help build out the rest of the other functions and capabilities that Love Bonito needed in order for it to be a large company.
Today it is a large company. Today it has sizeable revenues and today we want to IPO it on the Singapore stock exchange. So I think that's how I think about things and how I select founders.
Now, through the years, I have never really figured out a framework as to how to encapsulate what is a perfect founder. But over the past 12 months, I kind of have an idea and that's why I'm writing a book and that's why I'm trying to come up with a framework.
Keith 09:08
You have to tell me a little bit about the framework.
Hian 09:13
After doing this for 10 years, I have probably met about 3,000 founders. I meet about five founders or founding teams every week. This is after we filter thousands and thousands of companies. OpenSpace gets about 3,000 business proposals. We do due diligence on like 500. I end up meeting about 250 teams a year. So about five teams one-on-one. If you and I were having a conversation about investing and this takes an hour, this would be the kind of interaction I'm talking about. Not just a three-minute elevator pitch, but really talk about your background, why you're building the company and all that kind of stuff.
So over 3,000 conversations later, I have a theory about what it takes to be a great founder. I call this framework the TRUST framework. So TRUST. What are these things?
There are four qualities and a balancing fulcrum in the middle. T is first principle thinking. A lot of times something changes in our lives that disrupts all the assumptions that has happened. A very simple example is the invention of the motor vehicle. Once you had motorised transport, you could create suburbs and suddenly farmland became very valuable. Of course, you had to build the roads and you had to wait for cars to get really good. But this assumption that we all just live in nuclear cities and we walk around as opposed to have a suburb, if you're a real estate investor, that would be your first principle thinker. That's hard because you really need to have a conviction about a piece of the future where at this point in time nobody agrees with you.
You then need the second principle, which is resilience, because you now have to journey that first principle thinking and you never really know when it actually flips. Imagine the guy who did Zoom. The founder of Zoom said: video conferencing is kind of okay, but I'm going to make the best video conferencing. He did it for many, many years. Then it needed COVID to really prove to everybody that video conferencing was really the future. And then we have Zoom as a dominant company. You never know when that thing flips.
So the unreasonable thinking and the resilience creates what I call the unreasonable mindset. You're just an unreasonable human. You just have this conviction. Anything that is reasonable today, I would argue, was unreasonable previously. There are so many different examples. Even podcasting, like you and I, being able to record something like this. When I did Asian Food Channel, it took a lot of money to do this. Now we just walked in here and we did it. What is reasonable today was unreasonable at one point in time.
So those are the two fundamental qualities. Those are non-negotiable qualities. You have to find somebody like that. Sometimes people with great educational backgrounds have gone through a lot of wonderful things in life but they never really went through hardship. That's why I kind of like street smart type individuals in the first place.
The other last two qualities are self-learning organism and team curator. On the other hand, you have to also know how to take feedback. You also know how to think about things. You can't be such a megalomaniac, narcissistic, crazy person that you don't listen to people. By the way, this industry is littered with so many people who are completely delusional and they just don't listen to people and as a result blow up on a long-term basis.
The last thing, of course, is being a team curator, which is that over a period of time, if you are successful, your company actually scales and you actually become good at what you do, and then new kinds of people need to come on board. Even Mark Zuckerberg needed Cheryl Sandberg. Companies need to curate. So you need to curate your team for the right time and you need to know how to step back and say: actually, I shouldn't be the CEO anymore. Maybe I should be the head of product or maybe I should be the chairman. I should create a pathway for my project, my originally unreasonable project, to become a reasonably large, profitable, cash flow generative institution that the world can rely on.
So I think these are the four characteristics, of which the middle bit, unreasonable mindset, is the fulcrum because the first two always conflict with the last two. You want a megalomaniac who's also open to feedback. You want a dominant leader who will never give up, who also knows how to step aside when the time is right. That's why the industry always has stories of CEOs being changed or replaced or forcibly replaced, because by definition I think this is a very hard industry to pick up and to actually make successful.
Keith 14:32
Do you think there is a founder that you have backed in the past that kind of exemplifies this?
Hian 14:38
I've thought about that and I think my quick answer to that is by definition you'll never find, or I haven't yet found, somebody who when I first meet them has all four of these qualities. I think the whole point of this framework that I've come up with is to identify what the strong points of that person are and then see whether you can work to create the other two and also identify red flags.
One of the big red flags which I talk very openly about is the person who is a complete narcissist. They can raise a lot of money because they've got grandiose ideals, they're really good at presenting, but they never listen to anybody because the egos are so fragile. A person who's a narcissist, like a full-blown narcissist, has a fragile ego. So when you start trying to give that person feedback, it all falls apart.
But on the other hand, if you don't have narcissistic tendencies, if you don't have some delusions of grandeur, you also can't be an entrepreneur. So I think it's more about: I created a framework so that I can figure out from a metrics-based framework how to think about whether this founder has this quality or not and how to work with them to actually get them to the next level.
Keith 15:55
You're the Singaporean VC that funded one of Indonesia's biggest tech unicorns that came out in the last 20 years. That's Gojek, which is now Goto Group. That seems almost like an anomaly that you're not in the home market yet could still find such a deal.
Hian 16:16
OpenSpace is actually a group of about 40 individuals in six countries. I have a head of Vietnam, a head of Philippines, a head of Malaysia, a head of Thailand, obviously Singapore, and a head of Indonesia. When we started OpenSpace Capital, the thesis was that Southeast Asia was actually a very interesting region. China and Southeast Asia roughly is going to represent about half of the world's growth, GDP growth. Southeast Asia actually is slightly larger from a GDP perspective than India. People always never realise that.
Yet Southeast Asia has its complexities. It's not one contiguous country. So we've always been as a firm trying to figure out how to work. We actually have a very strong Indonesian franchise. It started by our partnership with an Indonesian private equity group and then today we have a very full-blown Indonesian team. When we do deals in Philippines, a lot of times our Filipino team speaks in Tagalog, and same in Vietnamese in Vietnam. We had the on-ground experience to actually find Nadim and those kind of folks. So we're not really a Singaporean VC in that regard.
Keith 17:35
But aside from that, you could still, among the sea of competitors, be able to find a unicorn like this and they're doing well. I wanted to know, how did that deal come about? What did you see in the founders in those early days?
Hian 17:48
Then we have to talk about Nadim. So Nadim Makarim is the founder of Gojek and when we first met him, and I still maintain this, if you ever meet Nadim, he's got the most amount of charisma. Does he have delusions of grandeur? Absolutely. But yet he's open to feedback. He fights his ground and so on and so forth.
It was very clear from day one that Nadim was, in my essence, founder material. He understood the first principle thinking and the first principle thinking in Gojek was that you could kind of do an Uber model but on motorcycles. The second theory was that the capacity utilisation was going to be higher because in the peak hours when there was taxi-like services you could do it on a motorcycle, and then off-peak hour you could deliver parcels and that kind of stuff. Obviously the model has changed significantly, but I'm bringing it back to that.
We did the economics and it was much cheaper to do a motorcycle. Indonesia is the only country where a person would get on the back of a motorcycle and use it as a taxi. He understood that. He communicated that very strongly and he was just a very stubborn individual. So based upon that, we made the bet.
I think although in this business you must remember, it's easy in hindsight for me to tell the story and say therefore we got it right. There are a lot of SEA founders which had the same qualities that we back and because of the industry dynamics or because of other things, it just doesn't work. So it is a bit of a numbers game. But essentially that was what we saw in Nadim and I still love the man to bits. Every time I meet him, the amount of energy he exudes is incredible.
Keith 19:33
If we looked at the past, you kind of rode on this e-commerce wave, this digitalisation wave. I like the fact that you talked about this mobile phone revolution where the access to the distribution of information, the cost of it, dramatically decreased. So now we're entering, as you put it and as many others have put it, you're having now widespread access to intelligence at a very cheap price and that has created a paradigm shift. Now we see this whole AI boom that's happening before our eyes.
Naturally the question is, if you ask anyone on Wall Street or any person in finance, is this a bubble? It's a futile question to ask. What I mean by that actually would be that it is a bubble. We can assume to a certain extent there are bubble-like qualities, but maybe there are other caveats that one should note and not just dismiss it as maybe a tulip mania because maybe that's not the right analogy.
Hian 20:31
Well, there are bubbles and then there are bubbles. Let's first and foremost exclude what it is not. It's not a tulip-like bubble. There's actual inherent value in artificial intelligence. We use ChatGPT, we talk. It's pervasive and the demand is going through the roof.
I just met with a friend of mine who's a senior person in one of the three hyperscalers. There's Amazon, Microsoft, and Google. So you can guess who he works for. But he was just telling me: demand is through the roof. Once we build a data centre, it's done. The capacity is booked and people just want it. I was like, you're kidding me. He goes: no. And he said this one specific thing: you don't understand, Hian. If just, say for example, everybody on WhatsApp starts using large language models, think of the amount of compute that we need to install to just proliferate that.
So the demand is there, the use case is there. Now whether people pay for it is another thing and that's exactly why people think of it as a bubble. But the way I look at it is that there's going to be one or two companies who essentially win a large part of this market. So I always say: where are you standing? Because if you're standing in the wrong place, you're in the bubble. If not, you're going to make money.
The guy who invested in the fourth version of YouTube did not make money. But if you actually invested in YouTube and YouTube became the dominant thing and Google paid like a billion dollars for it, you made money. So a bubble is simply because everybody will come through my door and say: I'm going to be the one that takes 80% market share, therefore you should pay $200 million for the company that I've only just started. That's called a bubble. But if it turns out that you're right, then you weren't in the bubble.
We are definitely in an environment where people are demanding very strong valuations for certain kinds of business models in AI. It might be useful to kind of unpack that a little more. Sorry, maybe I'll continue. It doesn't mean that it's wrong. It doesn't mean that it's all going to end up in tears. It doesn't mean that it's going to be like the tulip bubble and it goes away and we never use tulips again. I mean, there have been bubbles which are artificially generated which were hyped up. Was it the ape monkey painting, the NFTs? I don't know whether they're going to come back. But this thing has got value. I guess that's what I'm trying to say.
Keith 22:57
During the dot-com bubble, so many fibre optic companies went bust, but because they've laid the fibre optic cables, now today we enjoy the benefits. We in the moment perhaps underrate or are unable to conceive of a future where our lives are fundamentally different, this paradigm shift. Even an example of the internet is that now if you go to any Fortune 500 company, they can't run without the internet.
Hian 23:40
I was literally there in 1999. I'm 50 years old. I was literally there and you talk about companies like Global Crossing, Asia Global Crossing, World Com. Jack Grubman was a very famous research analyst that worked at Citigroup Salomon when I was there. The whole fibre laying was because of tremendous excitement of the fact that we're going to have a different way of doing communications. It turns out that after 15, 20 years, that revolution was actually correct. It's just people got too excited.
You know who actually made money? A Singaporean company, ST Telemedia. When Asia Global Crossing went into bankruptcy, they bought the assets of Asia Global Crossing quietly. ST Telemedia is one of the largest data centre operators in the world. They own significant minority stakes in many data centre companies in the world and they made money. So it's just really a pricing issue that happened during the 1990s. People got too excited about it. But at the end of the day, the revolution came through.
Now the other thing that was very interesting, and I was talking to a friend of mine who was a very senior Chinese venture capitalist, and he says this is why life is very tough, was that in the Chinese context, as the Chinese mobile phone revolution happened, you had companies like Meituan, you had companies like Baidu, you had many of these Chinese internet giants. Every time there was a round, it just felt like you were overpaying. But if you paid up for that price, you would have missed one of the greatest trajectories. I mean, he invested into JD.com when Richard Liu was not really that big. He made so much money.
So he said, on the other hand, if you didn't believe the hype, the mistake in that situation was you didn't actually lean into the valuation because you were too gun-shy. So then he says: I don't know how to do this because once in a while you get it horribly wrong, and even if you become too gun-shy, you also get it horribly wrong. So in both scenarios, you get horribly wrong.
Keith 25:47
The example you gave about YouTube being acquired for a billion dollars. At that point in time it seems insane, but if you look at the many billions that YouTube generates alone for Google today, but they didn't have that revenue at that time.
Hian 25:59
Precisely. Which is like a chicken and egg issue. So you come now, people go: oh, a trillion dollars worth of capital is going to go into AI data centres but only X amount of revenues are generating, so clearly this is a bubble. I've read so many of these analyses. I'll just give you one business model. I love this business model because it's the one that everyone sort of wakes up and goes: oh my god, Hian's saying this again.
AI girlfriend, Keith. AI girlfriend. Would you pay $100 a month for an AI girlfriend?
Keith 26:29
I think there are people who would do that.
Hian 26:31
And I use AI girlfriend because there's also AI boyfriend. By the way, in China, AI boyfriend is very big. So AI companion. Completely new business model, completely new revenue. If you try to analyse what is available now versus what could be in the future, it doesn't make sense.
Facebook, when it first came out, to think that Facebook was going to dominate the advertising business, destroy newspapers, destroy so many other traditional media, destroy television. I knew a lot of people who were original investors into Facebook back in the day. I knew Zhou Shouzi before he became the CEO of TikTok and he would just tell us: you don't understand, the pace at which this thing is growing, we will find a revenue model and we will basically generate significant revenues. Once in a while that really happens.
So this is why I always believe things that are reasonable today were unreasonable at one point in time.
Keith 27:31
So take me through the economics of, as you've described it, this AI arms race. Who are the major players? How are the people who are investing in it figuring out the unit cost economics?
Hian 27:46
The way I think about it is that your business model doesn't work until it works. So what do I mean by that? Say for example you're going to create an AI companion or, even better yet, say for example we're going to make a digital version of Keith, because people just want to talk to Keith.
A really dystopian way of looking at it is, say for example I have a friend who's diagnosed with cancer and I know that I'm going to die in like four years, but I want to make sure that my kids be able to talk to me in 20 years' time. Let's not talk about the grief part about it, how dystopian it is. But you could take everything that you have written, everything that you have said, your likeness, everything, and we can put it into a cold storage depository of information. We would wait until the time was right and then I would generate a digital version of you through large language models, video models, so on and so forth. And lo and behold, there is a digital version of my friend whom we can talk to in perpetuity.
At this point in time, you probably do it for Lee Kuan Yew, you probably do it for Nelson Mandela, you probably do it for Abraham Lincoln because the value of that is significant. Less so for Hian, because not everybody wants to have a digital version of Hian. If the cost drops to a point where it's affordable for everybody to just have a digital version, all hell breaks loose.
So that's what I mean by the economics. I'm just going to pause there. Does that make sense?
Keith 29:11
Yeah. Makes sense.
Hian 29:13
So then the priority is to reduce the cost of compute, to reduce the cost of intelligence. So how do you do that?
Right now in the AI world, there are sort of two thinkings. One thinking is: I'm going to make the one model that rules us all. So ChatGPT-5, ChatGPT-6, and we're going to have two trillion tokens and five trillion tokens. We're going to have massive training runs. You know, I'm going to be like Elon Musk and put in 250,000 GPUs and run a gigawatt of electricity. By the way, a gigawatt of electricity, just to give you a frame of reference, Singapore uses 13 gigawatts. When you say I'm going to use a gigawatt of electricity, that's the entire data and electricity usage of all the data centres in Singapore alone. People are now talking about doing a gigawatt training cluster.
Then what happens is that there's also this idea that we're going to make models that are smaller, leaner, more efficient and then put it on your mobile phone. So you'll have distilled versions of DeepSeek and distilled versions. There are two sort of methodologies there.
The third thing is that we'll train your information perhaps on the smaller model, and by the small models, like 70 billion, and those training runs will use say a thousand GPUs and you'll create a digital likeness of you. Now we haven't seen that before. Can you imagine if that's actually done right? Suddenly there's a completely new business model. I would pay $10,000 to make sure that my kids can talk to me after I'm gone. If that training run exists.
Now, why hasn't that happened? And this is where you have to be a bit technical for your audience, which is that it turns out that it's actually hard to do a training run. Specifically, everyone talks about this software called CUDA. We've heard of this idea called the CUDA moat and everyone says NVIDIA is dominant because they have a CUDA moat. The real reason for it is that actually NVIDIA is dominant right now in terms of revenues because they're the only people who can deliver clusters, tens of thousands, that you can orchestrate to do a training run.
If you actually want to run a model like inference, there's an argument that you can use AMD and soon you'll be able to use things like Tenstorrent and other new generations of technology to do it. But if you want to do a full-blooded training run, you pretty much have to use NVIDIA. If you're in China, you probably would be able to use Huawei or something like that. And then you have to run a training abstraction.
Here's the interesting thing. It turns out, and I just recently found out, that actually NVIDIA doesn't really provide you with the state-of-the-art training software abstraction. OpenAI has got its own version, Google's got its own version, DeepSeek's got its own version. It's not a well-known fact. I just figured this out. So even if today the Singapore government wanted to run a healthcare training model, specialised budget, there's an argument that it's very hard to do because you still need a specialisation, a team of software engineers who make that orchestration software.
Now NVIDIA's got one version which is above PyTorch which I've been told is not bad, but if you really want to drive down, you have to have your own capabilities. The day a software company comes up and says: I've actually got a software training package that you can use and people outside of OpenAI, Google and everybody else can use on say a thousand cluster training run and I can train your likeness in seven days as opposed to seven months, then the business model becomes very interesting. Then the revenue opportunity transforms. And then the criticism that this is all a bubble and where are the revenues changes.
Does that make sense?
Keith 33:09
That totally makes sense. It reminds me of the story of Pets.com. Everyone used that as the laughing joke because they failed primarily because the cost of distribution for me sending pet food to your house was too high, until the unit cost of delivery of those goods now comes down. It's come down. And now you have Chewy.com which bought the domain name. They earn like $6 billion a year just from that. So it turned out you could build a multi-billion dollar company from that when the cost of distribution dramatically lowers.
Hian 33:51
So I always joke, I think that somewhere in Silicon Valley someone already has a sentient AI in his basement. Some billionaire has it, but it's just that it's a little bit too expensive for everybody to have sentient AI.
But just to build on that point, think about what I just did and bringing it back to the founders. If you're trying to answer a very simple question, or well not a simple question but a simple to understand concept, is this a bubble? Then you go on LinkedIn and people go: oh, the revenue is not there. And then somebody goes: the revenue will come, don't worry about it. You have to actually go deeper and say: visualise what could be the things that generate the revenues.
So then I actually went and said, maybe it should be a training cluster which is a thousand GPUs. So then I asked people and they go: it can't be right now because it's too complicated. But then I go: if I found a company in the world who actually specialises only in making small model training software, I would invest in that company. So again you can see, first principle thinking becomes critical, even as an entrepreneur, as an investor, if you're trying to invest before the S-curve flips into that inflection point.
Keith 35:04
Yeah. In the current, say, AI race, what would the winner look like? I think you pointed out that maybe it's one that solves a very niche problem, you solve the software package that allows you to democratise access to the coordination layer?
Hian 35:25
At this point in time, I think what we're doing as a mental model is to build all the infrastructure layers. Somebody's got to figure out a chip that is even more powerful or even more power efficient. We've got to figure out cooling systems to get the energy out.
I'll give you a specific example. The H100 server, don't even talk about B200 or B300. The H100 server is a server that is roughly about this size. It's called 6U. It's about 60cm by I don't know, 45cm. That thing has got an energy load of 30 kilowatts. So 30,000 watts. A hairdryer is 1,000 watts. And essentially when you compute, that's all it does. The silicon is a resistor and it generates heat. So imagine 30 hairdryers turned on full blast. Put it in that little space. Congratulations. That's the heat that comes out of one H100 server.
Now put five of them there. So you look at the rack, the rack now has to be 150 kilowatts. 10 years ago when we were doing network equipment, it wasn't 150 kilowatt, probably like 12 kilowatt. So that infrastructure has to be changed. The cooling system has to be changed. The electrical system has to change. Everything has to be changed.
At this point in time, all the business opportunities you see are really building the infrastructure and I'm starting to see the software abstraction becoming important. Like a training cluster software orchestration software. Agentic is now becoming interesting.
Where I think the Cambrian moment for AI for Southeast Asia is when we have this thing called the agentic revolution. So what do I mean by that? When AI moves from just intelligence to actually being a robot. So if I tell the AI to book a ticket and it books a ticket for me, that's a robot. That's the definition of robot. It's not actually an IT, it's like somebody who does an action for me that is reliable that we can trust for it to execute well. You're starting to see things like that happen and that's what the agentic revolution is supposed to be about.
Here's the interesting thing about it. Today in Singapore, if you wanted to be a founder and a startup founder, you don't have to have 10,000 GPUs and train some sort of fundamental model. You just take the large language model, you add it into something called LangChain, which is an MCP agent, I mean there's a few of them, and then you can create robotic interactions.
So what I mean by that is, we for example invested in a company called Chope. It's a restaurant reservation business. Well, how do you reinvent a restaurant reservations business in the agent world? So I'll give you an example. I would call this business Hey Gorgeous. So Keith, you download literally just a phone number on WhatsApp and the service is called Hey Gorgeous. So you type in and Hey Gorgeous comes back and she says: Hi Keith.
Keith 38:21
Hey Gorgeous.
Hian 38:22
What do you want to do? See, I made you smile. See, emotion is very important when you deliver services. So you go: oh this is great. It's like, hey, I'm going out with my friends, four of us. Birthday party. Hian's coming. I want to take them out. Great. I'm going to spend $2,000. Can you recommend a place? And then Gorgeous comes back: no problems, Keith, I'll take care of you. You stay gorgeous.
Cheesy. And then she goes and pings. The agent goes and pings all these restaurants and says: private dining for four, $2,000. What can you deliver? Give me what you have. Are you going to give a free bottle of wine or whatever? And she gets three or four offers for you. And then she comes back in 10 minutes later and says: hey, I've got these three offers. Which one do you want? And you go: oh, I like this one. And she says: okay, I'll book it. And of course, basically she goes to the API at Chope. She books it for you and she pays a dollar per reservation. So she pays $4 or $5 to Chope. So Chope gets $4 or $5.
But behind the scenes, this business model could mean that I go to the restaurant and say: if you want this reservation, because I didn't tell you who it was, I need a 5% take of the $2,000. And then suddenly the revenue model changes. So that is what I mean by there will come a time when these AI business models will generate additional revenues, additional value, additional productivity, by the way. And it won't necessarily be these headlines that you hear about Mark Zuckerberg committing to the US government to invest billions and billions and billions of dollars in infrastructure.
That's going to get laid down, I think, in the next three to five years. But we're starting to see the Cambrian moment of these agentic models start to appear.
Keith 40:08
The internet revolution, the headlines back then were fundamentally not the ones that really changed the way we operate. You had Amazon.com and then you had Love Bonito. So you get this kind of scaled-down version of the same paradigm.
I wanted to ask a little bit more about maybe, I won't say the bottlenecks or constraints of the way we approach AI investing, or what would the worst case scenario look like. So if one was to use the dot-com era as an analogy, it's just the amount of, for example, CapEx or the amount of investment now for you to build an AI company. Now it's much more capital intensive than it was in the dot-com era. And also you're building out massive factories that are going to be, if let's say it goes bust, it's just going to be an idle factory there. What are your kind of views or your kind of counter arguments where people say that, hey, when it goes bust, it's really going to be a lot of tears everywhere?
Hian 41:11
Well, first of all, there's going to be tears. Let me tell you, hey, a lot of people are going to lose a lot of money investing in AI startups. Let's just be very clear about it. This has happened all the time because if there are four amazing AI startups, only one's going to be the big behemoth. If people invest in the other three, they're going to be tears.
I think the question you ask is: is this capacity going to be used and how fast is it going to get used? And I think the answer right now is it is totally going to get used and it's going to be there for a long time. The question of course is: is it going to be overbuilt? That's a very hard question to answer. So I think generally the challenge is not whether it's going to get overbuilt but what you're building right now becomes obsolete faster than you think it is.
So I'll give you an example. A lot of people started up, say for example you were going to generate tokens for a specific large language model and you use some GPUs and you model it out and you say: oh my gosh, I'm going to make a lot of money once it's all installed. But by the time you install, you put everything in, the next version of the GPU comes up and it's cheaper, it's faster, and then the price of the token drops. We've seen this. The prices of token economics are dropping like an order of 10x every like 12 months. I mean, don't quote me on this when we're on the podcast, but something crazy like that.
So then your original business model is screwed. I think that's what's going to happen. People, there are going to be situations where people have an optimistic view about how much a service is going to cost, but just by definition it's going to get cheaper faster.
So I'm a big fan of investing actually interestingly in the chip side because I think number one, compute is going to be absolutely needed and I feel like there are some companies in the world who are driving the cost dramatically and there are many ways of doing that. Inference is a very different value proposition than training and stuff like that. For the record, I just believe that NVIDIA, while it will be a very valuable company, won't have the market share that it has right now in two to three years' time. And it's actually ironically because there are certain markets which NVIDIA doesn't want to go into. There'll be other inference models, other inference abstractions, there'll be other deployments. There are a lot of companies who just want to spend $20 million to build a trading system or a bank. They want to spend $30 million to build an on-premise agentic customer service support. Maybe NVIDIA's in the business of rolling out huge abstractions. There will be customers and there will be space for more than one player. And so that's where I think it's going to move.
Keith 44:06
Naturally, the follow-on question would be: who are the other players that you see coming out? I mean, I spoke to someone like Stephen Witt who's looking at NVIDIA and his argument was that, yeah, you would have other players, but maybe they won't be able to compete with NVIDIA. And your argument is that maybe it's just a completely different segment altogether.
Hian 44:24
So I watched that Stephen Witt video and I think anybody who really wants a great abstraction of what Jensen's all about should really re-watch that. That was a great video. By the way, Keith, I think that the challenge with understanding this is you need to have a technical background and then you really understand that some of the representations he made really need to be teased out.
So one of the representations he made is this idea that CUDA is dominant. CUDA is the moat. And I would disagree. But on the other hand, he makes a very valid point which I think is not going to go away. And that is as follows.
Today if you're an AI engineer, you actually don't work on CUDA. You work on PyTorch. Have you heard of this thing called PyTorch? So PyTorch is the language that everyone thinks about. And then after you program in PyTorch, you accelerate the thing down to the GPU using something like CUDA. So PyTorch has got 4,000 operators. Plus, minus, times, divide, those are called operators. It has 4,000 operators, 4,000 mathematical instructions that you can call on Python. Of that 4,000 mathematical operations, a thousand are really critical when you're doing AI, you know, matrix multiplication, all that kind of stuff. And so CUDA accelerates those 1,000.
So the best way for your audience to understand this is like think of a GPU as your personal assistant. They can do every single thing. They can help you schedule your appointments. They can get your coffee, whatever. But then maybe 50% of your load is actually travel bookings. In that regard, your personal assistant calls a travel agent or uses Expedia. There's a specialist travel agent. So 50% of that workload goes to a specialist thing called a travel agency. That's exactly what CUDA does. 50% of a CPU workload when it's AI calls for matrix multiplications. They just devolve it down there. And CUDA is the software language, is that software library that is connected to Python and PyTorch that runs the GPU abstraction.
Now AMD has got the same thing. It's called ROCm. So if you wanted to use AMD, you just go install ROCm. So you load Python, you install ROCm, and to you if you just work on Python, it doesn't matter. You program in PyTorch, it doesn't matter how the engine runs. Is it an eight-cylinder engine, 12-cylinder engine, rotary engine? You don't care.
But what Stephen Witt made as a very interesting point is that the next version of PyTorch, every new version of PyTorch, which is a consortium and they figure it out, the first version they roll out is always the CUDA version. And because CUDA is the dominant language, if you're sitting on AMD and then you go: oh, there's a new version of PyTorch, and you go: I have to wait for the AMD guys to update it so that I can use the new version which has got the sexy, I don't know, mixture of experts and all that kind of stuff. But the people on NVIDIA are having a great time. So I think that is what is meant by the CUDA lock-in.
However, if it turns out that you kind of like AMD because AMD is cheaper equipment and price performance is better and that kind of stuff, actually you can live with AMD on the inference level. Do you know what I mean? So for a lot of times if you're trying to do new cutting edge things, NVIDIA's great. But if you're going to just run something on a price performance, there are many other platforms. So that is what I mean by other people who are trying to do that.
Okay, so we've got to call out a few names. In this neo-cloud, neo-GPU area, there's a couple of common names. But before we go there, let's just talk about some of the names that we always forget about. I always forget about them. So you've got Intel, you've got AMD, you've actually got Google. So Google's got its TPU and Google's starting to actually allow people to use TPU as a service provider. Amazon's got Trainium, but they're not doing that yet. And then you've got people like Grok, but the problem with Grok is they just sell tokens because you can't really access their hardware. There are specific reasons why they don't do that. And then there are companies like Cerebras which is doing a completely new thing where the whole computer is the whole wafer and it's all very closed in. Sambanova.
But my favourite one which I always talk about is Tenstorrent. So Tenstorrent, started by Jim Keller, and what they're taking as an approach is fully open sourced. So today if you bought a Tenstorrent card, you put it in, you download from GitHub, everything, you can basically control right to the lowest level the kernels that write to the Tensor NPU infrastructure. It's not GPUs, NPUs all the way. You can push it all the way up to PyTorch, but you better be a hardcore engineer. So for 99% of people, they don't do that. But if you're going to run a neo-cloud and you want to go for perf per dollar, you want to go for efficiency, there's an argument that AMD, there's an argument that Tenstorrent can also provide that.
There's an argument that you run a hybrid cluster with NVIDIA, AMD, Tenstorrent and you drop the workloads that work best for each architecture and you get a faster, cheaper and better service. So there's a lot of new ways of skinning the cat and it isn't just a pure green infrastructure, pure NVLink 72. That's all going to change. That's my thesis. So the future is hybrid. The future is heterogeneous. And that's not because NVIDIA is screwing up. It's just because the AI demand is so great. There's going to be many ways of skinning the cat and NVIDIA by definition doesn't want to have every single abstraction. They will focus on the abstractions they're great at that are the most profitable and they're going to deliver that at pace.
Keith 50:27
Yeah. On that note, I wanted to ask you a little bit about Tenstorrent. Jim Keller is known as, I think some people call him the chip ronin. Because his job, I mean, for my friends and I spoke to people in this space, they say that every time a company is failing, they will get him in and then he literally reverses their fortunes. So the famous example is how he literally revived AMD.
Hian 50:54
Twice. Twice. Yeah.
Keith 50:57
And now he's someone that you are working with as well.
Hian 51:00
I met Jim three years ago when he came to Singapore. He was looking at a wafer fab and then I spent some time with him. I famously cooked him a pizza, so he remembers me and so on and so forth. Again, OpenSpace Capital, we realised three years ago that we better be good at AI. If we're not good at AI, we're not going to understand it. So I spent time with Jim and some of his ecosystem in Silicon Valley, especially on the compute level. Not just him, but people like Raja Koduri. He's the godfather of the GPU at AMD. I met people like Debbie Marr at Hailo Computing trying to create the fastest RISC-V CPU. So it's not just Jim, but it's a whole group of people because by definition we need to bring that knowledge back into OpenSpace. Even just understanding the abstraction helps you understand the agentic, it helps you understand the whole thing.
But Jim's an interesting person because he is really a first principle thinker. First of all, he's a relatively introverted guy. He thinks very deeply and the first thing I notice about Jim is that his signal-to-noise ratio is relatively pure. In other words, he doesn't say anything until he says something. What he says is usually quite profound and you can watch him on YouTube. He's the only guy who not only did a lot of good work at AMD, but he's the only guy who's worked for Steve Jobs, because he did the Apple silicon, and Elon Musk. He actually started the Dojo project. So there's a Lex Fridman podcast. You can search for it on YouTube. The difference between Steve Jobs and Elon Musk as a manager, and Jim Keller will actually give that abstraction.
So I spent time with him because he's connected to my overall thesis that if you want to understand something, you've got to go to first principle thinkers. So that's why Jim is quite an interesting person. He's never been a commercial leader because he's been a chip designer. So to be specific, he did the K9 chip and the K9 chip was in 1999 for AMD. That was the first chip that was as fast as the Intel chips and that freaked Intel out. There's an argument that for the rest of the next 10 years, Intel was just super focused on killing AMD. So they sold the ARM division to Marvell and so they didn't do the iPhone chip in 2005. Intel did a lot of things which people have said: oh my god, you should have totally done it. But that's because Intel was so focused on making sure they did not lose the x86 war against AMD. That was Jim.
And then around 2010, 2011, if I'm not wrong, Intel actually succeeded and AMD was in a lot of financial difficulties. So Jim came back and did the Zen architecture which today has put AMD back on the map and has all that capability. So he's always been the chip designer but he's never been the CEO of a company. So Tenstorrent is the first time that he's the CEO of a company and he's not a young guy. He's above the age of 60. So in this business, he's got the chops and that's why I like spending time with him. I learned so much from him and the people that are around him and the technical excellence and the ability to always think about the future and just really just not worry about what is happening right now but paint a picture of where it's going to be in three to five years' time.
Keith 54:19
I asked about Jim Keller because I thought that if you think about talent in this AI era, a lot of times people look at the CEOs of the big tech companies. They think about the Sundar Pichais of the world, but there is also a Jim Keller of the world. And going back to Singapore, we need talent to thrive in the world that we live in. Regardless of which era we grew up in, for us to survive we need talent and you need talent density. So then the question becomes: how are we going to do it? How are we going to get the Jim Keller-like people to be in Singapore to build those companies?
Hian 54:59
The last thing I'll say is that to run a chip company, you better be a computer engineer. Jensen is one. I think Jim is another one. Lisa Su is one. So that goes without saying. We'll put that aside. I think what you're trying to say is: how does Singapore compete in the world now that AI is the dominant flavour?
Well, the good news is that we don't compete because of oil, because of natural resources, because we don't have any of that. Or because of I don't know, some sort of size of population or size of market. We need to make sure that the best and the brightest in the AI field choose Singapore as a place to reside. And I'll tell you the piece of good news. I believe that that is absolutely happening.
Because of the geopolitics of AI and because how it is so important to win the war, America is the left side and China's the right side and Singapore is really the Switzerland of this whole thing. We're seeing a lot of teams relocate here. MiniMax AI is a great example. So MiniMax AI was originally founded in China, now based in Singapore. Benchmark Capital, which is a very famous US venture capitalist, has invested in them and it's based out in Singapore.
I just met a team of people doing a video model. So like something like Sora 2 or the Google Veo, but they're a team, no guesses, from China. Really good. Have a strong track record in one of the big Chinese tech companies. They've spun off. They're in Singapore. This is not going to stop. We're going to see a lot of these companies.
I think the challenge is to make sure that Singapore is not perceived as a place where we're just a pass-through or we're some sort of geopolitical pawn in this whole thing. That somehow this is actually real and that the models and the abstractions that are created here are not geopolitically connected to one of the two giants, China and America. I think we've got the track record to do it. Singapore has always been able to balance the two large 600-pound gorillas and I look forward to working with these people. But are these some of the most talented individuals and the most excellent in their space? Absolutely. Do they live in Singapore? Absolutely. Will they increasingly live more in Singapore? I have no doubt about it.
Where else are you going to be able to conduct from a neutral platform in the world? You can't do it in Europe. I mean, Europe has got a lot of regulations as well. I'm literally thinking about it. India perhaps, but Singapore is cool. It's safe, good schools. The Switzerland of AI.
Now, if that happens over a period of three to five years, a lot of Singaporeans are going to learn that craft. And that skill transfers there. And then maybe the next iteration may include some really hardcore Singaporeans. And by the way, there are also hardcore Singaporeans who are doing this, but some of them live in the Valley. I met one guy which I was very impressed with and he's doing some crazy inference model which is really cutting edge. He's coding kernels, he's coding abstractions all the way down to the GPUs and CPUs of AMD and NVIDIA. And I was very shocked and he's a Singaporean guy.
I think Singapore needs to understand that we've got really great talent. And just because it's made in Singapore doesn't mean that it's not good. Israel has managed to convince everybody in the world that if you just say Israeli technology, I mean this is very good. This is really, we have to go like: this is Singapore technology. Oh wow. This is very good. It can be accepted. And I think over time, if we deliver, we will have more of that. But the mindset shift has to happen from the Singaporeans because sometimes, see, the world already understands that Singapore technology is awesome. Sometimes the people who really don't believe it are the Singaporeans. And I'll leave it at that. I just don't want to name names. Sometimes it might include me, by the way. I'll be very clear about it. I always worry about whether we're not the best in the world. But I've talked to so many people in the world that actually: opto-electronics research from A*STAR, NUS professors, people migrating here. We really have to start telling people: we're as cool as everybody else.
Keith 59:51
From a previous interview I did with Inderjit Singh, one of the things that I found to be extremely insightful was that it took him 20 years to go from being a Texas Instruments engineer to starting UTAC. And for the hard tech or deep tech, in his field especially back then, starting UTAC as this testing and assembling centre for a lot of these semicons, you really need time to build that technical expertise. Sometimes it just takes you 20 years to build that and then you can build a billion-dollar company out of it. And maybe it requires Singapore to kind of, Singaporeans to kind of understand that you need gestational periods or you need to accept that as part of the game.
Hian 1:00:31
You do and you don't. I'll give you the funny story. 20 years ago when I was a young man in my 20s, I'd go to China and everyone drove very badly because they only got their licences. Nobody had a license that was longer than five years. In that regard, it's beautiful. If you're talented at driving, you're probably going to be the best car driver in China.
So I think in certain fields, like especially in AI model, AI model training, and Stephen Witt actually made a good point, AI was not the cool thing that you did 10, 15 years ago. It was the graveyard of professions. You became a professor. I love that thing from Stephen. And it was only in 2015, I'm going to postulate, only in 2015 that the whole thing started. What I mean by that: OpenAI was formed in 2015. Then it took three years for them to figure things out. Then 2017 was the Attention Is All You Need paper from Google. And then OpenAI came out with GPT-1 in '18, GPT-2 in '19, and GPT-3 in 2020. By that time, people were still saying: I don't think this is really going to work. And it was only March of 2023 that ChatGPT-4 came out. And if you chart it, and we've done this at OpenSpace Capital, the CapEx just goes like that.
So my joke is always: if you invested in NVIDIA before March 2023, you were lucky. You weren't smart. But my point being is that in the field of AI, a 31 or 32-year-old guy or girl who actually focused on it when it was not cool is probably the world's leading researcher right now. So it's different from an industry construct than the person who's worked for TSMC for 20 years and so on and so forth.
What's interesting is in the chip world, the Jim Kellers of the world, all those kind of people, they're all above the age of 50. When you go to Tenstorrent, all the senior leaders, they're all above the age of 50. When you look at OpenAI, everybody's all below the age of 40. So I think that's the difference currently in the construct of the industry.
Keith 1:02:47
There's a point you made which is that you have to do things that are uncool. The challenge of being in Singapore is that once you've established yourself as a financial hub is that financial services tend to over-index on sucking the talent up. So there's an opportunity cost of going to, of having yourself set up as a financial hub, which is also why for example London is a financial hub but it's not known to be where all the great tech companies are born. Can Singapore balance that contradiction?
Hian 1:03:18
Look, one of the great things about Singapore is that we are actually essentially a centralised economy. And I'll prove it to you with the following statistics. I just went to Hungary. I went to Budapest. I went to talk and as part of the talk I did some research and ChatGPT would tell you that 40% of Hungary's GDP is controlled by state-owned enterprises. Then when you type in Singapore, they go: only 15% of Singapore's GDP is controlled by the government services. And I went: oh, that's quite cool. Oh, by the way, think about it. Singapore's GDP is like $600 billion. The government's budget is like $100 billion. So it does make sense, it's about right, 15%.
But what we forget is that Singapore Inc. also owns DBS and Singapore Airlines and CapitaLand. And I just thought: actually if you add all the corporatised state-owned companies that Singapore has, we're probably around 40%. So my point being is that Singapore is not really the free market environment where five families in Southeast Asia control a country or 100 people in America are equally important as the government of America. So Singapore Inc. and the Singapore government becomes a very pivotal decision maker in figuring out what's the next thing Singapore invests in. Would you agree?
So if Singapore decides that AI is very important and that we need to get AI as a core part of our strategy, then we could be able to execute that. But it takes the government to decide: well, finance is cool, but over the next 15 years, we better invest in AI-related things. And at first, in the first three to five years, people might be scratching their heads or maybe the population might be going: I don't know why you all do this AI thing. I guess the AI thing got no revenue. I think it's a bubble. But Singapore leadership has to have this belief that: that's fine. You'll figure out that we got it correct in about 5, 10 years' time. And by the way, some of the projects that we invest in will not succeed. And we have the resilience and also the belief that we're going to actually blow up some resources by going down wrong paths, but it's worth it for going down that path because if we don't, we lose. It'd be very hard to play catch-up.
And so I think just from that abstraction, I think it can be done. But it's not going to come from the Elon Musks of the world necessarily. I think there will be, obviously in Singapore, I think Singapore is very free market of course. You look at the Singapore stock exchange now, it needs to be revitalised, the government comes in and says: we're going to put together this $5 billion EDBI, everyone jumps on the bandwagon because we've got that reputation and the free market flywheel happens. But I think in Singapore the government does provide a lot more influence than many other countries in the world and I believe that they have identified AI is important because we do have a long-term planning approach to things. So I think it's possible.
What we need to avoid is over-optimising on success. So what do I mean by that? Don't take it from me. Take it from Clay Christensen. So Clay Christensen is the professor that wrote that famous book about the S-curve. And what he said is that any new technology by definition usually comes out less efficient than the current existing technology, but over time it will become better. The example he gives is the 1.8-inch hard disk drive. The hard drive, like five-inch, three-inch, and then somebody makes a 1.8-inch hard disk drive and nobody knew what to do with it until Steve Jobs came up and invented the iPod, which was the perfect thing for the 1.8-inch hard drive. If you were the top sales guy at Seagate, you were not going to be selling the 1.8-inch hard drive.
So again, similarly in the Singapore government context, we need to always invest in a certain amount of speculative things that we know in time will be the new S-curve and have the confidence to know that of the 10 projects that we do that are speculative, eight will not succeed. And don't be too hard on ourselves because it's worth being in the forefront of the two things. And this comes back to the Singaporean mentality. I was from Raffles, so must always get all the answers correct, always get the things correct, don't go and do this, don't go and do that. I could have easily been a banker, but I went off and started a TV network. By definition, I think Singapore should not over-optimise on efficiency. We should be strong enough to know that we have to make bets which currently at this point in time look unreasonable.
I fear if we don't do that, we don't keep our vibrancy in Singapore and we fail as a nation. That's it.
Keith 1:08:21
Yeah. The last question actually I have for you is the question I ask every guest, which is: knowing where you are at now, if you had to give one piece of advice to fresh graduates entering today's working world, what would that be?
Hian 1:08:40
I wrote down something for you, Keith. All right.
I think you do five things. All right. So I'm going to take my glasses off. It doesn't hurt to be ridiculously hard-working. You're in your 20s. Just be ridiculously hard-working. Work on the weekends. When I was an investment banker, I worked a lot because you're honing your craft. If you can learn five years of skill in two years because you're working 80 hours a week and you're just very good at PowerPoint or Excel spreadsheet, just work ridiculously hard. Okay.
The second thing is your reputation is everything. And what do I mean by this? I've known some of my peer group in their 20s who took shortcuts and maybe scammed one or two people and so on and so forth. And now they're in their 50s and nobody wants to do business with them. This is really true. In your cohort, Keith, there'll be one guy, I guarantee you, who's not in that WhatsApp group. They say: don't talk to that guy. That guy's a scam artist. And when you talk to him, most of the time that person doesn't actually know that everybody else doesn't want to do business with them. So they lament and they go: oh my god, life is so hard. It's really just for them. So your reputation is very important.
The third thing is actually likability. When I was young, and I still think I am, I'm just an arrogant guy. I speak very honestly and it comes from an insecurity. Don't be that person. Just be likable. And I think when you do all those three things, just remember the most important thing as a young person is to do a good job for your boss. And irrespective whether your boss is a good boss or a bad boss, just do a good job for your boss. You then create the final reputation which is that you're a reliable human being. You can deliver on whatever the task at hand is and then in your network you just be known as somebody that you could do business with.
And when I transitioned from being an entrepreneur at Asian Food Channel and wanted to be a venture capitalist, there were a couple of people who said: you know what? You've got no track record, but I'll back you. And these were actually people around the same age as me. And then a couple of people were like 10 years older than me. And so I think that if you do all those things—work really hard, be reliable, be likable, don't cheat anybody—invariably over time, people want to do business with you and then you will be successful because that's how success gets generated. It's never overnight. I've never seen it really. I've never seen sustainable success be generated overnight. That would be my advice to young people.
Keith 1:11:20
It takes years to build an overnight success. That's literally what I write on my website. The average time it takes to be an overnight success is about 10 years.
That's why with that, thank you for coming on.
Hian 1:11:33
I hope that was helpful.
Keith 1:11:33
Thank you for listening to today's episode. If you're watching this on YouTube, please consider subscribing and turning on notifications for whenever new episodes are out. If you're on Spotify or Apple, it would help us greatly if you could leave a five-star review on those platforms. Once again, thank you for tuning in to the Front Row.