00:00It's been a great run for Chinese chip designers as there's been a lot of state support for
00:05indigenous innovation and getting your own chips as opposed to having to import the most expensive
00:10and most advanced accelerators for the AI world from the likes of NVIDIA. Are you concerned though
00:17by the valuations? I mean some of the valuations of these chip companies are quite high right now
00:22including yours. Yeah. So there's an important day September 24, 2024. That day China government
00:34certainly have many policies trying to help the stock market because they cannot have housing
00:42in the bubble. You give the increased little salary just in the bank. You also have internal
00:48demand spending. They think maybe the only way to make the stock market more healthy.
00:54Yeah. Since that day, less than two years, our stock increased 10x. 10. Right? It's a high-tech company.
01:04Right? It's not easy. Now let me explain to you why is that. Okay? Why 10x? Is this too much
01:11or you
01:11have another growth? I need to explain this. You just say we have too much. I can tell you
01:18you know block on stock, right? Increase four times, revenue four times. Marvell maybe revenue
01:26two times, right? What do they do? For the AI, there's two kinds. One is general purpose, GPU, NVIDIA,
01:33right? Doing very well. However, not everything is general purpose. You want special purpose.
01:38AI, like Google TP, right? Now more and more, the other part getting more, maybe more than 50%
01:50reaching the market share, especially more to age, right? They call AI ASIC. Yeah. So who's helping
01:58those TPU, Google, Meta, block on, right? So that's how, now in China, also have both sides,
02:09GPU, but we don't, we have many GPUs, not one. Hong Kong Cup IPO, Shanghai Cup, they're quite high
02:16variation. And there's many GPUs on waiting to be IPO. So in block on didn't become NVIDIA, what they were
02:26doing
02:26is not, nothing needed to block on hell. Right. In China case, almost every GPU company is our
02:33customers. Right. Maybe we could do IPs or do some design. Our models, we never branded service.
02:41Either give you IP or doing the chip design, but it's your chip. But on the other hand, there's other
02:46side, special purpose. We call cloud service provider. Used to be internet company, right?
02:53So those people, also our customers, called the AI ASIC. The agentic AI. Yeah. The agentic AI,
03:02like open claw kind of services, right? Yeah. So the chip, chip between, I would say the chip
03:07has two kinds. One is general purpose. Yeah. Like NVIDIA. Yeah. China has many. Yeah. No, no,
03:12no, sorry. The NVIDIA, right? The general purpose. Yeah. China has many general purpose,
03:19GPS, right? But this part, the other part, AI ASIC part, like where the block on hell, like
03:27the Google TP type. We also have that type of cloud service provider. We also service that.
03:35So the opportunities are great. What is that going to mean for profitability? Because I talked
03:39about valuations earlier. I think you, according to Bloomberg data, you have an estimated price
03:44to earnings ratio over 12 months of 318 times. Yeah. You were not profitable in the first
03:50quarter. Net loss of 340.8 million RMB. Yeah. Yeah. But revenue is up 114%. Yeah. So how
03:56are you going to turn that to profitability before you list here in Hong Kong? Yeah. I can tell
04:00you about this. End of last year, we have a call backlog, right? The service is backlog. 715 million
04:10dollars backlog. This year, first, January 1st to April 2009, the first four months, new booking, 1.1 billion.
04:23Wow. So, so, so, so, so, so you know, what's the booking? If I have design win, then you know
04:29exactly how to
04:30look at revenue. You will place all the, I know when exactly. So mostly 80, 90% will be coming
04:36to revenue
04:37within one year. So this can kind of indication. So the thing is as a platform company, key thing is
04:43a skill.
04:45Platform is not a win big deal. We have 100 chip in production, right? So, so IP, we have 460
04:51customers.
04:52Chip, we have accumulated 350 customers. So scale is important. When you reach to scale, right?
05:00Non-profit becomes profit. If you not grow, you get profit and non-profit. So sometimes reach to a scale
05:06is very important.
05:07So now we're in the turning point for that. So you're in a turning point. So how much does, do
05:13the Chinese
05:14large language models and also the agentic AI suppliers, how much do they need the most advanced
05:20chips from the likes of NVIDIA? Yeah. As the indigenous chip making, like you, I mean, also
05:26we saw Huawei unveil the logic folding technology, you know, this tau scaling law, essentially replacing
05:32Moore's law to get more speed across the chip set than smaller size. Okay. I answer the two questions.
05:40Two questions. So first of all, like I say, when you go to the age, it's more and more ASIC
05:47rather than
05:47general purpose. Right? You have to see the trend. Okay. Google GPU, right now getting more market share than
05:54NVIDIA. Yeah. Even total speed, right? You see the China same move. So I say, my, I different from
06:01blockchain, I sell both sides. Yeah. Blockchain only serve the NVIDIA side, right? Right. We serve both sides. And we
06:08grow. And the age is very important. The age. Okay. I
06:16start using a tree. The root is computing power. The trunk is training. It's big model. Yeah. But you cannot
06:23have a leaf from
06:24trunk. So you have a branch. Right. And fine tuning medicals. Right. So it's a vertical model, right? Right. Small
06:33model or something. The fine tuning the, the, the, the financial, right, whatever it is, right? Right. Education. So this
06:41is called branch, fine
06:42training. Then you grow the leaf, the inference. Right. So two years ago, we say, everyone talk of a bigger
06:49model, building a big
06:51center, data center, training. I think fine tuning and the inference eventually will exceed. Eventually. After we build a
07:01many, many, many cloud training. So this is the truth. Right now we see that. And you make money on
07:07the age. If you don't, like
07:09without cell phone, we cannot internet coverage. Right. Yeah. Money is this device. So the age is very
07:15important. Okay. First I made the point. Yeah. My age is the air phone, air PC, but I'm talking about
07:22air
07:22glass. Yeah. Air glass. See people four years ago, we did the ASIC for Google. Yeah. Air glass. Very optimized.
07:32Right. Right now is one bigger, one smaller. If you know air glass. Yes. One from big is from cell
07:37phone
07:37change. Very powerful, but power is the power. You cannot turn on all the time. Right. And then if I'm
07:43a watch
07:43like low end. Yeah. It's very low power, but not enough. They put it one bigger, one small. Obviously not
07:49optimal, but no one wants willing to make a one chip because they don't know how many pair you sell.
07:55But now is the point. They call ASIC customer. Yes. Well, I think we did four years ago. I can
08:02write on
08:02a lot of people knows which glass can sell at least 500 million pairs. Once you know that they're willing
08:08to pay a nine. Like we do service unit, some will pay to that. Now is the time. So this
08:13age device is very
08:15important. But you realize the reason of Google announced some air glass, no display. Yeah. But
08:22again, I guess back to my point is like some of the innovations in the chip design by the likes
08:28of
08:28Huawei and others, like the logic folding technology. How much of a game changer is that
08:34potentially for the evolution of China's indigenous chip design? And I do need to, before we wrap up,
08:40I do need to ask you about the IPO here in Hong Kong. Okay. So I think this, yes, you
08:45shrink devices
08:46one wrong. There's another way, you know, this is just the biggest chip as you can, right? Actually,
08:53you make the 3D, 2D. You fold them instead of 2D. Then 3D also have power, the heat dissipation. Yeah.
09:00It's not for free, right? If you stack them together, heat dissipation also. But that's the way. So,
09:07so you used to call it systeming on chip, later called system in package. Let me open your multiple
09:13pieces. Yeah. Like covers or whatever. So you need to do two and a half of these stacking,
09:183D. This is called system in package. Yeah. It's not in package. You don't care how inside they're
09:24doing. 3D stack or two and a half D. Right. So that's one. So, but I think we need to
09:30point out,
09:31you need to make sure the power dissipation is not for free. Dissipation. Because you stack together,
09:36especially if each layer is not most advanced process, power dissipation also is power. Wayne,
09:43Wayne, we're running out of time, but I need to ask you about the Hong Kong IPO plan. Some
09:46reports about upwards of billion dollars you're going to raise here in Hong Kong. When do you hope
09:50to list? Yeah. There's some pressures in the market, bubble concerns. Yeah. When and how much are you
09:56going to raise? Yeah. So, so the thing is, right now, we're up to 10% of 1 billion. Not
10:02necessarily.
10:02Hong Kong, you put up 1 billion. Yeah. Right? With Hong Kong, that's the thing you can always refinance,
10:07you know, one time. But I think that the bubble that I explained to me, if compared currently the
10:13market we address, right? Yeah. I think we just started China. I don't think China has that much
10:19bubble compared with other places. So, so the advantage clearly is more international. By the way,
10:25our 35% revenue from outside China today, 98% in China. Gotcha. So for me, for me,
10:34is less export control power. So, so the thing is, they definitely provide more international
10:41market, right? So we already- 35% overseas revenue. Most from US.
Comments