No Priors: Artificial Intelligence | Technology | Startups
00:15:35 · Elad Gil
[00:15:35] S2: talking for a long time about eventually hitting a point of resolution where you can't really miniaturize things further. Like, the line width just gets too small to be able to keep going. When do you
00:15:47 · Lip-Bu Tan
[00:15:47] S0: think we actually hit that limit? Good question. So I think I can see, right now we have 18A, and then now going to production of 14A, I can see ten and seven. And so I think that path, I think we can get there, but couldn't it be more and more expensive and more difficult to do, and that's why we need partners. We cannot just do it ourselves alone, partner with a subscript vendor, partner with equipment vendors so that make sure that we can really drive those yield and performance. Yeah. And then the other part also really become the bottleneck is packaging, the advanced packaging. And so we all know about cohort by TSMC. Now we have a really good one called EMIP tea that is really next generation. I had to make sure that you become able to do in the production yield that meet the customer requirement. And now CMOS starting to run out of steam like you described. So right now, I also look at some new materials, so it become, going back to the material size or the chemical table. So gallium nitride, silicon carbide, and indium phosphide. So I invest in all three, and then looking at some of this new material, how can we really drive that? And then in terms of packaging, I starting to invest into glass. Glass is a very good heat insulator. So I invest a venture site called 3DGS, then I realized that Intel, we have like 1,000 pattern on the module. So how the, you know, subscript and the module put it together, and then we just announced a big program with Indian government to manufacturing in in India, plus in US and New Mexico. So I think this advanced packaging, very important. I also starting to look at artificial diamond, and that's another very good insulator. So I also invest into diamond foundry, and that's something is a next generation to look at. So new material, new subscript material, and new design methodology to drive that. So one thing good about being an engineer is you're always hitting the wall, then you find a way to either jump over the wall or you work around the wall, and then to get to the better result. And that's what I have been long time as an investor and building semiconductor from the EDA tool to design to manufacturing. It's kind of nice to have that experience. Now I can help find a way to make a small contribution to the industry. Yeah, and
00:18:16 · Elad Gil, Lip-Bu Tan
[00:18:16] S2: I mean, it's very exciting. And one of the reasons I'm asking about it as well is, to your point, there's always some things that you can vent around, but there are also physical limits where once you hit seven angstroms or whatever the limitation is, you start to run into [00:18:27] S0: New material. Yeah. You need to [00:18:28] S2: find new materials or find other workarounds. Yes. And then the interesting question is, and we've been talking about this for a long time. I remember twenty years ago, people were talking about how we'd eventually hit a point where we ran out of space on this, is do you run into some sort of asymptote that actually normalizes performance across different foundries or not? Yeah, good question. In terms of Moore's law, you say double and then [00:18:50] S0: the power and the cost. And then you can double the performance, but you cannot double down on the cost area. So those are the things you have to give way unless you find some new way of material, [00:19:05] S2: new way of design,
00:19:05 · Elad Gil, Lip-Bu Tan
[00:19:05] S2: new way of design, [00:19:07] S0: that become material science, I starting to hire more people in the material science. So that is kind of innovation in our area. How can we do that? And I still remember eighteen years ago, and I I still investing in semiconductor. And, actually, most of the VC firm, some of them are very nice tier one venture firm, a good friend of mine. And I initially the partners meeting, the whole partners in the room. Then after I talking about semiconductor, half make excuse to run out the room. Then eventually, the other half, they said, available, do you have any software service? So then they even left with only two sympathetically listen to me. So it's kind of the history have changed. And now semiconductor, if you look at it, Jensen is a 5,300,000,000,000 market cap company, and then Broadcom and TSMC is 2,000,000,000,000 market cap company, and Lisa Su, my good friend at AMD, is almost $800,000,000,000 and I'm close to $600,000,000,000 So in some ways, kind of semiconductor become hot again, and it become essential because fifteen years, twenty years ago, when I invest in semiconductor, no VC want to join me except some of the big corporations like Samsung, ARM and SoftBank and others, and investing with me. And then now I'm starting to see a lot of VC like to come investing in semi, so I'm very happy. Given
00:20:33 · Sarah Guo
[00:20:33] S1: the enormous interest in investing in this area that used to be considered too hard. Right? Yes. What do you think I mean, you've been a venture investor with Walden for a very long time as well as an operator. You know, the general fears I'm just gonna list a bunch of them. The general fears have been it's very capital intensive, and you should tell me what I'm missing. It's very unpredictable in terms of, you know, shipping a design that works missing tape out. And you need to understand the workload very well. I think there's another, which is just like it's very high risk for the customer to switch. Right? Think, you know, we've been involved in companies together where, you know, there's a design win, and then there's still the question of like scaling order volume. And then there's a cyclicality, right, of, you know, you build hard manufacturing capacity and demand may change or not in any given year. What is your view on how a bunch of, you know, what makes it hard as an industry and then the secular demand growth from a bunch of different areas, right? So you have the recognition of how important a more diverse supply chain is. And then you have this like explosive demand growth on the AI side. How do you you're still an investor, and then you're making the biggest bet ever. Like, go be CEO. How do you, like, think about these different risks and advise others about where to invest in this supply chain? I realize that's a very large question, but just given your history with it, I think there's a lot of like YOLO action of like, there's a memory shortage. Buy memory stocks as well as just an unwillingness to take on things that have a ten year timeline, like material science.
00:22:20 · Lip-Bu Tan, Sarah Guo
[00:22:20] S0: Good. You have quite a broad range of questions. Let me try to explain that. So first of all, I think the venture capital startup is in my blood, and I really enjoy it. And so I think this is not time to brag about it. And so there's some good exit. I still have 159 IPO, 126 ms and A, and that includes semiconductor. Just break down to semiconductor, I invest over the years February is in US. So what I usually look at some macro [00:22:57] S1: Just to be clear, that's incredible.