00:00So what parts of the supply chain do you see right now, the choke points at this point?
00:06Well, I would say it's really broad across the whole supply chain. Of course, Asia is a huge
00:12part of the AI and semi-supply chain, so a lot of Asian suppliers are really benefiting. Of course,
00:16you mentioned TSMC, SK Hynix, and Samsung in terms of logic and memory, respectively. But
00:22there are all these components. It's such a big supply chain, various materials upstream.
00:28One instance is what we're seeing that's interesting is AVF substrates, right? Substrates
00:34are the, they sit underneath the chip, they bring out signal and power out of that chip. And that's
00:41dominated by a lot of Asian manufacturers, such as Ibiden and Shinko in Japan, Unimicron, Nanya,
00:48Kinsis in Taiwan, and Korea as well, Semco. We're seeing a lot of demand driven not just by units,
00:56more AI chips being shipped, but the content value is increasing dramatically, right? So
01:02each chip, the chip is getting larger, needs more substrate area. It's also getting thicker.
01:09It's more complex to manufacture, yields are worse. So that's really causing dramatic tightness in that
01:14sector.
01:15To your point, I mean, you are seeing the dominance of just certain companies, especially
01:21perhaps even in certain regions and countries, near monopolies in these sectors where you have
01:26very important, critical materials and components, right? Where are you seeing those the most at this
01:32point?
01:33I would say, again, there's just so many. But another area is the raw silicon wafers, especially for the
01:41wafers used in leading edge logic manufacturing. They're called epitaxtual silicon wafers. Again,
01:47that's something that's dominated by Japanese companies, Shinetsu and Tsumko. And now we're
01:52seeing, again, because there's so much more logic demand, you need more of the raw silicon to etch the
01:58chips on. So we're seeing demand really outpacing the supply ads. And that's also flowing into better
02:05pricing for these players.
02:06So when you see right now the development of artificial intelligence and build out of data
02:11centers and what you project a trajectory of how this evolution could look like, where do you see
02:17the potential risks here?
02:20I would say the risks are, again, sort of lack of supply.
02:25In which parts of these components that could potentially see much more demand in the future?
02:30I would say probably it still goes back to one of the main bottlenecks we see is memory. DRAM, we're
02:38seeing
02:39just very, very limited supply additions. And there are structural reasons for this. But the ability to add
02:45both conventional DRAM supply, HBM supply, we think that's the sort of most difficult thing to ramp that we see
02:54is
02:54the thing that's really constraining your supply and demand is much higher. But I think that's going to be what's
03:02constraining all these shipments of the AI hot service.
03:05Is it the complexity of the manufacturing itself or also the lack of the critical minerals that go into it?
03:12I would say a lot of it is the complexity of the manufacturing as well as a long lead time.
03:18So, you know, really build a fab. To build a fab, it takes it's a multi-year process. You need
03:23to build clean rooms.
03:24And then also that goes down to you have a clean room, you need equipment. And there's only so much
03:31limited equipment supply as well, such as what we see in lithography, which is from the likes of like ASML.
03:38There's limited number of scanners that they can really manufacture each year.
03:42What about the workforce? Do we have enough talented engineers that can work around the clock to make all of
03:49this, to satisfy all of this demand out there?
03:52I'd say, again, that's another area that's really, really tight. There's only so much, you know, skilled labor out there.
04:01There's only so many, so few engineers that can really, you know, do all these things.
04:06So I think that's another critical bottleneck. And I think it's a combination of everything, right? With all the equipment,
04:14whether it be clean rooms, people, I think everything is very tight.
04:18But I think what we see is that the supply chain is also, you know, very good at adding capacity
04:25and especially when demand is so strong and when there's money to be made, I think the supply chain can
04:31really work to add capacity.
04:34Tell us a little bit about that. Is the concentration right now growing or diminishing, given that we're seeing more
04:42efforts globally in order to diversify supplies, whether it's the raw materials or even the components and who supplies them?
04:51Yeah, I think there's definitely more and more sort of spillover from the dominant players into, I'd say, players that
05:00probably didn't used to be as big in the industry.
05:04But because if they have spare capacity available, because your demand is so strong, that demand flows over to them.
05:12So we're seeing a lot more sort of outsourcing from or, you know, suppliers switching from the traditionally dominant players,
05:19such as, you know, away from TSMC to Intel and Samsung, for instance.
05:24When it comes to the bifurcation that we're seeing between the tech rivalries of the West versus China, what implications
05:30does this have long term for the supply chain?
05:33I think when it comes to China, they're obviously very, they emphasize having their own domestic supply chain in the
05:42same way that the U.S. is doing now.
05:44So they're, you know, ramping up their capabilities, building out logic, memory, all the components, all the material supply.
05:52But again, I think the quality is not as strong as the established global leaders, but they're certainly working very
06:01hard, investing a lot behind, ramping up their own domestic supply chains.
Comments