00:00What was yesterday? Was it a look for a safe haven trade or was it something a little bit different?
00:05Was it just that we had sold off enough and it was time to buy the dip?
00:08I think it's time to buy the dip. It looks to me, it has all the earmarks of a classic
00:13position unwinding,
00:15exacerbated by single-stock levered ETFs in Korea.
00:20And I think it's run its course roughly 20% drawdown in a lot of those names.
00:26And the Philadelphia Semiconductor Index is still well above levels that people would identify as technical support.
00:32It's probably about 7% below current levels.
00:36So a lift here doesn't surprise us and I think it's an opportunity to add exposure.
00:41But we had, you know, risen so much.
00:44I mean, some of these names are up several hundred percent since the beginning of the year.
00:47Would you not be a little bit nervous getting in even at these lower levels?
00:50So I think if you don't have to buy Micron and the new SK Hynix ADR and, you know, it's
00:58just not even necessary.
00:59The entire AI theme sold off.
01:02And some of those names are our highest conviction names in networking equipment,
01:06specifically the optical transceivers like Coherent and Lumentum.
01:10So opportunity to reload there.
01:11I think they're just a casualty in this position unwind and that's an opportunity.
01:16We also think that now is a decent time to expect the hyperscalers to show an improved return on invested
01:23capital.
01:24I think that's what this whole thing is about in the background, not just positioning unwind,
01:28but this concern or worry that during Q2 earnings season, these guys could lower their 2027 outlook for CapEx.
01:36I do not see that at all.
01:38Don't see it anywhere.
01:39And I would expect, if anything, it has an upward bias.
01:42At some point, they have to, right?
01:44Does there come a point for the market where it's not a problem that they dial back that CapEx spending?
01:49Yeah, I think maybe over time, particularly if you do it gradually, like, you know,
01:53if they are able to sort of widen that gap between operating cash flow minus CapEx.
02:00And that's really been the issue, right?
02:02That the semiconductors have underperformed, excuse me, that the hyperscalers have underperformed the semiconductors.
02:08And that looks a little bit like, I don't think a lot of people are talking about it, but it
02:11does look a little bit like the 99-2000 period when you have the communication services stocks lagging well behind
02:18the communication equipment names, right?
02:21The Nokias and the Ciscos and so forth.
02:23So that's the similarity.
02:25And you can maybe point to that and say that's a little eerie, right?
02:28And that bothers me.
02:29But the time is coming.
02:31Like, if it's not this quarter, it's next quarter.
02:32And I would expect it to happen sort of naturally.
02:35Token costs are falling.
02:37Token prices have stayed where they are, right?
02:40And that spread, that margin, leads to incremental revenues.
02:44It leads to incremental margin improvement.
02:46How do you know where to put your money or put your profits in the AI trade right now?
02:50Do you just, you know, splatter it across the AI trade?
02:53Or you picked out some areas.
02:56Why those areas?
02:57Is it a valuation thing?
02:59Is it that they're, you know, going to be next up for scarcity?
03:03It's, well, there's a scarcity component to it.
03:06We're talking about the optical transceivers and then there's a roadmap to co-package optics and so forth.
03:10The people, the company that controls that tech horizon is NVIDIA.
03:18And if you look at it, if you look at the roadmap, you're going to have to go to light
03:21at some point.
03:22Copper's not going to cut it, you know, in two or three more versions.
03:25And whether the AI trade collapses tomorrow or not, it may delay the inevitable.
03:30But the inevitable is that it's all going to go to light and you're going to be needing co-packaged
03:35optics when you go to scale up architecture, which is probably going to start happening in a real way next
03:41year.
03:42You know, the escalation of facilities, again, if it continues, does tech continue to be a little bit of a
03:47haven for investors?
03:49And why is that?
03:50Has it ever been the case before?
03:51Because it's a secular grower.
03:53You know, it's not really dependent as much on consumer behavior and things of that nature.
04:00So it's just money rotates that way.
04:02And so, yeah, part of this uptick we're seeing maybe this morning and a little bit yesterday has something to
04:08do with that.
04:09But I wouldn't make that a big part of the theme.
04:12These things are off 20%.
04:14That's pretty natural and normal.
04:16I think it's probably maybe bigger than what we realized last time in the momentum unwind.
04:21But that's just because you have all that added leverage.
04:23And that was the pain trade.
04:24It was the obvious pain trade.
04:25That's what we have to try to do is identify where people are, you know, everybody's sitting on one side
04:30of the boat.
04:31Something bad's going to happen.
04:32And so all that leverage in those single-stock ETFs, that had to unwind.
04:37And that's what's been happening.
04:38Andrew, you know, there was a question mark over whether it was the time to start getting into China AI
04:44plays, right?
04:45China 6, 12, 18 months behind the U.S. maybe.
04:48But, you know, catching up at some point.
04:50And it's cheaper.
04:51Would you buy into that argument or are you purely looking at U.S. stocks?
04:54Purely looking at U.S. stocks, you know, I have a hard enough time to do that.
04:58But what I thought was notable is that the open-source Chinese models are closing the gap with the U
05:04.S. frontier models.
05:05And the Brookings Institute's estimate is six to nine months behind our whatever, the best frontier labs.
05:12It's Claude Mythos.
05:13As I remember, we had to shut Claude Mythos down because we were worried about vulnerability discovery and things like
05:18that.
05:18So, I don't see much likelihood that U.S. corporates are going to skimp when it comes to spending on
05:26cybersecurity.
05:26So, names like Palo Alto and CrowdStrike probably even first.
05:31And then we recently added a small position in Tenable, which is really all about vulnerability discovery and management.
05:40What about Broadcom?
05:41I know it's something that you're very much looking at.
05:45It's only, and I put that in quotes, up 12% year to date.
05:47That would be fine in a normal year, right?
05:49But in a year where we're seeing 700 percentage points on stocks, 12% is nothing.
05:54Yeah.
05:55You know, I get that relative underperformer compared to the group.
05:58Okay.
05:59So, you can identify that.
06:00But really, it's the era or the age of compute ASIC silicon is here.
06:07The frontier labs decided to monetize at the end of the fourth quarter, right?
06:12We started the whole inferencing phase.
06:14And there's a whole trade that goes along with that.
06:16But inferencing, period, whether it's agentic or not, is just hungrier.
06:20It's two to three times more compute hungry.
06:23It's three to four times more memory hungry.
06:25And that's where some of these shortages are building into the channel.
06:29Broadcom is part of the solution.
06:31You could also say Marvell, any custom ASICs.
06:34Custom ASICs can reduce costs, overall cost of ownership, probably 20% to 30% less than a GPU.
06:41And energy usage could be as much as 50% less.
06:45So, that's where it's going.
06:47I would expect ASICs to outweigh, in terms of market exposure, would outweigh GPUs probably sometime in the middle of
06:59next year.
Comments