What's up with tech?

0:00Thank you. especially these days. But, you know, as questions continue to unfold about the strength of the AI trade and various players, what would you say to investors right now? I think what investors are, I think, dealing with this year is a bit of a tough situation, right? Because it's really been an interesting year. We had Q1 was a bit of a lull, and then Q2 was a huge rally with the stocks going up 87%, a little over 87%. And then Q3 starts, second half of the year begins. And then we have this downturn, right? It's a lot of profit taking, a lot of fear. We switch from this greed sentiment to a fear sentiment.

1:09And then we're seeing the market kind of struggle back and forth with it now. And I think last time when we spoke, that was around maybe mid-May. That was essentially mid the entire rally that happened in Q2. And what we were discussing is that in the rally, everyone looks like a winner, but not everyone's going to come out of a correction as a winner. And I think that's really what's becoming more pronounced in Q3. So the names that we saw as high flyers in Q2 are actually some of the underperformers in Q3 and vice versa. So if we take that and we think about the switch between first half and second half, you really see that kind of dynamic materialize.

1:45And alongside it, this theme in which earnings are acting more as a test rather than a catalyst, specifically after that Q2 run, everything is kind of coming out as not good enough. So I think that's a theme to be really cautious of and sensitive to. I mean, we saw TSM report great results. We saw ASML report great results and even raise their outlook. We saw Samsung report a 19-fold increase in profit and all three still traded down. So I think what we're seeing is a hypersensitive market in which everyone is waiting for the other shoe to drop. And I think we let out some steam since the beginning of Q3, specifically on the semi front.

2:23But it seems that the market is still in limbo and also dealing with new red flags, you know, from the circular financing to the Chinese competition. And so they're just dealing with new realities. A lot of these red flags existed in Q2, but I think they're getting a lot more attention in Q3. Does it make sense to start with the red flags or does it make sense to ask the question about the companies that do seem to be doing, in your words, good enough? Is it truly good enough? Like, will it turn back around? Like, how would you describe it in context, given that the reaction, the investor community reaction hasn't necessarily been reflective of what's going on with the stock?

3:04or would you put stocks in the even better category that have been reflected in investor sentiment being positive? I think, and let me know if I understood the question right, and I'm answering it from the right angle. But I think a big part of what we're seeing is this theme of great results being viewed as not good enough is because in Q2, the market preemptively pricing a lot of the positives to come and didn't really pay attention to the fact that at the end of the day, what we're talking about is a semi-rally. And at the end of the day, we're talking about components and a supply chain in which there's a shortage.

3:43And even if, you know, from our perspective, we believe that this shortage is a supply chain driven shortage rather than an end demand driven shortage, because we know that, you know, PC TAM and smartphone TAM are expected to contract by double digits this year. So end demand is pretty weak. Even though we think this is a supply chain led shortage. The idea that so much of the positive news got preemptively priced in in Q2 made it so that by the time we got to earnings in Q3, there was a combination of the results didn't look good enough against what was priced in. Investors were looking for more.

4:17And simultaneously, when we entered Q2, expectations were pretty low. And we were entering Q3, it's a different kind of expectation game. So that really does shift the difference. and then turning back to the supply chain factor in all of this, at the end of the day, there are physical constraints to achieving more financial upside, right? You have components. And we see this really clearly with the optical guys. There's a ton of demand, but there's not enough supply. There's a lot of bottlenecks within the semi-industry at the moment. So there's physical constraints that really prevent further upside that has already been priced into the stock.

4:49And I think that's part of what we're seeing get reset right now. And so even when we look further out, I don't think that we're going to see necessarily this AI bubble pop conversation happen in the second half. We're thinking about that more for 2027, first half of 2027 around then. But what we're seeing in the second half is the semi-correction. And a lot of it is because the supply chain inflated a lot of the pricing. A lot of the pricing went up because of this idea that there's a shortage. So there's a lot more leverage for these guys in the semi-supply chain to raise prices, that made their financials look extremely good in Q2 against what is a weak end demand environment.

5:26And then by the time that Q3 comes around, you have more pricing leverage, but then that begins to wane as end demand becomes more pronounced as a negative. So that's kind of how we're thinking about it in terms of how investors should view these earnings. A big part of what we're seeing get reset are the expectations that preemptively got built up in Q2 and now are being digested against reality. And at the same time, and I would say that's healthy, right? Because everyone knew in Q2 that this can't last. And that's why we saw such a mechanical shift, I would say, between Q2 and then Q3. But then looking into second half, I don't think that this is the end of it, right?

6:04The socks being down, I think it was as of this Monday, are roughly 13%. I don't think this is the end of the semi-correction. Instead, the way we're looking at it is in the second half of the year, we're going to see semis correct as the price increases in the supply chain also rebounds against the reality of end demand. So we saw a lot of double ordering, in our opinion. We're seeing a lot of customer pull-ins out of fear that there's shortage and you can't get components. And you need to get components for the outlook that you've guided and for the expectations that Wall Street is kind of holding now against the semiconductor industry.

6:40Does that, I hope that helps answer that question. As we continue to like have the question and answer, unfold, would you, does that mean that semis are a hold right now? How are you looking at the semis and are you looking at them as a group? Are you looking at any individual stocks separate from that group, if so? Yeah. So I would say that in general, I think there's an overarching kind of blanket on the semis in which we are going to see this, you know, we're going to see the socks. I think it has more downside to come. That doesn't make us, you know, ignore the semis entirely. Instead, I think there's a lot of potential still on the table in the semiconductor industry, But the idea is that you want to think about the guys who have more balanced expectations around them.

7:22And so, for example, and you want to think about the guys with more balanced expectations and equally as important, you want to stay away from the companies that have, you know, where the narrative has run ahead of where reality can really achieve that upside. And when I say that, I think mainly of memory, right? So I think part of what we're seeing is that the AI narrative this year has in large been built on that dynamic of the supply chain led shortages. And that really did prop up a lot of specific players as the winners. I think as we see these price increases moderate in the second half or there's less leverage for price increases as inventory builds up, we're going to see a lot of the expectations or the financial performance around these players that got propped up by these price increases come down.

8:05And that's where, you know, memory is center focus for us, because I think when we last spoke, memory was, you know, it was we were mid-May. Memory was gaining a lot of traction. Micron during Q2 entered the$1 trillion club. And a lot has happened since in the sense that a lot has happened to unpack this bull narrative around memory in which, you know, memory is regarded as no longer cyclical. So we saw, you know, the three big DRAM guys report. since the last time we spoke, we had SK Hynix come to the US market. And I want to highlight in specific that Micron's results, I think they underpin this kind of dynamic.

8:43So if there's a company that I would say I would stay clear of in the second half, I think it would be the memory kind of sector because I think those guys are at the highest risk because memory is the component that increased the most in terms of price and I think was most misunderstood by the market in terms of what was driving this memory rally. And I think we touched upon this briefly in May, but I'd love to get into it a bit more. Yeah, please do. Please do. Essentially, so a couple of things to get to here, right? So the bull argument is that memory is no longer cyclical, but all the data points since Q2 until today actually point to the exact opposite, right?

9:18So I think there's three main ones to point out here. The first is SK Hynix coming out in, I think it was late June, on what's looked at or called Black Tuesday and saying that, hey, we're going to reallocate capacity, manufacturing capacity from HBM to general purpose DRAM. And SK Hynix has the largest share in the HBM market, around something like 58%. And so them saying that we're going to shift from HBM, which is the hot AI memory, to general purpose DRAM is essentially highlighting this idea that general purpose DRAM is where the money is. And we already know this from Micron's own CEO when he noted that on the earnings call that, I think this was in December, that non-HBM gross margins, not in December, sorry, I think it was in Q2, non-HBM gross margins are actually higher than HBM gross margins.

10:11So SK Hynix making that shift as the largest shareholder in HBM does underpin this idea that they're following the money, right? And so I think that helps break down that misunderstanding the market had on the memory trade. And then the other part of that equation is that SK Hynix adding, you know, adding more supply, more DRAM supply into the market actually has the impact of reducing the ASP leverage that's happening right now on the DRAM front. So to take it back a minute, the reason why we think memory kind of popped and Micron had these great run and great outperformance, you know, huge margin expansion.

10:49We don't think that happened because of HBM. We think that happened because of the non-AI side of the memory, because of general, general purpose DRAM spot prices going up insanely high. And we think that's what underpins the gross margin expansion from Micron, in which they guided to, I think it's 86 for next quarter, and they printed 84.6 for this quarter. So they're really benefiting from the higher ASP on the DRAM side. So just going back to the quarter they reported at the end of June, their DRAM sales grew 67 % sequentially. And that was driven mainly by a 62 % in ASP. And bit shipments, on the other hand, were only up 2%.

11:30So it's really highlighting the results are telling us that the growth is coming from pricing. And if we believe that memory is cyclical, and it's historically proven to be cyclical, if we believe that memory is cyclical, then at the end of the day, when you have a shortage, you're going to have higher pricing, high ASP. But then when supply begins to catch up to demand, what you're going to have is that you're going to lose that ASP leverage. And so as SK Hynix shifts supply from HBM to DRAM, that's going to create a lot more output on the market. And where it becomes a bit more technical is this idea that HBM has tripled the diet size of DRAM.

12:06So when you shift capacity from HBM to DRAM, you actually have your output increase at a much faster pace. So we're talking about a lot more supply in the market. And so more supply in the market means you lose this ASP leverage. And that means that we're looking at Micron's gross margin trajectory. That's been something like a dream, right? 86%. That's the highest gross margin there is in the semiconductor industry. That's even higher than NVIDIA at its peak, which is crazy. We're looking at that gross margin expansion, moderating, or even U-turning back to more, I would say, realistic levels.

12:39And so that's just one factor of how we're looking at memory and the risk there thinking about, you know, what names we'd be cautious of in second half. The other factor is the Chinese DRAM supply also entering the market with CXMT IPO-ing in China. And then also, you know, talking about adding 85 ,000 wafer starts per month of DRAM. That's compared to SK Hynix adding 60 ,000 a month and MU adding 30 ,000 a month and Samsung adding 50 ,000 a month. And I specified these three because they're the largest DRAM players. And so the idea is that CXMT is actually going to be looking at an output of 350 ,000 wafer starts per month by the end of the year.

13:22And that puts it within, I think, roughly a 25 ,000 range of microns own output. So if we're seeing more supply from SK Hynix, we're seeing more supply from China DRAM. And we're seeing guys like Apple shop for Chinese memory and lobby at the Pentagon to be able to use Chinese memory, Chinese DRAM. I think this looks, this sets up the kind of conversation in which we're looking at more supply, the shortage will ease, ASP leverage will come down. And then to add just a cherry on top here, Trendforce is expecting now DRAM contract pricing to moderate to 13 to 18 % growth in Q3 this quarter, as opposed to what we saw in Q2, which was 53 % to 63 % growth.

14:06So I think that the conversation on memory, we were early, definitely on Micron, on the downgrade, but I think that we're seeing our thesis play out now. I think the conversation on memory is beginning to become more pronounced. I mean, Q2, no one was talking about the red flags in memory. And in Q3, we're seeing a lot more conversation about that. So I think it reaffirms what we discussed back in May. And it also, I think it speaks to looking into the second half as the supply comes online, a more risky dynamic around memory. Could you share more of how you see the coming months and the coming year playing out vis-a-vis the different lanes in tech?

14:41Yeah. So I think, you know, when we were thinking about the second half and then we're thinking about 2027, I think a big part of what we're seeing is the AI, you know, AI is maturing and it's segmentizing. This is a theme that we've been watching for a while. And I think we discussed on this podcast before. And as AI matures and segmentize, what we're seeing is that we're shifting from this conversation of token maxing to token efficiency. And, you know, this is underpinned by this idea that earlier this year and even last year, everyone was pushing for, you know, enterprises, for individuals to use AI as much as possible, right?

15:14Even Uber was pushing employees to use AI as much as possible. And then they blew through their budget in the first four months of the year. So what we're seeing is that compute is not free, but AI is free. And a lot more of the end users and enterprises are beginning to become more cost conscious. We're entering a more cost conscious, I would say, phase of the AI boom or the AI cycle. And as we enter this more cost-conscious phase, I think the conversation is a lot more about what kind of solutions you're using for compute. So we have AI GPUs, NVIDIA still dominates the scene, Verirubin is coming up, and the demand from all of our data points looks really great on that front, very strong demand.

15:56Case in point that Google and Tesla haven't spent the majority of their full capex. And I think that that's part of what we're going to see go fully to Verirubin and into NVIDIA's pocket with the ramp of Verirubin and into the second half and then early 27. But what we're seeing is that because of the AI conversation is much more cost conscious, we're seeing more hyperscalers gravitate to lower cost compute solutions. And that's where ASIC comes in. So I think as we think about, you know, second half of this year, I think Verirubin is really going to dominate the scene. But then in 2027, I think ASIC is going to become a much bigger part of the inference market because you need to focus on how to get, you know, how can you make your, how can you get the lowest cost solution to power what you're doing?

16:42And this becomes even more, I would say, important with the dynamics that we've seen kind of play out over the past two months from Chinese competition entering the scene. So it's kind of like we had a deep seek moment revival when, you know, Moonshots Kimi K3 came out last month because we're seeing China really compete with US frontier models on performance, but massively undercut them on price at a time when the market as a whole is very conscious about all this AI capex spend and whether or not, or when or not, we're going to see this AI return on investment and how these hyperscalers can show that.

17:21So it became a very, I would say, tangled web in which for, you know, the US frontier models for the hyperscalers to remain meaningfully competing and being able to show a return on their investment, they need to make sure that their compute is at the lowest cost possible. And so that's what ASIC, I think, is going to really provide. And I think, you know, again, the top of the ASIC race is Google with their TPUs, not only that they use for their internal use, and then they rent out to others, but also that they're now, as of this quarter, selling full TPU rack systems. So I think Google has executed the best on that front.

17:59And now we're seeing Amazon be, you know, a close second. And then Meta and Microsoft are really trying to catch up. OpenAI is working with Broadcom on their own ASIC chip. Anthropic has plans to work on its own ASIC chip. So we're seeing this kind of shift in terms of a shift to lower cost solutions as the idea that you need to get the lowest cost solution for compute available. So I think that's a big part of where the conversation is headed. So looking into 2027, that makes us a lot more, I would say, cautious on NVIDIA. Because I think NVIDIA has the most to lose from the switch, this shift towards more ASIC adoption.

18:39Because now for the first time ever, NVIDIA is going to have to be forced to share part of the AI accelerator market with ASICs, right? from a 80 to 90 percent those are the current estimates 80 to 90 percent market share uh is going to be pressured and so i think that's what keeps us more on the sidelines of nvidia thinking about it into next year uh because at the end of the day i think the ai capex is going to have to be spent across all of these players i i got a second thought i just want to emphasize that's where we like guys like marvell right because when these hyperscalers are going for asics they're multi-sourcing.

19:13No one wants to have all their eggs in one basket, similarly to how they were forced to have all their eggs in one basket at the beginning of the AI boom with NVIDIA. So everyone is trying to multi-source. Google went with Broadcom, then MediaTek, and then Marvell. And now we have new ASIC players entering the scene like Qualcomm. And so I think that's where the 2027 conversation is leading towards. So that's where I would pay attention for investors kind of trying to gauge how to think about the semis for next year. Much appreciated. Any other stocks or notes to mention? Yeah, I would say part of that ASIC conversation, I think really favors, you know, part of that ASIC shift really does favor ARM.

19:54Because ARM-based CPUs are actually, you know, I think they're the best fit for the ASIC conversation because they have the lower power consumption, which is increasingly important in the AI infrastructure build-out. And I think ARM CPU are already embedded into a lot of the tier one players via, you know, ARM's designs. And so I think that we're going to see ARM be the favorite choice on that front. It's not to say that the x86 market where, you know, Intel and AMD have their playground, it's not to say that these guys aren't going to benefit, but I think ARM is going to be the winner from that theme, both the ASIC theme and then this agentic AI theme, because at the end of the day, ARM was the first to show us what an agentic AI rack would actually look like.

20:35everyone else talked the talk and there's definitely you know benefit to go around but I would say I think Arm is the favorite choice there and the others still need to prove themselves the other stock I think it'd be kind of worth addressing is Intel because it also had this huge pop-out moment in in Q2 and for Intel they had this huge you know pop-out moment we had a lot of rumors about them getting external customers for their for their foundry business Interestingly enough, rumors that management never confirmed, but that Trump did. So we didn't really get a lot of clarity on those. But I would say for Intel, we're more cautious as well.

21:14The reason why we're more cautious is because of this last earnings call. Basically, on the last earnings call, management talked about, not on the last earnings call, excuse me, on ASML's earnings call, actually. So ASML announced that Intel is using their high NA tools, their high NA lithography tools for its 18A node. And that's that to us was a big flag, a big red flag. And the reason why is because it's a big red flag for the foundry business. And the reason why is because Intel was meant to use the high NA tool, which comes at a very high price, close to 400 million and almost double what a UV tool would cost.

21:55So Intel was supposed to use that more expensive tool on its 14A node rather than its 18A node. So the fact that they used it on their 18A node tells us that the yields on their 18A node weren't the greatest. And so they needed to use this far more expensive machine to be able to up the yields there. And so that makes us feel like there's a lot more uncertainty about whether Intel is going to get a meaningful external customer, because any customer coming on will have to, you know, will be harming their own business if they go to a foundry with bad yields, or at least with yields that don't compete with those of TSMC.

22:28So that's how we're also thinking about the Intel foundry side of the business, because we've been watching Intel for a sign of, you know, external customers coming on. And our indicator for that was if Libutan decided to raise CapEx, and we still didn't get that even as of, you know, this quarter, we got a very mild increase in CapEx. So I think that this underpins this idea that the Intel foundry business, it moves the stock more on headlines rather than the fundamentals. So we know that as of this call, they're raising their fiscal capex to 20 billion up from 18 last year, which isn't that much more meaningful.

23:07And they said that in 2027, they're going to raise it substantially, but we didn't get any numbers. So I think this is also management kind of having a foot in the door and then a foot outside of the door because there's some uncertainty on that front. So that makes us a bit more cautious as well on Intel in the second half after that run up that we saw in Q2. It strikes me that as we close out this conversation, and again, anything else that you feel like is worthy of investors' attention right now, happy for you to share that. But it strikes me that there's often talk of this bubble in tech and it strikes me that you you I believe you did mention um some bubble language at the beginning in terms of something coming to burst but do you see it a do you see the conversation as a useful conversation b do you see it as a large bubble do you see it as a series of kind of smaller bubbles in different lanes in the tech sector and also at different points in the cycle, like this is a bit of a bubble and then this becomes a bit of a bubble and then they kind of even things out and return to center.

24:11How would you respond to that? Or how do you think about that? That's a great question. I think that's a question more people should be asking. I think when it comes to the bubble conversation, there's definitely a bubble and it's growing. I think that's without a doubt the case. The way that we're looking at it is basically, yeah, there's going to be a semi-correction second half, but then the more pronounced real bubble conversation, I think it's more related to AI CapEx, right? And we had the bubble fears get revived when there was a really clear dynamic in which semis were being rewarded while the hyperscalers that are funding the semi rallies were being punished for that spend.

24:53And so then when you have China enter the conversation as well with cheaper alternatives and other models, making it so that essentially, as of now, China wins 45 % of US company AI use. That puts a lot of pressure, I think, on the AI capex spend and the pressure from investors on these tier one players to show return on investment. So I think that's at the core of when the bubble pop is going to really materialize. And I think it comes more from a financing perspective, that's, I think, what's going to really push us off the edge. And that's, you know, that's when we think about the things that we've seen NVIDIA and the broader market do and the creative ways, really the very creative ways that the market is beginning to try to finance this AI boom.

25:41So, you know, we had NVIDIA in talks to provide, I think it was about 250 billion financial backstop for OpenAI, which is unprofitable private company with the IPO likely pushed out until next year and rumors of missing internal targets. And we know getting pressured in terms of market share by competition from Anthropic, competition from other Chinese models. And then you have conversations or negotiations for financing up to 350 billion to fill up those data centers with chips that NVIDIA claims is not circular financing because there's no written commitment in which OpenAI has to buy NVIDIA GPUs.

26:15But I think it's kind of implied. And then NVIDIA stakes in a bunch of the the NeoCloud players, including Nebius, CoreWeave, and then Iron this year. So it looks like NVIDIA is increasingly funding its own customers. And so if we think about the circular financing structure that this creates, it creates essentially what I would say a scenario in which money moves in a loop, right? You invest it, and then it gets bought out from your chips, and then you invest it again, and then rinse and repeat. So I think that's a conversation to pay attention to for the bubble pop, because that underpins this idea of how much more money can we put into the AI infrastructure build out.