第009期 - 用开源数据驱动投资决策（ChipBook）| Chaim Eisenberg、Simi Sherman、Jordan Nanos

I'm here today with the guys from the ChipBook team. We're going to talk all about what ChipBook is, what open-source data is, some of the use cases that people are using it for, and some of the nice viral tweets that the guys have put out in the last couple of weeks. Guys, welcome to the show. Great to have you on.

It's good to be here, Jordan. Thanks for having us.

Thank you, Jordan, for having us.

All right, let's jump in. So, ChipBook and Chips and Wafers, can you give us a little bit of background about what you guys do at SemiAnalysis and what it is?

Yeah, sure. Jordan, great to be here. Longtime listener, first-time caller.

All three of them.

1. The Open Source Data Edge

Let me give you a little bit of background on what Chips and Wafers is and what the ChipBook is. Essentially, Chaim and I both came from the buy side. We worked at a hedge fund where we covered semis—not exclusively semis—but one thing is true of any sector you cover on the investment side: You're constantly looking for data.

Data is valuable to generate investment ideas. You're looking to validate ideas. Let's say you have some idea and you want to know whether it makes sense or not. You're looking for some sort of alternative-data platform or information you could use to validate an idea.

Sometimes you want to track an idea. Let's say you have a great thesis; it seems to check out, but how do you time that position? How long do you hold it? When do you size it? When do you get out? So you have to track the idea.

Everybody in the hedge fund world is looking for alternative data. In fact, just a funny anecdote: A few weeks ago, we attended a conference for alternative data, where alternative-data providers came to a conference center, and they had dozens of hedge funds looking to buy some sort of alternative-data platform. Of the maybe 60 vendors at the conference, we were the only guys selling semiconductor-related data.

The reason I'm pointing that out is to say that we recognized on the buy side as well that there is a dearth of data and information out there. Semiconductor investors are thirsty and hungry for some kind of data.

What we developed an expertise in is identifying open-source data that can help inform your investment decisions and your view of the industry. It's all open source and public, but it's messy and distributed all over the world. It's in different languages, and it's coded. Oftentimes—most of the time—you don't even know what you're looking for or how to use it.

There's all this data out there, and what we do is take that interesting information and turn it into actionable intelligence. The way we do that is by learning how to gather all of this data and put it together in a package that can help inform your idea generation, tracking and monitoring, and investment-thesis validation.

The Chips and Wafers ChipBook platform is a way for the investment community and semiconductor companies themselves to gain insight into the big picture of where the industry is going, and, on a more granular level, how that impacts individual companies, trends, themes, and inflections within the space.

Just to give you an idea of what the data is, most of it is open-source import data, export data, production statistics, and inventory data. All that stuff is out there, but it's a pain in the neck to find, and we try to bring that to our ChipBook customers.

If I could interject here to add a little bit to what Simi is saying: Simi used the word “intelligence,” and a helpful parable that I like to think about comes from the fact that we both came from the buy side. Before the buy side, I was in the military for a few years.

An interesting parable to think about in this context is the concept of what people like to call open-source intelligence, or OSINT for short. For years, any respectable military operation had its entire intelligence directorate working across organizations. They had their intelligence silos: signals intelligence, imagery intelligence, and human intelligence.

Then, at the beginning of the 21st century, this whole new theme of open-source intelligence emerged. At first, it was reserved for the nerds on the internet who were scraping IP addresses and random YouTube videos. Within the intelligence community, it was totally disregarded at first because people thought, “What's interesting about open-source intelligence? If it's open source and anyone can access it, why is that valuable? Why do I need that?”

All the intelligence agencies would say, “I don't need this. If everyone can access it, what use do I have for it?” But what everyone has come to understand over time is that one thing open-source intelligence has over all these other sources of intelligence is that it's real-time. Yes, it's messy, but it provides super-valuable signals super early, and a lot of times in places that are really hard to get to with all the other forms of intelligence.

Obviously, if you could have a human asset inside somewhere, that's super valuable. You'd want to get that, but it's not easy to do. However, if there's a guy on a street corner recording a video that he then uploads to YouTube, and you're able to see that video, then you have access to something that in the past would have been super hard to get.

To bring this full circle, when people say, “Wait, open-source data—is that valuable?” the obvious answer is yes, it's super valuable. But you need to know how to find it. You need to know where to find it. You need to know how to sift through it, and you need to be able to distinguish between what's noise and what's a true signal.

That's a helpful parable for how to think about this whole idea and concept of open-source data to help inform decision-making, especially in the semiconductor industry.

I think this is obviously about trends. If you look back at historical data and then try to inform the current day, you need to stay up to date. It needs to be current. More than anything, it's about establishing a process to get access to this data and then making sure it's current and up to date. You have monthly or quarterly updates.

Can you give me some examples of real data in the Chips and Wafers context? We've seen you guys put out some interesting teasers on the SemiAnalysis account or the Chips and Wafers account. Maybe we can run through some of those as examples.

2. Why Granularity Matters

Yeah. Let me take one step back and give the value proposition, so I can explain where the examples come in. Everybody knows there's open-source data available. People are aware it's out there, and you have a lot of banks that put out a generic statistic as an indication that WFE is up or WFE is down, for example. People are looking at a lot of the WFE equipment-type imports and exports, things like that.

I think the secret to using it properly is having the ability to get as granular as possible. Just looking at the WFE side, for example, WFE as a category: I've seen statistics saying that WFE is up 10%, so let's assume Aixtron is going to go up, or Maymays going up, or whatever equipment manufacturer you're invested in.

The problem is that WFE is the broadest category in the world. It includes wafer-manufacturing equipment, such as tools to make boules and slice wafers. It includes deposition tools, etch tools, lithography tools, and ion implanters. It even includes packaging equipment, like flip-chip tools and wire bonders. Oftentimes, it even includes inspection tools and metrology tools.

If you're tracking KLA metrology and looking broadly at WFE as a metric to monitor your investment, you're looking at a metric that's far too broad to be meaningful for your investment, and you can make mistakes.

One thing I like to say is, “The only thing worse than having no map at all is having the wrong map.” Investing based on the wrong bit of information is very, very dangerous.

The value comes from trying to be as granular and targeted as possible. That applies to WFE and to the AI supply chain, using AI supply-chain supporting companies as a tracker, both for the big guys and as investment opportunities on their own.

When you look at shipments, production, and inventory levels around the world, you need to be able to distinguish between logic and memory. Within memory, you need to be able to look at flash, DRAM, and HBM from different countries and know who's making what and for which customers.

I think granularity is the way we try, as best as possible, to make our information targeted. Maybe I'm trying to think, Chaim—maybe you want to do one example or talk a little bit about some use cases from the data.

We could do some use cases.

The only thing I would add to what you were saying up to now, which I think is important to expand on, is this idea that a lot of the investors we talk to—everyone knows semiconductors are huge, especially over the past 3 years since AI has entered everybody’s lives. Everyone knows that semiconductors are a super-hot market right now, right? Most people really do not understand just how massive the supply chain is.

They do not understand where it starts, and they really can’t even understand where it ends. You’re thinking about a token output on the chatbot that you’re using, whether it’s OpenAI or Claude—it doesn’t matter. That’s what the consumer is seeing at the end right now with AI. That started 4,000 steps earlier, when some random Japanese chemicals company was making a polysilicon boule. Everything flows from there, right?

It’s this massive supply chain that’s super-duper hard to comprehend, and another really good value proposition. Again, we do open-source data, and we track the entire supply chain. We start all the way, as I said, from the Japanese company and try to go as far as we can, to where the data allows us to go.

The ability to see that entire value chain—to see what’s moving up and what’s moving down—is super valuable because they also always impact each other, right? As Simi mentioned, WFE: if WFE is an input, then obviously the output is chips that the fabs and foundries produce after that. If it’s chips, those go to the server ODMs, and if it’s the server ODMs, they’re going to the hyperscaler data centers. From the data centers, that’s token output.

Just being able to see that whole wave, that whole supply chain move, and noticing where the different signals hit to understand where that’s happening upstream, then how that impacts the downstream, is super valuable.

3. Catching The Memory Cycle

Yeah, let me give you an example. Let me give one example of memory, and maybe we can talk more about early signals. Obviously, we’re in this memory supercycle, and everybody in the world wants to know about memory. It was probably a year ago that we called out the memory cycle. How did we see it early?

Obviously, you can look at memory sales numbers once Samsung and Hynix report, but then you’re already looking backward. What we were looking at, for example, was Taiwanese DRAM inventory levels, and we had seen that inventory levels of Taiwanese DRAM were rising for months. They were building, building, building.

By conventional DRAM here, by the way—not HBM. Just conventional DRAM, non-HBM, non-AI-related, at least at the time. Sorry, Simi. Go.

Yeah. We were watching the Taiwan DRAM levels go up, and they were hitting historic highs. Then, about a year ago—probably in the summer of 2025—we saw the inventory levels drop for the first time in a year. We called that out and said, “Something’s happening, but let’s wait to see. It’s only been 1 month.”

The next month, we saw the inventory levels drop again, and then a 3rd month, and we realized we were starting a trend. Sure enough, I think we’ve now been at 11 or 12 months in a row where Taiwanese DRAM inventory levels have dropped.

What that told us, as inventory levels started to drop, was that we were beginning to see a demand-supply imbalance where demand was now outstripping supply. That was a very early call we were able to make by tracking a relatively obscure, specific data point that was available. It was open-source, but you had to know how to find it and, more importantly, what it was telling you about the memory cycle.

Then we started to track it. Everyone is going to be looking at Korean memory exports, but what are we seeing when it comes to Chinese memory exports or Taiwanese memory exports? That gives you a better sense of the broader market demand beyond the specific HBM chips, which we’re also tracking.

Those were all ways that we were able to get early identification of a trend within memory and, number 2, continue to foster that tracker going forward. Now, what do we do today? Everyone has made huge investments in memory companies, and you’re sitting on huge positions.

If you’re a hedge fund right now sitting on a huge position in a memory company, what you’re really nervous about is when this cycle ends. You need to know before Hynix reports on a company call, “Oh, by the way, demand is now falling off and supply outstrips demand,” or somebody comes online with huge capacity and we’re no longer in a supply-constrained environment.

So what we’re doing now as a tracker is not generating the investment idea, because we did that a year ago. Now we’re tracking and monitoring your investments for you by making sure we’re following memory shipments globally, both from Korea, Taiwan, and China. We’re following DRAM, Flash, and HBM.

All of those are invaluable trackers because the size of the position you’re sitting on is huge, and the risk of not getting out of those positions—or selling too early—is significant. Those positions are worth a tremendous amount of money to a hedge fund. Staying on top of that data on a monthly basis for our customers, I think, is a huge value proposition.

Yeah. Makes perfect sense. We had Sravan on a few weeks ago. He was talking about the allocation of TSMC’s 3-nanometer capacity between NVIDIA, Apple, and some of the smaller smartphone players. I think just this morning we saw Xiaomi smartphones from China—they’re down 35% on shipments or something like that.

Some of these calls can happen in weeks, but they can also happen over months. It can take months or even a year-plus for some of this stuff to play out when it comes to inventories or things shipping and moving around. Do you have anything currently that you’re tracking?

To some extent, looking at historical data is different from forecasting the future, but historical data on something so far back in the supply chain is like predicting the future for things downstream of it in the supply chain.

Maybe, Chaim, you want to talk about WFE? WFE is a great example because WFE equipment is shipped 6, 12, or 18 months before production even starts.

Yeah. Correct.

Totally. I’ll speak to that in a minute, but just to address what Simi was talking about beforehand, with the examples we were giving about memory, the odds are, in investing in general, it’s going to be rare that you’re going to have that silver-bullet piece of information—just 1 thing that gives you everything you need to know: “This is the time to go long this company.” The odds of that happening with 1 piece of information are highly unlikely, basically 0.

Going on the examples Simi was giving, the Taiwan DRAM inventory we were able to call out, or memory exports coming from non-HBM geographies like South Korea—those were signals. Again, going back to what I was saying beforehand, you get these signals, and you need to be paying attention to them and be aware of them before they hit the market.

If it’s already in the company’s print when results come out, it’s too late, right? This idea of being able to track those signals is important. Sometimes, going back to the intelligence parable that I gave, it’s noise; sometimes, it’s signal. But you need to be aware of the signal so that you can have that on your radar and understand: Does this translate into actual intelligence?

Suddenly, it gives you an even heightened awareness of the questions you need to be asking or the areas that you need to be focusing on. That’s just to speak to what Simi was talking about beforehand.

4. WFE Predicts Future Capacity

You were talking about certain examples of upstream things that you could look at that then impact the downstream. We talked about WFE. WFE is massively important because, going back to what I was saying beforehand, WFE moves from the equipment manufacturer into the fab.

Once it’s installed in the fab, that’s what allows you to output wafers and output chips, which ultimately ends up in the data centers, which ends up in the memory modules that go into the packages. Again, you have lead times on this. If you order it today, it’s only showing up at your fab 6 to 12 months from now.

Then you have an installation period, and then you have time until it ramps. Being able to track the WFE movements, again, is giving you a good 12- to 24-month preview into what wafer capacity is going to look like down the road.

And again, once the company says it—once you already know that it's coming—it's already too late. I think a really good example of this is something that we track super closely because we understand just what an impact it has on the market: China WFE. Not the Chinese manufacturing of WFE, although I'll touch on that in a second also, but primarily the amount of WFE flowing into China.

Because, first of all, when AI and memory were down, everyone's talking right now about how much South Korea—the revenue composition of South Korea—was in the ASML print earlier today. But memory was really muted for 2 years, and you had TSMC, but everyone else wasn't really putting in orders. What was really propping up all of these WFE companies was just the China demand. Companies were getting to 25%, 35%, 45%, even 50% of quarterly revenue coming from China.

Sorry, this is companies like SMIC or YMTC importing equipment?

Yeah, that's the example that I'm giving right now. That's one of the things that we track super closely. We track WFE imports into China at the provincial level, because you have YMTC, SMIC, Hua Hong, and CXMT, and they're all operating in different provinces. We're tracking the WFE inflow into all of those.

This has 2 massive consequences. Number 1 is: What does that mean for the WFE company revenues? If we're thinking about KLA, Tokyo Electron, or ASML, it obviously impacts them, because their China revenue and how that fits in is obviously material to how people are going to be looking at their results.

But also, once Chinese capacity comes online at all these companies, everyone wants to know: When is there going to be this massive adoption of Chinese memory? When are we going to see SMIC being adopted by a ton of people? When is the memory capacity going to come online, and then suddenly we'll start seeing more of a supply-demand balance, unlike what we're going through right now?

That's a really good example where we're tracking this super-upstream thing because it's also informing you how you need to think about what WFE revenue is going to look like for these companies that are selling into China. But it's also telling you what this is going to mean for Chinese capacity moving forward. That's something to keep in mind for the Chinese companies, but it always has an impact. If people are adopting more SMIC, then maybe that's coming out of UMC or Texas Instruments. It's just another thing to keep in mind.

Yeah.

Sorry.

Maybe you can look backward and give an example of the impact of tariffs, or the impact of some of the regulations that the US has explored in terms of trying to restrict companies from actually selling equipment into China?

Yeah, that was the example I was going to give. We had observed—this was a while ago—but anyone who looked at the China WFE revenue saw that the massive ramp happened in 2024, because we're still talking about the Biden administration, and people were talking about all these restrictions that were going to come online.

You basically saw this massive lithography buildup happening inside China. People were like, “Oh my God.”

Not just lithography. We saw deposition, etch—everything.

Right.

I mean, deposition, etch, everything.

We saw everything. Lithography was definitely more prominent. I think they were probably thinking that lithography meant there was going to be a lot more focus put on ASML, and they would probably be able to get deposition and etch even with some of the restrictions in place. So there was a lot more focus on lithography.

But yes, we saw this happening. 2024 was a ridiculous year in terms of Chinese WFE demand. If you look back at all of the WFE companies' transcripts at the end of 2024, when they were guiding into 2025, they were all saying, “We realize that was just stockpiling ahead of tariffs, ahead of regulation, ahead of whatever the administration was going to do. We're guiding that 2025 is going to be somewhere between 20% and 25% down.”

Everyone was saying that. We were tracking this data super closely, and month after month, it was on par with 2024, at the same levels. This was already when TSMC was ramping a little bit, so you were starting to have all these other orders. At the same time, Chinese demand was literally remaining exactly the same.

We were tracking all this month after month, and at year's end, if you look at China revenue for these WFE players, they were not down 25% year over year. It was basically flat year over year. I can't remember if it was slightly up or slightly down, but 2024 was massive—absolutely massive.

Now, when you look at all the WFE company commentary going into 2026, what are they all saying? China's going to be down 20% to 25% in 2026. That's what they're guiding. Maybe it's true, but, for example, ASMI was saying that they're actually seeing China go up a little bit in 2026.

Something that's on our radar is: Wait a second. We already saw this play out in 2025, when people said it was going to be down, but we tracked it and it was actually the same. That obviously impacts what 2026 is going to look like. That's something that we're tracking super closely.

That sort of speaks to the importance of having a granular tracker. I do think, to some degree, we're going to see a slowdown in China WFE imports, but the degree of slowdown may not be consistent across the supply chain. For certain tools, you may see a slowdown; for certain tools, you may not.

I don't know exactly how this is going to play out, but we did see that inspection equipment into China was holding on a little bit longer than some of the front-end tools. I don't know if it's holding on anymore. We're going to find out more in the next week or so, when more data comes out. But our ChipBook comes out once a month. So next week we're going to put out the new one, and I think there's going to be some really important data about that split: which tools are slowing in China and which ones are not. That also creates a huge opportunity, because if everybody is assuming something about WFE writ large, but there are exceptions within that, being able to identify the exceptions that are going to be hit by the macro noise but ultimately will continue to perform on an earnings level—that's a huge buying opportunity.

That's something we'll probably have more insight into next week. To Chaim's point, there's a lot of value in tracking each one of those things on a granular level.

Yeah. In the back of my mind, I'm thinking about how it all connects at a high level. When you were walking through that, it's the AI Diffusion Rule that Biden was exploring, that people were saying Trump was going to repeal—and then he did—and how that actually flows through to shipments and earnings.

A lot of it is also trying to figure out what the normal run rate is. I think Biden comes out with these rules so that everybody in the world rushes to pull forward orders, because people want to get things in before the restrictions kick in. You see a huge ramp in imports—let's say, China WFE equipment.

Looking forward, is there now an elevated capacity expansion in China that will be at an elevated level, albeit maybe not as high as 2024 and 2025? Is there a new norm, or do we have to look back at historical run-rate levels to get a sense of how far it can drop?

I think people sometimes have a hard time visualizing how far something can drop after it's already down. You'll say, “Shipments are already down 10%, so I guess we've bottomed out.” Companies love to say that. Every single company call will talk about, “Now we've hit the bottom.” They've always hit the bottom.

They're constantly hitting the bottom. So what is the bottom? About 10% of a 180% historic high is the bottom or something like that? Yeah.

Exactly.

Yeah.

I remember when wire bonders—like a very obscure packaging tool, not obscure but unexciting, unsexy packaging equipment—were flying into China during COVID. It was so out of whack with the historic norm, which we've now kind of returned to. The order levels dropped off 10%, 15%, 20%, so you see KLA saying, “Okay, now we've kind of hit the bottom again.” But if you looked at the historical data, it was clear that we were far from the bottom. Having that perspective, even though it's outdated data, can certainly inform your investment decision going forward.

I love what Chaim said about that: It's like the mosaic theory. Sophisticated hedge fund investors are very, very smart. They're not looking for somebody to feed them the answer. They need pieces to the puzzle. They're going to put the puzzle together, but every single puzzle piece you offer them, they will incorporate into their mental model when they develop their thoughts about the business. None of these data points is the answer, but each one is a puzzle piece. When you know how to put those together, I'll give you an example.

Yeah, I'm itching for more puzzle pieces here.

How can you cut me off? What?

I'm itching for more puzzle pieces here, man. Give me some more puzzle pieces.

Yeah, I'll give you one. This is an obscure one, but it's interesting. One of the things we were looking at was photomask writers going into China. 2 years ago, there was a big buildup. China was building up its own mask shops, and we were watching photomasks go into China. All of a sudden, they started to slow down, which made sense because they had built up more capacity than they needed.

But watching the slowdown of photomask writer imports into China made us start asking questions: Why aren't they going into China anymore? What's happening? What we found was actually counterintuitive. At first, I thought, “Okay, they're not going into China because they're not utilizing the capacity they already have available.” But then it's like, why aren't they utilizing that capacity? If they're able to create photomasks at a cheaper price than Western suppliers, why don't they just do it?

What we discovered is that a lot of the chipmakers weren't comfortable buying masks from China because, in order to get a mask made for you in China, you need to share your chip designs with the Chinese companies, and they didn't want to do that. That made us realize that photomasks were a huge, huge onshoring motivation. Of all things, you don't want your photomask to be made in China.

That's why, even though China had built up huge capacity in mask writers, the customers weren't comfortable buying photomasks from China. So we said, “Okay, well, who are the photomask makers in the Western world?” Photronics (PLAB) is an American company, even with a Chinese subsidiary, which is a win-win on both sides. Then we realized that PLAB is a huge long. Sure enough, once they came out with that story, the stock doubled.

Nobody was handing you that story, but by tracking the data, you begin to know what questions to ask. It leads you down the road of discovery, and you figure out winners and losers that way.

Makes sense. Great story. Can you walk me through how that actually gets incorporated into a ChipBook release? What does it actually look like? When people subscribe, what do they actually get access to? How clearly are you spelling out the long and short positions that you're recommending to people, or are you just providing data? What format does that data come in?

Okay, that's a great question. Chaim, jump in if I'm missing anything. The ChipBook looks like this: It's basically a PDF with 35 pages of charts. The first 10 charts are the same charts every single month. It comes out on a monthly basis.

The first 10 are fundamental building blocks of the semiconductor industry—things that every semiconductor analyst, every investor, and every company needs to keep their eye on. Basic things: What are the hyperscalers spending? What do the main silicon-content product shipments look like, whether that's PCs, smartphones, autos, wafer shipments, or PCBs? We're not just looking at the end market; we're also looking very early in the supply chain. Those are 10 slides that appear every single month.

The next 25 slides rotate on a monthly basis. We track probably 200 to 300 different data sets, but those data sets aren't interesting or actionable every month. Sometimes they're just boring and nothing happens that month. As opposed to sending our customers a 250-page ChipBook that would completely overwhelm any analyst—nobody would even look at it—we don't have a specific number. It could be 15, it could be 25. Whatever we think is actually interesting, we pick out another 25 or so slides and append those to the first 10.

Now we have 35 pages: the 10 that are always there, and the 25 that vary every month. Every single page of the ChipBook has the chart, and at the bottom it has 3 things. Number 1, what is this data? Number 2, what are the stocks—not all of them, but what public companies are connected to this data? Number 3, it gives an update every month of what we're seeing in the data.

You can scroll through the Chipbook or flip through it and say, “Here are the companies that are connected. Here's what happened this month. Here's my update.” That's how you read it.

What we now add to the Chipbook, which I think is a super valuable tool, is an executive summary at the beginning of every Chipbook. The front page of the Chipbook is the executive summary. In the executive summary, Chaim and I tell you the 2, 3, 4, or 5 most important implications of the Chipbook. We say, “These are the trends that we're seeing this month. These are actionable ideas.” These are either investment ideas, ways of tracking very important components, or ways of tracking the overall industry.

We call out in the executive summary what page in the Chipbook you can find that data. We'll say, “This is what we're seeing in PCs, page 7, which relates to our view on China exports, page 13, which has implications for silicon content from this company, page 15.” You could just read that 1-page executive summary, and it's a very, very valuable piece of research that allows customers to then go through the 35 pages of the Chipbook in depth.

Mm.

That's the overview of what it looks like.

Makes sense. Super-good summary. It begs the question: Where do people go from there? Let's say they've got a chart or a theme that they're really interested in, which they focused on from their review of the Chipbook. You guys have been doing this for a long time, before you were part of SemiAnalysis, right? Where's the connection to the rest of the SemiAnalysis organization?

Maybe you could tell me a little bit about what it's like working with other teams at SemiAnalysis, other data, and other research that we do beyond the 35 pages that a lot of people treat as maybe the entry point to this industry, even if they end up wanting to go bigger, deeper, or whatever it is.

I'll start off here. Ray, our memory analyst, Ray Wang—I think he was here a few weeks ago. Was it 3 or 4 weeks ago, Jordan?

Yeah, yeah.

I think he did. He put out a tweet—I can't remember if it was earlier this week or last week—and I'm going to botch exactly what he said, but it was something along these lines: “Every single day, I am amazed at the quality of people and research that are at SemiAnalysis.”

I have to echo that because those who are fortunate enough to be inside our chaotic Slack channels know that these are all people who are in it for the love of the game. Everyone's always sharing ideas, sharing data, and sharing what they're seeing. It's obviously massively encouraged to talk about what it is that you're seeing.

Again, SemiAnalysis covers the entire supply chain. The SemiAnalysis product portfolio spans the entire semiconductor value chain, right? You have industrials, energy, data center, accelerators, and the core research team doing a great job. Everyone's doing a great job.

We, in our small part as the Chipbook team, have had the great fortune of being able to feed that into the different people that we talk to. We talked a lot about WFE in this conversation.

Obviously, we're talking to Jeff and the great WFE team about what we're seeing in the different WFE arenas to help inform them in their research process. If we see something interesting that we think is a good output to put into a core research piece, we'll put it there. I'll give an example.

One of the datasets that we were tracking was this whole idea of ABF substrates, which has exploded over the past 2 weeks. We put it in the Chipbook, and in addition, we also put out a Core Research piece because we thought that was a good platform to put it out, talking about the massive tailwinds that were coming to the ABF substrate space.

You basically had GPUs that were increasing in size and layer count, and that was driving demand. Suddenly, CPU shortages were coming up, and everyone was talking about CPU substrates as well. This was all leading to the fact that we knew and were seeing really interesting data on what was happening in ABF production, both production value and production volume, out of Japan and Taiwan, which are the 2 biggest ABF suppliers.

We put that out there. Having our hands in all these different datasets across the value chain has been really fun, in the sense that we've been able to contribute and cross-pollinate across other teams.

Yeah.

That's essentially what the integration looks like.

As Chaim said, this is the smartest group of semi analysts anywhere assembled in the world. We worked on the buy side, covered semis, and thought we were really smart. You show up here, and you're just surrounded by people who get it on a very deep level.

I think what we maybe add to the team—and what we all add, as Chaim said—is that we're contributing to Core Research. We're talking to Ray, we're talking to Shravan, and all of us are sharing information that helps build up what we're able to provide our customers. Very broadly, ours is an objective, quantitative set of data that complements other qualitative research.

Mm.

To me, it's a no-brainer: you're subscribing to Core Research and Chipbook. Those are the basic building blocks and table stakes for understanding what's going on in the industry. You have Core Research, which is a brilliant research tool that every hedge fund should have their hands on.

Then you want to have the complementary dataset that helps inform and gives more depth, more granularity, more color, or more whatever finance word you want to use. It makes it more meaningful. So you have the qualitative and the quantitative before you even get to the models, right? Then you're in a different world entirely.

I think that we complement each other on that qualitative-quantitative basis.

I'm going to jump back a few topics because I think we wanted to say something about this and then forgot because we got onto something else. Jordan, you asked about the diffusion rules and where you see geopolitics playing into the supply chain, and how we were able to catch up on that.

5. Tracking Geopolitical Supply Shifts

One of the tweets that we put out earlier this week, which kind of went viral, was talking about one of the things we're tracking: smartphone imports into the US. Since the Trump nomination in November—I'm pretty sure November 2025—we saw this massive drop in US imports of smartphones coming from China.

That was interesting because, again, we talk about the supply chain: how do supply chains move and react when there's geopolitical instability? What is changing? What's happening? How could you track that? How could you know?

Obviously, it would be nice if you had a guy who went to the Foxconn assembly and test facility in China and then reported back, saying, "Yep, it's still here," or, "Nope, they moved it out of here." But the reality is, you don't have that all the time.

We try to supplement that vacuum of information with the sources that we tap into. This is something interesting. It's important to highlight, specifically for this dataset, that this exploded. It's almost at 600,000 views.

The smartphone supply chain is obviously so massive. Every single processor that goes into an Apple iPhone is coming from TSMC in Taiwan, and then it's going to China. When you look at a smartphone imported into the US and the country it came from, that's not representing the entire phone and everything that's there. There are different rules about the percentage of components that need to be in there.

But the fact is, it used to always be coming from China, even though different components came from different countries. Then that basically dropped to 25%, and suddenly it's coming from other countries like Vietnam and India. Clearly, the supply chain shifted and realized that even if it's the last step of FATP—final assembly, test, and packaging—it needs to move out of China so that we're able to respond to whatever is going to be happening in the political landscape in the US.

There was a very clear response to that, and that was something that we tracked here. Simi, maybe you want to talk about the other tweet that we put out earlier this week on Qatari helium, right? Everyone's talking about the war in the Middle East and how it impacts the semiconductor supply chain. We put out something about that.

Yeah.

Maybe you want to respond about what this is, why it's important, and why this didn't exist until we put it out—and that's why it blew up.

Let me add one more thing about what Chaim was saying about the smartphones, because the shift that Chaim was describing is actually even more pronounced for PCs. What I think I like best about the tweet is that, in addition to the hundreds of thousands of people who saw it, there was almost a lively debate that we started because of that tweet.

Basically, is it real or not? Some people are like, "It's not real. It's just final assembly. It's just putting a sticker made in Vietnam on it so it won't be made in China." Some people are like, "This shows that we're moving in a direction. Once the ball gets rolling, who knows what's going to happen next?"

Maybe the whole thing is a shenanigan. Maybe it's not even real. They're just making it look that way so it's not coming from China. I don't know for sure. I don't think any of us really know for sure right now, but I think the data opens up a conversation that forces the world to look at this and say, "What's happening? Is this real or not? Can a supply chain shift that quickly?"

I wish I could show you the PC one. It's just the opposite direction. It was about 90% coming out of China, and now it's about 6% in terms of PC imports to the US.

Yeah, significant trend.

The question is, not only does this data allow us to give answers, but it provides us with a direction in terms of asking questions. That's also what you're saying, Jordan: that's how we work in SemiAnalysis. We're going to put that data out there, and then we're going to say, "Doug, Shravan, Ray, Dylan, Dan, what do you guys think?"

As a result, the conversation is a lot more meaningful within the company and within the entire ecosystem. I just wanted to add that. Chaim said something about the Qatari helium. That was an interesting one.

Yeah.

As soon as the Iran war started, one of the implications recognized by the industry was that Qatar, which was getting bombed by Iran, provides a large portion of the semiconductor industry's helium, which is used in chip manufacturing.

What happens when the Qataris' facilities are blown up or shut down, or shipping lanes are closed? What happens to all the helium that's coming from Qatar? There was one camp that was like, "Who cares? Don't worry. The supply chain will be replaced elsewhere." There was another camp that was like, "Oh, we're in big trouble because a lot of the helium comes from Qatar, and we need that."

What bothered us was: where's the data? What is the actual number? How much of the helium for semiconductor manufacturing comes from Qatar? The first thing we did was figure out, when it comes to Korea, Taiwan, and China, what's the answer? What percentage comes from Qatar?

What we found is that well over 50% in all 3 countries—meaning the major manufacturing facilities—of their helium is coming from Qatar.

So it's a problem. But then the question becomes: How quickly can that supply chain reinvent itself and start getting helium from the U.S. or from Russia? The chart over here shows that very quickly, the supply chain was able to make an about-face. Whereas the Taiwanese had stopped importing helium from the U.S. and relied almost entirely on Qatar, all of a sudden, they were able to shift their supply chains over to the U.S., which is a good sign.

Now, can they go all the way to 100% from the U.S.? I think so, but that's something we're definitely going to want to track. The next question I would ask is: What does the pricing look like? There was some reason why the Taiwanese decided to stop importing helium from the U.S. and start importing it from Qatar. Is that a pricing question?

If they now have to shift back to U.S. facilities, what does that do to pricing? It could be that helium is such a small part of the total BOM that it won't have a huge impact, but that's definitely something to ask. I thought this was a great example. As Chaim said, for some reason, nobody else in the world had gone through the trouble to actually look at the data. Everybody was talking vaguely.

Yeah.

They have 6 months of supply. They have 2 months of inventory. Does it really come from Qatar? Does it come from Russia? Our question always is: Show us the data. Let's just look at the facts, and then we at least know what we're talking about to have an intelligent conversation. Then we can begin to ask the next questions.

In some ways, what's jumping out to me is the fact that this is a process. You guys have a system. You have access to the data; you know where to look. You have the software built to be able to build these charts at, effectively, a moment's notice when something happens in the world.

This Qatar helium chart, for those just listening, is from April 12, 3 days before we're recording this conversation, and the tweets about the Foxconn China assembly network stuff were from April 13. You guys had a big week of viral tweets talking about this stuff.

But I think it goes to show again that we don't know what's coming. We don't know what geopolitical trends are coming. We don't know what macro, big-picture stuff is coming in terms of demand for tokens or constraints in the supply chain. Having the process established to be able to build a chart and get some insight from that data is, in some ways, more valuable than actually having 1 individual data point.

You need to be able to adapt to whatever area of the market is in focus. I don't think a lot of people forecast helium being a big focus 3 months ago or 6 months ago, right?

If you told us we'd go viral on a helium post—

Yeah.

Last week, I don't think we would've called that one.

Yes. Right.

Yeah. Yeah.

There's the meme template of the guy who's like, “Wait, you're talking about this?” Today, it came out because of this Allbirds thing and their transition to now being an AI company. It's the guy in the podcast saying, “Wait, did you say Allbirds? You know, the AI company?”

When I hear helium for the first time, I'm like, “Helium? You know, like the inflatable balloon substance, gas?” But no, apparently it's for semiconductors.

To go back to what Simi started with in this whole thing, we were from the buy side. The reason we were talking about the process, how we do this, is that when you're on the buy side, you're looking for data. You're looking for things to inform you: How much of an impact is this actually?

The reason I love that helium tweet was because—I don't know if this is true about your Twitter timelines—but my Twitter timeline was dominated by helium.

It was literally that chaotic, because there were those who were saying, “Guys, calm down. This is totally negligible. Stop getting worked up over it.” And then there was the other side that was like, “TSMC is going to zero tomorrow. It's over. It's so over.”

Then Simi and I look at each other and we're just like, “Can someone please give me an intelligent answer as to what it has been up until now? Are there other suppliers who get in there?” And then, of course, as Simi said, there are the questions after that: How does this impact pricing? Who are those other suppliers who could benefit from it?

But let's answer the question: Is this a big deal or not? We've developed this process: Let's answer the question. Who's impacted by this? Where are these different sources coming from? Can we get an answer that's in the data?

Again, the data is objective. We don't manipulate the data. We don't change the data. The data is what the data is. That's why I particularly like that chart, because there was so much noise about helium. I was just like, “Let's cut through the noise and find numbers that could either back this up or not back this up.”

It would be nice if you got the procurement manager at TSMC on the phone to just tell you, “This is the amount that we get in. We are or are not able to get it from other suppliers.” But you don't have access to that. You can't. It would be nice if you could, but you don't.

Where can you find alternatives to that in order to inform your decision-making process and your investment process? That's where Simi and I, and the ChipBook product, try to be handy.

6. TSMC And The Pacific Theater

I have to add one more thing, though, because right now I'm going to be an armchair geopolitical strategist. I think that Simi talked about the war in the Middle East right now, and there's one thing that we need to be talking about that people aren't thinking about.

I think it's important to realize—and obviously, there's a lot in the air right now with this war and what it means—that the largest winner or loser of this war, funnily enough, from my perspective, is actually TSMC. Nobody's talking about that. I'm going to give my wacko perspective as to why I think that is.

Okay.

Let's frame this war right now, and then I'll say how this ties into TSMC. If you think about the war right now, you have the U.S. fighting with a partner nation—in this case, Israel—who is a technological ally to the U.S. and provides a lot of technology that goes into the U.S. They are fighting what is, for the U.S., an adversary and, for Israel, an existential threat. They're managing a campaign together in order to try to take out this enemy.

How this war is going to play out is going to massively impact a new theater—the next theater. You always need to be thinking from the U.S. perspective. There's obviously a very big reason why Israel would want to be in this war, but you always need to be thinking about what's happening on the Pacific front with China. I think that the U.S. perspective has been and always will be what's happening there.

Now, how does this relate to TSMC? We framed what the situation is like right now in the Middle East; now let's move that to the Pacific Theater. You have China, which is an adversary to the U.S. and, in a way, an existential threat to a small island there, Taiwan, which is a technological ally to the U.S. because they're providing a lot of the backbone to the largest U.S. companies.

SemiAnalysis put out that table showing that 8 out of the largest 10 companies by market cap all rely on TSMC. Putting this into the framing, you have China, you have the technological partner, which is Taiwan, and the U.S.

If there were ever to be a future campaign in the Pacific Theater, if the U.S. is able to effectively execute this campaign and show that it is actually able to fight with a partner nation in the theater against an adversary, that could be seen as a massive deterrent in other theaters also. In this case, moving to the Pacific Theater, if the U.S. is able to come out of this war as the proclaimed winner—and I think that's still up in the air; people still don't know—but if they are able to create that effective deterrent against the Chinese adversary, that obviously means there's a big deterrent against the Chinese making a move on Taiwan.

Everyone's talking about 2027, and the biggest winner from that is obviously TSMC.

Now again, it could also be the biggest loser, because if the deterrent is not effective and it doesn’t end up repelling China, then that’s going to be an issue. But when you think about what the outcome of this war in the Middle East is going to be, yes, obviously there are massive implications for what the Middle East is going to look like. But if you’re thinking two steps ahead, one of the big companies that’s probably going to have the biggest impact from this is TSMC, and that’s something people need to be thinking about.

And when you see the outcome of this war, because again, this entire conversation has been about signals, right, how this war ends is going to have some downstream effect on what’s going to happen to TSMC. So that’s me taking off my SemiAnalysis nerdy semi-analyst hat, and suddenly I’m a geopolitical analyst. But whatever, just a random tangent that came to mind.

Yeah, moving the focus from the Strait of Hormuz to the Strait of Malacca going forward or something like that.

Chips and wafers and war.

Okay. I don’t know if we want to coin that one on this podcast. All right, guys. Well, look, I learned a lot. I appreciate the overview of chips and wafers, and I definitely appreciate the walkthrough of some of these examples.

We’ve got to do this again soon, because I think some people need to listen to this, check out some of the data, and then watch to see how things play out in the next few months to see if we have the same track record while they’re paying attention that we claim to have as we look back at some of the previous tweets or previous calls that we’ve made. Anyway, thanks so much.

Thanks, Jordan. Appreciate it. If people are interested, you can find the ChipBook on the SemiAnalysis website. You can even download a sample there. Maybe we'll put that in the show notes. It's there; it exists. And yeah, thanks for the time, Jordan. This was awesome.

Do we have show notes?

We have a transcript, yeah. But we can definitely put—

Put it in the show notes.

We can definitely put links to SemiAnalysis.com, email address sales@semianalysis.com, SemiAnalysis.com/chipbook or /institutional/chipbook. That's probably the place to go.

/chipbook.

Yeah. We'll put the links in the show notes and the description on YouTube and Spotify and wherever everybody else is listening to this.

Awesome.

Yeah. Cool. Thanks, guys.