Ep. 009 - Using Open Source Data To Drive Investment Decisions (ChipBook) | Chaim Eisenberg, Simi Sherman, Jordan Nanos

Summary

• ChipBook made an early call on the memory supercycle via Taiwanese conventional DRAM inventory — which built to historic highs, then began dropping in summer 2025 and has now fallen “eleven or twelve months in a row.” Simi Sherman’s read: demand outstripping supply, visible long before Samsung or Hynix printed it. The job has now flipped from idea generation to exit-timing — tracking Korean, Taiwanese, and Chinese memory shipments across DRAM, Flash, and HBM so funds sitting on huge positions know the cycle’s end “before Hynix reports on a company call.”
• The team is watching China WFE guidance closely after it missed for a second straight year. After the 2024 stockpiling ramp (heaviest in lithography) ahead of Biden-era restrictions, every WFE company guided 2025 China revenue down 20-25% — but monthly import data ran “on par at the same levels with what twenty twenty-four was like” and the year ended basically flat. Now the same 20-25% down guidance is out for 2026, while ASMI is “actually seeing China go up a little bit” — and inspection tools have held up longer than front-end, a split next week’s ChipBook will detail.
• The Photronics story is the mosaic-theory showcase: photomask writer imports into China slowed, the “why” turned out to be that chipmakers refused to share chip designs with Chinese mask shops, making photomasks “a huge, huge onshoring motivation” — so Photronics (PLAB) was “a huge long” and “the stock like doubled.” Simi’s point: “nobody was handing you that story, but by tracking the data, you begin to know what questions to ask.”
• US import data shows the smartphone supply chain visibly fleeing China — the China share dropped to roughly 25% since the Trump nomination, shifting to Vietnam and India, and PCs are starker still: from ~90% out of China to ~6%. Whether it’s a real FATP migration or a “made in Vietnam” sticker shenanigan is openly debated — Simi concedes “I don’t think any of us really know for sure right now,” but the data forces the question.
• On the viral Qatari helium panic, the team’s answer to both the doomers and the dismissers was the same: “show us the data.” Well over 50% of helium in Korea, Taiwan, and China comes from Qatar — a real exposure — but Taiwan’s imports shifted back to the US very quickly once the Iran war started. Open questions worth tracking: whether the US can supply 100%, and what the forced switch does to pricing.
• Chaim Eisenberg’s self-declared “wacko perspective”: the biggest winner or loser of the Middle East war is TSMC, “and nobody’s talking about that.” His deterrence logic maps the US-Israel campaign onto the Pacific theater — China as adversary, Taiwan as technological ally with 8 of the 10 largest companies by market cap relying on TSMC. A US “proclaimed winner” outcome deters a Chinese move on Taiwan (“everyone’s talking about 2027”); a failed deterrent makes TSMC the biggest loser.
• The product thesis: semiconductor alternative data is scarce — at a recent conference of ~60 alt-data vendors pitching dozens of hedge funds, “we were the only guys who were selling semiconductor-related data.” The edge isn’t the openness of the data but the OSINT-style craft of finding, cleaning, and reading it: granular, real-time signals (WFE ships 6-18 months before production, followed by installation and ramp) that turn “interesting information into actionable intelligence” before it hits the print.

Deep dive

1. Two ex-buy-siders selling the only semi alt-data in the room

• Simi’s origin story: both he and Chaim covered semis at a hedge fund, where “you’re constantly looking for data” to generate ideas, validate a thesis, and then time it — “how long do you hold it? When do you size it? When do you get out?” At a recent alt-data conference, roughly 60 vendors pitched dozens of hedge funds, and “we were the only guys who were selling semiconductor-related data.” Semi investors, in his words, are “thirsty.”
• The core claim: the data is all open source and public, “but it’s messy, and it’s distributed all over the world. It’s in different languages. It’s coded” — and mostly “you don’t even know what you’re looking for.” The mission statement: “turn that interesting information into actionable intelligence” from import/export data, production statistics, and inventory.
• Chaim’s framing from his military years — the OSINT parable: intelligence agencies initially dismissed open-source intel (“if everyone can access it, what use do I have for it?”), then learned it’s real-time and reaches where signals and human intelligence can’t — the guy filming a street corner and uploading to YouTube. Same answer applies to open-source semi data: “yes, it’s super valuable, but you need to know how to find it… to distinguish between what’s noise and what’s a true signal.”

2. Granularity, or the danger of the wrong map

• Simi’s attack on headline WFE metrics: banks publish “WFE is up ten percent,” but WFE is “the broadest category in the world” — boule and wafer-slicing tools, deposition, etch, litho, ion implanters, flip-chip and wire bonders, even inspection and metrology. Tracking KLA metrology off aggregate WFE means investing on a metric “far too broad to be meaningful.”
• His signature line: “the only thing worse than having no map at all is having the wrong map” — so the value is maximum granularity: logic versus memory, and within memory flash versus DRAM versus HBM, by country, knowing who’s making what for which customers.
• Chaim on why investors underestimate the problem: nobody comprehends the supply chain’s scale. A chatbot token “started four thousand steps earlier when some random Japanese chemicals company was making a polysilicon boule” — and seeing the whole wave move, upstream signal to downstream impact, is the product.

3. The memory call: Taiwanese DRAM inventory rang the bell a year early

• The mechanism, as Simi tells it: Taiwanese conventional DRAM inventory (Chaim’s clarification: non-HBM, at the time non-AI) built “to historic highs” for months — then in summer 2025 dropped for the first time in a year. They waited: a second month, a third, “and then we realize we’re starting a trend.” Now 11 or 12 consecutive monthly declines — demand outstripping supply, called from “a relatively obscure, specific data point” long before it hit company prints.
• The role has now inverted from idea generation to position monitoring: funds sitting on huge memory positions are “really nervous about when does this cycle end,” so the tracker follows memory shipments from Korea, Taiwan, and China across DRAM, Flash, and HBM — because the risk of exiting late, or too early, on positions that size is enormous.
• Chaim’s caveat on epistemics: the silver-bullet data point “is highly unlikely to basically zero.” These are signals that sharpen the questions you ask — and timing matters because “if it’s already on the print of the company when results come out, it’s too late.”

4. China WFE: guidance said down 20-25%, the imports said flat

• Setup: WFE ships 6, 12, or 18 months before production; after installation and ramp, tracking it gives “a good 12 to 24-month preview into what wafer capacity is going to look like.” The team tracks WFE imports into China at the provincial level — YMTC, SMIC, Hua Hong, CXMT each sit in different provinces — because during the memory downturn China demand was what propped up equipment makers, reaching “25, 35, 45, even 50% of quarterly revenue.”
• The 2024-25 case study: massive stockpiling in 2024 (lithography most of all, though Simi interjects “deposition, etch, everything”) ahead of Biden-era restrictions. Every WFE company guided 2025 China down 20-25%; the monthly data ran level with 2024 all year, and China revenue finished “basically flat year over year.” Now 2026 guidance repeats the down-20-25% line — “maybe it’s true,” Chaim hedges, but ASMI is “actually seeing China go up in twenty twenty-six a little bit,” and the team is watching closely.
• Simi’s granular refinement: the slowdown may not be uniform — inspection equipment into China held on longer than front-end tools (“I don’t know if it’s holding on anymore. We’re gonna find out more in the next week”). Names hit by macro noise but performing on earnings are “a huge buying opportunity,” and next week’s ChipBook will show the tool-by-tool split.
• On calling bottoms: wire bonder orders into China during COVID were “so out of whack with the historic norm” that when orders fell 10%, 15%, or 20% the market called a bottom prematurely. Simi’s rule: “They’ve always hit the bottom. They’re constantly hitting the bottom” — only historic run rates tell you how far something can still fall.

5. Mosaic theory and the Photronics double

• The chain of questions, as told: photomask writer imports into China slowed after a big buildup → why isn’t the cheap capacity being used? → because getting a mask made in China means sharing your chip designs with Chinese companies, which customers refused → therefore photomasks are “a huge, huge onshoring motivation” → therefore Photronics (PLAB), an American maker even with a Chinese subsidiary, “is a huge long.” “Once they came out with that story, the stock like doubled.”
• Simi on how sophisticated clients consume this: hedge fund investors “are not looking for somebody to feed them the answer. They need pieces to the puzzle” — every data point is a puzzle piece for the mental model, and “by tracking the data, you begin to know what questions to ask.”

6. What a ChipBook actually is — and how it plugs into SemiAnalysis

• The format: a monthly ~35-page PDF of charts. The first 10 are fixed fundamentals — hyperscaler capex, PC/smartphone/auto shipments, wafer shipments, PCBs. The next ~25 rotate from 200-300 tracked datasets, chosen when interesting or actionable (“sometimes they’re just boring and nothing happens that month”). Every page carries three things: what the data is, which public companies connect to it, and the month’s read — plus a one-page executive summary flagging the top implications with page references.
• The cross-pollination example: ABF substrates, published both in ChipBook and as a core research piece — GPUs growing in size and layer count plus CPU shortages coming up and driving discussion of CPU substrates, tracked through production value and volume out of Japan and Taiwan, “the two biggest ABF suppliers.”
• Simi’s positioning of the product internally: “ours is an objective quantitative set of data which complements other qualitative research” — subscribing to core research plus ChipBook is, to him, “table stakes” for understanding what’s going on in the industry.

7. The viral week: smartphones fleeing China, and the helium panic

• The smartphone tweet (nearly 600,000 views): since the Trump nomination — “I’m pretty sure November, November 2025” per Chaim — US smartphone imports from China dropped massively, to roughly 25%, with Vietnam and India picking up share; the read is that final assembly, test and package moved out of China to hedge US politics. Simi adds PCs are “just like opposite directions… it was like 90% coming out of China, and now it’s like 6%.”
• The honest uncertainty, kept intact: is it real or “just putting a sticker made in Vietnam”? “Maybe the whole thing is a shenanigan… I don’t think any of us really know for sure right now, but the data opens up a conversation that forces the world to look at this.”
• Qatari helium: when the Iran war started, Chaim’s timeline split between “this is totally negligible” and “TSMC is going to zero tomorrow. It’s so over” — with nobody showing numbers. The team’s finding: well over 50% of helium into Korea, Taiwan, and China comes from Qatar — a genuine exposure — but Taiwan, which had stopped importing US helium and relied almost entirely on Qatar, “very quickly” shifted back to the US. Next questions: can it reach 100%, and what does the switch do to pricing? “Our question always is, show us the data.”
• Jordan’s synthesis — the process is the product: the helium chart was built three days before recording, the Foxconn tweet the day before. Nobody forecast helium being a focus; having the system to chart any theme on a moment’s notice beats any single data point. Simi: “If you told us we’d go viral on a helium post last week, I don’t think we would’ve called that one.”

8. Chaim’s armchair-geopolitics closer: the war’s biggest potential winner or loser is TSMC

• The self-labeled “wacko perspective”: frame the Middle East war as the US fighting alongside a technological partner (Israel) against an adversary, then transpose to the Pacific — China the adversary, Taiwan the small-island technological ally, with “eight out of the largest ten companies by market cap” relying on TSMC. If the US emerges “the proclaimed winner,” that’s “a massive deterrent” against a Chinese move on Taiwan (“everyone’s talking about 2027”) — biggest winner TSMC; if the deterrent fails, TSMC is the biggest loser. His hedge stays in: “I think that’s still up in the air; people still don’t know.”
• The tie back to the episode’s thesis — everything is signals: “how this war ends is going to have some downstream effect on what’s going to happen to TSMC.” Simi jokes, “Chips and wafers and war”; Jordan says he doesn’t know if they should coin it.