Semis Memo: Muscle Memory - AI 半導體供應鏈知識門檻正在提高¶

English Summary¶

Citrini Research launches a new "Semis Memo" series with the hiring of two semiconductor analysts (Zephyr and Jukan) to address a critical insight: as expected returns from mainstream AI semiconductor "layups" (Nvidia, Micron, SK Hynix, Broadcom) moderate, the next wave of alpha requires deeper supply chain knowledge. The introduction articulates why catching the most asymmetric trades in the "AI Bubble" demanded attention to details complex for generalists, plus strong views on supply chain dynamics 6-12 months forward. Three historical examples illustrate this: (1) Teradyne's robotics opportunity required understanding memory testing tailwinds; (2) Amkor's upside required recognizing chiplet architecture/advanced packaging demand driving OSAT growth; (3) the SK Hynix/Micron Q1 2024 outperformance call (while they languished in post-COVID RAM glut doldrums) required understanding HBM supply chain constraints and agentic AI's nature. The series promises to go "a layer deeper" on AI infrastructure trends while ensuring readers don't need a PhD in semiconductor design to keep up—positioning it as a tour of the most interesting, potentially overlooked opportunities in one of the planet's hottest sectors.

The Core Thesis: Knowledge Threshold Is Rising¶

The inflection point articulated: Citrini quotes their April 2025 piece "AI Phase Two":

"While one has indeed so far been able to get away with an understanding that only goes GPU-deep in AI, I believe that will rapidly begin changing...

In the meantime, generalists are left to keep up with a rapidly advancing technological landscape in semiconductors that can easily leave you behind... The shame in that is that it often robs you of the ability to truly understand what's behind the whole picture while ensuring you're properly positioned – a picture which will only become increasingly important when these narratives take hold in the real world."

The competitive reality: In order to have caught the most asymmetric trades of the "AI Bubble," investors had to: 1. Pay attention to things that were "a bit complex for a generalist" 2. Have a strong view on what would happen in the supply chain 6-12 months from now due to whatever was happening right now

The bad news: It is not going to get any easier from here. The firm can't sit idly by while expected returns from mainstream "layups" like Nvidia, Micron, SK Hynix, Broadcom etc. moderate.

Why Citrini Hired Semiconductor Analysts¶

The strategic response: Citrini hired "talented semiconductor analysts" who can: - Explain to "ignorant generalists" what's going on behind the Wall Street curtain - Translate supply chain dynamics back into thematic upside

The team addition: Two new members—Zephyr and Jukan—add deep semiconductor knowledge to Citrini's existing thematic and trading expertise. The combination: existing penchant for strong, occasionally non-consensus views on AI's continued proliferation + in-depth semiconductor supply chain knowledge.

What they've navigated so far: AI Data Centers, GPUs, Memory, Optical Interconnects, ASICs, and more.

Three Historical Examples Requiring Deep Knowledge¶

Example 1: Teradyne (TER) - Robotics Opportunity - What was required: Keep abreast of tailwinds for players in memory testing - Why it mattered: To underwrite Teradyne's robotics opportunity, understanding the memory testing sector's favorable dynamics was essential - Implication: A generalist looking only at "robotics" would miss the memory testing connection

Example 2: Amkor (AMKR) - Advanced Packaging Play - What was required: Recognize increasing demand for chiplet architectures and advanced packaging - Key insight: This would increase demand for OSAT (Outsourced Semiconductor Assembly and Testing) - Translation barrier: "If you haven't been paying attention" to OSAT, you'd miss the Amkor opportunity entirely - Implication: Supply chain knowledge = competitive edge

Example 3: SK Hynix & Micron - Q1 2024 Outperformance Call - Context: Both companies were "languishing in the doldrums of the post-COVID RAM glut" - What was required: Understand the constraints HBM placed on the supply chain AND the nature of truly agentic AI - Result: Citrini made the call that SK Hynix and Micron would outperform in Q1 2024 while most investors saw them as commodity memory plays stuck in oversupply - Validation: This became one of the most asymmetric trades as HBM demand exploded - Implication: Seeing around corners requires understanding supply chain bottlenecks (HBM) + demand drivers (agentic AI)

The "Semis Memo" Series Framework¶

Positioning: This is the first of a new series where Citrini will: 1. Go "a layer deeper" on current trends in the infrastructure powering AI 2. Ensure the audience doesn't need a PhD in semiconductor design to keep up 3. Provide a "tour" of the most interesting, potentially overlooked opportunities in one of the hottest sectors on the planet

What this inaugural memo covers (per the introduction): - Existing trends and second-order winners of the Memory Supercycle - Semicap Subsystems - Silicon Photonics - Early thoughts on watchlist opportunities: SpaceX supply chain beneficiaries to opportunities arising from the AI-driven SaaS selloff - List of single name stories worth keeping on radar

Recent Performance Validates the Thesis¶

Context setting: There is "a lot going on in AI and the semiconductor supply chain. Almost too much, it seems at times."

Recent validation: The wild performance in the most prominent players has reinforced the supply/demand balance Citrini described in "Post-Traumatic Supply Disorder" from their "26 Trades for 2026" report.

Specific sectors outperforming: Memory, storage, semicap, advanced packaging—these sectors have performed incredibly well over the past month (as of January 2026).

The moderating tailwind: As these mainstream names have run, their expected forward returns are moderating. This creates the necessity for deeper supply chain knowledge to find the next layer of opportunities.

The Knowledge Gap as Competitive Moat¶

The generalist's dilemma: Generalists are "left to keep up with a rapidly advancing technological landscape in semiconductors that can easily leave you behind."

The cost of ignorance: Being left behind "robs you of the ability to truly understand what's behind the whole picture while ensuring you're properly positioned."

The increasing importance: This picture "will only become increasingly important when these narratives take hold in the real world."

The inflection: While GPU-deep understanding was sufficient in AI's early phase, that era is ending. The next phase demands supply chain depth.

Strategic Positioning: Where Citrini Sits¶

Existing strengths: - Thematic and trading expertise - Strong, occasionally non-consensus views on AI's trajectory - Track record of identifying asymmetric opportunities early (Teradyne, Amkor, SK Hynix/Micron)

New capability addition: - Deep semiconductor supply chain knowledge (via Zephyr and Jukan) - Ability to translate complex semiconductor dynamics into actionable thematic upside - Capacity to see 6-12 months forward in the supply chain

The synthesis: Combining thematic vision with supply chain granularity to find second-order and third-order winners as AI infrastructure build-out continues.

繁體中文總結¶

Citrini Research 推出新的「Semis Memo」系列，聘請兩位半導體分析師（Zephyr 和 Jukan）來應對一個關鍵洞察：隨著主流 AI 半導體「穩贏球」（Nvidia、Micron、SK Hynix、Broadcom）的預期報酬趨緩，下一波 alpha 需要更深的供應鏈知識。Introduction 闡明了為何捕捉「AI 泡沫」中最不對稱的交易機會需要關注對通才來說有點複雜的細節，加上對供應鏈 6-12 個月後動態的強烈觀點。三個歷史案例說明這一點：(1) Teradyne 機器人機會需要理解記憶體測試順風；(2) Amkor 上漲空間需要認識到 chiplet 架構/先進封裝需求推動 OSAT 成長；(3) SK Hynix/Micron 2024 年第一季超額表現預測（當時還在 COVID 後 RAM 過剩低谷徘徊）需要理解 HBM 供應鏈限制和 agentic AI 的本質。該系列承諾在 AI 基礎設施趨勢上「深入一層」，同時確保讀者不需要半導體設計博士學位也能跟上——將其定位為這個星球上最熱門領域中最有趣、可能被忽視的機會巡禮。

核心論點：知識門檻正在提高¶

轉折點表述： Citrini 引用他們 2025 年 4 月的文章「AI Phase Two」：

"雖然到目前為止，只需要 GPU 深度的理解就能在 AI 領域生存，但我相信這將迅速改變...

與此同時，通才投資人被留在快速進步的半導體技術景觀中，很容易被拋在後面...遺憾的是，這往往剝奪了你真正理解整體圖景背後內容的能力，同時確保你正確定位——當這些敘事在現實世界中站穩腳跟時，這幅圖景只會變得越來越重要。"

競爭現實： 為了捕捉「AI 泡沫」中最不對稱的交易機會，投資人必須： 1. 關注「對通才來說有點複雜」的事情 2. 對現在發生的事情將在 6-12 個月後如何影響供應鏈有強烈觀點

壞消息： 從現在開始不會變得更容易。當 Nvidia、Micron、SK Hynix、Broadcom 等主流「穩贏球」的預期報酬趨緩時，公司不能袖手旁觀。

為何 Citrini 聘請半導體分析師¶

戰略回應： Citrini 聘請「有才華的半導體分析師」，他們可以： - 向「無知的通才」解釋華爾街幕後發生的事情 - 將供應鏈動態轉譯回主題性上漲空間

團隊增員： 兩位新成員——Zephyr 和 Jukan——為 Citrini 現有的主題和交易專業知識增添深厚的半導體知識。組合：對 AI 持續擴散的強烈、偶爾非共識觀點 + 深入的半導體供應鏈知識。

到目前為止導航過的領域： AI 資料中心、GPU、記憶體、光學互連、ASIC 等。

三個需要深度知識的歷史案例¶

案例 1：Teradyne (TER) - 機器人機會 - 需要什麼： 掌握記憶體測試領域參與者的順風 - 為何重要： 要承保 Teradyne 的機器人機會，理解記憶體測試領域的有利動態至關重要 - 含義： 只看「機器人」的通才會錯過記憶體測試連結

案例 2：Amkor (AMKR) - 先進封裝標的 - 需要什麼： 認識到 chiplet 架構和先進封裝需求增加 - 關鍵洞察： 這將增加對 OSAT（外包半導體組裝測試）的需求 - 轉譯障礙： 「如果你沒有關注」OSAT，你會完全錯過 Amkor 機會 - 含義： 供應鏈知識 = 競爭優勢

案例 3：SK Hynix & Micron - 2024 年第一季超額表現預測 - 背景： 兩家公司都「在 COVID 後 RAM 過剩的低谷中徘徊」 - 需要什麼： 理解 HBM 對供應鏈施加的限制 AND 真正 agentic AI 的本質 - 結果： Citrini 在大多數投資人將它們視為陷入供過於求的商品記憶體標的時，預測 SK Hynix 和 Micron 將在 2024 年第一季超額表現 - 驗證： 隨著 HBM 需求爆發，這成為最不對稱的交易之一 - 含義： 看到轉角需要理解供應鏈瓶頸（HBM）+ 需求驅動因素（agentic AI）

「Semis Memo」系列框架¶

定位： 這是新系列的第一篇，Citrini 將： 1. 在支撐 AI 的基礎設施當前趨勢上「深入一層」 2. 確保觀眾不需要半導體設計博士學位也能跟上 3. 提供這個星球上最熱門領域中最有趣、可能被忽視的機會「巡禮」

本期 memo 涵蓋內容（根據 introduction）： - 現有趨勢和記憶體超級週期的二階受益者 - Semicap Subsystems（半導體資本設備子系統） - Silicon Photonics（矽光子） - 關注清單機會的早期想法：從 SpaceX 供應鏈受益者到 AI 驅動的 SaaS 拋售機會 - 值得關注的單一個股故事清單

近期表現驗證論點¶

背景設定： AI 和半導體供應鏈「發生了很多事情。有時似乎太多了。」

近期驗證： 最突出參與者的瘋狂表現強化了 Citrini 在「26 Trades for 2026」報告中「Post-Traumatic Supply Disorder」所描述的供需平衡。

具體超額表現領域： 記憶體、儲存、semicap、先進封裝——這些領域在過去一個月（截至 2026 年 1 月）表現極佳。

趨緩的順風： 隨著這些主流標的上漲，它們的預期前瞻報酬正在趨緩。這創造了尋找下一層機會所需的更深供應鏈知識的必要性。

知識差距作為競爭護城河¶

通才的困境： 通才「被留在快速進步的半導體技術景觀中，很容易被拋在後面。」

無知的代價： 被拋在後面「剝奪了你真正理解整體圖景背後內容的能力，同時確保你正確定位。」

日益增長的重要性： 這幅圖景「當這些敘事在現實世界中站穩腳跟時，只會變得越來越重要。」

轉折點： 雖然 GPU 深度理解在 AI 早期階段是足夠的，但那個時代正在結束。下一階段需要供應鏈深度。

戰略定位：Citrini 的位置¶

現有優勢： - 主題和交易專業知識 - 對 AI 軌跡的強烈、偶爾非共識觀點 - 早期識別不對稱機會的記錄（Teradyne、Amkor、SK Hynix/Micron）

新增能力： - 深厚的半導體供應鏈知識（透過 Zephyr 和 Jukan） - 將複雜半導體動態轉譯為可行動主題性上漲空間的能力 - 在供應鏈中看到 6-12 個月前景的能力

綜合： 將主題願景與供應鏈細粒度結合，隨著 AI 基礎設施建設持續，找到二階和三階受益者。

Key Quotes¶

"While one has indeed so far been able to get away with an understanding that only goes GPU-deep in AI, I believe that will rapidly begin changing... In the meantime, generalists are left to keep up with a rapidly advancing technological landscape in semiconductors that can easily leave you behind... The shame in that is that it often robs you of the ability to truly understand what's behind the whole picture while ensuring you're properly positioned – a picture which will only become increasingly important when these narratives take hold in the real world." — AI Phase Two, April 2025

"In order to have caught the most asymmetric trades of the 'AI Bubble', you had to be paying attention to things that were a bit complex for a generalist. You also had to have a strong view on what would happen in the supply chain 6-12 months from now due to whatever was happening right now."

"Some bad news: it is not going to get any easier from here. We can't sit idly by while the expected returns from mainstream 'layups' like Nvidia, Micron, SK Hynix, Broadcom etc. moderate."

"That's why we've hired some talented semiconductor analysts, so they can explain to us ignorant generalists what's going on behind the Wall Street curtain (and we can then translate it back into thematic upside)."

"We're going to combine our existing penchant for having strong, occasionally non-consensus views on what the continued proliferation of Artificial Intelligence looks like with their in-depth knowledge of the semiconductor supply chain."

"For example, for us to underwrite Teradyne (TER) and its robotics opportunity, we had to keep abreast of the tailwinds for players in memory testing. To recognize the opportunity in Amkor (AMKR), we needed to recognize the increasing demand for chiplet architectures and advanced packaging that would increase demand for OSAT (Outsourced Semiconductor Assembly and Testing, if you haven't been paying attention)."

"Going back even further, for us to make the call that SK Hynix and Micron would outperform in Q1 2024 (while they were languishing in the doldrums of the post-COVID RAM glut), we needed to understand the constraints HBM placed on the supply chain and the nature of truly agentic AI."

"This is the first of our new 'Semis Memo' series, where we'll go a layer deeper on current trends in the infrastructure powering AI while also ensuring our audience doesn't need to go get a PhD in semiconductor design to keep up."

"Consider this a tour around what we view as the most interesting, potentially overlooked opportunities in one of the hottest sectors on the planet."

Investment Framework: From Examples to Principles¶

Principle 1: Supply Chain Bottlenecks → Second-Order Winners¶

Teradyne case study logic: - Primary trend: AI training/inference drives demand for advanced memory (HBM, GDDR) - Supply chain constraint: These memory types require more sophisticated testing - Second-order winner: Memory testing equipment providers like Teradyne - Generalist miss: Focusing only on memory manufacturers (Micron, SK Hynix) without seeing testing equipment upside

Replicable pattern: - Identify primary AI infrastructure trend - Map supply chain dependencies - Find bottlenecks or constraint points - Identify specialized players serving those bottlenecks

Principle 2: Architecture Shifts → Ecosystem Beneficiaries¶

Amkor case study logic: - Architecture trend: Chiplet-based designs becoming standard (driven by Moore's Law slowdown + heterogeneous integration) - Packaging requirement: Chiplets require advanced packaging (CoWoS, fan-out, etc.) - Ecosystem shift: Increased outsourcing to OSAT providers due to complexity - Second-order winner: OSAT players with advanced packaging capabilities - Generalist miss: Seeing chiplets as "just another chip design" without understanding packaging implications

Replicable pattern: - Track architectural transitions (monolithic → chiplet, 2D → 3D stacking) - Identify new requirements these architectures impose - Find specialized service providers addressing those requirements - Assess competitive moats (technical expertise, capital intensity, customer stickiness)

Principle 3: Constraint Recognition → Counter-Consensus Positioning¶

SK Hynix/Micron case study logic: - Market consensus: Memory sector in post-COVID glut, commodity dynamics, avoid - Hidden constraint: HBM production bottlenecked (limited suppliers, technical complexity, long qualification cycles) - Demand catalyst: Agentic AI architectures require massive memory bandwidth (HBM essential) - Constraint + Catalyst = Supply shortage despite apparent glut in commodity DRAM - Counter-consensus positioning: Buy memory stocks while "languishing in doldrums" - Result: Asymmetric upside as HBM shortage becomes obvious

Replicable pattern: - Distinguish between commodity segments and constrained segments within same sector - Identify demand catalysts that specifically hit constrained segments - Position before market consensus shifts - Requires deep supply chain knowledge to separate "real glut" from "apparent glut masking specific shortage"

Meta-Principle: 6-12 Month Forward View Requires Today's Supply Chain Understanding¶

The time horizon challenge: Investment returns come from seeing what happens 6-12 months forward, but that future state is determined by today's supply chain constraints, capacity additions, and technology transitions.

Knowledge requirement: Understanding today's supply chain in depth provides the foundation for forecasting tomorrow's bottlenecks and beneficiaries.

Competitive edge: As mainstream trades (Nvidia, Broadcom) become crowded, edge migrates to those who can see second-order and third-order effects early.

The Broader Context: Why This Series Matters Now¶

Timing: The Moderating Tailwind¶

What's happened: Memory, storage, semicap, advanced packaging sectors have performed incredibly well over the past month (as of January 2026). This validates Citrini's "Post-Traumatic Supply Disorder" thesis from "26 Trades for 2026."

What's next: As these obvious plays have run, expected forward returns are moderating. The "easy money" phase (buying Nvidia, Micron, SK Hynix) is ending.

Implication: Investors need to go deeper to find the next layer of opportunities. This requires more specialized knowledge than the GPU-focused AI investment thesis of 2023-2025.

The Knowledge Arms Race¶

Phase 1 (2023-2024): Understanding "AI = GPU demand" was sufficient. Buy Nvidia, ride the wave.

Phase 2 (2024-2025): Recognizing "AI = memory/storage/power constraints" separated winners. SK Hynix HBM play, data center power infrastructure, cooling solutions.

Phase 3 (2026+): Understanding supply chain bottlenecks, second-order winners, and architectural transitions becomes essential. Memory testing, OSAT, semicap subsystems, silicon photonics.

The escalation: Each phase requires more specialized knowledge. Generalists who thrived in Phase 1 struggle in Phase 3 without deepening their understanding.

Why Citrini's Approach Matters¶

The translation layer: Most semiconductor analysts write for other semiconductor specialists. Their insights are often inaccessible to generalist investors without deep technical background.

Citrini's value proposition: 1. Hire semiconductor experts (Zephyr, Jukan) 2. Have them identify supply chain dynamics and opportunities 3. Translate insights into thematic investment narratives accessible to informed generalists 4. Provide the "go a layer deeper" knowledge without requiring a PhD

The positioning: Bridge between technical expertise and actionable investment strategy.

What the Full Article Likely Covers (Based on Introduction)¶

⚠️ Speculation based on stated topics—actual content is paywalled

Memory Supercycle Second-Order Winners¶

Likely coverage: Beyond Micron/SK Hynix/Samsung, who benefits from HBM production ramp?
Potential angles: Testing equipment (Advantest, Teradyne mentioned), substrate suppliers, packaging materials, memory controller IP
Investment thesis: As HBM supply scales up, look for equipment/materials providers with exposure

Semicap Subsystems¶

Likely coverage: Components and subsystems that go into semiconductor manufacturing equipment
Potential angles: Precision motion control, vacuum systems, chemical delivery, metrology tools
Investment thesis: Semicap capex flowing to subsystem specialists with high barriers to entry

Silicon Photonics¶

Likely coverage: Optical interconnects for AI data centers, replacing copper for high-speed chip-to-chip communication
Potential angles: Photonic chip manufacturers, optical transceiver companies, laser suppliers
Investment thesis: As data center AI interconnect bandwidth requirements exceed copper limits, silicon photonics adoption accelerates

SpaceX Supply Chain Beneficiaries¶

Likely coverage: Companies supplying components to SpaceX's Starlink and Starship programs
Potential angles: Could tie into AI infrastructure (Starlink for edge AI), or simply diversification beyond pure AI plays
Investment thesis: SpaceX's production ramp creates opportunities for specialized suppliers

AI-Driven SaaS Selloff Opportunities¶

Likely coverage: SaaS companies punished by market for AI disruption risk, but actual resilience or AI integration potential
Potential angles: Vertical SaaS with proprietary data moats, infrastructure SaaS benefiting from AI workloads
Investment thesis: Market overreaction to "AI will replace SaaS" creates mispricing

Single Name Stories¶

Likely coverage: Specific stock ideas with detailed investment theses
Potential approach: Deep dives on 3-5 companies with catalysts in 2026

Personal Reflection¶

This introduction is a masterclass in positioning a new research product. Citrini accomplishes several strategic objectives simultaneously:

1. Establishes credibility through track record: The three historical examples (Teradyne, Amkor, SK Hynix/Micron) aren't just success stories—they're proof of concept for the investment approach. Each required non-obvious supply chain knowledge that generalists lacked. The SK Hynix/Micron call is particularly powerful because it was counter-consensus (buying during "doldrums") and spectacularly validated (HBM shortage became obvious 6-12 months later).

2. Creates urgency through diminishing returns narrative: "Expected returns from mainstream layups... moderate" is a polite way of saying "if you're still just buying Nvidia, you're late." This motivates readers to seek deeper insights. The timing aligns with real market dynamics—memory/storage/semicap have indeed performed well recently, and forward multiples have compressed.

3. Justifies team expansion as strategic necessity: Hiring Zephyr and Jukan isn't just scaling up—it's acknowledging that the knowledge threshold has risen beyond what thematic generalists can handle alone. This frames the semiconductor analysts as essential infrastructure, not optional add-ons.

4. Positions the series as accessible depth: "Go a layer deeper" while "ensuring our audience doesn't need a PhD" is the sweet spot. Citrini recognizes that their audience wants edge but can't/won't become full-time semiconductor specialists. The "translation layer" is the product.

5. Foreshadows non-obvious opportunities: The topics (memory testing, OSAT, semicap subsystems, silicon photonics, SpaceX supply chain) are deliberately one layer removed from obvious plays. These aren't "buy more Nvidia" recommendations—they're "here's what Nvidia's growth means for the supply chain" insights.

The meta-lesson for investors: The introduction itself is evidence of the thesis. If a sophisticated research firm is investing in semiconductor-specific talent and launching a dedicated series, that's a signal that the knowledge threshold really is rising. Firms don't hire expensive specialists for content that generalists can produce.

What's likely valuable in the full article: Given Citrini's track record, the specific company names and investment theses in the paywalled sections probably identify genuine second-order opportunities. The Memory Supercycle section likely names specific testing/materials/substrate companies. The Semicap Subsystems section probably highlights motion control or vacuum system suppliers. The Silicon Photonics section may identify emerging photonic chip companies before they're widely covered.

The strategic question for non-subscribers: Is the knowledge gap between this introduction and the full article worth the subscription cost? If you're allocating capital in AI/semiconductor supply chain, probably yes. If you're a generalist trying to stay informed, the introduction's framework (supply chain bottlenecks → second-order winners, architecture shifts → ecosystem beneficiaries, constraint recognition → counter-consensus positioning) may be sufficient to apply independently.

Final observation: The irony is that by making this argument publicly in the introduction, Citrini is simultaneously helping and competing with their audience. Helping: providing the framework for thinking about supply chain opportunities. Competing: demonstrating that without dedicated semiconductor expertise, you're at a disadvantage. The resolution: subscribe for the specific insights, or hire your own Zephyr and Jukan.