P1-C10 · 5 Real Cases to Build Intuition¶
Core Takeaway
Real cases build intuition 10x faster than theory — 9 chapters of framework, this chapter applies it all to 5 real events.
AI Industry Knowledge — History → Technology → Supply Chain → Business → Applications → Geopolitics
P1-C10 (Part 1's final chapter). After this chapter, you can apply the C1-C9 framework to 5 real cases, each dissected from 6 dimensions: history/technology/industry/business/applications/geopolitics.
1. The Problem: 9 Chapters of Framework in Your Head, But Can't Use It = Haven't Learned¶
From C1-C9 you already know: - C1 History + Winter Cycles - C2 Transformer + Scaling Laws - C3 NVDA vs Intel Positioning - C4 LLM Intuition - C5 Hardware Stack - C6 Supply Chain Map - C7 Value Capture - C8 Application Layer - C9 Geopolitics
But not applying the framework to real events = learning nothing. This chapter gives you 5 real cases, each dissected across 6 dimensions — forcing you to ground the framework.
2. The Solution: 5 Cases × 6 Dimensions¶
Each case has a 5-section structure:
- Background: Industry state before the event
- Event: What happened (specific time / numbers)
- Market Reaction: How prices moved (short-term / long-term)
- How Your Thesis Applies (mapping to C1-C9 chapters)
- Lesson Learned
3. How It Works: 5 Cases in Detail¶
3.1 Case 1: DeepSeek Selloff — NVDA -17% / -$593B in One Day¶
Background: 2024-2025 hyperscaler $725B capex consensus, NVDA Q4 FY26 earnings $39B revenue. Market believed "scaling laws + GPU = NVDA always goes up."
Event: 2025/01/27 DeepSeek released V3 (later R1) — trained on H800 (crippled version) to achieve near GPT-4 capability, with estimated training cost of only ~$6M (vs GPT-4 $100M+).
Market Reaction: - 1/27 single day: NVDA -17%, SK Hynix -10%, Samsung -8% - 1 week later: NVDA recovered 60-80% - 1 month later: NVDA recovered 90%+, hit new highs
How Your Thesis Applies (mapping C1-C9): - C2 Scaling Laws: DeepSeek proves efficiency can improve, but Jevons Paradox — efficiency up → price down → demand up - C7 Value Capture: NVDA's moat = CUDA ecosystem + scaling laws positioning. DeepSeek doesn't break CUDA, doesn't break NVDA's moat - C9 Geopolitics: DeepSeek is a Chinese breakthrough, intensifying US-China chip war, long-term NVDA's China market is invisible
Lesson: Short-term price volatility ≠ long-term fundamental change. To break a thesis, check if supports fail (e.g., capex guide cut), not price moves.
3.2 Case 2: Samsung Loses HBM Market Share¶
Background: 2024 NVDA H100/H200 required HBM3e. Only 3 global players: SK Hynix / Micron / Samsung.
Event: - SK Hynix qualified early for NVDA H200 in 2024 → primary supplier, 70%+ share - Micron ramped in 2024, gained share - Samsung kept failing qualification (technology + yield issues), lost market - 2025 Samsung worker strike further impacted HBM capacity
Market Reaction (1 year): - SK Hynix: +130% YoY - Samsung: -20% YoY - Gap: Whether a company passes NVDA qualification determines ±150% stock price difference
How Your Thesis Applies: - C5 Hardware Stack: HBM is NVDA's shipment ceiling. Monitoring HBM capacity = monitoring NVDA revenue - C6 Supply Chain: SK Hynix / Micron / Samsung are an oligopoly; micro differences determine macro market share - C7 Value Capture: HBM customers have high stickiness (qualification is hard). Once market share is set, it's hard to reverse
Lesson: Within the same industry, micro differences in "can you pass technical qualification" can determine stock price differences of 100%+.
3.3 Case 3: Stargate Phase 1 — $500B AI Infrastructure¶
Background: 2024 US election Trump won. OAI / Oracle / MSFT / SoftBank had been discussing a joint mega-project.
Event: 2025/01/21 (Trump's 2nd day in office), Stargate plan announced: $500B over 5 years to build AI data centers. First site in Abilene, Texas, $50B Phase 1.
Market Reaction (7 days after announcement): - ORCL: +15% (Oracle Cloud running OpenAI workloads) - CRWV (CoreWeave): +20% - VRT (Data center electrical): +12% - CEG (Nuclear PPA): +10% - GEV (Gas turbines): +18% - ETN (Power distribution): +8%
How Your Thesis Applies: - C7 Value Capture: Stargate benefits go to shovel sellers + electricity providers, not OpenAI itself. Value capture is in the supporting layer - C8 Applications: $500B capex needs application ROI to materialize. But short-term, capex drives the supply chain - C9 Geopolitics: Stargate is a sovereign AI statement (US first, excluding China)
Lesson: "Shovel sellers" have better odds than "gold miners". AI isn't just a chip story — electrical contractors (ETN/HUBB) / data center REITs (EQIX/DLR) / nuclear (CEG/VST) / gas turbines (GEV) are all winners.
3.4 Case 4: Export Controls Revive SMIC / Huawei¶
Background: 2022-2023 US export controls on AI chips to China — H100/H200 not sold to China. Market thought this was bearish for NVDA.
Actual Development (2 years): - NVDA lost ~25% of China revenue - But NVDA's other overseas customers (Stargate + Europe + Middle East sovereign) filled the gap → total revenue still grew (+126% YoY 2024) - Meanwhile Huawei Ascend 910B/C rose as "China's AI chip" - SMIC revenue jumped 30%+ (accelerated localization) - DeepSeek trained near GPT-4 using H800 + algorithmic efficiency (Case 1)
How Your Thesis Applies: - C9 Geopolitics: Export controls have a lag (6-24 months). Short-term "obvious losers" (NVDA) may be offset by other factors - C2 Scaling Laws: Banned regions will use algorithmic + efficiency breakthroughs to substitute compute - C8 Applications: Banned regions' application layers develop independently (DeepSeek's China market stickiness)
Lesson: Regulatory actions have a 6-24 month lag. Short-term "obvious losers" may be offset by other factors, and local substitutes in banned regions become new theses.
3.5 Case 5: 13F Holdings ≠ Short-Term Price Signal¶
Background: May 2026 13F filings disclosed (institutional Q1 holdings). CIEN (Ciena, optical networking) data:
- Bridgewater reduced CIEN -32%
- Lone Pine reduced CIEN -60%
- Citadel increased CIEN +39%
Assuming you see this data and form a bull thesis (Citadel / Point72 increasing is a positive signal).
Actual Development: - 7 days later: CIEN -4.6% (net institutional selling had more impact than net buying) - 14 days later: CIEN +3% (reverted to fundamentals, earnings expectations improved)
How Your Thesis Applies: - C7 Value Capture: CIEN is a networking/optical mid-layer with moderate moat. Valuation is driven by both 13F and earnings - 13F Caveat (Layer 2 C3 taught): It's a quarter-end snapshot, disclosed 45 days later. What you see on 5/15 is holdings as of 3/31 — already 6 weeks old
Lesson: 13F is a quarter-end snapshot, not current holdings. Its value is in seeing institutional thesis trends (consensus / divergence), not timing signals. Any thesis using "institutional buying" as support must add the caveat: 6-week lag.
4. vs C9 What You Already Know¶
| Dimension | C9 Gives You | C10 Adds |
|---|---|---|
| Geopolitics Framework | ✓ (3-line theory) | Not applied |
| Real Case Application | ✗ | 5 cases × 6 dimensions |
| Investment Meaning | Know geopolitics is a wildcard | Know which case teaches which lesson in practice |
C9 = theory. C10 = forced application. Without C10, your 9-chapter framework sits gathering dust.
5. Try It: Recount 1 Case, Walk Through with a Friend¶
Task (30 minutes): 1. Pick 1 case (recommend Case 1 DeepSeek — most attention) 2. Without looking at the doc, recount to a friend: background / event / market reaction / how your thesis applies / lesson 3. Friend can follow + ask 1 good question = pass
Self-check (3 items met = Part 1 complete → move to Part 2):
- You can recount at least 3 of the 5 cases without looking at the doc
- You can extract 2 recurring patterns from the 5 cases (e.g., "shovel seller odds > gold miner" / "short-term price ≠ long-term thesis")
- You can find 1 recent AI event (current 2026) and dissect it across 6 dimensions (similar to the 5 cases)
6. What's Next (Part 1 Complete)¶
🎉 Completed AI Industry Knowledge — 10 chapters. You now have:
- ✅ Historical anchor (4 winters + 5-item checklist for "this time is different")
- ✅ Technical intuition (Transformer + scaling + LLM + hardware)
- ✅ Industry map (5 roles + 60 tickers + causality)
- ✅ Value capture (5-dimension scoring for strength)
- ✅ Application ROI reality (revenue realization across 4 major layers)
- ✅ Geopolitical wildcard (3 lines)
- ✅ 5 real-case intuition
Part 1 Complete: You have a complete cognitive map of the AI industry. Pure understanding before any investment intent.
AI Investing teaches you how to turn this understanding into investment decisions:
- General Investment Models: Valuation DCF / Mental Models / Historical Comps / Portfolio / Behavioral Finance
- AI Industry-Specific Analysis: Thesis 4 Dimensions / KPI Predictions / Multi-PM Perspectives / Real Case Analysis / Continuous Review
→ AI Investing (In progress — you can currently view the old versions of Layers 2-5)
7. Deep Dive (optional): 5 Cases' Common Lessons + Dotcom Base Rate Comparison¶
Click to see Part 1's Final Lessons
5 Cases' Common Lessons¶
| Lesson | Case Source |
|---|---|
| Short-term price volatility ≠ long-term fundamentals | Case 1 DeepSeek + Case 5 13F |
| Micro differences = huge divergence | Case 2 Samsung HBM |
| Shovel sellers > gold miners | Case 3 Stargate |
| Regulation has lag, there's always a beneficiary | Case 4 Export Controls |
| Data ≠ signal, check the timestamp | Case 5 13F |
Dotcom 1999 vs AI 2026 Base Rate Comparison¶
| Dimension | Dotcom 1999 | AI 2026 |
|---|---|---|
| Infrastructure capex | Cisco / Lucent / Nortel | NVDA / Hyperscaler |
| Application layer | Amazon / eBay (had revenue) + many 0-revenue companies | OAI / Anthropic + many verticals |
| Valuation | Cisco PE 200x | NVDA PE 30-35x |
| Real demand | Internet penetration < 5% | AI penetration truly taking off (300M MAU) |
| Outcome | 2000-2002 Nasdaq -80% | TBD |
Key difference: Dotcom valuations were crazy (Cisco PE 200x). AI's current valuations are still in a reasonable range (NVDA PE 30-35x), because revenue is actually growing.
Risk: If hyperscaler capex turns (any capex guide cut), the AI chain could re-rate -30~-50%. But unlikely to be as bad as dotcom's -80%.
Part 1's Final Insight¶
"Understanding the industry isn't about predicting stock prices; it's about knowing, when the unexpected happens, which thesis still holds and which one to close."