P2A-C3 · Historical Analogies (dotcom / mobile / Industrial Revolution)¶
Core Takeaway
Anchor AI to historical analogies; base rates don't betray — history doesn't repeat but it rhymes.
Universal Investment Model — Applicable to Any Industry
P2A-C3 (Part 2.A, Chapter 3). After this chapter, you'll be able to identify how AI resembles each of 3 historical paradigm shifts and what base rates suggest for the next move.
1. The Problem: AI "This Time Is Different" — How to Verify?¶
Current AI consensus: "AI changes everything, NVDA goes up forever."
History's 4 "this time is different" episodes (covered in P1-C1) all crashed. How do we know AI isn't the 5th?
→ Find historical analogies + base rates. Not prediction, but risk pricing.
3 candidate historical paradigms: - Dotcom 1995-2002 (Internet bubble) - Mobile 2008-2018 (Smartphone revolution) - Industrial Revolution 1760-1840 (Steam + railways)
Each has lessons, and AI partially resembles each.
2. The Solution: 3 Historical Paradigms + 5-Dimension Comparison¶
| Dimension | Dotcom (1999) | Mobile (2010) | Industrial Revolution (1840) | AI (2026) |
|---|---|---|---|---|
| Infrastructure | Cisco / Lucent / Sun | iPhone / TSM / ARM | Railways / Steam engines / Steel | NVDA / HBM / Data centers |
| Applications | Amazon (with revenue) + massive 0-revenue | App Store + Google / Facebook | Factories + massive transport | OpenAI / Copilot + vertical |
| Capex / GDP Ratio | High (1.8% peak) | Medium (0.9%) | High (3%+ peak) | Currently ~1.2%, rising |
| Real Users | <5% Internet penetration | 50%+ smartphones (2015) | Massive manufacturing workforce | 300M ChatGPT MAU (real demand) |
| Valuation Insanity | Cisco PE 200x | iPhone OEM PE 15-25x | Railway PE 20-30x | **NVDA PE 30-35x** |
| Post-Crash Decline | -80% Nasdaq | None (continued rising) | 1840s+ railway bubble -50% | TBD |
→ AI most resembles mobile (real users + moderate valuation) + partially resembles dotcom (soaring capex).
3. How It Works: 3 Historical Paradigms in Detail¶
3.1 Dotcom (1995-2002)¶
Boom (1995-1999): - Internet penetration from 1% → 5% (1999) - Cisco / Lucent / Sun (basic infrastructure) PE soared to 100-200x - Amazon / eBay / Yahoo (applications) had insane valuations - Massive 0-revenue companies IPO'd (Pets.com / Webvan / Boo.com)
Crash (2000-2002): - Nasdaq -80% peak to trough (5132 → 1108) - Cisco -90% - Amazon -94% (2.5-year low)
Why It Crashed: - Capex far outpaced application ROI realization speed - Real user base (5% Internet penetration) couldn't support valuation assumptions - First "infrastructure bubble → application layer cooling" cycle
Why Amazon / Google Later Became Biggest Winners: - Had real revenue + customers during the crisis - Capex truly converted to ROI (Amazon built AWS) - Survival bias — among 100 application-layer companies, 99 died; Amazon was the 1%
AI Analogy: - AI current capex / GDP ~1.2%, close to dotcom peak 1.8% (not there yet, but fast) - ChatGPT 300M MAU far exceeds dotcom's real users (at Internet penetration) - NVDA PE 30-35x far below Cisco 200x — valuation is much more reasonable
→ AI is healthier than dotcom, but the capex trend warrants caution.
3.2 Mobile (2008-2018)¶
Boom (2008-2015): - iPhone launched in 2007 - Smartphone penetration 0% → 50% (2015) - TSM / ARM / Qualcomm (infrastructure) rose steadily - App Store + Google + Facebook (applications) dominated
Key Differences vs Dotcom: - Real users + real revenue (App Store 2010 $10B GMV) - Application layer diversified (Uber / Instagram / WhatsApp each became giants) - No crash — TSM / ARM / AAPL rose steadily, with occasional pullbacks
Why It Didn't Crash: - Capex / GDP ratio smaller than dotcom (0.9% vs 1.8%) - Application layer ROI truly realized (iPhone sales + App Store revenue) - Customers paid long-term (vs dotcom free + ad-dependent)
AI Analogy: - ChatGPT / Copilot application layer has real revenue (like mobile) - Application diversity (foundation + SaaS + vertical + agentic, like mobile) - Capex higher than mobile (like dotcom)
→ AI is a hybrid of dotcom + mobile.
3.3 Industrial Revolution (1760-1840)¶
Boom (1820-1840 railway boom): - Steam engines + factories + railways + steel - UK GDP +200% over 100 years - Railway bubble 1845-46 (Cisco-like)
Crash (1846-1850 railway panic): - Railway stocks -50% (peak to trough) - Massive small railway company bankruptcies - Major railways (LMR / GWR) later became century winners
Key Lesson: - True technological revolution (steam + railways) lasted 80+ years - Short-term bubble + crash → long-term winners still rose 5-10x - "Winner survive base rate" extremely low (10-20%)
AI Analogy: - True technological revolution (LLM + scaling) may last 20-50 years - Short-term bubble + crash possible (e.g., 2027-2028 hyperscaler peak?) - Long-term winners (NVDA / MSFT) still rise, but short-term -30~50% risk exists
4. AI Comprehensive Base Rate Assessment¶
| Dimension | AI 2026 vs History | Base Rate Suggestion |
|---|---|---|
| Real Users | Like mobile (strong) | Won't crash like dotcom's 5% penetration |
| Application Layer Revenue | Like mobile (App Store-like emerging) | Real ROI being realized |
| Capex Speed | Like dotcom + Industrial Revolution | Short-term bubble risk |
| Valuation | Healthier than dotcom, like mobile | Not extreme |
| Infrastructure vs Application Lag | Like dotcom (infrastructure ahead 2-3 years) | Applications need to catch up, risk window 2026-2028 |
Overall Assessment: - AI is not a dotcom replay (valuation + users healthier) - But soaring capex + lagging application ROI realization = -30~50% risk within 2-3 years - Long-term (10+ years) should resemble Industrial Revolution — winners keep rising, with cycles in between
→ Your thesis should: - Long-term hold winners (MSFT / partial NVDA) - Keep cash short-term for potential cycle peak retreat - Monitor capex / revenue ratio (>35% warning)
5. Try It: Find 1 Most Similar Historical Paradigm for AI¶
Task (~30 minutes): For the sub-sector your thesis involves (e.g., AI chips / AI applications / AI power), find the most similar historical paradigm:
| Sub-sector | Historical Analogy | Which Period Most Similar |
|---|---|---|
| AI Chips (NVDA) | Cisco 1999 / Intel 1990s / TSM 2000s | ? |
| AI Applications (OAI / Copilot) | Amazon 1999 / Google 2004 / Salesforce 2000s | ? |
| AI Power (CEG) | 1990s power deregulation / 1980 oil & gas | ? |
| AI Data Center REIT (EQIX) | 2000s data center REIT (EQIX 2007 IPO) | ? |
Self-check (3 items checked → proceed to P2A-C4):
- You can identify 1 most similar historical paradigm for your thesis ticker
- You can state the base rate (winner proportion / long-term returns in that paradigm)
- You can derive the current biggest risk from the historical base rate
6. What's Next¶
You now have a historical anchor. But with a thesis library of 5-10 stocks, how to size them? Portfolio construction.
→ P2A-C4 · Portfolio Construction — sizing + concentration + risk parity.
7. Deep Dive (Optional): 1840 Railway Panic + 1929 / 1973 Historical Comparisons¶
Click to see details on 3 historical bubbles
1845-1846 Railway Panic: - UK 1840s railway stocks PE 30-50x (reasonable by today's standards, insane then) - 6,000 miles of railways approved, only 1,000 miles actually built (over-building) - 1846 banks raised rates → railway stocks -50% - 1850s truly became the "railway era"; winners rose 5-10x - Lesson: Bubbles occur within true revolutions; winners emerge stronger after the bubble
1929 Great Depression: - 1920s US GDP +50%, stock market +200% - 1929 Dow -89% (peak to trough) - True technologies (automobiles / electricity / radio) were long-term winners, but underperformed Treasuries for 25 years in between - Lesson: Even with a true technological revolution, extreme valuations can lead to a crash that takes 25 years to recover (1954 to return to 1929 peak)
1973-1974 Nifty Fifty: - 50 "growth at any price" stocks (Polaroid / Avon / Xerox / IBM) - 1973 PE 50-100x - 1973-74 -45% (Nifty Fifty fell even more) - Lesson: Even large companies with real growth, PE 50-100x is unsustainable
AI Currently: - NVDA PE 30-35x — not Nifty Fifty - Real users / revenue — not dotcom - But capex soaring — like dotcom + Industrial Revolution
Summary: AI is not at a bubble peak, but capex retreat risk exists in the next 2-3 years.