P4-C5 · Buffett Didn't Buy NVDA — A Forensic Analysis¶
Core Takeaway
Buffett didn't buy NVDA from 2023-2025 — not because he missed it, but because his 5-step framework genuinely excluded it. Learn the discipline of non-participation.
Real Public Case — Using Part 3 Tools to Dissect 5 Real Events
P4-C5. All based on public data (Berkshire 13F + Buffett's shareholder letters + CNBC interviews).
1. The Event: NVDA +1500% from 2023-2025, Berkshire Bought Zero¶
| Time | NVDA Performance | Berkshire Action |
|---|---|---|
| 2023 Q1 | +90% | Didn't buy |
| 2023 Full Year | +240% | Didn't buy |
| 2024 Full Year | +180% | Didn't buy |
| 2025 (Post-DeepSeek) | -17% (Jan 27 selloff) | Didn't buy (didn't use entry opportunity) |
| 2025 to Present | Hit new highs | Still didn't buy |
→ Buffett missed one of history's biggest winners. Was it a miss? Or discipline?
2. Verifying Step-by-Step with P3-C2 Buffett's 5 Steps — 4 Steps Clearly Fail¶
2.1 Step 1 — Circle of Competence¶
Buffett's 2024 Annual Meeting Quote (Public CNBC):
"I don't really know much about AI specifically. I know about people, and I know there's a lot of money being spent on AI. Some of it will work out, some won't. But I don't have any edge in figuring out which is which."
Translation: I have no edge in AI. I don't know which companies will still be around in 10 years.
Applied to NVDA**: - AI economics uncertain in 10 years (will scaling laws hold? ASIC replacement?) - NVDA's single business concentrated on AI — vs AAPL (iPhone + Mac + Services + Watch + WAR) - Finances easy to understand ($75% margin), but future not easy to understand**
→ Step 1 ❌ Buffett admits he's outside his circle of competence.
2.2 Step 2 — Durable Moat¶
Buffett's 4 Moat Types: - Brand ✗ (NVDA is B2B brand, unlike Apple's global consumer brand) - Switching Cost ✓ (CUDA's 20-year ecosystem is strong) - Network Effect ✗ (B2B lacks user-to-user network) - Scale + Cost Advantage ✓ (Annual capex $5B+)
2/4 items strong — weaker than AAPL (4/4).
From Buffett's perspective: Not "no moat," but "moat isn't multiple enough." AAPL has 4 items, NVDA has 2, so long-term holding favors AAPL.
→ Step 2 ⚠️ Moderate — passes but not strong.
2.3 Step 3 — Honest, Capable Management¶
Jensen Huang: - 33-year founder, platform strategy long-term (CUDA / Mellanox / Stargate investments) - Capital allocation: heavy capex ($5B+ annual R&D, strategic investment in CRWV $36B), less dividends/buybacks (vs AAPL $200B buybacks + dividends)
From Buffett's perspective: Capital allocation isn't his preference (he likes cash hoarders + buybacks).
→ Step 3 ⚠️ Jensen is capable but capital allocation style doesn't align with Buffett.
2.4 Step 4 — Reasonable Price (Margin of Safety)¶
**Key step — Buffett's main reason for excluding NVDA**:
| Time | NVDA fwd PE | Buffett "Value" Estimate | Margin of Safety |
|---|---|---|---|
| 2023 Q1 | 25x | Estimate 15-18x | Negative |
| 2023 Q4 | 32x | 18-20x | Negative |
| 2024 Q4 | 35x | 18-22x (supported by 50%+ growth) | Negative |
| 2025 Q1 (Post-DeepSeek) | 27x | 18-22x | Near zero |
Buffett doesn't require cheap valuation (PE 10x), but requires: - 10-year discounted FCF > current price - Growth assumptions conservative (doesn't assume +50% forever)
→ Step 4 ❌ Never reached margin of safety.
2.5 Step 5 — Hold Forever¶
If Steps 1-4 already fail, Step 5 automatically fails.
**NVDA score under Buffett's framework**: ⅕ (only Step 2 partially passes).
→ Shouldn't invest. Buffett didn't miss it; his framework genuinely excluded it.
3. Berkshire's Cash Hoard Forensic¶
Buffett didn't buy NVDA from 2023-2025, but during the same period hoarded $350B+ in cash (Berkshire 2025 Q1 financials, public).
Implications: - Not buying AI isn't because of lack of money - Not buying AI is because the 5-step framework excluded it - Cash is waiting for the next "AAPL moment" (2016 PE 11x + 4/4 moat + reasonable price)
Public Buffett 2024 Letter:
"We continue to prefer holding cash and short-term Treasuries to making large equity purchases when we cannot identify compelling value."
4. Buffett Framework Applicability — Which AI Stocks Might Pass¶
Using P3-C2's 5 steps to score mainstream AI stocks:
| Stock | Step 1 Circle | Step 2 Moat | Step 3 Mgmt | Step 4 Price | Step 5 Hold | Total |
|---|---|---|---|---|---|---|
| AAPL (Berkshire holds) | ✓ | 4/4 | ⭐⭐⭐⭐⭐ | 11x (entered 2016) | ✓ | 5/5 |
| MSFT | ⚠️ (cloud OK, AI uncertain) | 3.5/4 | ⭐⭐⭐⭐ Satya | 30x (warmish) | ⚠️ | 3.5/5 |
| GOOGL | ⚠️ (search threatened by AI) | ¾ | ⭐⭐⭐ (Sundar) | 20x | ⚠️ | ⅗ |
| NVDA | ❌ | 2/4 | ⭐⭐⭐ | 30-35x | ❌ | ⅕ |
| OpenAI | ❌ | 2/4 | ⭐⭐ (Altman opaque) | High primary valuation | ❌ | 0/5 |
→ **Most Buffett-friendly in the AI era is AAPL (but trimmed in 2024) + partial MSFT**.
5. Lesson — The Discipline of Non-Participation¶
| Lesson | How to Use in Your Thesis |
|---|---|
| Acknowledging circle of competence boundaries = wisdom | Don't invest in what doesn't pass all 5 steps |
| Missing a winner isn't a mistake | Buffett missed NVDA's $3T gain, didn't break his record |
| Cash waits for opportunities | Cash hoard isn't laziness; it's waiting for the next AAPL moment |
| Only hold long-term if all 5 steps pass | If any step fails, it shouldn't be a long-term core |
Key Reflection: Does your thesis pass all 5 steps? If not (like NVDA with 2 steps), it's a trade, not an investment. The position sizes should differ (trade < 5%, invest can be 10%+).
6. Part 4 Conclusion¶
🎉 Completed 5 chapters of real public cases. You can now:
- ✅ Use Part 3's 5 tools to dissect the DeepSeek selloff (P4-C1) → short-term panic ≠ broken thesis
- ✅ Use the 5-step bottleneck finder to analyze Samsung HBM (P4-C2) → micro differences = huge divergence
- ✅ Use the beneficiary chain framework to dissect Stargate (P4-C3) → selling shovels > mining gold
- ✅ Use the regulatory lag framework to analyze export controls (P4-C4) → 6-24 month lag + substitution effects
- ✅ Use Buffett's 5 steps to analyze why he didn't buy NVDA (P4-C5) → discipline of non-participation
Next Step:
→ Further Reading — Free source system map + 6-week reading roadmap.