NVIDIA: when frameworks disagree

The quality side (Fisher, Smith, Lynch optimistic)

NVIDIA is a textbook quality compounder by every business-quality metric. ROCE above 70%, gross margins above 70%, cash conversion above 95%, R&D investment as a percentage of revenue maintained even at scale, a dominant developer ecosystem (CUDA) that creates switching costs no competitor has matched in 15 years. The Fisher framework, which weights product cadence and technical leadership above starting valuation, scores NVDA at 84. Smith's framework, which is built precisely on the kind of high-ROCE quality compounder NVDA represents, scores 82. Lynch's GARP framework passes if you accept consensus growth assumptions; the PEG sits comfortably below 1.0 on consensus 3-year growth.

The narrative around each of these frameworks is the same: NVDA owns the accelerator generation that powers every modern AI workload, the moat compounds with each new architecture (Hopper, Blackwell, Rubin), and the earnings trajectory justifies the multiple if any significant portion of the consensus AI capex forecasts materialises.

The valuation side (Graham, Munger, Buffett-Fit)

Graham's defensive criteria fail on NVDA structurally. P/E above 60, P/B above 30, and a market cap that has appreciated 10x in three years all sit outside what Graham would have called a margin-of-safety starting point. Graham did not weight quality above starting valuation; his framework scores NVDA at 48, well below the C/B boundary.

Munger's discipline is more nuanced. The mental- model framework does not refuse to own expensive compounders. What it does refuse is to own companies with the obvious bubble characteristics: cap-weighted index dominance, retail-investor concentration in the momentum cohort, and the kind of consensus narrative where everyone is on the same side. Munger's behavioral-economics edge requires the marginal investor to be wrong; on NVDA at scale, the marginal investor is everyone. The framework dings the case for that reason.

Buffett-Fit scores B (64). The composite reflects the quality of the business (moat, durability, financial health all high) pulled down by valuation (low). The B is exactly the right composite for a high-quality business at a stretched price.

What the disagreement teaches

The 35-point Smith-Graham spread is the consensus screen's most useful signal. When frameworks disagree this sharply, it means the asset's attractiveness genuinely depends on which philosophy the investor applies. The 5+/7 bull count of 3 is the right honest verdict: NVDA is not a consensus pass. A quality-first investor (Smith, Fisher, Lynch optimistic) can buy NVDA with the framework's approval. A value-first investor (Graham, Buffett-Fit valuation, Greenblatt earnings yield) cannot buy NVDA within their own discipline.

The lesson generalises: the cross-framework consensus screen is most useful when it disagrees, not when it agrees. Consensus passes at 7-of-7 are the high- conviction picks; deep disagreement at 3-of-7 to 4-of-7 is the philosophical fork in the road. NVDA is the archetype.