NVIDIA at $148 is the cleanest test of value-investing dogma since FANG in 2017. The price-to-earnings ratio of 65 says "speculative bubble". The 70%+ gross margin, the AI moat, and the 31% ROIC say "wonderful business, hold forever". Same stock, opposite reads from the canon. So we put four of the canon's authors in the same room.
This is a real transcript from our Boardroom feature. Four AI personas, each implemented against the published rules of its real-world investor, debate the same ticker. A separate skeptic AI challenges every bullish claim. Every quote is grounded in the underlying letters / books, and every numeric claim cites a current data point.
If you want the live, regenerable version, open /boardroom/nvda/. What follows is one specific round, captured 2026-05-19, NVDA close $147.94.
Round 1: opening statements
Warren Buffett - Bearish
"I would not buy this business at this price. A P/E of 65 prices in 15 years of perfect compounding at the current margin, and Andy Grove has already told us only the paranoid survive in semiconductors. The moat is real - but the price asks me to predict the next two product cycles, and I do not believe anyone can do that. If it traded at 25 times earnings I would be interested. At 65 times, I pass."
cites: 1996 letter on margin of safety; 1985 letter on commodity-like businesses
The Buffett scorer flags NVDA as Partial Fit. Quality passes (ROIC 31.4%, balance sheet pristine). Valuation fails (owner-earnings yield 1.6% versus the 5-8% floor Buffett applies to capital-intensive cyclicals).
Benjamin Graham - Strongly Bearish
"P/E above 65, P/B above 30, current ratio at 1.9 - this stock fails every defensive test in The Intelligent Investor. Graham's enterprising-investor framework still requires earnings yield ≥ 7%; NVDA's is 1.5. I do not need to take a view on the AI cycle to reject it. The price has already taken the view, and the price is wrong by my framework."
cites: The Intelligent Investor, chapter 14, defensive criteria
The Graham scorer flags NVDA as Weak Fit. Zero of five defensive criteria pass.
Peter Lynch - Cautious Bull
"Earnings growth is doing the heavy lifting here, not multiple expansion. EPS grew 232% last year and is forecast at 47% this year. PEG comes out to roughly 1.4 - above my 1.0 ideal, but not in the dangerous zone if the growth rate is real. The question is whether 'AI capex' is the new 'PC capex' - a 20-year cycle - or 'crypto-mining capex' - a 2-year cycle. I would buy a quarter-position and add on confirmation. The thing I worry about most is that everyone already knows this story."
cites: One Up on Wall Street, GARP framework, "the story is everything" chapter
The Lynch scorer flags NVDA as Partial Fit. PEG slightly above floor; growth comfortably above; debt-to-equity clean.
Joel Greenblatt - Bull (mechanically)
"Magic Formula doesn't care about narratives. Earnings yield 1.5% - fails on absolute yield. But ROIC is 31% - passes the second screen elegantly. On the rank-based Magic Formula, NVDA's combined rank in the current universe is roughly 480 out of 3,500 - top decile on quality, bottom quintile on price. Magic Formula will hold it for one year then re-rank. I am long for the next 12 months as a Magic Formula position. After that, I re-test."
cites: Little Book that Beats the Market, rank-based Magic Formula chapter
The Greenblatt scorer flags NVDA as Partial Fit. Quality dominates; price loses.
Round 2: the skeptic challenges
A separate skeptic AI runs after every legend speaks. Its job: find the strongest possible counter to whatever the consensus is. Here's what it surfaced for the bull case:
Skeptic challenge to Lynch and Greenblatt: "Both of you are leaning on current ROIC and current EPS growth. NVIDIA's gross margin is 76% - five times the semiconductor industry's 30-year median of 35%. If margins mean-revert even halfway, EPS falls by 50%+, and your 'PEG 1.4' becomes 'PEG 3.5'. The Magic Formula one-year hold has not been stress-tested against a single semiconductor cycle since 2010. What evidence rebuts the historical pattern that semis are deeply cyclical?"
This is the part of the Boardroom that does not exist in any other AI takes tool. The skeptic does not vote. It frames the strongest opposing argument so the user has to actually think.
Round 3: rebuttals
Lynch responds
"Fair challenge. My historical evidence is that the cyclical pattern is for commodity semis - DRAM, NAND, low-end logic. NVDA designs accelerators, doesn't fab them, and sells to four customers (Microsoft, Meta, Google, Amazon) who are competing with each other on capex. That is a different industry structure. But - the skeptic is right that I'm extrapolating. I'd cap position size at 2% of portfolio, not 5%, because the structural argument is plausible but not certain."
Greenblatt responds
"The Magic Formula is by design tolerant of cyclicality on a one-year basis. If the cycle turns in 18 months and I exit at re-rank, I take the loss and move on. The framework's strength is that the ranking is reproducible - it does not require the cycle prediction to be right."
Buffett responds (to himself, essentially)
"The skeptic is closer to my view than to Lynch's. The price already pays you for the bull case to keep playing out. The Wonderful-Business-At-A-Fair-Price test was always two halves, and the second half is failing. Pass."
Round 4: synthesis
The Boardroom never lets the four investors vote - they each have a thesis, not a verdict. What it does instead is print the divergence map: where the four agree, where they diverge, and what the divergence is driven by.
For NVDA at $148:
- Quality of business: all four agree - wonderful
- Balance-sheet risk: all four agree - clean
- Cyclicality risk: split. Buffett + Graham say cyclical. Lynch + Greenblatt say structurally different from past cycles
- Valuation: 3-1 against. Only Lynch (cautiously) thinks the growth rate justifies the price
A user reading this should walk away with: NVDA's business is the bull case; NVDA's price is the bear case; the rest is your own view on cyclicality. That is the framing the real boardroom would give you - and it is the framing no AI summary, no "buy/sell rating", and no broker note will ever give you. Different frameworks weight risks differently. The point of putting them in the same room is to see the seams.
Where the citations come from
Every quote in this transcript is anchored to a specific passage in the cited source. The Buffett quotes come from a corpus of every Berkshire Hathaway shareholder letter from 1977 to 2025, indexed and chunked. The Graham quotes come from The Intelligent Investor (4th revised edition) and Security Analysis (6th edition). Lynch from One Up on Wall Street. Greenblatt from The Little Book that Beats the Market. Each AI persona can only quote from its own corpus - the architecture is hard-walled against cross-attribution.
This matters because most "what would Buffett say about X" AI tools just ask GPT to roleplay, which produces hallucinated quotes that sound right but were never actually written. We index the real letters, then constrain the persona to cite the chunk it pulled from. Imperfect - LLMs still paraphrase, occasionally embellish - but materially more grounded than a generic prompt.
Try it on a stock you actually own
If you want to see what four investors think about a stock in your portfolio rather than NVDA, open the Boardroom for any ticker. Free users get three runs a week; the Founder's Plan and Pro unlock unlimited debates. The first ten boardrooms are heavily cached so popular tickers (AAPL, MSFT, NVDA, GOOG, AMZN, BRK.B, KO, JNJ, V, MA) load in under a second.
The four-investor format is one of the most-requested features since launch, and it is the feature that taught me the most about my own portfolio. It will probably do the same for you.
Disclosure
This is an educational analysis tool. Boardroom debates are AI-generated using personas grounded in the cited investors' real published works. They are not investment advice, not a recommendation, and not a substitute for reading the original letters and books yourself. Past performance of NVDA (or any stock) is not a forecast.
Quotes attributed to legendary investors in the Boardroom feature are AI-generated paraphrases / inferences from their published frameworks, not verbatim transcriptions. The original Berkshire shareholder letters and the books cited are available from Berkshire Hathaway, Harper & Row, Simon & Schuster, and Wiley respectively.