My first valuation of NVIDIA said the company was worth $55.88 per share. I looked at that number, looked at the evidence sitting right next to it, and realized I had a problem. Not with the evidence. With the number.
The number was wrong. And the bias that made it wrong is one most investors will never catch, because it hides behind the word “conservative.”
The $32 Mistake Hiding Behind “Conservative”
The GAMMERS process has a final step called the S-step, where you take everything you’ve learned across five prior steps of analysis and convert it into a single number: the Next Level Sale Price. The NLSP. It’s the price at which the business becomes a buy in my framework. Below that price, the evidence says you’re getting a wonderful business at a wonderful price. Above it, you wait.
The S-step is where the math meets the conviction. And the most important input to that math is the EFGR, the Estimated Future Growth Rate. Get the EFGR right and the valuation is reliable. Get it wrong and everything downstream is off.
My first pass at NVIDIA’s EFGR landed at 15-18%. That’s the base case growth rate the R-step recommended after stress-testing the thesis. It felt conservative. It felt responsible. It produced an NLSP of $55.88.
The problem: the evidence didn’t support it.
NVIDIA’s 10-year EPS compound annual growth rate is 56.1%. Not 15%. Not 18%. Fifty-six percent. Their quarterly revenue was re-accelerating, not decelerating, for three straight quarters: Q2 at 122% year-over-year, Q3 at 94%, Q4 at 78%. Management had just guided Q1 FY2027 at $78 billion, implying 77% year-over-year growth. Jensen Huang stood on stage at GTC 2026 and declared $1 trillion in cumulative demand through 2027.
Nine separate growth tailwinds. Re-accelerating quarterly results. A 56% historical EPS CAGR. And I was modeling 15-18%.
(This is the part where I tell you that “being conservative” and “being wrong” are not mutually exclusive.)
The single-rate EFGR was treating NVIDIA like a steady-state compounder, the kind of company that grows 15% a year like clockwork for a decade. NVIDIA is not that company. NVIDIA is in the middle of the largest infrastructure buildout in computing history, with hyperscalers pouring $600 billion annually into AI, 75% of it aimed directly at the company’s products. Modeling it as a smooth 15% grower was like measuring a rocket at cruising altitude and assuming it was always at that speed.
The fix was a two-phase model. Phase 1, covering years one through five, uses a 28% base growth rate to reflect the evidence of continued hyper-growth: the re-accelerating revenue, the $78 billion guidance, the nine tailwinds with no visible demand ceiling. Phase 2, covering years six through ten, decelerates to 17% to account for the law of large numbers, custom silicon erosion, and potential CEO succession disruption.
The blended 10-year CAGR comes out to 22.4%. Still below every historical growth window the company has posted. Still conservative relative to the actual trajectory. But far closer to what the evidence supports than 15%.
The revised valuation: $88.00 per share. A 57% increase from the original $55.88.
That $32 difference is not academic. At $55.88, NVIDIA needs to fall 68% from $175.20 before it becomes a buy in my framework. At $88, it needs to fall 50%. One number defines a reasonable watchlist entry. The other defines a fantasy.
The mechanics of where that $32 came from are revealing. I run three separate valuation engines on every company: two discounted cash flow models (NLCV and NLEV) and a Rule #1 margin-of-safety calculation. The DCF engines barely moved. NLCV went from roughly $47 to $49. NLEV stayed near $64. The higher near-term growth got eaten by the heavier discounting at my 16% hurdle rate.
The Rule #1 engine exploded. It went from approximately $56 to $173.50. The reason: Rule #1 projects a Year 10 EPS and multiplies it by a future PE ratio. Under the original EFGR, Year 10 EPS was around $20. Under the two-phase model, it’s $38.27. Nearly double. And that doubling gets multiplied by a 40x PE, which NVIDIA has routinely exceeded throughout the AI era. One input change. Cascading impact.
The final NLSP of $88 weighs all three engines, with the DCF models at 30% each and Rule #1 at 40%, then triangulates against five additional pricing methods that cluster around $40 to $60. The number is not aggressive. It’s calibrated. And it only became calibrated when I stopped assuming that “lower” automatically meant “better.”
The Bathroom Scale Problem
Think of a valuation model like a bathroom scale. You can set the scale to read 10 pounds light. That will make you feel responsible. Disciplined. You’ll say, “Better to be surprised by a lower number than fooled by a higher one.”
But the scale’s job is to be accurate, not pessimistic. If the scale says 150 and you actually weigh 170, you’re not being conservative. You’re being wrong. And wrong in a specific direction that makes you feel virtuous while costing you information.
That’s what a 15% EFGR does when the evidence says 22%. It doesn’t protect you. It blinds you. It creates a buy price so low that you’ll never reach it, and then you’ll congratulate yourself for your discipline as the stock compounds without you.
The margin of safety, the 50% discount I apply to fair value, already exists to protect against overestimation. That’s literally its job. Reducing the growth rate and applying a 50% margin of safety for the same risk is double-counting the same fear. It’s the valuation equivalent of wearing two seatbelts. [PERMANENT]
Why This Matters for Investors
Most investors have a specific, well-practiced instinct: when in doubt, round down. Lower the growth rate. Raise the discount rate. Shrink the multiple. And most of the time, that instinct serves them well. Optimism kills more portfolios than pessimism ever has.
But there’s a version of this instinct that stops being caution and starts being bias. When the evidence says 22% and you model 15%, you’re not being careful. You’re being systematically wrong in a direction that feels safe.
I have bought great companies at the wrong price. The kind of mistake that teaches you more than any book. That experience built an instinct in me to be cautious, and I am grateful for it every day. But I also know that the same instinct, unchecked, produces a different kind of mistake: watching a wonderful business compound at 20%+ per year while you sit on the sidelines waiting for a price that the math, your own math, says was never the right target.
The discipline is not to be conservative. The discipline is to be accurate. And accuracy sometimes means revising upward, not just downward.
For my process, the revised NLSP of $88.00 still says NVIDIA is not on sale at $175.20. The stock sits 99% above the buy threshold. The verdict didn’t change. But the confidence in the verdict changed, because now it’s built on an EFGR that matches the evidence instead of one that hides behind the word “conservative.”
And that confidence matters for a specific reason: when NVIDIA does eventually have a bad quarter, or a regulatory scare, or a custom silicon panic that drops it 30-40%, I need to know whether $88 is a real number or a fake one. The original $55.88 was a number I wouldn’t have trusted in the moment of crisis, because some part of me would have known the inputs were too low. The revised $88 is a number I can act on, because I built it from evidence, not from anxiety.
One Thing to Watch
The next time you finish a valuation and feel good about how conservative it is, run it a second time. Not to make it more aggressive. To check whether the inputs actually match the evidence you collected. Look at the growth rate you used. Compare it to the company’s actual historical growth, its recent trajectory, its forward guidance. If there’s a gap, ask yourself a hard question: did I model what the evidence supports, or did I model what makes me comfortable?
Comfort and accuracy are not the same thing. And the gap between them is where returns go to hide.
If you want to learn how to run the full GAMMERS valuation process, including the two-phase EFGR model and the bias checks that catch errors like this, the OIA Research Lab walks through every step on real companies. Learn more at oialabs.com.