I owned Nvidia before the current AI boom, sold it in 2022 and bought it again in 2025 at a much higher valuation.
***To be clear, I sold with 300% profit. I bought at $5 and sold at $15. I bought back in at $100. Now it’s $230..***
I've thought about that mistake quite a bit because I don't think the useful lesson is simply “never sell your winners.”
There were perfectly rational reasons to be worried about Nvidia in 2022. The stock was getting crushed, gaming was going through an inventory correction, crypto-related GPU demand had collapsed, rates were rising rapidly and Nvidia eventually reported essentially flat fiscal 2023 revenue with net income down 55%.
My mistake was more specific.
I didn't sell because I had concluded that accelerated computing was becoming less important or because CUDA's moat was weakening. I became less confident in my long-term technological view while the market was falling.
That turned out to be an expensive distinction.
Fiscal 2023 revenue was about $27 billion. Nvidia's latest quarter produced $96.2 billion of revenue, including $89 billion from Data Center alone.
Its market cap ended 2022 around $360 billion. Today it is around $5.5 trillion.
Obviously nobody in 2022 could have known those numbers in advance. That's not the standard I'm applying to myself. What I think I should have understood was that the right tail was enormous and that the reason I originally owned the company hadn't actually disappeared.
By the time I bought back in 2025, there was much more evidence. AI infrastructure spending was exploding, Blackwell demand was visible and Data Center revenue had become enormous.
The problem is that I had to pay for that certainty.
I've changed my selling rule because of this.
For strategic positions, I now want a reason connected to one of three things:
1. **The thesis has weakened.** The competitive position, market opportunity or underlying technological assumptions have changed.
2. **Valuation has outrun the distribution of plausible outcomes.** A fantastic company can still become a bad investment if the market prices the bull case as though it is guaranteed.
3. **The position threatens portfolio survival.** Concentration risk is real, particularly if you fund your life from the portfolio.
What I don't want to use as a reason is simply that the stock has risen a lot, or that a large drawdown has made me emotionally less confident.
This matters more in a power-law portfolio than in a conventional diversified portfolio.
Imagine starting with twenty equal 5% positions. One eventually becomes a 20x winner while most of the others are mediocre. If you continuously rebalance the winner back toward 5%, you systematically transfer capital away from the investment where you were most right.
Sometimes that's still the correct decision. A position can become dangerously large.
But rebalancing isn't free.
There's a famous behavioral-finance literature on the disposition effect, where investors sell winners too readily and hold losers too long. Terrance Odean studied 10,000 brokerage accounts and found that the winners investors sold subsequently outperformed the losers they kept.
My Nvidia sale wasn't a textbook disposition-effect trade because I sold during a major decline. But I think the deeper psychology was related. It is surprisingly difficult to hold an investment when the amount of money at stake becomes large and the market is making you question your original reasoning.
There's an obvious danger in taking this too far. Every bagholder thinks he has “conviction.” Refusing to update is not a virtue.
The distinction I'm trying to make now is between information about the **thesis** and information about the **stock price**.
A 50% decline should make me investigate. It doesn't automatically tell me that the thesis is broken.
Likewise, a fivefold gain should make me redo the valuation. It doesn't automatically tell me that the stock should be trimmed.
I suspect this is particularly important in technological discontinuities because the addressable market can change faster than the share price. A company can triple while new information increases its plausible long-term opportunity by much more than three times.
Nvidia is an extreme example, but that's partly the point. If investment returns are power-law distributed, the extreme examples matter disproportionately.
I would rather occasionally hold a winner too long and give back some gains than systematically sell the small number of companies where my original thesis keeps becoming larger.
Disclosure: I own Nvidia again. I wrote a longer version of this argument on my Substack, K-Shaped: [https://kshaped.substack.com/p/i-sold-nvidia-too-early](https://kshaped.substack.com/p/i-sold-nvidia-too-early)
