The stock I got right — and the year my own screen would have thrown it away

I wrote a piece a few weeks ago about the mistakes that cost me the most. This is the other side of the ledger, and it's the harder one to write honestly — a winner invites you to explain how clever you were, and it invites the opposite error too, where you shrug the whole thing off as luck because that sounds modest.

Neither is what happened. I started buying NVIDIA in 2022 at about $230 a share, before the ten-for-one split. It is now my largest position by a wide margin and I have never sold a share. I bought it for a specific, checkable reason; that reason turned out to be right; and then something I had not underwritten at all made it right by a much larger margin than I had any business expecting.

Those are three different claims, and they deserve to be kept apart.

What I was actually doing in 2022

The model this site runs on did not exist yet. What I had was the manual process I've written about before — pulling up a ticker and working through metrics one at a time. Free cash flow. Return on equity. Debt to equity. The PEG ratio. Noting them down, weighing them against what I knew about the business, arriving at something like a grade.

NVIDIA had been on my list for a while, mostly because I follow gaming and it was impossible to avoid in that world. But watching a company is not owning one. What moved it from a name I knew to a position I held was the growth record, and that record was not ambiguous.

That is what was on the table in early 2022. Revenue had grown more than five-fold in six years. It had grown 53% and then 61% in the two most recent full years. It also had a genuine down year in FY2020, which mattered to me at the time and matters more in hindsight: I was not buying a flawless story; I was buying a company whose growth had already survived a cycle and reaccelerated out of it.

The balance sheet was in no danger either. Debt to equity was about 0.41 when I started buying.

The trade-off I made on purpose

By most of the value measures I was using, it was expensive. It was not close. That was the whole argument against it, and I want to be clear that I saw it and overrode it rather than missed it.

My reasoning was that I'm young, my horizon is measured in decades, and a high multiple on a company growing north of 50% a year is a different proposition from a high multiple on a company growing at 8%. If the growth continued, the multiple would be earned. If it didn't, I'd be wrong in the ordinary way and I could live with that.

That judgment is not a clever one. It is the oldest argument in growth investing and it is wrong often enough to have ruined plenty of people. But it is a judgment, made in advance, for a stated reason — and it is close to the exact trade-off the screen I later built now makes mechanically. Growth at a reasonable price is a formalization of the thing I was already doing by hand and doing inconsistently.

Buying it down, which I've said elsewhere is a mistake

It fell hard after my first buy, and I kept buying it the whole way down. After the first purchase at $230 I added at $190, $170, $155, $151, $143 and $130 — seven purchases, all pre-split, over roughly a year.

I list them individually because the number of them is the point. This was not one flinch and it was not one act of conviction. It was the same decision taken seven times, each time at a price that had moved further against me, and a decision you are willing to repeat is a decision you should be able to state. Mine was that the growth record was intact and the price I was paying for it kept improving.

In the piece about my mistakes I described averaging down into AT&T after its dividend cut as the sharper of the two errors in that story — adding to a loser feels like conviction and is usually a refusal to re-underwrite the position.

So which is it?

The two aren't in conflict, though it would be convenient for me if they simply weren't comparable. The honest distinction is what the business was doing while the price fell. AT&T had cut its dividend, and the dividend was the thesis. NVIDIA's revenue was still growing more than 35% year over year when I was buying it in the low $100s, and the multiple was falling — about 69 times earnings when I started, about 36 times by that October.

The business was intact and the price was getting better. That's the only version of averaging down I'd defend. The version I got wrong with AT&T was adding to a position whose reason for existing had already been withdrawn. The falling price is not the signal in either case. The question is whether the thing you underwrote is still there, and answering that honestly is much harder when you are already down.

What I did not underwrite

My thesis was esports getting big and every device around us needing more capable chips than the generation before. That thesis was fine and it was not what happened.

What happened was data centers, inference, and the whole GPU software stack becoming the thing the company sold. I was aware of AI as an idea. It was not in my near-term reasoning in any load-bearing way.

I want to be precise about what that does and doesn't mean, because this is where a piece like this usually goes wrong in one direction or the other. The company I identified was the right company, and I identified it on the right evidence. A business compounding revenue north of 50% a year with a clean balance sheet is a good thing to own, and it was a good thing to own for reasons that were visible in the filings before anyone was saying the word inference on an earnings call.

What I did not and could not underwrite was the size of what came next. I own a position that worked far harder than my reasoning entitled me to expect. The selection was mine. The magnitude was not.

Averaging up, and the two numbers that disagreed

Through 2023 the stock ran to roughly $500 pre-split and then spent a few months pulling back into the low $400s. I looked at the numbers, decided the pullback was profit-taking rather than anything real, and bought more. My only regret about that decision is that I didn't buy considerably more.

Here's the part I only understood properly later.

On trailing twelve-month figures, the numbers in October 2023 were much weaker than when I'd been buying in the low $100s — revenue growth of about 10% against 36%, and 104 times earnings against 36 times. On the most recent reported quarter they were extraordinary: revenue had just come in 101% higher than the same quarter a year before.

Both numbers were real. They pointed in opposite directions for the better part of a year, because a trailing twelve-month figure still had three collapsed quarters inside it and the latest quarter did not.

I was reading the quarter, and I want to be plain about that, because it would be easy to dress it up after the fact. I did not weigh the trailing figure against the quarterly one and decide which deserved more weight. I saw the growth and I wanted in.

That turned out to be the right read and I don't think I get much credit for it. This is worth separating from the 2022 purchases, which were not like this — those came out of working through the metrics one at a time and concluding the growth record justified the multiple. This was a top-up on a number that impressed me.

The lesson isn't that instinct beat the arithmetic. It's that two numbers describing the same company were telling opposite stories for the better part of a year, and I did not notice a choice was being made.

What my own model would have said

This is the part I was most curious about, and the site let me check it rather than guess. The rankings are built on trailing twelve-month fundamentals, so I ran the model's inputs against the data it would actually have had — filings as filed, dated when they were filed, with no benefit of knowing what came next.

Look at the top two rows first, because they are the vindication and I'll take it: when I was buying, this was a textbook growth-at-a-reasonable-price profile. A PEG right around 1.0, revenue compounding in the thirties to sixties, debt under control. My hand-rolled process and the model I later built agree about 2022, which is reassuring about both.

Now look at the middle. In June 2023, with the stock up roughly four-fold from where I'd been buying, NVIDIA's trailing revenue was shrinking by 12% and it traded at 218 times trailing earnings. Growth at a reasonable price is the entire premise of my screen. At that moment NVIDIA had neither the growth nor the reasonable price.

So the answer to "would my model have found it" is not yes and it is not no. It would have liked it in 2022 while I was buying, thrown it out through most of 2023, and liked it again from early 2024 — dropping it precisely during the run that made the position what it is.

I should be exact about what that table does and doesn't show. Those are the model's inputs, not a rank. A rank depends on how every other company scored on the same day, and I have not rebuilt the whole 2022 market to find out. What I can say is which way each pillar pointed, and in mid-2023 the value pillar and the growth pillar pointed hard at the exit together.

That is not a defect I intend to fix. A screen built on reported figures is built on the past, and the past is the only part you can audit. The cost of that choice is that it will be late to any story where the future shows up in the guidance before it shows up in the filings. I would rather run a model that is honestly late than one that is speculatively early, and this is the clearest example I own of what that preference costs.

The times I nearly sold

Two of them.

The run into the $800s and then past $1,000 pre-split was the stronger pull, and it was the ordinary one — the position had become large and locking in a gain that size is genuinely tempting. I didn't, because by then a second thesis had replaced my original one and it was unfolding faster than I could talk myself out of.

The split in June 2024 was the second, and it's the one I find most interesting in hindsight, because nothing had happened. Ten times the shares at a tenth the price is the same company and the same ownership. But it doesn't feel that way. Selling some of a position you now count in hundreds of shares feels smaller than selling the same value of one you count in dozens. That's an illusion, and having more shares to sell is not a reason to sell any of them. I looked at growth that was still accelerating and at multiples that had compressed rather than expanded, and it didn't make sense.

What I did instead, on a pullback to around $100 post-split, was buy more.

Where it stands

The model has NVDA at number seven in the top thirty today, scoring 85.5 out of 100. At $217.55 it trades at about 27 times earnings with a PEG near 0.59 — close to the median for its sector, which is a strange sentence to be able to write about this company. Revenue growth is still accelerating rather than fading, and margins with it.

I do reinvest the dividend, and it has only just become worth mentioning. For nearly all the time I have held NVIDIA the dividend was a cent a share per quarter or less — four cents a year, a rounding error on a $200 stock, and no part of anybody's reason for owning it. In the quarter ended July 2026 the company raised it to twenty-five cents, an annualized dollar a share.

That is a yield of about 0.46% at $217.55, so it is still not income and I am not treating it as any. But it is a twenty-five-fold raise, and there is something quietly funny about it from where I sit. The growth company I bought for its growth — after spending years buying dividends I did not need, which is the third of the three mistakes in that other piece — has started paying me a dividend I still do not need. I reinvest it, which at this size is closer to a rounding instruction than a strategy.

What I'd take from it

Not "buy NVIDIA," and not "hold your winners," which sounds wise and tells you nothing about which ones.

Two things, and the first is the smaller of them. A high multiple is a question, not an answer. It asks whether the growth will arrive. That question has a real answer. It is sometimes yes, and refusing to ask it is how I spent years buying cheap companies that deserved to be cheap.

The second is the one I'd actually pass on. Know which number you're reading, and know what it can't see. The trailing figures and the latest quarter disagreed for the better part of a year, and that disagreement was the whole story of this position. A screen reading the past will be honest and occasionally slow. A person reading the present will be quicker and much easier to fool. I use both, and this is the clearest case I have of the two reaching different conclusions.

It is not a claim that I'll be on the right side the next time they disagree. Read the piece about my mistakes for the counter-examples.

The obligatory caveat, which I mean sincerely

I own NVDA, it is my largest position, and I have never sold a share — so treat all of the above as the account of an interested party. I'm not a licensed financial advisor and none of this is a recommendation to buy or sell NVIDIA or anything else. Prices and figures are as of the dates shown; the historical numbers are reconstructed from filings as they stood at the time, and the current ones come from the model's live data. One position that has worked for four years is a very small sample, and nothing here should be read as a claim about what it does next.