Conviction · NVIDIA
The hundred and two who bought NVIDIA at the bottom and never left
In the six months to September 2022, four hundred and twenty-five institutions stopped reporting NVIDIA. A hundred and sixty-seven started. A central bank and an insurance company are still holding.
HolderBook · from Form 13F filings, Q1 2022 to Q2 2026
NVIDIA closed the third quarter of 2022 at an implied $121.39 a share, on the agreement of 3,674 institutional filers. Adjusted for the ten-for-one split of June 2024, that is $12.14 in the share count the company uses today. It closed the second quarter of 2026 at $200.09.
What follows is not a story about that. It is a story about the six months before it, when the register was shrinking.
The quarter nobody wanted it
Institutional ownership of a large company does not usually move much. The same index funds hold the same securities, quarter after quarter, and the count drifts up as new managers cross the reporting threshold.
NVIDIA in 2022 was not that.
| Quarter | Holders | Reported value | Implied price |
|---|---|---|---|
| Q1 2022 | 3,118 | $520bn | $272.86 |
| Q2 2022 | 2,834 | $278bn | $151.59 |
| Q3 2022 | 2,693 | $226bn | $121.39 |
| Q4 2022 | 2,914 | $269bn | $146.14 |
Prices are pre-split. Implied from the filings themselves: every 13F states a value and a share count per position, and the median across thousands of filers is a solid quarter-end price.
Four hundred and twenty-five managers left the register in two quarters, a seventh of everyone who held it. Five hundred and ninety-six sold out completely at some point in that window, some having arrived only recently. The reported value more than halved.
The hundred and sixty-seven
In the same quarter that 425 managers stopped reporting NVIDIA, 167 reported it for the first time.
Seventeen of them bought options rather than shares. That distinction matters more than it sounds. A 13F reports a call option as a position in the underlying security, so a manager who bought contracts appears in the same table as one who bought stock. An option expires. Buying calls into a drawdown is a trade with a deadline, which is a different act from buying the company.
Who they were
The names are not the ones a story like this usually produces.
| Manager | Bought, Q3 2022 | Value, Q2 2026 | Shares held now |
|---|---|---|---|
| T. Rowe Price Investment Management | $278m | $3,375m | 16,865,766 |
| Sands Capital Management | $194m | $2,813m | 14,059,454 |
| INTECH Investment Management | $76m | $727m | 3,633,381 |
| New York Life Investment Management | $48m | $728m | 3,640,286 |
| Czech National Bank | $35m | $1,379m | 6,894,089 |
| Triodos Investment Management | $32m | $77m | 384,570 |
| Zurich Insurance Group | $25m | $965m | 4,824,127 |
Largest new share positions opened in Q3 2022, and what they were worth at the most recent quarter end. Managers who bought options rather than shares are excluded.
No hedge fund made a famous call here. The list is an insurance company, a pension manager, a quantitative equity shop, a Dutch bank that runs sustainability mandates, and a central bank.
The Czech National Bank opened a $35m position in the quarter NVIDIA bottomed and reported $1,379m of it in June 2026, holding 6.9 million shares. Central banks hold equities as reserve assets and rebalance mechanically rather than opportunistically, which is worth saying plainly: this is very probably an index allocation that happened to be struck at the right moment rather than a judgement about accelerated computing.
That is the honest reading of most of this table. Zurich Insurance, New York Life and INTECH run large systematic mandates. A systematic manager buying into weakness is doing what its rules say, and the rules did not foresee 2023.
Two of the names are different. T. Rowe Price Investment Management and Sands Capital both run discretionary equity strategies, both bought shares rather than contracts, and both still hold eight-figure share counts. Whatever was decided in those two buildings in the third quarter of 2022 was decided by someone.
What this does not show
A 13F is a quarter-end photograph. A manager who bought in July, sold in August and bought again in September reports the same thing as one who bought in July and never moved. Nothing here proves any of these managers held continuously, only that they reported a position at both ends of a fifteen quarter gap.
Nor does the form say what anything cost. The value on a 13F is the position's worth at quarter end, not what was paid for it, so the figures above are not returns. A manager who bought at $121.39 and one who bought at $146.14 report identically.
And the split is a trap for anyone reading these numbers quickly. NVIDIA split ten for one in June 2024, which is why the implied price falls from $903.56 to $123.54 between March and June of that year without anything having gone wrong. Every share count before June 2024 is a tenth of its equivalent today. Any comparison that ignores this overstates the move by a factor of ten.
The other side
Five hundred and ninety-six managers sold NVIDIA out entirely during those six months. Some were rebalancing, some were redeeming, some had mandates that forced them out of a position that had fallen far enough. The form does not say which, and that is the honest limit of what a 13F can tell anybody.
What it does tell us is that in the quarter when the largest number of institutions in four years gave up on this company, a hundred and fifty others bought shares in it. A hundred and two of them are still there.
Holder counts, reported values and share counts are from Form 13F filings as reported, corrected for amendments. Implied prices are the median value-per-share across all filers reporting the security in that quarter, not licensed market data. Positions reported as derivatives are identified separately and excluded where the text says so. Not investment advice.